[X86] Lowering FMA intrinsics to native IR (LLVM part)

[tkrupa]

Summary

Suuport for Clang lowering of fused intrinsics. This patch:

1. Removes bindings to clang fma intrinsics.
2. Introduces new LLVM unmasked intrinsics with rounding mode:

int_x86_avx512_vfmadd_pd_512
int_x86_avx512_vfmadd_ps_512
int_x86_avx512_vfmaddsub_pd_512
int_x86_avx512_vfmaddsub_ps_512
supported with a new intrinsic type (INTR_TYPE_3OP_RM).

3. Introduces new x86 fmaddsub/fmsubadd folding.
4. Introduces new tests for code emitted by sequentions introduced in Clang part.

Corresponding Clang patch: D47444

[craig.topper]

lib/Target/X86/X86ISelLowering.cpp

30595 ↗

(On Diff #148787)

Can we detect this pattern after the ISD::FMA+FNEG has been turned into X86ISD::FMSUB? It might simplify the code a little.

[RKSimon]

You're going to hate me for this, but we could do with -fast-isel test files covering the builtin test files codegen on the clang side

[tkrupa]

In D47443#1114187, @RKSimon wrote:

You're going to hate me for this, but we could do with -fast-isel test files covering the builtin test files codegen on the clang side

Not really, thankfully these are all auto-generated from clang tests. So do you mean we could do with fast-isel tests instead of these or in addition to these?

[RKSimon]

In D47443#1114199, @tkrupa wrote:

In D47443#1114187, @RKSimon wrote:

You're going to hate me for this, but we could do with -fast-isel test files covering the builtin test files codegen on the clang side

Not really, thankfully these are all auto-generated from clang tests. So do you mean we could do with fast-isel tests instead of these or in addition to these?

We need both please - the fast-isel tests are there to ensure non-optimized builds do not deviate too much from what coders were trying to accomplish with intrinsics - which is quite different in purpose to the regular dag tests which concern optimization combines.

[tkrupa]

Added fast-isel tests - there are some XORs/MOVs which aren't combined but it doesn't look terrible overall.

[craig.topper]

LGTM

llvm/trunk/include/llvm/IR/IntrinsicsX86.td

def int_x86_avx2_mpsadbw : GCCBuiltin<"__builtin_ia32_mpsadbw256">,
               Intrinsic<[llvm_v16i16_ty], [llvm_v32i8_ty, llvm_v32i8_ty,
                          llvm_i8_ty], [IntrNoMem, Commutative]>;
 }
 
 //===----------------------------------------------------------------------===//
 // FMA3 and FMA4
 
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
-  def int_x86_fma_vfmadd_ss : GCCBuiltin<"__builtin_ia32_vfmaddss3">,
+  def int_x86_fma_vfmadd_ss : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v4f32_ty],
                         [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmadd_sd : GCCBuiltin<"__builtin_ia32_vfmaddsd3">,
+  def int_x86_fma_vfmadd_sd : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v2f64_ty],
                         [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty],
                         [IntrNoMem]>;
   def int_x86_fma4_vfmadd_ss : GCCBuiltin<"__builtin_ia32_vfmaddss">,
               Intrinsic<[llvm_v4f32_ty],
                         [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty],
                         [IntrNoMem]>;
   def int_x86_fma4_vfmadd_sd : GCCBuiltin<"__builtin_ia32_vfmaddsd">,
               Intrinsic<[llvm_v2f64_ty],
                         [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmadd_ps : GCCBuiltin<"__builtin_ia32_vfmaddps">,
+  def int_x86_fma_vfmadd_ps : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v4f32_ty],
                         [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmadd_pd : GCCBuiltin<"__builtin_ia32_vfmaddpd">,
+  def int_x86_fma_vfmadd_pd : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v2f64_ty],
                         [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmadd_ps_256 : GCCBuiltin<"__builtin_ia32_vfmaddps256">,
+  def int_x86_fma_vfmadd_ps_256 : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v8f32_ty],
                         [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmadd_pd_256 : GCCBuiltin<"__builtin_ia32_vfmaddpd256">,
+  def int_x86_fma_vfmadd_pd_256 : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v4f64_ty],
                         [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty],
                         [IntrNoMem]>;
 
-  def int_x86_fma_vfmaddsub_ps : GCCBuiltin<"__builtin_ia32_vfmaddsubps">,
+  def int_x86_fma_vfmaddsub_ps : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v4f32_ty],
                         [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmaddsub_pd : GCCBuiltin<"__builtin_ia32_vfmaddsubpd">,
+  def int_x86_fma_vfmaddsub_pd : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v2f64_ty],
                         [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmaddsub_ps_256 :
-               GCCBuiltin<"__builtin_ia32_vfmaddsubps256">,
+  def int_x86_fma_vfmaddsub_ps_256 : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v8f32_ty],
                         [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty],
                         [IntrNoMem]>;
-  def int_x86_fma_vfmaddsub_pd_256 :
-              GCCBuiltin<"__builtin_ia32_vfmaddsubpd256">,
+  def int_x86_fma_vfmaddsub_pd_256 : // FIXME: remove this intrinsic.
               Intrinsic<[llvm_v4f64_ty],
                         [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty],
                         [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmadd_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd128_mask">,
+  def int_x86_avx512_mask_vfmadd_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmadd_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd128_mask3">,
+  def int_x86_avx512_mask3_vfmadd_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmadd_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd128_maskz">,
+  def int_x86_avx512_maskz_vfmadd_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmadd_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd256_mask">,
+  def int_x86_avx512_mask_vfmadd_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmadd_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd256_mask3">,
+  def int_x86_avx512_mask3_vfmadd_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmadd_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd256_maskz">,
+  def int_x86_avx512_maskz_vfmadd_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmadd_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd512_mask">,
+  def int_x86_avx512_vfmadd_pd_512 :
+          Intrinsic<[llvm_v8f64_ty],
+          [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i32_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_vfmadd_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
-          [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
+          [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask3_vfmadd_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd512_mask3">,
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmadd_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddpd512_maskz">,
+  def int_x86_avx512_maskz_vfmadd_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmadd_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps128_mask">,
+  def int_x86_avx512_mask_vfmadd_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmadd_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps128_mask3">,
+  def int_x86_avx512_mask3_vfmadd_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmadd_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps128_maskz">,
+  def int_x86_avx512_maskz_vfmadd_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmadd_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps256_mask">,
+  def int_x86_avx512_mask_vfmadd_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmadd_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps256_mask3">,
+  def int_x86_avx512_mask3_vfmadd_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmadd_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps256_maskz">,
+  def int_x86_avx512_maskz_vfmadd_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmadd_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps512_mask">,
+  def int_x86_avx512_vfmadd_ps_512 :
+          Intrinsic<[llvm_v16f32_ty],
+          [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i32_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_vfmadd_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmadd_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps512_mask3">,
+  def int_x86_avx512_mask3_vfmadd_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmadd_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddps512_maskz">,
+  def int_x86_avx512_maskz_vfmadd_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmaddsub_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd128_mask">,
+  def int_x86_avx512_mask_vfmaddsub_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmaddsub_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd128_mask3">,
+  def int_x86_avx512_mask3_vfmaddsub_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmaddsub_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd128_maskz">,
+  def int_x86_avx512_maskz_vfmaddsub_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmaddsub_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd256_mask">,
+  def int_x86_avx512_mask_vfmaddsub_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmaddsub_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd256_mask3">,
+  def int_x86_avx512_mask3_vfmaddsub_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmaddsub_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd256_maskz">,
+  def int_x86_avx512_maskz_vfmaddsub_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmaddsub_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd512_mask">,
+  def int_x86_avx512_vfmaddsub_pd_512 :
+          Intrinsic<[llvm_v8f64_ty],
+          [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty, llvm_i32_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_vfmaddsub_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmaddsub_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd512_mask3">,
+  def int_x86_avx512_mask3_vfmaddsub_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmaddsub_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubpd512_maskz">,
+  def int_x86_avx512_maskz_vfmaddsub_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmaddsub_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps128_mask">,
+  def int_x86_avx512_mask_vfmaddsub_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmaddsub_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps128_mask3">,
+  def int_x86_avx512_mask3_vfmaddsub_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmaddsub_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps128_maskz">,
+  def int_x86_avx512_maskz_vfmaddsub_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmaddsub_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps256_mask">,
+  def int_x86_avx512_mask_vfmaddsub_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmaddsub_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps256_mask3">,
+  def int_x86_avx512_mask3_vfmaddsub_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmaddsub_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps256_maskz">,
+  def int_x86_avx512_maskz_vfmaddsub_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfmaddsub_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps512_mask">,
+  def int_x86_avx512_vfmaddsub_ps_512 :
+          Intrinsic<[llvm_v16f32_ty],
+          [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty, llvm_i32_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_vfmaddsub_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmaddsub_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps512_mask3">,
+  def int_x86_avx512_mask3_vfmaddsub_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_maskz_vfmaddsub_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmaddsubps512_maskz">,
+  def int_x86_avx512_maskz_vfmaddsub_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
 
   def int_x86_avx512_mask_vfmadd_sd :
          GCCBuiltin<"__builtin_ia32_vfmaddsd3_mask">,
           Intrinsic<[llvm_v2f64_ty],
def int_x86_avx512_mask3_vfmsub_sd :
           llvm_i32_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask3_vfmsub_ss :
          GCCBuiltin<"__builtin_ia32_vfmsubss3_mask3">,
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsub_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmsubpd128_mask3">,
+  def int_x86_avx512_mask3_vfmsub_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsub_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmsubpd256_mask3">,
+  def int_x86_avx512_mask3_vfmsub_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsub_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmsubpd512_mask3">,
+  def int_x86_avx512_mask3_vfmsub_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsub_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmsubps128_mask3">,
+  def int_x86_avx512_mask3_vfmsub_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsub_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmsubps256_mask3">,
+  def int_x86_avx512_mask3_vfmsub_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsub_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmsubps512_mask3">,
+  def int_x86_avx512_mask3_vfmsub_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsubadd_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfmsubaddpd128_mask3">,
+  def int_x86_avx512_mask3_vfmsubadd_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsubadd_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfmsubaddpd256_mask3">,
+  def int_x86_avx512_mask3_vfmsubadd_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsubadd_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfmsubaddpd512_mask3">,
+  def int_x86_avx512_mask3_vfmsubadd_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsubadd_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfmsubaddps128_mask3">,
+  def int_x86_avx512_mask3_vfmsubadd_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsubadd_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfmsubaddps256_mask3">,
+  def int_x86_avx512_mask3_vfmsubadd_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfmsubadd_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfmsubaddps512_mask3">,
+  def int_x86_avx512_mask3_vfmsubadd_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmadd_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfnmaddpd128_mask">,
+  def int_x86_avx512_mask_vfnmadd_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmadd_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfnmaddpd256_mask">,
+  def int_x86_avx512_mask_vfnmadd_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmadd_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfnmaddpd512_mask">,
+  def int_x86_avx512_mask_vfnmadd_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmadd_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfnmaddps128_mask">,
+  def int_x86_avx512_mask_vfnmadd_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmadd_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfnmaddps256_mask">,
+  def int_x86_avx512_mask_vfnmadd_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmadd_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfnmaddps512_mask">,
+  def int_x86_avx512_mask_vfnmadd_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask3_vfnmsub_sd :
          GCCBuiltin<"__builtin_ia32_vfnmsubsd3_mask3">,
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask3_vfnmsub_ss :
          GCCBuiltin<"__builtin_ia32_vfnmsubss3_mask3">,
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmsub_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubpd128_mask">,
+  def int_x86_avx512_mask_vfnmsub_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfnmsub_pd_128 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubpd128_mask3">,
+  def int_x86_avx512_mask3_vfnmsub_pd_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v2f64_ty],
           [llvm_v2f64_ty, llvm_v2f64_ty, llvm_v2f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmsub_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubpd256_mask">,
+  def int_x86_avx512_mask_vfnmsub_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfnmsub_pd_256 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubpd256_mask3">,
+  def int_x86_avx512_mask3_vfnmsub_pd_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f64_ty],
           [llvm_v4f64_ty, llvm_v4f64_ty, llvm_v4f64_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmsub_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubpd512_mask">,
+  def int_x86_avx512_mask_vfnmsub_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfnmsub_pd_512 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubpd512_mask3">,
+  def int_x86_avx512_mask3_vfnmsub_pd_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f64_ty],
           [llvm_v8f64_ty, llvm_v8f64_ty, llvm_v8f64_ty,  llvm_i8_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmsub_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubps128_mask">,
+  def int_x86_avx512_mask_vfnmsub_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfnmsub_ps_128 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubps128_mask3">,
+  def int_x86_avx512_mask3_vfnmsub_ps_128 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v4f32_ty],
           [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmsub_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubps256_mask">,
+  def int_x86_avx512_mask_vfnmsub_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfnmsub_ps_256 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubps256_mask3">,
+  def int_x86_avx512_mask3_vfnmsub_ps_256 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v8f32_ty],
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_v8f32_ty,  llvm_i8_ty],
           [IntrNoMem]>;
 
-  def int_x86_avx512_mask_vfnmsub_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubps512_mask">,
+  def int_x86_avx512_mask_vfnmsub_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
-  def int_x86_avx512_mask3_vfnmsub_ps_512 :
-         GCCBuiltin<"__builtin_ia32_vfnmsubps512_mask3">,
+  def int_x86_avx512_mask3_vfnmsub_ps_512 : // FIXME: remove this intrinsic.
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_v16f32_ty,  llvm_i16_ty,
           llvm_i32_ty], [IntrNoMem]>;
 
   def int_x86_avx512_vpmadd52h_uq_128 :
               GCCBuiltin<"__builtin_ia32_vpmadd52huq128">,
               Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty,
                          llvm_v2i64_ty], [IntrNoMem]>;

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

case INTR_TYPE_3OP_MASK: {
                                       Src1, Src2, Src3, Rnd),
                                       Mask, PassThru, Subtarget, DAG);
         }
       }
       return getVectorMaskingNode(DAG.getNode(IntrData->Opc0, dl, VT,
                                               Src1, Src2, Src3),
                                   Mask, PassThru, Subtarget, DAG);
     }
+    case INTR_TYPE_3OP_RM: {
+      SDValue Src1 = Op.getOperand(1);
+      SDValue Src2 = Op.getOperand(2);
+      SDValue Src3 = Op.getOperand(3);
+
+      // We specify 2 possible opcodes for intrinsics with rounding modes.
+      // First, we check if the intrinsic may have non-default rounding mode,
+      // (IntrData->Opc1 != 0), then we check the rounding mode operand.
+      unsigned IntrWithRoundingModeOpcode = IntrData->Opc1;
+      if (IntrWithRoundingModeOpcode != 0) {
+        SDValue Rnd = Op.getOperand(4);
+        if (!isRoundModeCurDirection(Rnd)) {
+          return DAG.getNode(IntrWithRoundingModeOpcode,
+                             dl, Op.getValueType(),
+                             Src1, Src2, Src3, Rnd);
+        }
+      }
+      return DAG.getNode(IntrData->Opc0, dl, VT, Src1, Src2, Src3);
+    }
     case VPERM_2OP : {
       SDValue Src1 = Op.getOperand(1);
       SDValue Src2 = Op.getOperand(2);
 
       // Swap Src1 and Src2 in the node creation
       return DAG.getNode(IntrData->Opc0, dl, VT,Src2, Src1);
     }
     case FMA_OP_MASK3:
if (SDValue NewN = combineRedundantDWordShuffle(N, Mask, DAG))
       return NewN;
 
     break;
   }
 
   return SDValue();
 }
 
+/// Checks if the shuffle mask takes subsequent elements
+/// alternately from two vectors.
+/// For example <0, 5, 2, 7> or <8, 1, 10, 3, 12, 5, 14, 7> are both correct.
+static bool isAddSubOrSubAddMask(ArrayRef<int> Mask, int ParitySrc[2]) {
+
+  unsigned Size = Mask.size();
+  for (unsigned i = 0; i != Size; ++i) {
+    int M = Mask[i];
+    if (M < 0)
+      continue;
+
+    // Make sure we are using the matching element from the input.
+    if ((M % Size) != i)
+      return false;
+
+    // Make sure we use the same input for all elements of the same parity.
+    int Src = M / Size;
+    if (ParitySrc[i % 2] >= 0 && ParitySrc[i % 2] != Src)
+      return false;
+    ParitySrc[i % 2] = Src;
+  }
+
+  // Make sure each input is used.
+  if (ParitySrc[0] < 0 || ParitySrc[1] < 0 || ParitySrc[0] == ParitySrc[1])
+    return false;
+
+  return true;
+}
+
 /// Returns true iff the shuffle node \p N can be replaced with ADDSUB(SUBADD)
 /// operation. If true is returned then the operands of ADDSUB(SUBADD) operation
 /// are written to the parameters \p Opnd0 and \p Opnd1.
 ///
 /// We combine shuffle to ADDSUB(SUBADD) directly on the abstract vector shuffle nodes
 /// so it is easier to generically match. We also insert dummy vector shuffle
 /// nodes for the operands which explicitly discard the lanes which are unused
 /// by this operation to try to flow through the rest of the combiner
if (V1.getOpcode() == ISD::FSUB) {
     assert(V2.getOpcode() == ISD::FSUB && "Unexpected opcode");
     LHS = V2->getOperand(0); RHS = V2->getOperand(1);
     if ((V1->getOperand(0) != LHS || V1->getOperand(1) != RHS) &&
         (V1->getOperand(0) != RHS || V1->getOperand(1) != LHS))
       return false;
   }
 
   ArrayRef<int> Mask = cast<ShuffleVectorSDNode>(N)->getMask();
-
   int ParitySrc[2] = {-1, -1};
-  unsigned Size = Mask.size();
-  for (unsigned i = 0; i != Size; ++i) {
-    int M = Mask[i];
-    if (M < 0)
-      continue;
-
-    // Make sure we are using the matching element from the input.
-    if ((M % Size) != i)
-      return false;
-
-    // Make sure we use the same input for all elements of the same parity.
-    int Src = M / Size;
-    if (ParitySrc[i % 2] >= 0 && ParitySrc[i % 2] != Src)
-      return false;
-    ParitySrc[i % 2] = Src;
-  }
-
-  // Make sure each input is used.
-  if (ParitySrc[0] < 0 || ParitySrc[1] < 0 || ParitySrc[0] == ParitySrc[1])
+  if (!isAddSubOrSubAddMask(Mask, ParitySrc))
     return false;
 
   // It's a subadd if the vector in the even parity is an FADD.
   IsSubAdd = ParitySrc[0] == 0 ? V1->getOpcode() == ISD::FADD
                                : V2->getOpcode() == ISD::FADD;
 
   Opnd0 = LHS;
   Opnd1 = RHS;
   return true;
 }
 
+/// Combine shuffle of two fma nodes into FMAddSub or FMSubAdd.
+static SDValue combineShuffleToFMAddSub(SDNode *N,
+                                        const X86Subtarget &Subtarget,
+                                        SelectionDAG &DAG) {
+  // We only handle target-independent shuffles.
+  // FIXME: It would be easy and harmless to use the target shuffle mask
+  // extraction tool to support more.
+  if (N->getOpcode() != ISD::VECTOR_SHUFFLE)
+    return SDValue();
+
+  MVT VT = N->getSimpleValueType(0);
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  if (!Subtarget.hasAnyFMA() || !TLI.isTypeLegal(VT))
+    return SDValue();
+
+  // We're trying to match (shuffle fma(a, b, c), X86Fmsub(a, b, c).
+  SDValue Op0 = N->getOperand(0);
+  SDValue Op1 = N->getOperand(1);
+  SDValue FMAdd = Op0, FMSub = Op1;
+  if (FMSub.getOpcode() != X86ISD::FMSUB)
+    std::swap(FMAdd, FMSub);
+
+  if (FMAdd.getOpcode() != ISD::FMA || FMSub.getOpcode() != X86ISD::FMSUB ||
+      FMAdd.getOperand(0) != FMSub.getOperand(0) || !FMAdd.hasOneUse() ||
+      FMAdd.getOperand(1) != FMSub.getOperand(1) || !FMSub.hasOneUse() ||
+      FMAdd.getOperand(2) != FMSub.getOperand(2))
+    return SDValue();
+
+  // Check for correct shuffle mask.
+  ArrayRef<int> Mask = cast<ShuffleVectorSDNode>(N)->getMask();
+  int ParitySrc[2] = {-1, -1};
+  if (!isAddSubOrSubAddMask(Mask, ParitySrc))
+    return SDValue();
+
+  // FMAddSub takes zeroth operand from FMSub node.
+  SDLoc DL(N);
+  bool IsSubAdd = ParitySrc[0] == 0 ? Op0 == FMAdd : Op1 == FMAdd;
+  unsigned Opcode = IsSubAdd ? X86ISD::FMSUBADD : X86ISD::FMADDSUB;
+  return DAG.getNode(Opcode, DL, VT, FMAdd.getOperand(0), FMAdd.getOperand(1),
+                     FMAdd.getOperand(2));
+}
+
 /// Try to combine a shuffle into a target-specific add-sub or
 /// mul-add-sub node.
 static SDValue combineShuffleToAddSubOrFMAddSub(SDNode *N,
                                                 const X86Subtarget &Subtarget,
                                                 SelectionDAG &DAG) {
+  if (SDValue V = combineShuffleToFMAddSub(N, Subtarget, DAG))
+    return V;
+
   SDValue Opnd0, Opnd1;
   bool IsSubAdd;
   if (!isAddSubOrSubAdd(N, Subtarget, DAG, Opnd0, Opnd1, IsSubAdd))
     return SDValue();
 
   MVT VT = N->getSimpleValueType(0);
   SDLoc DL(N);
 

llvm/trunk/lib/Target/X86/X86IntrinsicsInfo.h

 #include "X86InstrInfo.h"
 
 namespace llvm {
 
 enum IntrinsicType : uint16_t {
   INTR_NO_TYPE,
   GATHER, SCATTER, PREFETCH, RDSEED, RDRAND, RDPMC, RDTSC, XTEST, XGETBV, ADX, FPCLASS, FPCLASSS,
   INTR_TYPE_1OP, INTR_TYPE_2OP, INTR_TYPE_3OP, INTR_TYPE_4OP,
+  INTR_TYPE_3OP_RM,
   CMP_MASK, CMP_MASK_CC,CMP_MASK_SCALAR_CC, VSHIFT, COMI, COMI_RM,
   CVTPD2PS, CVTPD2PS_MASK,
   INTR_TYPE_1OP_MASK, INTR_TYPE_1OP_MASK_RM,
   INTR_TYPE_2OP_MASK, INTR_TYPE_2OP_MASK_RM, INTR_TYPE_2OP_IMM8_MASK,
   INTR_TYPE_3OP_MASK, INTR_TYPE_3OP_IMM8_MASK,
   FMA_OP_MASK, FMA_OP_MASKZ, FMA_OP_MASK3,
   FMA_OP_SCALAR_MASK, FMA_OP_SCALAR_MASKZ, FMA_OP_SCALAR_MASK3,
   IFMA_OP, VPERM_2OP, INTR_TYPE_SCALAR_MASK,
static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_vcvtsd2si32, INTR_TYPE_2OP, X86ISD::CVTS2SI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtsd2si64, INTR_TYPE_2OP, X86ISD::CVTS2SI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtsd2usi32, INTR_TYPE_2OP, X86ISD::CVTS2UI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtsd2usi64, INTR_TYPE_2OP, X86ISD::CVTS2UI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtss2si32, INTR_TYPE_2OP, X86ISD::CVTS2SI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtss2si64, INTR_TYPE_2OP, X86ISD::CVTS2SI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtss2usi32, INTR_TYPE_2OP, X86ISD::CVTS2UI_RND, 0),
   X86_INTRINSIC_DATA(avx512_vcvtss2usi64, INTR_TYPE_2OP, X86ISD::CVTS2UI_RND, 0),
+  X86_INTRINSIC_DATA(avx512_vfmadd_pd_512, INTR_TYPE_3OP_RM, ISD::FMA, X86ISD::FMADD_RND),
+  X86_INTRINSIC_DATA(avx512_vfmadd_ps_512, INTR_TYPE_3OP_RM, ISD::FMA, X86ISD::FMADD_RND),
+  X86_INTRINSIC_DATA(avx512_vfmaddsub_pd_512, INTR_TYPE_3OP_RM, X86ISD::FMADDSUB,
+                     X86ISD::FMADDSUB_RND),
+  X86_INTRINSIC_DATA(avx512_vfmaddsub_ps_512, INTR_TYPE_3OP_RM, X86ISD::FMADDSUB,
+                     X86ISD::FMADDSUB_RND),
 
   X86_INTRINSIC_DATA(avx512_vpdpbusd_128,  INTR_TYPE_3OP, X86ISD::VPDPBUSD, 0),
   X86_INTRINSIC_DATA(avx512_vpdpbusd_256,  INTR_TYPE_3OP, X86ISD::VPDPBUSD, 0),
   X86_INTRINSIC_DATA(avx512_vpdpbusd_512,  INTR_TYPE_3OP, X86ISD::VPDPBUSD, 0),
   X86_INTRINSIC_DATA(avx512_vpdpbusds_128, INTR_TYPE_3OP, X86ISD::VPDPBUSDS, 0),
   X86_INTRINSIC_DATA(avx512_vpdpbusds_256, INTR_TYPE_3OP, X86ISD::VPDPBUSDS, 0),
   X86_INTRINSIC_DATA(avx512_vpdpbusds_512, INTR_TYPE_3OP, X86ISD::VPDPBUSDS, 0),
   X86_INTRINSIC_DATA(avx512_vpdpwssd_128,  INTR_TYPE_3OP, X86ISD::VPDPWSSD, 0),

llvm/trunk/test/CodeGen/X86/avx512-fma-intrinsics.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f | FileCheck %s
 
-declare <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-declare <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
+declare <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i32)
+declare <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i32)
 
 define <16 x float> @test_x86_vfnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_x86_vfnmadd_ps_z:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfnmadd213ps %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
   ret <16 x float> %res
; CHECK-NEXT:    retq
   ret <8 x double> %res
 }
 
 define <16 x float> @test_x86_vfmaddsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_x86_vfmaddsubps_z:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmaddsub213ps %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) nounwind
   ret <16 x float> %res
 }
 
 define <16 x float> @test_mask_fmaddsub_ps(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
 ; CHECK-LABEL: test_mask_fmaddsub_ps:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmaddsub132ps %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 4)
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i32 4)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a
+  ret <16 x float> %sel
 }
 
-declare <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
+declare <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) nounwind readnone
 
 define <8 x double> @test_x86_vfmaddsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
 ; CHECK-LABEL: test_x86_vfmaddsubpd_z:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmaddsub213pd %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
   ret <8 x double> %res
 }
-declare <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
+declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) nounwind readnone
 
 define <8 x double> @test_mask_vfmaddsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; CHECK-LABEL: test_mask_vfmaddsub_pd:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
+  ret <8 x double> %sel
 }
 
 define <8 x double>@test_int_x86_avx512_mask_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_pd_512:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vmovapd %zmm0, %zmm3
 ; CHECK-NEXT:    vfmaddsub132pd %zmm1, %zmm2, %zmm3 {%k1}
 ; CHECK-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vaddpd %zmm1, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  %res1 = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 -1, i32 0)
-  %res2 = fadd <8 x double> %res, %res1
+  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i32 4)
+  %bc = bitcast i8 %x3 to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %x0
+  %res1 = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i32 0)
+  %res2 = fadd <8 x double> %sel, %res1
   ret <8 x double> %res2
 }
 
 declare <8 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
 
 define <8 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_512:
 ; CHECK:       ## %bb.0:

 ; CHECK-LABEL: test_int_x86_avx512_mask_vfmaddsub_ps_512:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vmovaps %zmm0, %zmm3
 ; CHECK-NEXT:    vfmaddsub132ps %zmm1, %zmm2, %zmm3 {%k1}
 ; CHECK-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vaddps %zmm1, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  %res1 = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 -1, i32 0)
-  %res2 = fadd <16 x float> %res, %res1
+  %res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i32 4)
+  %bc = bitcast i16 %x3 to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %x0
+  %res1 = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i32 0)
+  %res2 = fadd <16 x float> %sel, %res1
   ret <16 x float> %res2
 }
 
 declare <16 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
 
 define <16 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_512:
 ; CHECK:       ## %bb.0:

 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rne:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 0) nounwind
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 0) nounwind
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
+  ret <16 x float> %sel
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtn(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtn:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132ps {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 1) nounwind
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 1) nounwind
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
+  ret <16 x float> %sel
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtp(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtp:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132ps {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 2) nounwind
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 2) nounwind
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
+  ret <16 x float> %sel
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtz(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_rtz:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132ps {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 3) nounwind
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 3) nounwind
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
+  ret <16 x float> %sel
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrb_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrb_current:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132ps %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) nounwind
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
+  ret <16 x float> %sel
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rne:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 0) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 0) nounwind
   ret <16 x float> %res
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtn(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtn:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213ps {rd-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 1) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 1) nounwind
   ret <16 x float> %res
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtp(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtp:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 2) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 2) nounwind
   ret <16 x float> %res
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtz(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtz:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213ps {rz-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 3) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 3) nounwind
   ret <16 x float> %res
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_current:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) nounwind
   ret <16 x float> %res
 }
 
 declare <8 x double> @llvm.x86.avx512.mask3.vfmsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
 
 define <8 x double>@test_int_x86_avx512_mask3_vfmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_512:
 ; CHECK:       ## %bb.0:

 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rne:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 0) nounwind
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 0) nounwind
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
+  ret <8 x double> %sel
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtn:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132pd {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 1) nounwind
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 1) nounwind
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
+  ret <8 x double> %sel
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtp:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132pd {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 2) nounwind
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 2) nounwind
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
+  ret <8 x double> %sel
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_rtz:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132pd {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 3) nounwind
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 3) nounwind
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
+  ret <8 x double> %sel
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrb_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrb_current:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1}
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
+  ret <8 x double> %sel
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rne:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 0) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 0) nounwind
   ret <8 x double> %res
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtn:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213pd {rd-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 1) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 1) nounwind
   ret <8 x double> %res
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtp:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213pd {ru-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 2) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 2) nounwind
   ret <8 x double> %res
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtz:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 3) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 3) nounwind
   ret <8 x double> %res
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
 ; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_current:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
   ret <8 x double> %res
 }
 
 define <8 x double>@test_int_x86_avx512_mask_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_pd_512:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vmovapd %zmm0, %zmm3
 ; CHECK-NEXT:    vfmadd132pd %zmm1, %zmm2, %zmm3 {%k1}
 ; CHECK-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vaddpd %zmm1, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  %res1 = call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 -1, i32 0)
-  %res2 = fadd <8 x double> %res, %res1
+  %res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i32 4)
+  %bc = bitcast i8 %x3 to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %x0
+  %res1 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i32 0)
+  %res2 = fadd <8 x double> %sel, %res1
   ret <8 x double> %res2
 }
 
 declare <8 x double> @llvm.x86.avx512.mask3.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
 
 define <8 x double>@test_int_x86_avx512_mask3_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_512:
 ; CHECK:       ## %bb.0:

 ; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_ps_512:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    kmovw %edi, %k1
 ; CHECK-NEXT:    vmovaps %zmm0, %zmm3
 ; CHECK-NEXT:    vfmadd132ps %zmm1, %zmm2, %zmm3 {%k1}
 ; CHECK-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vaddps %zmm1, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  %res1 = call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 -1, i32 0)
-  %res2 = fadd <16 x float> %res, %res1
+  %res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i32 4)
+  %bc = bitcast i16 %x3 to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %x0
+  %res1 = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i32 0)
+  %res2 = fadd <16 x float> %sel, %res1
   ret <16 x float> %res2
 }
 
 declare <16 x float> @llvm.x86.avx512.mask3.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
 
 define <16 x float>@test_int_x86_avx512_mask3_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_512:
 ; CHECK:       ## %bb.0:

llvm/trunk/test/CodeGen/X86/avx512-intrinsics-canonical.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl --show-mc-encoding | FileCheck %s
+
+; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512f-builtins.c
+
+define <8 x double> @test_mm512_fmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xa8,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  ret <8 x double> %0
+}
+
+declare <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) #1
+
+define <8 x double> @test_mm512_mask_fmadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x19,0x98,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x19,0xb8,0xd1]
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x99,0xa8,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xaa,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x19,0x9a,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x99,0xaa,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fnmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xac,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask3_fnmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x19,0xbc,0xd1]
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fnmadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x99,0xac,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xae,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %sub1, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_maskz_fnmsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x99,0xae,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %sub1, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa8,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x98,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231pd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0xb8,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xc9,0xa8,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsub213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xaa,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132pd %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x9a,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213pd %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xc9,0xaa,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xac,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask3_fnmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231pd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0xbc,0xd1]
+; CHECK-NEXT:    ## zmm2 = -(zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fnmadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213pd %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xc9,0xac,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmsub213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xae,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %sub1.i) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_maskz_fnmsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213pd %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xc9,0xae,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %sub1.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_fmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  ret <16 x float> %0
+}
+
+declare <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) #1
+
+define <16 x float> @test_mm512_mask_fmadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x19,0x98,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x19,0xb8,0xd1]
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x99,0xa8,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xaa,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x19,0x9a,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x99,0xaa,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fnmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xac,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask3_fnmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x19,0xbc,0xd1]
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fnmadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x99,0xac,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xae,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %sub1, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_maskz_fnmsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x99,0xae,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %sub1, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa8,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x98,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231ps %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x49,0xb8,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xc9,0xa8,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsub213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xaa,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x9a,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ps %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xc9,0xaa,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xac,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask3_fnmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231ps %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x49,0xbc,0xd1]
+; CHECK-NEXT:    ## zmm2 = -(zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fnmadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ps %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xc9,0xac,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfnmsub213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xae,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %sub1.i) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_maskz_fnmsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ps %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xc9,0xae,0xc2]
+; CHECK-NEXT:    ## zmm0 = -(zmm1 * zmm0) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %sub1.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <8 x double> @test_mm512_fmaddsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmaddsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xa6,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  ret <8 x double> %0
+}
+
+declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) #1
+
+define <8 x double> @test_mm512_mask_fmaddsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmaddsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x19,0x96,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmaddsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmaddsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x19,0xb6,0xd1]
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmaddsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmaddsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x99,0xa6,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmsubadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmsubadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x18,0xa7,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmsubadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsubadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x19,0x97,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmsubadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsubadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x99,0xa7,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmaddsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmaddsub213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa6,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) +/- zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x double> %3
+}
+
+define <8 x double> @test_mm512_mask_fmaddsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x96,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) +/- zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %__A
+  ret <8 x double> %5
+}
+
+define <8 x double> @test_mm512_mask3_fmaddsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231pd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0xb6,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) +/- zmm2
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %__C
+  ret <8 x double> %5
+}
+
+define <8 x double> @test_mm512_maskz_fmaddsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213pd %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xc9,0xa6,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) +/- zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> zeroinitializer
+  ret <8 x double> %5
+}
+
+define <8 x double> @test_mm512_fmsubadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsubadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa7,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) -/+ zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask_fmsubadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132pd %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x97,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) -/+ zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__A
+  ret <8 x double> %4
+}
+
+define <8 x double> @test_mm512_maskz_fmsubadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213pd %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xc9,0xa7,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) -/+ zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
+  ret <8 x double> %4
+}
+
+define <16 x float> @test_mm512_fmaddsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmaddsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xa6,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  ret <16 x float> %0
+}
+
+declare <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) #1
+
+define <16 x float> @test_mm512_mask_fmaddsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmaddsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x19,0x96,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmaddsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmaddsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x19,0xb6,0xd1]
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmaddsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmaddsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x99,0xa6,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmsubadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmsubadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsubadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x18,0xa7,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmsubadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsubadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x19,0x97,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmsubadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsubadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x99,0xa7,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmaddsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmaddsub213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa6,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) +/- zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x float> %3
+}
+
+define <16 x float> @test_mm512_mask_fmaddsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x96,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) +/- zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %4 = bitcast i16 %__U to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %__A
+  ret <16 x float> %5
+}
+
+define <16 x float> @test_mm512_mask3_fmaddsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231ps %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x49,0xb6,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) +/- zmm2
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %4 = bitcast i16 %__U to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %__C
+  ret <16 x float> %5
+}
+
+define <16 x float> @test_mm512_maskz_fmaddsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213ps %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xc9,0xa6,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) +/- zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %4 = bitcast i16 %__U to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> zeroinitializer
+  ret <16 x float> %5
+}
+
+define <16 x float> @test_mm512_fmsubadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vfmsubadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa7,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) -/+ zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask_fmsubadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x97,0xc1]
+; CHECK-NEXT:    ## zmm0 = (zmm0 * zmm1) -/+ zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %3 = bitcast i16 %__U to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__A
+  ret <16 x float> %4
+}
+
+define <16 x float> @test_mm512_maskz_fmsubadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_maskz_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213ps %zmm2, %zmm1, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xc9,0xa7,0xc2]
+; CHECK-NEXT:    ## zmm0 = (zmm1 * zmm0) -/+ zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %3 = bitcast i16 %__U to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
+  ret <16 x float> %4
+}
+
+define <8 x double> @test_mm512_mask3_fmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x19,0xba,0xd1]
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231pd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0xba,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x19,0xba,0xd1]
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ps %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x49,0xba,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmsubadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsubadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x19,0xb7,0xd1]
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmsubadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231pd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0xb7,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) -/+ zmm2
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__C
+  ret <8 x double> %4
+}
+
+define <16 x float> @test_mm512_mask3_fmsubadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsubadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x19,0xb7,0xd1]
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmsubadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231ps %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x49,0xb7,0xd1]
+; CHECK-NEXT:    ## zmm2 = (zmm0 * zmm1) -/+ zmm2
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %3 = bitcast i16 %__U to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__C
+  ret <16 x float> %4
+}
+
+define <8 x double> @test_mm512_mask_fnmadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmadd_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x19,0x9c,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask_fnmadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
+; CHECK-NEXT:    ## zmm0 = -(zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmadd_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x19,0x9c,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
+; CHECK-NEXT:    ## zmm0 = -(zmm0 * zmm1) + zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <8 x double> @test_mm512_mask_fnmsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x19,0x9e,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %sub, <8 x double> %sub1, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fnmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmsub_round_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x19,0xbe,0xd1]
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %sub, <8 x double> %sub1, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask_fnmsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
+; CHECK-NEXT:    ## zmm0 = -(zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub2.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %sub.i, <8 x double> %sub2.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fnmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231pd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0xbe,0xd1]
+; CHECK-NEXT:    ## zmm2 = -(zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    vmovapd %zmm2, %zmm0 ## encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub2.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %sub.i, <8 x double> %sub2.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x19,0x9e,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %sub, <16 x float> %sub1, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fnmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmsub_round_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x19,0xbe,0xd1]
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %sub, <16 x float> %sub1, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_mask_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
+; CHECK-NEXT:    ## zmm0 = -(zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %sub.i, <16 x float> %sub1.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fnmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; CHECK-LABEL: test_mm512_mask3_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231ps %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x49,0xbe,0xd1]
+; CHECK-NEXT:    ## zmm2 = -(zmm0 * zmm1) - zmm2
+; CHECK-NEXT:    vmovaps %zmm2, %zmm0 ## encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %sub.i, <16 x float> %sub1.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <4 x float> @test_mm_mask_fmadd_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fmadd_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__A, i64 0
+  %2 = extractelement <4 x float> %__B, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fmadd_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fmadd_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %__A, <4 x float> %__B, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_maskz_fmadd_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmadd_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %1 = extractelement <4 x float> %__B, i64 0
+  %2 = extractelement <4 x float> %__C, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fmadd_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmadd_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_mask3_fmadd_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmadd_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xb9,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__X, i64 0
+  %2 = extractelement <4 x float> %__Y, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fmadd_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmadd_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xb9,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_mask_fmsub_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fmsub_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__A, i64 0
+  %.rhs.i = extractelement <4 x float> %__B, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fmsub_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fmsub_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %__A, <4 x float> %sub, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_maskz_fmsub_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsub_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %1 = extractelement <4 x float> %__B, i64 0
+  %.rhs.i = extractelement <4 x float> %__C, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fmsub_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsub_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask3_fmsub_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsub_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbb,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__X, i64 0
+  %.rhs.i = extractelement <4 x float> %__Y, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fmsub_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsub_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbb,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_mask_fnmadd_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmadd_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__A, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %__B, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fnmadd_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmadd_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__B, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_maskz_fnmadd_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmadd_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %.rhs.i = extractelement <4 x float> %__B, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %__C, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fnmadd_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmadd_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %sub, <4 x float> %__C, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask3_fnmadd_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmadd_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbd,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__X, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %__Y, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fnmadd_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmadd_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbd,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__X
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask_fnmsub_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmsub_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__A, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <4 x float> %__B, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs7.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext2.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fnmsub_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmsub_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %sub1, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_maskz_fnmsub_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmsub_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %.rhs.i = extractelement <4 x float> %__B, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs5.i = extractelement <4 x float> %__C, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs5.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fnmsub_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmsub_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %sub, <4 x float> %sub1, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask3_fnmsub_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmsub_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbf,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__X, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <4 x float> %__Y, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs7.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext2.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fnmsub_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmsub_round_ss:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; CHECK-NEXT:    ## encoding: [0xc4,0xe2,0x79,0x18,0x1d,A,A,A,A]
+; CHECK-NEXT:    ## fixup A - offset: 5, value: LCPI119_0-4, kind: reloc_riprel_4byte
+; CHECK-NEXT:    vxorps %xmm3, %xmm1, %xmm1 ## encoding: [0xc5,0xf0,0x57,0xcb]
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ss %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbb,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__X
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_mask_fmadd_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fmadd_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__A, i64 0
+  %2 = extractelement <2 x double> %__B, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fmadd_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fmadd_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %__A, <2 x double> %__B, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_maskz_fmadd_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmadd_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %1 = extractelement <2 x double> %__B, i64 0
+  %2 = extractelement <2 x double> %__C, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fmadd_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmadd_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xa9,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_mask3_fmadd_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmadd_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xb9,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__X, i64 0
+  %2 = extractelement <2 x double> %__Y, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fmadd_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmadd_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xb9,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_mask_fmsub_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fmsub_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__A, i64 0
+  %.rhs.i = extractelement <2 x double> %__B, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fmsub_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fmsub_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %__A, <2 x double> %sub, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_maskz_fmsub_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsub_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %1 = extractelement <2 x double> %__B, i64 0
+  %.rhs.i = extractelement <2 x double> %__C, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fmsub_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsub_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xab,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask3_fmsub_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsub_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbb,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__X, i64 0
+  %.rhs.i = extractelement <2 x double> %__Y, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fmsub_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsub_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbb,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_mask_fnmadd_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmadd_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__A, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %__B, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fnmadd_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmadd_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__B, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_maskz_fnmadd_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmadd_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %.rhs.i = extractelement <2 x double> %__B, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %__C, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fnmadd_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmadd_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xad,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %sub, <2 x double> %__C, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask3_fnmadd_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmadd_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbd,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__X, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %__Y, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fnmadd_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmadd_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbd,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__X
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask_fnmsub_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmsub_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__A, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <2 x double> %__B, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs7.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext2.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fnmsub_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; CHECK-LABEL: test_mm_mask_fnmsub_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %sub1, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_maskz_fnmsub_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmsub_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %.rhs.i = extractelement <2 x double> %__B, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs5.i = extractelement <2 x double> %__C, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs5.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fnmsub_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmsub_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xaf,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %sub, <2 x double> %sub1, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask3_fnmsub_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmsub_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbf,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__X, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <2 x double> %__Y, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs7.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext2.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fnmsub_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmsub_round_sd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm1, %xmm1 ## encoding: [0xc5,0xf1,0x57,0x0d,A,A,A,A]
+; CHECK-NEXT:    ## fixup A - offset: 4, value: LCPI143_0-4, kind: reloc_riprel_4byte
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231sd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbb,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__X
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>) #9
+declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>) #9
+declare float @llvm.fma.f32(float, float, float) #9
+declare double @llvm.fma.f64(double, double, double) #9

llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll

 entry:
   %0 = tail call <8 x i64> @llvm.x86.avx512.vpermi2var.q.512(<8 x i64> %__A, <8 x i64> %__I, <8 x i64> %__B)
   %1 = bitcast i8 %__U to <8 x i1>
   %2 = select <8 x i1> %1, <8 x i64> %0, <8 x i64> zeroinitializer
   ret <8 x i64> %2
 }
 
 
+define <8 x double> @test_mm512_fmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  ret <8 x double> %0
+}
+
+declare <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) #1
+
+define <8 x double> @test_mm512_mask_fmadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2
+; X32-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; X64-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fnmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxorq {{\.LCPI.*}}{1to8}, %zmm0, %zmm0
+; X32-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0
+; X64-NEXT:    vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask3_fnmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fnmadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fnmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0]
+; X32-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X32-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X32-NEXT:    vfmadd231pd {rn-sae}, %zmm4, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0]
+; X64-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X64-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X64-NEXT:    vfmadd231pd {rn-sae}, %zmm4, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %sub1, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_maskz_fnmsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %sub1, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2
+; X32-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; X64-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxorq {{\.LCPI.*}}{1to8}, %zmm0, %zmm0
+; X32-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0
+; X64-NEXT:    vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask3_fnmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fnmadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fnmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0]
+; X32-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X32-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X32-NEXT:    vfmadd231pd {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0]
+; X64-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X64-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X64-NEXT:    vfmadd231pd {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %sub1.i) #10
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_maskz_fnmsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %sub1.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_fmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  ret <16 x float> %0
+}
+
+declare <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) #1
+
+define <16 x float> @test_mm512_mask_fmadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2
+; X32-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; X64-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fnmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxord {{\.LCPI.*}}{1to16}, %zmm0, %zmm0
+; X32-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm0
+; X64-NEXT:    vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask3_fnmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fnmadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fnmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0]
+; X32-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X32-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X32-NEXT:    vfmadd231ps {rn-sae}, %zmm4, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0]
+; X64-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X64-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X64-NEXT:    vfmadd231ps {rn-sae}, %zmm4, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %sub1, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_maskz_fnmsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %sub1, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2
+; X32-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; X64-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxord {{\.LCPI.*}}{1to16}, %zmm0, %zmm0
+; X32-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm0
+; X64-NEXT:    vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask3_fnmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmadd231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fnmadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmadd213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fnmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0]
+; X32-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X32-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X32-NEXT:    vfmadd231ps {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0]
+; X64-NEXT:    vpxorq %zmm3, %zmm0, %zmm4
+; X64-NEXT:    vpxorq %zmm3, %zmm2, %zmm0
+; X64-NEXT:    vfmadd231ps {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %sub1.i) #10
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_maskz_fnmsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmsub213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %sub1.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <8 x double> @test_mm512_fmaddsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmaddsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmaddsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  ret <8 x double> %0
+}
+
+declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) #1
+
+define <8 x double> @test_mm512_mask_fmaddsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmaddsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmaddsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmaddsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmaddsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmaddsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmaddsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmaddsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmaddsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmsubadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmsubadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2
+; X32-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsubadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; X64-NEXT:    vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  ret <8 x double> %0
+}
+
+define <8 x double> @test_mm512_mask_fmsubadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmsubadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsubadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_maskz_fmsubadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsubadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsubadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_fmaddsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x double> %3
+}
+
+define <8 x double> @test_mm512_mask_fmaddsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %__A
+  ret <8 x double> %5
+}
+
+define <8 x double> @test_mm512_mask3_fmaddsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %__C
+  ret <8 x double> %5
+}
+
+define <8 x double> @test_mm512_maskz_fmaddsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10
+  %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> zeroinitializer
+  ret <8 x double> %5
+}
+
+define <8 x double> @test_mm512_fmsubadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask_fmsubadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__A
+  ret <8 x double> %4
+}
+
+define <8 x double> @test_mm512_maskz_fmsubadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer
+  ret <8 x double> %4
+}
+
+define <16 x float> @test_mm512_fmaddsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmaddsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmaddsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  ret <16 x float> %0
+}
+
+declare <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) #1
+
+define <16 x float> @test_mm512_mask_fmaddsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmaddsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmaddsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmaddsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmaddsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmaddsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmaddsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmaddsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmaddsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmaddsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmaddsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmsubadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmsubadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2
+; X32-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsubadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; X64-NEXT:    vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  ret <16 x float> %0
+}
+
+define <16 x float> @test_mm512_mask_fmsubadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmsubadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsubadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsubadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_maskz_fmsubadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsubadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsubadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsubadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_fmaddsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x float> %3
+}
+
+define <16 x float> @test_mm512_mask_fmaddsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmaddsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %4 = bitcast i16 %__U to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %__A
+  ret <16 x float> %5
+}
+
+define <16 x float> @test_mm512_mask3_fmaddsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmaddsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %4 = bitcast i16 %__U to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %__C
+  ret <16 x float> %5
+}
+
+define <16 x float> @test_mm512_maskz_fmaddsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10
+  %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %4 = bitcast i16 %__U to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> zeroinitializer
+  ret <16 x float> %5
+}
+
+define <16 x float> @test_mm512_fmsubadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask_fmsubadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsubadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %3 = bitcast i16 %__U to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__A
+  ret <16 x float> %4
+}
+
+define <16 x float> @test_mm512_maskz_fmsubadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_maskz_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_maskz_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %3 = bitcast i16 %__U to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer
+  ret <16 x float> %4
+}
+
+define <8 x double> @test_mm512_mask3_fmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmsubadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsubadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsubadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fmsubadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10
+  %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10
+  %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__C
+  ret <8 x double> %4
+}
+
+define <16 x float> @test_mm512_mask3_fmsubadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsubadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsubadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsubadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fmsubadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfmsubadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10
+  %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10
+  %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %3 = bitcast i16 %__U to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__C
+  ret <16 x float> %4
+}
+
+define <8 x double> @test_mm512_mask_fnmadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fnmadd_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmadd_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask_fnmadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fnmadd_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmadd_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmadd132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <8 x double> @test_mm512_mask_fnmsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fnmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %sub, <8 x double> %sub1, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fnmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmsub_round_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmsub_round_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %sub, <8 x double> %sub1, i32 8)
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask_fnmsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X32-LABEL: test_mm512_mask_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub2.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %sub.i, <8 x double> %sub2.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A
+  ret <8 x double> %2
+}
+
+define <8 x double> @test_mm512_mask3_fnmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2
+; X32-NEXT:    vmovapd %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2
+; X64-NEXT:    vmovapd %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub2.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %sub.i, <8 x double> %sub2.i) #10
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C
+  ret <8 x double> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fnmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %sub, <16 x float> %sub1, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fnmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmsub_round_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmsub_round_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %sub, <16 x float> %sub1, i32 8)
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask_fnmsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X32-LABEL: test_mm512_mask_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmsub132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %sub.i, <16 x float> %sub1.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A
+  ret <16 x float> %2
+}
+
+define <16 x float> @test_mm512_mask3_fnmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X32-LABEL: test_mm512_mask3_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    kmovw {{[0-9]+}}(%esp), %k1
+; X32-NEXT:    vfnmsub231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2
+; X32-NEXT:    vmovaps %zmm2, %zmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm512_mask3_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2
+; X64-NEXT:    vmovaps %zmm2, %zmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %sub.i, <16 x float> %sub1.i) #10
+  %1 = bitcast i16 %__U to <16 x i1>
+  %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C
+  ret <16 x float> %2
+}
+
+define <4 x float> @test_mm_mask_fmadd_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fmadd_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmadd_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__A, i64 0
+  %2 = extractelement <4 x float> %__B, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fmadd_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fmadd_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmadd_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %__A, <4 x float> %__B, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_maskz_fmadd_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmadd_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmadd_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %1 = extractelement <4 x float> %__B, i64 0
+  %2 = extractelement <4 x float> %__C, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fmadd_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmadd_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmadd_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_mask3_fmadd_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmadd_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmadd_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__X, i64 0
+  %2 = extractelement <4 x float> %__Y, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fmadd_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmadd_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmadd_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_mask_fmsub_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fmsub_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__A, i64 0
+  %.rhs.i = extractelement <4 x float> %__B, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fmsub_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fmsub_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %__A, <4 x float> %sub, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_maskz_fmsub_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmsub_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %1 = extractelement <4 x float> %__B, i64 0
+  %.rhs.i = extractelement <4 x float> %__C, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fmsub_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmsub_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask3_fmsub_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsub_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsub_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %1 = extractelement <4 x float> %__X, i64 0
+  %.rhs.i = extractelement <4 x float> %__Y, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fmsub_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsub_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsub_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1
+
+define <4 x float> @test_mm_mask_fnmadd_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fnmadd_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__A, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %__B, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fnmadd_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fnmadd_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__B, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_maskz_fnmadd_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fnmadd_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %.rhs.i = extractelement <4 x float> %__B, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %__C, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fnmadd_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fnmadd_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %sub, <4 x float> %__C, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask3_fnmadd_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmadd_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__X, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %__Y, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext1.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fnmadd_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmadd_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__X
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask_fnmsub_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fnmsub_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__A, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <4 x float> %__B, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs7.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <4 x float> %__W, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext2.i, float %3
+  %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask_fnmsub_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) {
+; X32-LABEL: test_mm_mask_fnmsub_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X32-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %sub1, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_maskz_fnmsub_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fnmsub_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__A, i64 0
+  %.rhs.i = extractelement <4 x float> %__B, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs5.i = extractelement <4 x float> %__C, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs5.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3
+  %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_maskz_fnmsub_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fnmsub_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X32-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X64-NEXT:    vxorps %xmm3, %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %sub, <4 x float> %sub1, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <4 x float> @test_mm_mask3_fnmsub_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmsub_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %__W, i64 0
+  %.rhs.i = extractelement <4 x float> %__X, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <4 x float> %__Y, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs7.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <4 x float> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, float %vecext2.i, float %3
+  %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0
+  ret <4 x float> %vecins.i
+}
+
+define <4 x float> @test_mm_mask3_fnmsub_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmsub_round_ss:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X32-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_round_ss:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0]
+; X64-NEXT:    vxorps %xmm3, %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__X
+  %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__Y, i8 %__U, i32 4)
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_mask_fmadd_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fmadd_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmadd_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__A, i64 0
+  %2 = extractelement <2 x double> %__B, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fmadd_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fmadd_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmadd_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %__A, <2 x double> %__B, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_maskz_fmadd_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmadd_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmadd_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %1 = extractelement <2 x double> %__B, i64 0
+  %2 = extractelement <2 x double> %__C, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fmadd_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmadd_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmadd_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_mask3_fmadd_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmadd_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmadd_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__X, i64 0
+  %2 = extractelement <2 x double> %__Y, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fmadd_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmadd_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmadd_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_mask_fmsub_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fmsub_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__A, i64 0
+  %.rhs.i = extractelement <2 x double> %__B, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fmsub_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fmsub_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vxorpd {{\.LCPI.*}}, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vxorpd {{.*}}(%rip), %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %__A, <2 x double> %sub, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_maskz_fmsub_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmsub_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %1 = extractelement <2 x double> %__B, i64 0
+  %.rhs.i = extractelement <2 x double> %__C, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fmsub_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmsub_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vxorpd {{\.LCPI.*}}, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vxorpd {{.*}}(%rip), %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask3_fmsub_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsub_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsub_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %1 = extractelement <2 x double> %__X, i64 0
+  %.rhs.i = extractelement <2 x double> %__Y, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fmsub_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsub_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsub_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1
+
+define <2 x double> @test_mm_mask_fnmadd_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fnmadd_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__A, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %__B, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fnmadd_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fnmadd_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vxorpd {{\.LCPI.*}}, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vxorpd {{.*}}(%rip), %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__B, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_maskz_fnmadd_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fnmadd_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %.rhs.i = extractelement <2 x double> %__B, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %__C, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fnmadd_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fnmadd_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vxorpd {{\.LCPI.*}}, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vxorpd {{.*}}(%rip), %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %sub, <2 x double> %__C, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask3_fnmadd_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmadd_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__X, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %__Y, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext1.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext1.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fnmadd_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmadd_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vxorpd {{\.LCPI.*}}, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vxorpd {{.*}}(%rip), %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__X
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask_fnmsub_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fnmsub_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__A, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <2 x double> %__B, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs7.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <2 x double> %__W, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext2.i, double %3
+  %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask_fnmsub_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) {
+; X32-LABEL: test_mm_mask_fnmsub_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
+; X32-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
+; X32-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
+; X64-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
+; X64-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %sub1, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_maskz_fnmsub_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fnmsub_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__A, i64 0
+  %.rhs.i = extractelement <2 x double> %__B, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs5.i = extractelement <2 x double> %__C, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs5.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3
+  %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_maskz_fnmsub_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fnmsub_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
+; X32-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
+; X32-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
+; X64-NEXT:    vxorpd %xmm3, %xmm1, %xmm1
+; X64-NEXT:    vxorpd %xmm3, %xmm2, %xmm2
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %sub, <2 x double> %sub1, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+define <2 x double> @test_mm_mask3_fnmsub_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmsub_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %__W, i64 0
+  %.rhs.i = extractelement <2 x double> %__X, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs7.i = extractelement <2 x double> %__Y, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs7.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10
+  %4 = and i8 %__U, 1
+  %tobool.i = icmp eq i8 %4, 0
+  %vecext2.i = extractelement <2 x double> %__Y, i32 0
+  %cond.i = select i1 %tobool.i, double %vecext2.i, double %3
+  %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0
+  ret <2 x double> %vecins.i
+}
+
+define <2 x double> @test_mm_mask3_fnmsub_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmsub_round_sd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    vxorpd {{\.LCPI.*}}, %xmm1, %xmm1
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_round_sd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    vxorpd {{.*}}(%rip), %xmm1, %xmm1
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__X
+  %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__Y, i8 %__U, i32 4)
+  ret <2 x double> %0
+}
+
+declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>) #9
+declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>) #9
+declare float @llvm.fma.f32(float, float, float) #9
+declare double @llvm.fma.f64(double, double, double) #9
+
 !0 = !{i32 1}
 

llvm/trunk/test/CodeGen/X86/avx512vl-intrinsics-canonical.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512vl --show-mc-encoding| FileCheck %s
+
+; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512vl-builtins.c
+
+define <2 x double> @test_mm_mask_fmadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_mask_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132pd %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x98,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask_fmsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_mask_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132pd %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x9a,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask3_fmadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231pd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xb8,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask3_fnmadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231pd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbc,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fmadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xa8,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fmsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213pd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xaa,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fnmadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xac,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fnmsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xae,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %__B, <2 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fmadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132pd %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x98,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fmsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132pd %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x9a,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fmadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231pd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0xb8,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    vmovapd %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fnmadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231pd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0xbc,0xd1]
+; CHECK-NEXT:    ## ymm2 = -(ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    vmovapd %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fmadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0xa8,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fmsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213pd %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0xaa,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fnmadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0xac,0xc2]
+; CHECK-NEXT:    ## ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fnmsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0xae,0xc2]
+; CHECK-NEXT:    ## ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %__B, <4 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask_fmadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_mask_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132ps %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x98,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask_fmsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_mask_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132ps %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x9a,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask3_fmadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231ps %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xb8,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask3_fnmadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231ps %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbc,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fmadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xa8,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fmsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ps %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xaa,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fnmadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xac,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fnmsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xae,0xc2]
+; CHECK-NEXT:    ## xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %__B, <4 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fmadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd132ps %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x29,0x98,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fmsub_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub132ps %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x29,0x9a,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fmadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd231ps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0xb8,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    vmovaps %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fnmadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd231ps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0xbc,0xd1]
+; CHECK-NEXT:    ## ymm2 = -(ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    vmovaps %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fmadd_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0xa8,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fmsub_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub213ps %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0xaa,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fnmadd_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0xac,0xc2]
+; CHECK-NEXT:    ## ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fnmsub_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0xae,0xc2]
+; CHECK-NEXT:    ## ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %__B, <8 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <2 x double> @test_mm_mask_fmaddsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_mask_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132pd %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x96,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) +/- xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %1) #9
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract.i, <2 x double> %3, <2 x double> %__A
+  ret <2 x double> %5
+}
+
+define <2 x double> @test_mm_mask_fmsubadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_mask_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132pd %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x97,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) -/+ xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> %__A
+  ret <2 x double> %4
+}
+
+define <2 x double> @test_mm_mask3_fmaddsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231pd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xb6,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) +/- xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %1) #9
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract.i, <2 x double> %3, <2 x double> %__C
+  ret <2 x double> %5
+}
+
+define <2 x double> @test_mm_maskz_fmaddsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xa6,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %1) #9
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract.i, <2 x double> %3, <2 x double> zeroinitializer
+  ret <2 x double> %5
+}
+
+define <2 x double> @test_mm_maskz_fmsubadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213pd %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0x89,0xa7,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> zeroinitializer
+  ret <2 x double> %4
+}
+
+define <4 x double> @test_mm256_mask_fmaddsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132pd %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x96,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) +/- ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %1) #9
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x double> %3, <4 x double> %__A
+  ret <4 x double> %5
+}
+
+define <4 x double> @test_mm256_mask_fmsubadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132pd %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x97,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) -/+ ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> %__A
+  ret <4 x double> %4
+}
+
+define <4 x double> @test_mm256_mask3_fmaddsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231pd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0xb6,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) +/- ymm2
+; CHECK-NEXT:    vmovapd %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %1) #9
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x double> %3, <4 x double> %__C
+  ret <4 x double> %5
+}
+
+define <4 x double> @test_mm256_maskz_fmaddsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmaddsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0xa6,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %1) #9
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x double> %3, <4 x double> zeroinitializer
+  ret <4 x double> %5
+}
+
+define <4 x double> @test_mm256_maskz_fmsubadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213pd %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xf5,0xa9,0xa7,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> zeroinitializer
+  ret <4 x double> %4
+}
+
+define <4 x float> @test_mm_mask_fmaddsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_mask_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132ps %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x96,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) +/- xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %1) #9
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x float> %3, <4 x float> %__A
+  ret <4 x float> %5
+}
+
+define <4 x float> @test_mm_mask_fmsubadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_mask_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132ps %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x97,0xc1]
+; CHECK-NEXT:    ## xmm0 = (xmm0 * xmm1) -/+ xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> %__A
+  ret <4 x float> %4
+}
+
+define <4 x float> @test_mm_mask3_fmaddsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231ps %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xb6,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) +/- xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %1) #9
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x float> %3, <4 x float> %__C
+  ret <4 x float> %5
+}
+
+define <4 x float> @test_mm_maskz_fmaddsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xa6,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %1) #9
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x float> %3, <4 x float> zeroinitializer
+  ret <4 x float> %5
+}
+
+define <4 x float> @test_mm_maskz_fmsubadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_maskz_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213ps %xmm2, %xmm1, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0x89,0xa7,0xc2]
+; CHECK-NEXT:    ## xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> zeroinitializer
+  ret <4 x float> %4
+}
+
+define <8 x float> @test_mm256_mask_fmaddsub_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub132ps %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x29,0x96,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) +/- ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %1) #9
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> %__A
+  ret <8 x float> %5
+}
+
+define <8 x float> @test_mm256_mask_fmsubadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_mask_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd132ps %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x29,0x97,0xc1]
+; CHECK-NEXT:    ## ymm0 = (ymm0 * ymm1) -/+ ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %__A
+  ret <8 x float> %4
+}
+
+define <8 x float> @test_mm256_mask3_fmaddsub_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub231ps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0xb6,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) +/- ymm2
+; CHECK-NEXT:    vmovaps %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %1) #9
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> %__C
+  ret <8 x float> %5
+}
+
+define <8 x float> @test_mm256_maskz_fmaddsub_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmaddsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0xa6,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %1) #9
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> zeroinitializer
+  ret <8 x float> %5
+}
+
+define <8 x float> @test_mm256_maskz_fmsubadd_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_maskz_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd213ps %ymm2, %ymm1, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0x75,0xa9,0xa7,0xc2]
+; CHECK-NEXT:    ## ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> zeroinitializer
+  ret <8 x float> %4
+}
+
+define <2 x double> @test_mm_mask3_fmsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231pd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xba,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fmsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231pd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0xba,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    vmovapd %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask3_fmsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ps %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xba,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fmsub_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsub231ps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0xba,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    vmovaps %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+define <2 x double> @test_mm_mask3_fmsubadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231pd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xb7,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) -/+ xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> %__C
+  ret <2 x double> %4
+}
+
+define <4 x double> @test_mm256_mask3_fmsubadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmsubadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231pd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0xb7,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) -/+ ymm2
+; CHECK-NEXT:    vmovapd %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> %__C
+  ret <4 x double> %4
+}
+
+define <4 x float> @test_mm_mask3_fmsubadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231ps %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xb7,0xd1]
+; CHECK-NEXT:    ## xmm2 = (xmm0 * xmm1) -/+ xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> %__C
+  ret <4 x float> %4
+}
+
+define <8 x float> @test_mm256_mask3_fmsubadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fmsubadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfmsubadd231ps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0xb7,0xd1]
+; CHECK-NEXT:    ## ymm2 = (ymm0 * ymm1) -/+ ymm2
+; CHECK-NEXT:    vmovaps %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %__C
+  ret <8 x float> %4
+}
+
+define <2 x double> @test_mm_mask_fnmadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_mask_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132pd %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x9c,0xc1]
+; CHECK-NEXT:    ## xmm0 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %sub.i, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fnmadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_mask_fnmadd_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132pd %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x9c,0xc1]
+; CHECK-NEXT:    ## ymm0 = -(ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %sub.i, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask_fnmadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_mask_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132ps %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x9c,0xc1]
+; CHECK-NEXT:    ## xmm0 = -(xmm0 * xmm1) + xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %sub.i, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fnmadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_mask_fnmadd_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmadd132ps %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x29,0x9c,0xc1]
+; CHECK-NEXT:    ## ymm0 = -(ymm0 * ymm1) + ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %sub.i, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <2 x double> @test_mm_mask_fnmsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; CHECK-LABEL: test_mm_mask_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132pd %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x9e,0xc1]
+; CHECK-NEXT:    ## xmm0 = -(xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %sub.i, <2 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask3_fnmsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231pd %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0xbe,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovapd %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %sub.i, <2 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fnmsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; CHECK-LABEL: test_mm256_mask_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132pd %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x9e,0xc1]
+; CHECK-NEXT:    ## ymm0 = -(ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %sub.i, <4 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fnmsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fnmsub_pd:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231pd %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0xbe,0xd1]
+; CHECK-NEXT:    ## ymm2 = -(ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    vmovapd %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %sub.i, <4 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask_fnmsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; CHECK-LABEL: test_mm_mask_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132ps %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x9e,0xc1]
+; CHECK-NEXT:    ## xmm0 = -(xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %sub.i, <4 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask3_fnmsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm_mask3_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231ps %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0xbe,0xd1]
+; CHECK-NEXT:    ## xmm2 = -(xmm0 * xmm1) - xmm2
+; CHECK-NEXT:    vmovaps %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %sub.i, <4 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fnmsub_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; CHECK-LABEL: test_mm256_mask_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub132ps %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x29,0x9e,0xc1]
+; CHECK-NEXT:    ## ymm0 = -(ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %sub.i, <8 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fnmsub_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; CHECK-LABEL: test_mm256_mask3_fnmsub_ps:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vfnmsub231ps %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0xbe,0xd1]
+; CHECK-NEXT:    ## ymm2 = -(ymm0 * ymm1) - ymm2
+; CHECK-NEXT:    vmovaps %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %sub.i, <8 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) #8
+declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) #8
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) #8
+declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) #8

llvm/trunk/test/CodeGen/X86/avx512vl-intrinsics-fast-isel.ll

entry:
   %0 = tail call <4 x i64> @llvm.x86.avx512.vpermi2var.q.256(<4 x i64> %__A, <4 x i64> %__I, <4 x i64> %__B)
   %1 = bitcast i8 %__U to <8 x i1>
   %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
   %2 = select <4 x i1> %extract.i, <4 x i64> %0, <4 x i64> zeroinitializer
   ret <4 x i64> %2
 }
 
 
+define <2 x double> @test_mm_mask_fmadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_mask_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd132pd {{.*#+}} xmm0 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132pd {{.*#+}} xmm0 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask_fmsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_mask_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub132pd {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132pd {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask3_fmadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231pd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231pd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask3_fnmadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fmadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fmsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fnmadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_maskz_fnmsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %__B, <2 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> zeroinitializer
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fmadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_mask_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm1) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132pd {{.*#+}} ymm0 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fmsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_mask_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132pd {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fmadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231pd {{.*#+}} ymm2 = (ymm0 * ymm1) + ymm2
+; X32-NEXT:    vmovapd %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231pd {{.*#+}} ymm2 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fnmadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) + ymm2
+; X32-NEXT:    vmovapd %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fmadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_maskz_fmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fmsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_maskz_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fnmadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_maskz_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_maskz_fnmsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_maskz_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213pd {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %__B, <4 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> zeroinitializer
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask_fmadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_mask_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd132ps {{.*#+}} xmm0 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132ps {{.*#+}} xmm0 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask_fmsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_mask_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub132ps {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132ps {{.*#+}} xmm0 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask3_fmadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231ps {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ps {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask3_fnmadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231ps {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231ps {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fmadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fmsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fnmadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_maskz_fnmsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %__B, <4 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> zeroinitializer
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fmadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_mask_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm1) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd132ps {{.*#+}} ymm0 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fmsub_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_mask_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub132ps {{.*#+}} ymm0 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fmadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd231ps {{.*#+}} ymm2 = (ymm0 * ymm1) + ymm2
+; X32-NEXT:    vmovaps %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd231ps {{.*#+}} ymm2 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fnmadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd231ps {{.*#+}} ymm2 = -(ymm0 * ymm1) + ymm2
+; X32-NEXT:    vmovaps %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd231ps {{.*#+}} ymm2 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fmadd_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_maskz_fmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fmsub_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_maskz_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fnmadd_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_maskz_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_maskz_fnmsub_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_maskz_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %__B, <8 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
+  ret <8 x float> %2
+}
+
+define <2 x double> @test_mm_mask_fmaddsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_mask_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub132pd {{.*#+}} xmm0 = (xmm0 * xmm1) +/- xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132pd {{.*#+}} xmm0 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %1) #9
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract.i, <2 x double> %3, <2 x double> %__A
+  ret <2 x double> %5
+}
+
+define <2 x double> @test_mm_mask_fmsubadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_mask_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd132pd {{.*#+}} xmm0 = (xmm0 * xmm1) -/+ xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132pd {{.*#+}} xmm0 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> %__A
+  ret <2 x double> %4
+}
+
+define <2 x double> @test_mm_mask3_fmaddsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub231pd {{.*#+}} xmm2 = (xmm0 * xmm1) +/- xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231pd {{.*#+}} xmm2 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %1) #9
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract.i, <2 x double> %3, <2 x double> %__C
+  ret <2 x double> %5
+}
+
+define <2 x double> @test_mm_maskz_fmaddsub_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %1) #9
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract.i, <2 x double> %3, <2 x double> zeroinitializer
+  ret <2 x double> %5
+}
+
+define <2 x double> @test_mm_maskz_fmsubadd_pd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_maskz_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> zeroinitializer
+  ret <2 x double> %4
+}
+
+define <4 x double> @test_mm256_mask_fmaddsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_mask_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm1) +/- ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132pd {{.*#+}} ymm0 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %1) #9
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x double> %3, <4 x double> %__A
+  ret <4 x double> %5
+}
+
+define <4 x double> @test_mm256_mask_fmsubadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_mask_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm1) -/+ ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132pd {{.*#+}} ymm0 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> %__A
+  ret <4 x double> %4
+}
+
+define <4 x double> @test_mm256_mask3_fmaddsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub231pd {{.*#+}} ymm2 = (ymm0 * ymm1) +/- ymm2
+; X32-NEXT:    vmovapd %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231pd {{.*#+}} ymm2 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %1) #9
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x double> %3, <4 x double> %__C
+  ret <4 x double> %5
+}
+
+define <4 x double> @test_mm256_maskz_fmaddsub_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_maskz_fmaddsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmaddsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %1) #9
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x double> %3, <4 x double> zeroinitializer
+  ret <4 x double> %5
+}
+
+define <4 x double> @test_mm256_maskz_fmsubadd_pd(i8 zeroext %__U, <4 x double> %__A, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_maskz_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> zeroinitializer
+  ret <4 x double> %4
+}
+
+define <4 x float> @test_mm_mask_fmaddsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_mask_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub132ps {{.*#+}} xmm0 = (xmm0 * xmm1) +/- xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132ps {{.*#+}} xmm0 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %1) #9
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x float> %3, <4 x float> %__A
+  ret <4 x float> %5
+}
+
+define <4 x float> @test_mm_mask_fmsubadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_mask_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd132ps {{.*#+}} xmm0 = (xmm0 * xmm1) -/+ xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132ps {{.*#+}} xmm0 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> %__A
+  ret <4 x float> %4
+}
+
+define <4 x float> @test_mm_mask3_fmaddsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub231ps {{.*#+}} xmm2 = (xmm0 * xmm1) +/- xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231ps {{.*#+}} xmm2 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %1) #9
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x float> %3, <4 x float> %__C
+  ret <4 x float> %5
+}
+
+define <4 x float> @test_mm_maskz_fmaddsub_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %1) #9
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %4, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract.i, <4 x float> %3, <4 x float> zeroinitializer
+  ret <4 x float> %5
+}
+
+define <4 x float> @test_mm_maskz_fmsubadd_ps(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_maskz_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_maskz_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> zeroinitializer
+  ret <4 x float> %4
+}
+
+define <8 x float> @test_mm256_mask_fmaddsub_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_mask_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm1) +/- ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub132ps {{.*#+}} ymm0 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %1) #9
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> %__A
+  ret <8 x float> %5
+}
+
+define <8 x float> @test_mm256_mask_fmsubadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_mask_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm1) -/+ ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd132ps {{.*#+}} ymm0 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %__A
+  ret <8 x float> %4
+}
+
+define <8 x float> @test_mm256_mask3_fmaddsub_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub231ps {{.*#+}} ymm2 = (ymm0 * ymm1) +/- ymm2
+; X32-NEXT:    vmovaps %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub231ps {{.*#+}} ymm2 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %1) #9
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> %__C
+  ret <8 x float> %5
+}
+
+define <8 x float> @test_mm256_maskz_fmaddsub_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_maskz_fmaddsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmaddsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2
+; X64-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %1) #9
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %4 = bitcast i8 %__U to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> zeroinitializer
+  ret <8 x float> %5
+}
+
+define <8 x float> @test_mm256_maskz_fmsubadd_ps(i8 zeroext %__U, <8 x float> %__A, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_maskz_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_maskz_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> zeroinitializer
+  ret <8 x float> %4
+}
+
+define <2 x double> @test_mm_mask3_fmsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231pd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231pd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fmsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231pd {{.*#+}} ymm2 = (ymm0 * ymm1) - ymm2
+; X32-NEXT:    vmovapd %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231pd {{.*#+}} ymm2 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask3_fmsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231ps {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ps {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fmsub_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsub231ps {{.*#+}} ymm2 = (ymm0 * ymm1) - ymm2
+; X32-NEXT:    vmovaps %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsub231ps {{.*#+}} ymm2 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+define <2 x double> @test_mm_mask3_fmsubadd_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd231pd {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231pd {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub.i) #9
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C) #9
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract.i, <2 x double> %2, <2 x double> %__C
+  ret <2 x double> %4
+}
+
+define <4 x double> @test_mm256_mask3_fmsubadd_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmsubadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd231pd {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X32-NEXT:    vmovapd %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmsubadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231pd {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %sub.i) #9
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C) #9
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x double> %2, <4 x double> %__C
+  ret <4 x double> %4
+}
+
+define <4 x float> @test_mm_mask3_fmsubadd_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd231ps {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231ps {{.*#+}} xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub.i) #9
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C) #9
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %3, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract.i, <4 x float> %2, <4 x float> %__C
+  ret <4 x float> %4
+}
+
+define <8 x float> @test_mm256_mask3_fmsubadd_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fmsubadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfmsubadd231ps {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X32-NEXT:    vmovaps %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fmsubadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfmsubadd231ps {{.*#+}} ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %sub.i) #9
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C) #9
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %3 = bitcast i8 %__U to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %__C
+  ret <8 x float> %4
+}
+
+define <2 x double> @test_mm_mask_fnmadd_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_mask_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %sub.i, <2 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fnmadd_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_mask_fnmadd_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmadd_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %sub.i, <4 x double> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask_fnmadd_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_mask_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %sub.i, <4 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fnmadd_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_mask_fnmadd_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmadd_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmadd132ps {{.*#+}} ymm0 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %sub.i, <8 x float> %__C) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <2 x double> @test_mm_mask_fnmsub_pd(<2 x double> %__A, i8 zeroext %__U, <2 x double> %__B, <2 x double> %__C) {
+; X32-LABEL: test_mm_mask_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132pd {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %sub.i, <2 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__A
+  ret <2 x double> %2
+}
+
+define <2 x double> @test_mm_mask3_fnmsub_pd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovapd %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231pd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %__A, <2 x double> %sub.i, <2 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %2 = select <2 x i1> %extract.i, <2 x double> %0, <2 x double> %__C
+  ret <2 x double> %2
+}
+
+define <4 x double> @test_mm256_mask_fnmsub_pd(<4 x double> %__A, i8 zeroext %__U, <4 x double> %__B, <4 x double> %__C) {
+; X32-LABEL: test_mm256_mask_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132pd {{.*#+}} ymm0 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %sub.i, <4 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__A
+  ret <4 x double> %2
+}
+
+define <4 x double> @test_mm256_mask3_fnmsub_pd(<4 x double> %__A, <4 x double> %__B, <4 x double> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fnmsub_pd:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) - ymm2
+; X32-NEXT:    vmovapd %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fnmsub_pd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231pd {{.*#+}} ymm2 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %__A, <4 x double> %sub.i, <4 x double> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x double> %0, <4 x double> %__C
+  ret <4 x double> %2
+}
+
+define <4 x float> @test_mm_mask_fnmsub_ps(<4 x float> %__A, i8 zeroext %__U, <4 x float> %__B, <4 x float> %__C) {
+; X32-LABEL: test_mm_mask_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132ps {{.*#+}} xmm0 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %sub.i, <4 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__A
+  ret <4 x float> %2
+}
+
+define <4 x float> @test_mm_mask3_fnmsub_ps(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm_mask3_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231ps {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X32-NEXT:    vmovaps %xmm2, %xmm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm_mask3_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231ps {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %__A, <4 x float> %sub.i, <4 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %extract.i = shufflevector <8 x i1> %1, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = select <4 x i1> %extract.i, <4 x float> %0, <4 x float> %__C
+  ret <4 x float> %2
+}
+
+define <8 x float> @test_mm256_mask_fnmsub_ps(<8 x float> %__A, i8 zeroext %__U, <8 x float> %__B, <8 x float> %__C) {
+; X32-LABEL: test_mm256_mask_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm1) - ymm2
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub132ps {{.*#+}} ymm0 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %sub.i, <8 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__A
+  ret <8 x float> %2
+}
+
+define <8 x float> @test_mm256_mask3_fnmsub_ps(<8 x float> %__A, <8 x float> %__B, <8 x float> %__C, i8 zeroext %__U) {
+; X32-LABEL: test_mm256_mask3_fnmsub_ps:
+; X32:       # %bb.0: # %entry
+; X32-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X32-NEXT:    kmovw %eax, %k1
+; X32-NEXT:    vfnmsub231ps {{.*#+}} ymm2 = -(ymm0 * ymm1) - ymm2
+; X32-NEXT:    vmovaps %ymm2, %ymm0
+; X32-NEXT:    retl
+;
+; X64-LABEL: test_mm256_mask3_fnmsub_ps:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    kmovw %edi, %k1
+; X64-NEXT:    vfnmsub231ps {{.*#+}} ymm2 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0
+; X64-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %__A, <8 x float> %sub.i, <8 x float> %sub1.i) #9
+  %1 = bitcast i8 %__U to <8 x i1>
+  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %__C
+  ret <8 x float> %2
+}
+
+declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) #8
+declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) #8
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) #8
+declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) #8
+
 declare <4 x i32> @llvm.x86.avx512.mask.cvtpd2dq.128(<2 x double>, <4 x i32>, i8)
 declare <4 x i32> @llvm.x86.avx.cvt.pd2dq.256(<4 x double>)
 declare <4 x float> @llvm.x86.avx512.mask.cvtpd2ps(<2 x double>, <4 x float>, i8)
 declare <4 x float> @llvm.x86.avx.cvt.pd2.ps.256(<4 x double>)
 declare <4 x i32> @llvm.x86.avx512.mask.cvtpd2udq.128(<2 x double>, <4 x i32>, i8)
 declare <4 x i32> @llvm.x86.avx512.mask.cvtpd2udq.256(<4 x double>, <4 x i32>, i8)
 declare <4 x i32> @llvm.x86.sse2.cvtps2dq(<4 x float>)
 declare <8 x i32> @llvm.x86.avx.cvt.ps2dq.256(<8 x float>)

llvm/trunk/test/CodeGen/X86/fma-fneg-combine.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512bw -mattr=+avx512vl -mattr=+avx512dq  | FileCheck %s  --check-prefix=CHECK --check-prefix=SKX
 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f -mattr=+fma | FileCheck %s --check-prefix=CHECK --check-prefix=KNL
 
 ; This test checks combinations of FNEG and FMA intrinsics on AVX-512 target
 ; PR28892
 
 define <16 x float> @test1(<16 x float> %a, <16 x float> %b, <16 x float> %c)  {
 ; CHECK-LABEL: test1:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfmsub213ps %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
 entry:
   %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
-  %0 = tail call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %sub.i, i16 -1, i32 4) #2
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %sub.i, i32 4) #2
   ret <16 x float> %0
 }
 
-declare <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
+declare <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i32)
 declare <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
 declare <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
 
 
 define <16 x float> @test2(<16 x float> %a, <16 x float> %b, <16 x float> %c) {
 ; CHECK-LABEL: test2:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfnmsub213ps %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
 entry:
-  %0 = tail call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 -1, i32 4) #2
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i32 4) #2
   %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %0
   ret <16 x float> %sub.i
 }
 
 define <16 x float> @test3(<16 x float> %a, <16 x float> %b, <16 x float> %c)  {
 ; CHECK-LABEL: test3:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfmsub213ps %zmm2, %zmm1, %zmm0

 
 define <16 x float> @test5(<16 x float> %a, <16 x float> %b, <16 x float> %c) {
 ; CHECK-LABEL: test5:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfmsub213ps {ru-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
 entry:
   %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
-  %0 = tail call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %sub.i, i16 -1, i32 2) #2
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %sub.i, i32 2) #2
   ret <16 x float> %0
 }
 
 define <16 x float> @test6(<16 x float> %a, <16 x float> %b, <16 x float> %c) {
 ; CHECK-LABEL: test6:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq

 
 
 define <8 x double> @test9(<8 x double> %a, <8 x double> %b, <8 x double> %c) {
 ; CHECK-LABEL: test9:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfnmsub213pd %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
 entry:
-  %0 = tail call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 -1, i32 4) #2
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i32 4) #2
   %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %0
   ret <8 x double> %sub.i
 }
 
-declare <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8, i32)
+declare <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i32)
 
 define <2 x double> @test10(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: test10:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0
 ; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm0, %xmm0
 ; CHECK-NEXT:    retq
 entry:

 ;
 ; KNL-LABEL: test12:
 ; KNL:       # %bb.0: # %entry
 ; KNL-NEXT:    kmovw %edi, %k1
 ; KNL-NEXT:    vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1}
 ; KNL-NEXT:    vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0
 ; KNL-NEXT:    retq
 entry:
-  %0 = tail call <8 x double> @llvm.x86.avx512.mask.vfmadd.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 4) #2
-  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %0
+  %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i32 4) #2
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %0, <8 x double> %a
+  %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %sel
   ret <8 x double> %sub.i
 }
 
 define <2 x double> @test13(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
 ; SKX-LABEL: test13:
 ; SKX:       # %bb.0: # %entry
 ; SKX-NEXT:    vxorpd {{.*}}(%rip), %xmm0, %xmm0
 ; SKX-NEXT:    kmovd %edi, %k1

 ; KNL-NEXT:    kmovw %edi, %k1
 ; KNL-NEXT:    vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm3
 ; KNL-NEXT:    vfnmadd213ps {ru-sae}, %zmm2, %zmm0, %zmm1
 ; KNL-NEXT:    vmovaps %zmm1, %zmm3 {%k1}
 ; KNL-NEXT:    vfnmadd132ps {rd-sae}, %zmm0, %zmm2, %zmm3 {%k1}
 ; KNL-NEXT:    vmovaps %zmm3, %zmm0
 ; KNL-NEXT:    retq
 entry:
+  %bc = bitcast i16 %mask to <16 x i1>
   %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
-  %0 = tail call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %sub.i, <16 x float> %b, <16 x float> %c, i16 %mask, i32 2)
-  %1 = tail call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %0, <16 x float> %sub.i, <16 x float> %c, i16 %mask, i32 1)
-  ret <16 x float> %1
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub.i, <16 x float> %b, <16 x float> %c, i32 2)
+  %sel = select <16 x i1> %bc, <16 x float> %0, <16 x float> %sub.i
+  %1 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sel, <16 x float> %sub.i, <16 x float> %c, i32 1)
+  %sel2 = select <16 x i1> %bc, <16 x float> %1, <16 x float> %sel
+  ret <16 x float> %sel2
 }
 
 define <16 x float> @test16(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
 ; SKX-LABEL: test16:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    kmovd %edi, %k1
 ; SKX-NEXT:    vfmsubadd132ps {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; SKX-NEXT:    retq
 ;
 ; KNL-LABEL: test16:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    kmovw %edi, %k1
 ; KNL-NEXT:    vfmsubadd132ps {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1}
 ; KNL-NEXT:    retq
   %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %sub.i, i16 %mask, i32 1)
-  ret <16 x float> %res
+  %res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %sub.i, i32 1)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a
+  ret <16 x float> %sel
 }
-declare <16 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
+declare <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i32)
 
 define <8 x double> @test17(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
 ; SKX-LABEL: test17:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    kmovd %edi, %k1
 ; SKX-NEXT:    vfmsubadd132pd %zmm1, %zmm2, %zmm0 {%k1}
 ; SKX-NEXT:    retq
 ;
 ; KNL-LABEL: test17:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    kmovw %edi, %k1
 ; KNL-NEXT:    vfmsubadd132pd %zmm1, %zmm2, %zmm0 {%k1}
 ; KNL-NEXT:    retq
   %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %sub.i, i8 %mask, i32 4)
-  ret <8 x double> %res
+  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %sub.i, i32 4)
+  %bc = bitcast i8 %mask to <8 x i1>
+  %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a
+  ret <8 x double> %sel
 }
-declare <8 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
+declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32)
 
 define <4 x float> @test18(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 zeroext %mask) local_unnamed_addr #0 {
 ; SKX-LABEL: test18:
 ; SKX:       # %bb.0: # %entry
 ; SKX-NEXT:    kmovd %edi, %k1
 ; SKX-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 {%k1}
 ; SKX-NEXT:    retq
 ;

 define <16 x float> @test25(<16 x float> %a, <16 x float> %b, <16 x float> %c)  {
 ; CHECK-LABEL: test25:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0
 ; CHECK-NEXT:    retq
 entry:
   %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %b
   %sub.i.2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
-  %0 = tail call <16 x float> @llvm.x86.avx512.mask.vfmadd.ps.512(<16 x float> %a, <16 x float> %sub.i, <16 x float> %sub.i.2, i16 -1, i32 8) #2
+  %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a, <16 x float> %sub.i, <16 x float> %sub.i.2, i32 8) #2
   ret <16 x float> %0
 }

llvm/trunk/test/CodeGen/X86/fma-intrinsics-canonical.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX512VL
+; RUN: llc < %s -mtriple=x86_64-pc-windows -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
+
+; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/fma-builtins.c
+
+define <4 x float> @test_mm_fmadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fmadd_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmadd_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmadd_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa8,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fmadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fmadd_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmadd_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmadd_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa8,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fmadd_ss:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmadd_ss:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmadd_ss:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero,zero,zero
+; CHECK-FMA-WIN-NEXT:    vfmadd132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x99,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm0 * mem) + xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %1 = extractelement <4 x float> %b, i64 0
+  %2 = extractelement <4 x float> %c, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fmadd_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fmadd_sd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa9,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmadd_sd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmadd_sd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero
+; CHECK-FMA-WIN-NEXT:    vfmadd132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x99,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm0 * mem) + xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %1 = extractelement <2 x double> %b, i64 0
+  %2 = extractelement <2 x double> %c, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fmsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fmsub_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xaa,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmsub_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xaa,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmsub_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xaa,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fmsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fmsub_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xaa,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmsub_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xaa,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmsub_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xaa,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fmsub_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fmsub_ss:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xab,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmsub_ss:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xab,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmsub_ss:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero,zero,zero
+; CHECK-FMA-WIN-NEXT:    vfmsub132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9b,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm0 * mem) - xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %1 = extractelement <4 x float> %b, i64 0
+  %.rhs.i = extractelement <4 x float> %c, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fmsub_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fmsub_sd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xab,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmsub_sd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xab,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmsub_sd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero
+; CHECK-FMA-WIN-NEXT:    vfmsub132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9b,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm0 * mem) - xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %1 = extractelement <2 x double> %b, i64 0
+  %.rhs.i = extractelement <2 x double> %c, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fnmadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmadd_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xac,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmadd_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xac,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmadd_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xac,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm1 * xmm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %b, <4 x float> %c) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fnmadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmadd_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmadd_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmadd_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xac,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm1 * xmm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %b, <2 x double> %c) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fnmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmadd_ss:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xad,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmadd_ss:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xad,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmadd_ss:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero,zero,zero
+; CHECK-FMA-WIN-NEXT:    vfnmadd132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9d,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm0 * mem) + xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %.rhs.i = extractelement <4 x float> %b, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %c, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fnmadd_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmadd_sd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xad,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmadd_sd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xad,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmadd_sd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero
+; CHECK-FMA-WIN-NEXT:    vfnmadd132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9d,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm0 * mem) + xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %.rhs.i = extractelement <2 x double> %b, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %c, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fnmsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmsub_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xae,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmsub_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xae,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmsub_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xae,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm1 * xmm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %b, <4 x float> %sub1.i) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fnmsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmsub_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmsub_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmsub_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xae,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm1 * xmm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %b, <2 x double> %sub1.i) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fnmsub_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmsub_ss:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xaf,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmsub_ss:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xaf,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmsub_ss:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero,zero,zero
+; CHECK-FMA-WIN-NEXT:    vfnmsub132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9f,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm0 * mem) - xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %.rhs.i = extractelement <4 x float> %b, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs2.i = extractelement <4 x float> %c, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs2.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fnmsub_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fnmsub_sd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xaf,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fnmsub_sd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xaf,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fnmsub_sd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
+; CHECK-FMA-WIN-NEXT:    vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
+; CHECK-FMA-WIN-NEXT:    # xmm1 = mem[0],zero
+; CHECK-FMA-WIN-NEXT:    vfnmsub132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9f,0x02]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = -(xmm0 * mem) - xmm1
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %.rhs.i = extractelement <2 x double> %b, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs2.i = extractelement <2 x double> %c, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs2.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fmaddsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fmaddsub_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmaddsub_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmaddsub_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa6,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) +/- mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %1) #2
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %3
+}
+
+define <2 x double> @test_mm_fmaddsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fmaddsub_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmaddsub_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmaddsub_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa6,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) +/- mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %1) #2
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %3
+}
+
+define <4 x float> @test_mm_fmsubadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-FMA-LABEL: test_mm_fmsubadd_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsubadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa7,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmsubadd_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsubadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa7,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmsubadd_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa7,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) -/+ mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i) #2
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %2
+}
+
+define <2 x double> @test_mm_fmsubadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-FMA-LABEL: test_mm_fmsubadd_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsubadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa7,0xc2]
+; CHECK-FMA-NEXT:    # xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm_fmsubadd_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsubadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa7,0xc2]
+; CHECK-AVX512VL-NEXT:    # xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm_fmsubadd_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa7,0x00]
+; CHECK-FMA-WIN-NEXT:    # xmm0 = (xmm1 * xmm0) -/+ mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i) #2
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %2
+}
+
+define <8 x float> @test_mm256_fmadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmadd_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmadd_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmadd_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa8,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fmadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmadd_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmadd_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmadd_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa8,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fmsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmsub_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xaa,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) - ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmsub_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xaa,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) - ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmsub_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xaa,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fmsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmsub_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xaa,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) - ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmsub_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xaa,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) - ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmsub_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xaa,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %sub.i) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fnmadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-FMA-LABEL: test_mm256_fnmadd_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xac,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fnmadd_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xac,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fnmadd_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xac,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = -(ymm1 * ymm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %b, <8 x float> %c) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fnmadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-FMA-LABEL: test_mm256_fnmadd_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fnmadd_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fnmadd_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xac,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = -(ymm1 * ymm0) + mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %b, <4 x double> %c) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fnmsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-FMA-LABEL: test_mm256_fnmsub_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xae,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fnmsub_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xae,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fnmsub_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xae,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = -(ymm1 * ymm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %b, <8 x float> %sub1.i) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fnmsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-FMA-LABEL: test_mm256_fnmsub_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fnmsub_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fnmsub_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfnmsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xae,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = -(ymm1 * ymm0) - mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %b, <4 x double> %sub1.i) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fmaddsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmaddsub_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmaddsub_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmaddsub_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa6,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) +/- mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %1) #2
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %3
+}
+
+define <4 x double> @test_mm256_fmaddsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmaddsub_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmaddsub_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmaddsub_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa6,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) +/- mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %1) #2
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %3
+}
+
+define <8 x float> @test_mm256_fmsubadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmsubadd_ps:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsubadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa7,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmsubadd_ps:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsubadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa7,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmsubadd_ps:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa7,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) -/+ mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %2
+}
+
+define <4 x double> @test_mm256_fmsubadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-FMA-LABEL: test_mm256_fmsubadd_pd:
+; CHECK-FMA:       # %bb.0: # %entry
+; CHECK-FMA-NEXT:    vfmsubadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa7,0xc2]
+; CHECK-FMA-NEXT:    # ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-FMA-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-AVX512VL-LABEL: test_mm256_fmsubadd_pd:
+; CHECK-AVX512VL:       # %bb.0: # %entry
+; CHECK-AVX512VL-NEXT:    vfmsubadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa7,0xc2]
+; CHECK-AVX512VL-NEXT:    # ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-AVX512VL-NEXT:    retq # encoding: [0xc3]
+;
+; CHECK-FMA-WIN-LABEL: test_mm256_fmsubadd_pd:
+; CHECK-FMA-WIN:       # %bb.0: # %entry
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa7,0x00]
+; CHECK-FMA-WIN-NEXT:    # ymm0 = (ymm1 * ymm0) -/+ mem
+; CHECK-FMA-WIN-NEXT:    retq # encoding: [0xc3]
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %sub.i) #2
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %2
+}
+
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) #1
+declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) #1
+declare float @llvm.fma.f32(float, float, float) #1
+declare double @llvm.fma.f64(double, double, double) #1
+declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) #1
+declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) #1

llvm/trunk/test/CodeGen/X86/fma-intrinsics-fast-isel.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+fma,-fma4 | FileCheck %s --check-prefix=CHECK
+
+define <4 x float> @test_mm_fmadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fmadd_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fmadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fmadd_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fmadd_ss:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %1 = extractelement <4 x float> %b, i64 0
+  %2 = extractelement <4 x float> %c, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fmadd_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fmadd_sd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %1 = extractelement <2 x double> %b, i64 0
+  %2 = extractelement <2 x double> %c, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fmsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fmsub_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorps {{.*}}(%rip), %xmm2, %xmm2
+; CHECK-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fmsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fmsub_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm2, %xmm2
+; CHECK-NEXT:    vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fmsub_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fmsub_ss:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %1 = extractelement <4 x float> %b, i64 0
+  %.rhs.i = extractelement <4 x float> %c, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fmsub_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fmsub_sd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %1 = extractelement <2 x double> %b, i64 0
+  %.rhs.i = extractelement <2 x double> %c, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fnmadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fnmadd_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorps {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-NEXT:    vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %b, <4 x float> %c) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fnmadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fnmadd_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorpd {{.*}}(%rip), %xmm0, %xmm0
+; CHECK-NEXT:    vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %b, <2 x double> %c) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fnmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fnmadd_ss:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %.rhs.i = extractelement <4 x float> %b, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %2 = extractelement <4 x float> %c, i64 0
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fnmadd_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fnmadd_sd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %.rhs.i = extractelement <2 x double> %b, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %2 = extractelement <2 x double> %c, i64 0
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fnmsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fnmsub_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmovaps {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00]
+; CHECK-NEXT:    vxorps %xmm3, %xmm0, %xmm4
+; CHECK-NEXT:    vxorps %xmm3, %xmm2, %xmm0
+; CHECK-NEXT:    vfmadd231ps {{.*#+}} xmm0 = (xmm1 * xmm4) + xmm0
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %b, <4 x float> %sub1.i) #2
+  ret <4 x float> %0
+}
+
+define <2 x double> @test_mm_fnmsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fnmsub_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00]
+; CHECK-NEXT:    vxorpd %xmm3, %xmm0, %xmm4
+; CHECK-NEXT:    vxorpd %xmm3, %xmm2, %xmm0
+; CHECK-NEXT:    vfmadd231pd {{.*#+}} xmm0 = (xmm1 * xmm4) + xmm0
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a
+  %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %b, <2 x double> %sub1.i) #2
+  ret <2 x double> %0
+}
+
+define <4 x float> @test_mm_fnmsub_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fnmsub_ss:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %a, i64 0
+  %.rhs.i = extractelement <4 x float> %b, i64 0
+  %1 = fsub float -0.000000e+00, %.rhs.i
+  %.rhs2.i = extractelement <4 x float> %c, i64 0
+  %2 = fsub float -0.000000e+00, %.rhs2.i
+  %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2
+  %4 = insertelement <4 x float> %a, float %3, i64 0
+  ret <4 x float> %4
+}
+
+define <2 x double> @test_mm_fnmsub_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fnmsub_sd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <2 x double> %a, i64 0
+  %.rhs.i = extractelement <2 x double> %b, i64 0
+  %1 = fsub double -0.000000e+00, %.rhs.i
+  %.rhs2.i = extractelement <2 x double> %c, i64 0
+  %2 = fsub double -0.000000e+00, %.rhs2.i
+  %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2
+  %4 = insertelement <2 x double> %a, double %3, i64 0
+  ret <2 x double> %4
+}
+
+define <4 x float> @test_mm_fmaddsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fmaddsub_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmaddsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %1) #2
+  %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %3
+}
+
+define <2 x double> @test_mm_fmaddsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fmaddsub_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmaddsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %1) #2
+  %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %3
+}
+
+define <4 x float> @test_mm_fmsubadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_mm_fmsubadd_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmsubadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i) #2
+  %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2
+  %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %2
+}
+
+define <2 x double> @test_mm_fmsubadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_mm_fmsubadd_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmsubadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i) #2
+  %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2
+  %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %2
+}
+
+define <8 x float> @test_mm256_fmadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_mm256_fmadd_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fmadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_mm256_fmadd_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fmsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_mm256_fmsub_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorps {{.*}}(%rip), %ymm2, %ymm2
+; CHECK-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fmsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_mm256_fmsub_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorpd {{.*}}(%rip), %ymm2, %ymm2
+; CHECK-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %sub.i) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fnmadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_mm256_fnmadd_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorps {{.*}}(%rip), %ymm0, %ymm0
+; CHECK-NEXT:    vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %b, <8 x float> %c) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fnmadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_mm256_fnmadd_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vxorpd {{.*}}(%rip), %ymm0, %ymm0
+; CHECK-NEXT:    vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %b, <4 x double> %c) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fnmsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_mm256_fnmsub_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmovaps {{.*#+}} ymm3 = [-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00]
+; CHECK-NEXT:    vxorps %ymm3, %ymm0, %ymm4
+; CHECK-NEXT:    vxorps %ymm3, %ymm2, %ymm0
+; CHECK-NEXT:    vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm4) + ymm0
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a
+  %sub1.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %b, <8 x float> %sub1.i) #2
+  ret <8 x float> %0
+}
+
+define <4 x double> @test_mm256_fnmsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_mm256_fnmsub_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmovapd {{.*#+}} ymm3 = [-0.000000e+00,-0.000000e+00,-0.000000e+00,-0.000000e+00]
+; CHECK-NEXT:    vxorpd %ymm3, %ymm0, %ymm4
+; CHECK-NEXT:    vxorpd %ymm3, %ymm2, %ymm0
+; CHECK-NEXT:    vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm4) + ymm0
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a
+  %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %b, <4 x double> %sub1.i) #2
+  ret <4 x double> %0
+}
+
+define <8 x float> @test_mm256_fmaddsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_mm256_fmaddsub_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %1) #2
+  %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %3
+}
+
+define <4 x double> @test_mm256_fmaddsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_mm256_fmaddsub_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-NEXT:    retq
+entry:
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %1) #2
+  %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %3
+}
+
+define <8 x float> @test_mm256_fmsubadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_mm256_fmsubadd_ps:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2
+  %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2
+  %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %2
+}
+
+define <4 x double> @test_mm256_fmsubadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_mm256_fmsubadd_pd:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2
+; CHECK-NEXT:    retq
+entry:
+  %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c
+  %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %sub.i) #2
+  %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2
+  %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %2
+}
+
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) #1
+declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) #1
+declare float @llvm.fma.f32(float, float, float) #1
+declare double @llvm.fma.f64(double, double, double) #1
+declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) #1
+declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) #1