AVX512 : valign intrinsic implementation

[igorb]

Summary

AVX512: Implemented intrinsics for valign.
Added tests for intrinsics.

[delena]

LGTM

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

let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
 
   def int_x86_avx512_mask_pshuf_b_512 : 
         GCCBuiltin<"__builtin_ia32_pshufb512_mask">,
           Intrinsic<[llvm_v64i8_ty],
           [llvm_v64i8_ty, llvm_v64i8_ty, llvm_v64i8_ty,  llvm_i64_ty],
           [IntrNoMem]>;
 }
 
-
 // Vector blend
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx_blendv_pd_256 : GCCBuiltin<"__builtin_ia32_blendvpd256">,
         Intrinsic<[llvm_v4f64_ty], [llvm_v4f64_ty,
                   llvm_v4f64_ty, llvm_v4f64_ty], [IntrNoMem]>;
   def int_x86_avx_blendv_ps_256 : GCCBuiltin<"__builtin_ia32_blendvps256">,
         Intrinsic<[llvm_v8f32_ty], [llvm_v8f32_ty,
                   llvm_v8f32_ty, llvm_v8f32_ty], [IntrNoMem]>;
def int_x86_avx512_mask_valign_q_512 :
                    llvm_i8_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask_valign_d_512 :
         GCCBuiltin<"__builtin_ia32_alignd512_mask">,
         Intrinsic<[llvm_v16i32_ty],
                   [llvm_v16i32_ty, llvm_v16i32_ty, llvm_i32_ty, llvm_v16i32_ty,
                    llvm_i16_ty], [IntrNoMem]>;
 
+  def int_x86_avx512_mask_valign_q_256 :
+        GCCBuiltin<"__builtin_ia32_alignq256_mask">,
+        Intrinsic<[llvm_v4i64_ty],
+                  [llvm_v4i64_ty, llvm_v4i64_ty, llvm_i32_ty, llvm_v4i64_ty,
+                   llvm_i8_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_mask_valign_d_256 :
+        GCCBuiltin<"__builtin_ia32_alignd256_mask">,
+        Intrinsic<[llvm_v8i32_ty],
+                  [llvm_v8i32_ty, llvm_v8i32_ty, llvm_i32_ty, llvm_v8i32_ty, 
+                   llvm_i8_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_mask_valign_q_128 :
+        GCCBuiltin<"__builtin_ia32_alignq128_mask">,
+        Intrinsic<[llvm_v2i64_ty],
+                  [llvm_v2i64_ty, llvm_v2i64_ty, llvm_i32_ty, llvm_v2i64_ty, 
+                   llvm_i8_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_mask_valign_d_128 :
+        GCCBuiltin<"__builtin_ia32_alignd128_mask">,
+        Intrinsic<[llvm_v4i32_ty],
+                  [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i32_ty, llvm_v4i32_ty,
+                   llvm_i8_ty], [IntrNoMem]>;
+
   def int_x86_avx512_mask_palignr_128 :
         GCCBuiltin<"__builtin_ia32_palignr128_mask">,
         Intrinsic<[llvm_v16i8_ty],
-                   [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i32_ty, llvm_v16i8_ty,
-                    llvm_i16_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask_palignr_256 :
         GCCBuiltin<"__builtin_ia32_palignr256_mask">,
         Intrinsic<[llvm_v32i8_ty],
                   [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty, llvm_v32i8_ty,
                    llvm_i32_ty], [IntrNoMem]>;
 
   def int_x86_avx512_mask_palignr_512 :

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

static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_unpckl_pd_512, INTR_TYPE_2OP_MASK,
                      X86ISD::UNPCKL, 0),
   X86_INTRINSIC_DATA(avx512_mask_unpckl_ps_128, INTR_TYPE_2OP_MASK,
                      X86ISD::UNPCKL, 0),
   X86_INTRINSIC_DATA(avx512_mask_unpckl_ps_256, INTR_TYPE_2OP_MASK,
                      X86ISD::UNPCKL, 0),
   X86_INTRINSIC_DATA(avx512_mask_unpckl_ps_512, INTR_TYPE_2OP_MASK,
                      X86ISD::UNPCKL, 0),
+  X86_INTRINSIC_DATA(avx512_mask_valign_d_128, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::VALIGN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_valign_d_256, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::VALIGN, 0),
   X86_INTRINSIC_DATA(avx512_mask_valign_d_512, INTR_TYPE_3OP_IMM8_MASK,
                      X86ISD::VALIGN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_valign_q_128, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::VALIGN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_valign_q_256, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::VALIGN, 0),
   X86_INTRINSIC_DATA(avx512_mask_valign_q_512, INTR_TYPE_3OP_IMM8_MASK,
                      X86ISD::VALIGN, 0),
   X86_INTRINSIC_DATA(avx512_mask_vfmadd_pd_128, FMA_OP_MASK, X86ISD::FMADD, 0),
   X86_INTRINSIC_DATA(avx512_mask_vfmadd_pd_256, FMA_OP_MASK, X86ISD::FMADD, 0),
   X86_INTRINSIC_DATA(avx512_mask_vfmadd_pd_512, FMA_OP_MASK, X86ISD::FMADD,
                      X86ISD::FMADD_RND),
   X86_INTRINSIC_DATA(avx512_mask_vfmadd_ps_128, FMA_OP_MASK, X86ISD::FMADD, 0),
   X86_INTRINSIC_DATA(avx512_mask_vfmadd_ps_256, FMA_OP_MASK, X86ISD::FMADD, 0),

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

 ; CHECK: vrndscaleps {{.*}}{%k1} 
 ; CHECK: vrndscaleps
 define <8 x float>@test_int_x86_avx512_mask_rndscale_ps_256(<8 x float> %x0, <8 x float> %x2, i8 %x3) {
   %res = call <8 x float> @llvm.x86.avx512.mask.rndscale.ps.256(<8 x float> %x0, i32 5, <8 x float> %x2, i8 %x3)
   %res1 = call <8 x float> @llvm.x86.avx512.mask.rndscale.ps.256(<8 x float> %x0, i32 66, <8 x float> %x2, i8 -1)
   %res2 = fadd <8 x float> %res, %res1
   ret <8 x float> %res2
 }
+
+declare <4 x i32> @llvm.x86.avx512.mask.valign.d.128(<4 x i32>, <4 x i32>, i32, <4 x i32>, i8)
+
+define <4 x i32>@test_int_x86_avx512_mask_valign_d_128(<4 x i32> %x0, <4 x i32> %x1,<4 x i32> %x3, i8 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_valign_d_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    kmovw %eax, %k1
+; CHECK-NEXT:    valignd $22, %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT:    valignd $22, %xmm1, %xmm0, %xmm3 {%k1} {z}
+; CHECK-NEXT:    valignd $22, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vpaddd %xmm0, %xmm2, %xmm0
+; CHECK-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
+; CHECK-NEXT:    retq
+  %res = call <4 x i32> @llvm.x86.avx512.mask.valign.d.128(<4 x i32> %x0, <4 x i32> %x1, i32 22, <4 x i32> %x3, i8 %x4)
+  %res1 = call <4 x i32> @llvm.x86.avx512.mask.valign.d.128(<4 x i32> %x0, <4 x i32> %x1, i32 22, <4 x i32> %x3, i8 -1)
+    %res2 = call <4 x i32> @llvm.x86.avx512.mask.valign.d.128(<4 x i32> %x0, <4 x i32> %x1, i32 22, <4 x i32> zeroinitializer,i8 %x4)
+  %res3 = add <4 x i32> %res, %res1
+    %res4 = add <4 x i32> %res3, %res2
+  ret <4 x i32> %res4
+}
+
+declare <8 x i32> @llvm.x86.avx512.mask.valign.d.256(<8 x i32>, <8 x i32>, i32, <8 x i32>, i8)
+
+define <8 x i32>@test_int_x86_avx512_mask_valign_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x3, i8 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_valign_d_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    kmovw %eax, %k1
+; CHECK-NEXT:    valignd $22, %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT:    valignd $22, %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpaddd %ymm0, %ymm2, %ymm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i32> @llvm.x86.avx512.mask.valign.d.256(<8 x i32> %x0, <8 x i32> %x1, i32 22, <8 x i32> %x3, i8 %x4)
+  %res1 = call <8 x i32> @llvm.x86.avx512.mask.valign.d.256(<8 x i32> %x0, <8 x i32> %x1, i32 22, <8 x i32> %x3, i8 -1)
+  %res2 = add <8 x i32> %res, %res1
+  ret <8 x i32> %res2
+}
+
+declare <2 x i64> @llvm.x86.avx512.mask.valign.q.128(<2 x i64>, <2 x i64>, i32, <2 x i64>, i8)
+
+define <2 x i64>@test_int_x86_avx512_mask_valign_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x3, i8 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_valign_q_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    kmovw %eax, %k1
+; CHECK-NEXT:    valignq $22, %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT:    valignq $22, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vpaddq %xmm0, %xmm2, %xmm0
+; CHECK-NEXT:    retq
+  %res = call <2 x i64> @llvm.x86.avx512.mask.valign.q.128(<2 x i64> %x0, <2 x i64> %x1, i32 22, <2 x i64> %x3, i8 %x4)
+  %res1 = call <2 x i64> @llvm.x86.avx512.mask.valign.q.128(<2 x i64> %x0, <2 x i64> %x1, i32 22, <2 x i64> %x3, i8 -1)
+  %res2 = add <2 x i64> %res, %res1
+  ret <2 x i64> %res2
+}
+
+declare <4 x i64> @llvm.x86.avx512.mask.valign.q.256(<4 x i64>, <4 x i64>, i32, <4 x i64>, i8)
+
+define <4 x i64>@test_int_x86_avx512_mask_valign_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x3, i8 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_valign_q_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    kmovw %eax, %k1
+; CHECK-NEXT:    valignq $22, %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT:    valignq $22, %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
+; CHECK-NEXT:    retq
+  %res = call <4 x i64> @llvm.x86.avx512.mask.valign.q.256(<4 x i64> %x0, <4 x i64> %x1, i32 22, <4 x i64> %x3, i8 %x4)
+  %res1 = call <4 x i64> @llvm.x86.avx512.mask.valign.q.256(<4 x i64> %x0, <4 x i64> %x1, i32 22, <4 x i64> %x3, i8 -1)
+  %res2 = add <4 x i64> %res, %res1
+  ret <4 x i64> %res2
+}
+