[X86][AVX512] add convert float to half

[AsafBadouh]

Summary

convert float to half with mask/maskz for the reg to reg version and mask for the reg to mem version (there is no maskz version for reg to mem).

made naming fix in GCC builtin for cvtph2ps in its 128bit width version (in order to be align with GCC)

[delena]

LGTM

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

def int_x86_vcvtps2ph_256 : GCCBuiltin<"__builtin_ia32_vcvtps2ph256">,
               Intrinsic<[llvm_v8i16_ty], [llvm_v8f32_ty, llvm_i32_ty],
                         [IntrNoMem]>;
   def int_x86_avx512_mask_vcvtph2ps_512 : GCCBuiltin<"__builtin_ia32_vcvtph2ps512_mask">,
               Intrinsic<[llvm_v16f32_ty], [llvm_v16i16_ty, llvm_v16f32_ty,
                                            llvm_i16_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_vcvtph2ps_256 : GCCBuiltin<"__builtin_ia32_vcvtph2ps256_mask">,
               Intrinsic<[llvm_v8f32_ty], [llvm_v8i16_ty, llvm_v8f32_ty,
                                            llvm_i8_ty], [IntrNoMem]>;
-  def int_x86_avx512_mask_vcvtph2ps_128 : GCCBuiltin<"__builtin_ia32_vcvtph2ps128_mask">,
+  def int_x86_avx512_mask_vcvtph2ps_128 : GCCBuiltin<"__builtin_ia32_vcvtph2ps_mask">,
               Intrinsic<[llvm_v4f32_ty], [llvm_v8i16_ty, llvm_v4f32_ty,
                                            llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_vcvtps2ph_512 : GCCBuiltin<"__builtin_ia32_vcvtps2ph512_mask">,
               Intrinsic<[llvm_v16i16_ty], [llvm_v16f32_ty, llvm_i32_ty,
                                            llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_vcvtps2ph_256 : GCCBuiltin<"__builtin_ia32_vcvtps2ph256_mask">,
+              Intrinsic<[llvm_v8i16_ty], [llvm_v8f32_ty, llvm_i32_ty,
+                                           llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_vcvtps2ph_128 : GCCBuiltin<"__builtin_ia32_vcvtps2ph_mask">,
+              Intrinsic<[llvm_v8i16_ty], [llvm_v4f32_ty, llvm_i32_ty,
+                                           llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
 }
 
 //===----------------------------------------------------------------------===//
 // TBM
 
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_tbm_bextri_u32 : GCCBuiltin<"__builtin_ia32_bextri_u32">,
         Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;

llvm/trunk/lib/Target/X86/X86InstrAVX512.td

multiclass avx512_cvtph2ps<X86VectorVTInfo _dest, X86VectorVTInfo _src,
   let hasSideEffects = 0, mayLoad = 1 in {
     defm rm : AVX512_maskable<0x13, MRMSrcMem, _dest, (outs _dest.RC:$dst), (ins x86memop:$src),
                       "vcvtph2ps", "$src", "$src", 
                       (X86cvtph2ps (_src.VT (bitconvert (ld_frag addr:$src))),
                                        (i32 FROUND_CURRENT))>, T8PD;
   }
 }
 
-multiclass avx512_cvtph2ps_sae<X86VectorVTInfo _dest, X86VectorVTInfo _src,
-                               X86MemOperand x86memop> {
+multiclass avx512_cvtph2ps_sae<X86VectorVTInfo _dest, X86VectorVTInfo _src> {
   defm rb : AVX512_maskable<0x13, MRMSrcReg, _dest ,(outs _dest.RC:$dst), (ins _src.RC:$src),
                     "vcvtph2ps", "{sae}, $src", "$src, {sae}",
                    (X86cvtph2ps (_src.VT _src.RC:$src),
                                                 (i32 FROUND_NO_EXC))>, T8PD, EVEX_B;
 
 }
 
 let Predicates = [HasAVX512] in {
   defm VCVTPH2PSZ : avx512_cvtph2ps<v16f32_info, v16i16x_info, f256mem, loadv4i64>,
-                    avx512_cvtph2ps_sae<v16f32_info, v16i16x_info, f256mem>, 
+                    avx512_cvtph2ps_sae<v16f32_info, v16i16x_info>, 
                     EVEX, EVEX_V512, EVEX_CD8<32, CD8VH>;
   let Predicates = [HasVLX] in {
     defm VCVTPH2PSZ256 : avx512_cvtph2ps<v8f32x_info, v8i16x_info, f128mem, 
                          loadv2i64>,EVEX, EVEX_V256, EVEX_CD8<32, CD8VH>;
     defm VCVTPH2PSZ128 : avx512_cvtph2ps<v4f32x_info, v8i16x_info, f64mem,
                          loadv2i64>, EVEX, EVEX_V128, EVEX_CD8<32, CD8VH>;
   }
 }
 
-multiclass avx512_cvtps2ph<RegisterClass destRC, RegisterClass srcRC,
-                             X86MemOperand x86memop> {
-  def rr : AVX512AIi8<0x1D, MRMDestReg, (outs destRC:$dst),
-               (ins srcRC:$src1, i32u8imm:$src2),
-               "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
-               []>, EVEX;
-  let hasSideEffects = 0, mayStore = 1 in
+multiclass avx512_cvtps2ph<X86VectorVTInfo _dest, X86VectorVTInfo _src, 
+  defm rr : AVX512_maskable<0x1D, MRMDestReg, _dest ,(outs _dest.RC:$dst),
+               (ins _src.RC:$src1, i32u8imm:$src2),
+                    "vcvtps2ph", "$src2, $src1", "$src1, $src2", 
+                   (X86cvtps2ph (_src.VT _src.RC:$src1),
+                                (i32 imm:$src2), 
+                                (i32 FROUND_CURRENT))>, AVX512AIi8Base;
+  let hasSideEffects = 0, mayStore = 1 in {
     def mr : AVX512AIi8<0x1D, MRMDestMem, (outs),
-               (ins x86memop:$dst, srcRC:$src1, i32u8imm:$src2),
-               "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, EVEX;
+               (ins x86memop:$dst, _src.RC:$src1, i32u8imm:$src2),
+               "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", 
+               [(store (_dest.VT (X86cvtps2ph (_src.VT _src.RC:$src1),
+                                       (i32 imm:$src2), (i32 FROUND_CURRENT) )),
+                                       addr:$dst)]>;
+    def mrk : AVX512AIi8<0x1D, MRMDestMem, (outs),
+               (ins x86memop:$dst, _dest.KRCWM:$mask, _src.RC:$src1, i32u8imm:$src2),
+               "vcvtps2ph\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}", 
+                []>, EVEX_K;
+  }
+}
+multiclass avx512_cvtps2ph_sae<X86VectorVTInfo _dest, X86VectorVTInfo _src> {
+  defm rb : AVX512_maskable<0x1D, MRMDestReg, _dest ,(outs _dest.RC:$dst),
+               (ins _src.RC:$src1, i32u8imm:$src2),
+                    "vcvtps2ph", "$src2, {sae}, $src1", "$src1, $src2, {sae}", 
+                   (X86cvtps2ph (_src.VT _src.RC:$src1),
+                                (i32 imm:$src2), 
+                                (i32 FROUND_NO_EXC))>, EVEX_B, AVX512AIi8Base;
+}
+let Predicates = [HasAVX512] in {
+  defm VCVTPS2PHZ : avx512_cvtps2ph<v16i16x_info, v16f32_info, f256mem>,
+                    avx512_cvtps2ph_sae<v16i16x_info, v16f32_info>,
+                      EVEX, EVEX_V512, EVEX_CD8<32, CD8VH>;
+  let Predicates = [HasVLX] in {
+    defm VCVTPS2PHZ256 : avx512_cvtps2ph<v8i16x_info, v8f32x_info, f128mem>,
+                        EVEX, EVEX_V256, EVEX_CD8<32, CD8VH>;
+    defm VCVTPS2PHZ128 : avx512_cvtps2ph<v8i16x_info, v4f32x_info, f128mem>,
+                        EVEX, EVEX_V128, EVEX_CD8<32, CD8VH>;
+  }
 }
-
-defm VCVTPS2PHZ : avx512_cvtps2ph<VR256X, VR512, f256mem>, EVEX_V512,
-                                    EVEX_CD8<32, CD8VH>;
-
-def : Pat<(v16i16 (int_x86_avx512_mask_vcvtps2ph_512 (v16f32 VR512:$src),
-           imm:$rc, (bc_v16i16(v8i32 immAllZerosV)), (i16 -1))),
-           (VCVTPS2PHZrr VR512:$src, imm:$rc)>;
-
 let Defs = [EFLAGS], Predicates = [HasAVX512] in {
   defm VUCOMISSZ : sse12_ord_cmp<0x2E, FR32X, X86cmp, f32, f32mem, loadf32,
                                  "ucomiss">, PS, EVEX, VEX_LIG,
                                  EVEX_CD8<32, CD8VT1>;
   defm VUCOMISDZ : sse12_ord_cmp<0x2E, FR64X, X86cmp, f64, f64mem, loadf64,
                                   "ucomisd">, PD, EVEX,
                                   VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>;
   let Pattern = []<dag> in {

llvm/trunk/lib/Target/X86/X86InstrFragmentsSIMD.td

 def X86cvtpd2UInt     : SDNode<"X86ISD::FP_TO_UINT_RND",  SDTDoubleToInt>;
 
 def X86cvtph2ps     : SDNode<"ISD::FP16_TO_FP",
                               SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
                                                    SDTCVecEltisVT<0, f32>,
                                                    SDTCVecEltisVT<1, i16>,
                                                    SDTCisFP<0>, SDTCisInt<2>]> >;
 
+def X86cvtps2ph   : SDNode<"ISD::FP_TO_FP16",
+                        SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisVec<1>,
+                                             SDTCVecEltisVT<0, i16>,
+                                             SDTCVecEltisVT<1, f32>,
+                                             SDTCisFP<1>, SDTCisInt<2>, SDTCisInt<3>]> >;
 def X86vfpextRnd  : SDNode<"X86ISD::VFPEXT",
                         SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
                                              SDTCisFP<0>, SDTCisFP<1>,
                                              SDTCisOpSmallerThanOp<1, 0>,
                                              SDTCisInt<2>]>>;
 def X86vfproundRnd: SDNode<"X86ISD::VFPROUND",
                         SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
                                              SDTCisFP<0>, SDTCisFP<1>,

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

   X86_INTRINSIC_DATA(avx512_mask_valign_q_512, INTR_TYPE_3OP_IMM8_MASK,
                      X86ISD::VALIGN, 0),
   X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_128, INTR_TYPE_1OP_MASK_RM,
                      ISD::FP16_TO_FP, 0),
   X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_256, INTR_TYPE_1OP_MASK_RM,
                      ISD::FP16_TO_FP, 0),
   X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_512, INTR_TYPE_1OP_MASK_RM,
                      ISD::FP16_TO_FP, 0),
+  X86_INTRINSIC_DATA(avx512_mask_vcvtps2ph_128, INTR_TYPE_2OP_MASK_RM, 
+                     ISD::FP_TO_FP16, 0),
+  X86_INTRINSIC_DATA(avx512_mask_vcvtps2ph_256, INTR_TYPE_2OP_MASK_RM, 
+                     ISD::FP_TO_FP16, 0),
+  X86_INTRINSIC_DATA(avx512_mask_vcvtps2ph_512, INTR_TYPE_2OP_MASK_RM, 
+                     ISD::FP_TO_FP16, 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),
   X86_INTRINSIC_DATA(avx512_mask_vfmadd_ps_512, FMA_OP_MASK, X86ISD::FMADD,
                      X86ISD::FMADD_RND),

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

 define <8 x float> @test_x86_vcvtph2ps_256_rrkz(<8 x i16> %a0, i8 %mask) {
   ; CHECK: test_x86_vcvtph2ps_256_rrkz
   ; CHECK: vcvtph2ps  %xmm0, %ymm0 {%k1} {z}
   %res = call <8 x float> @llvm.x86.avx512.mask.vcvtph2ps.256(<8 x i16> %a0, <8 x float> zeroinitializer, i8 %mask)
   ret <8 x float> %res
 }
 
 declare <8 x float> @llvm.x86.avx512.mask.vcvtph2ps.256(<8 x i16>, <8 x float>, i8) nounwind readonly
+
+define <8 x i16> @test_x86_vcvtps2ph_128(<4 x float> %a0) {
+  ; CHECK: test_x86_vcvtps2ph_128
+  ; CHECK: vcvtps2ph $2, %xmm0, %xmm0
+  %res = call <8 x i16> @llvm.x86.avx512.mask.vcvtps2ph.128(<4 x float> %a0, i32 2, <8 x i16> zeroinitializer, i8 -1)
+  ret <8 x i16> %res
+}
+
+
+declare <8 x i16> @llvm.x86.avx512.mask.vcvtps2ph.128(<4 x float>, i32, <8 x i16>, i8) nounwind readonly
+
+define <8 x i16> @test_x86_vcvtps2ph_256(<8 x float> %a0) {
+  ; CHECK: test_x86_vcvtps2ph_256
+  ; CHECK: vcvtps2ph $2, %ymm0, %xmm0
+  %res = call <8 x i16> @llvm.x86.avx512.mask.vcvtps2ph.256(<8 x float> %a0, i32 2, <8 x i16> zeroinitializer, i8 -1)
+  ret <8 x i16> %res
+}
+
+declare <8 x i16> @llvm.x86.avx512.mask.vcvtps2ph.256(<8 x float>, i32, <8 x i16>, i8) nounwind readonly

llvm/trunk/test/MC/X86/avx512-encodings.s

 
 // CHECK: vcvtph2ps -4096(%rdx), %zmm13
 // CHECK:  encoding: [0x62,0x72,0x7d,0x48,0x13,0x6a,0x80]
           vcvtph2ps -4096(%rdx), %zmm13
 
 // CHECK: vcvtph2ps -4128(%rdx), %zmm13
 // CHECK:  encoding: [0x62,0x72,0x7d,0x48,0x13,0xaa,0xe0,0xef,0xff,0xff]
           vcvtph2ps -4128(%rdx), %zmm13
+
+// CHECK: vcvtps2ph $171, %zmm14, %ymm11
+// CHECK:  encoding: [0x62,0x53,0x7d,0x48,0x1d,0xf3,0xab]
+          vcvtps2ph $0xab, %zmm14, %ymm11
+
+// CHECK: vcvtps2ph $171, %zmm14, %ymm11 {%k6}
+// CHECK:  encoding: [0x62,0x53,0x7d,0x4e,0x1d,0xf3,0xab]
+          vcvtps2ph $0xab, %zmm14, %ymm11 {%k6}
+
+// CHECK: vcvtps2ph $171, %zmm14, %ymm11 {%k6} {z}
+// CHECK:  encoding: [0x62,0x53,0x7d,0xce,0x1d,0xf3,0xab]
+          vcvtps2ph $0xab, %zmm14, %ymm11 {%k6} {z}
+
+// CHECK: vcvtps2ph $171, {sae}, %zmm14, %ymm11
+// CHECK:  encoding: [0x62,0x53,0x7d,0x18,0x1d,0xf3,0xab]
+          vcvtps2ph $0xab,{sae}, %zmm14, %ymm11
+
+// CHECK: vcvtps2ph $123, %zmm14, %ymm11
+// CHECK:  encoding: [0x62,0x53,0x7d,0x48,0x1d,0xf3,0x7b]
+          vcvtps2ph $0x7b, %zmm14, %ymm11
+
+// CHECK: vcvtps2ph $123, {sae}, %zmm14, %ymm11
+// CHECK:  encoding: [0x62,0x53,0x7d,0x18,0x1d,0xf3,0x7b]
+          vcvtps2ph $0x7b,{sae}, %zmm14, %ymm11
+
+// CHECK: vcvtps2ph $171, %zmm19, (%rcx)
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x48,0x1d,0x19,0xab]
+          vcvtps2ph $0xab, %zmm19, (%rcx)
+
+// CHECK: vcvtps2ph $171, %zmm19, (%rcx) {%k6}
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x4e,0x1d,0x19,0xab]
+          vcvtps2ph $0xab, %zmm19, (%rcx) {%k6}
+
+// CHECK: vcvtps2ph $123, %zmm19, (%rcx)
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x48,0x1d,0x19,0x7b]
+          vcvtps2ph $0x7b, %zmm19, (%rcx)
+
+// CHECK: vcvtps2ph $123, %zmm19, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0xa3,0x7d,0x48,0x1d,0x9c,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vcvtps2ph $0x7b, %zmm19, 291(%rax,%r14,8)
+
+// CHECK: vcvtps2ph $123, %zmm19, 4064(%rdx)
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x48,0x1d,0x5a,0x7f,0x7b]
+          vcvtps2ph $0x7b, %zmm19, 4064(%rdx)
+
+// CHECK: vcvtps2ph $123, %zmm19, 4096(%rdx)
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x48,0x1d,0x9a,0x00,0x10,0x00,0x00,0x7b]
+          vcvtps2ph $0x7b, %zmm19, 4096(%rdx)
+
+// CHECK: vcvtps2ph $123, %zmm19, -4096(%rdx)
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x48,0x1d,0x5a,0x80,0x7b]
+          vcvtps2ph $0x7b, %zmm19, -4096(%rdx)
+
+// CHECK: vcvtps2ph $123, %zmm19, -4128(%rdx)
+// CHECK:  encoding: [0x62,0xe3,0x7d,0x48,0x1d,0x9a,0xe0,0xef,0xff,0xff,0x7b]
+          vcvtps2ph $0x7b, %zmm19, -4128(%rdx)
 No newline at end of file

llvm/trunk/test/MC/X86/x86-64-avx512f_vl.s

 
 // CHECK: vcvtph2ps -2048(%rdx), %ymm30
 // CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x13,0x72,0x80]
           vcvtph2ps -2048(%rdx), %ymm30
 
 // CHECK: vcvtph2ps -2064(%rdx), %ymm30
 // CHECK:  encoding: [0x62,0x62,0x7d,0x28,0x13,0xb2,0xf0,0xf7,0xff,0xff]
           vcvtph2ps -2064(%rdx), %ymm30
+
+// CHECK: vcvtps2ph $171, %xmm25, %xmm21
+// CHECK:  encoding: [0x62,0x23,0x7d,0x08,0x1d,0xcd,0xab]
+          vcvtps2ph $0xab, %xmm25, %xmm21
+
+// CHECK: vcvtps2ph $171, %xmm25, %xmm21 {%k5}
+// CHECK:  encoding: [0x62,0x23,0x7d,0x0d,0x1d,0xcd,0xab]
+          vcvtps2ph $0xab, %xmm25, %xmm21 {%k5}
+
+// CHECK: vcvtps2ph $171, %xmm25, %xmm21 {%k5} {z}
+// CHECK:  encoding: [0x62,0x23,0x7d,0x8d,0x1d,0xcd,0xab]
+          vcvtps2ph $0xab, %xmm25, %xmm21 {%k5} {z}
+
+// CHECK: vcvtps2ph $123, %xmm25, %xmm21
+// CHECK:  encoding: [0x62,0x23,0x7d,0x08,0x1d,0xcd,0x7b]
+          vcvtps2ph $0x7b, %xmm25, %xmm21
+
+// CHECK: vcvtps2ph $171, %ymm28, %xmm23
+// CHECK:  encoding: [0x62,0x23,0x7d,0x28,0x1d,0xe7,0xab]
+          vcvtps2ph $0xab, %ymm28, %xmm23
+
+// CHECK: vcvtps2ph $171, %ymm28, %xmm23 {%k3}
+// CHECK:  encoding: [0x62,0x23,0x7d,0x2b,0x1d,0xe7,0xab]
+          vcvtps2ph $0xab, %ymm28, %xmm23 {%k3}
+
+// CHECK: vcvtps2ph $171, %ymm28, %xmm23 {%k3} {z}
+// CHECK:  encoding: [0x62,0x23,0x7d,0xab,0x1d,0xe7,0xab]
+          vcvtps2ph $0xab, %ymm28, %xmm23 {%k3} {z}
+
+// CHECK: vcvtps2ph $123, %ymm28, %xmm23
+// CHECK:  encoding: [0x62,0x23,0x7d,0x28,0x1d,0xe7,0x7b]
+          vcvtps2ph $0x7b, %ymm28, %xmm23
+
+// CHECK: vcvtps2ph $171, %xmm27, (%rcx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x08,0x1d,0x19,0xab]
+          vcvtps2ph $0xab, %xmm27, (%rcx)
+
+// CHECK: vcvtps2ph $171, %xmm27, (%rcx) {%k4}
+// CHECK:  encoding: [0x62,0x63,0x7d,0x0c,0x1d,0x19,0xab]
+          vcvtps2ph $0xab, %xmm27, (%rcx) {%k4}
+
+// CHECK: vcvtps2ph $123, %xmm27, (%rcx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x08,0x1d,0x19,0x7b]
+          vcvtps2ph $0x7b, %xmm27, (%rcx)
+
+// CHECK: vcvtps2ph $123, %xmm27, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0x23,0x7d,0x08,0x1d,0x9c,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vcvtps2ph $0x7b, %xmm27, 291(%rax,%r14,8)
+
+// CHECK: vcvtps2ph $123, %xmm27, 1016(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x08,0x1d,0x5a,0x7f,0x7b]
+          vcvtps2ph $0x7b, %xmm27, 1016(%rdx)
+
+// CHECK: vcvtps2ph $123, %xmm27, 1024(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x08,0x1d,0x9a,0x00,0x04,0x00,0x00,0x7b]
+          vcvtps2ph $0x7b, %xmm27, 1024(%rdx)
+
+// CHECK: vcvtps2ph $123, %xmm27, -1024(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x08,0x1d,0x5a,0x80,0x7b]
+          vcvtps2ph $0x7b, %xmm27, -1024(%rdx)
+
+// CHECK: vcvtps2ph $123, %xmm27, -1032(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x08,0x1d,0x9a,0xf8,0xfb,0xff,0xff,0x7b]
+          vcvtps2ph $0x7b, %xmm27, -1032(%rdx)
+
+// CHECK: vcvtps2ph $171, %ymm30, (%rcx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0x31,0xab]
+          vcvtps2ph $0xab, %ymm30, (%rcx)
+
+// CHECK: vcvtps2ph $171, %ymm30, (%rcx) {%k2}
+// CHECK:  encoding: [0x62,0x63,0x7d,0x2a,0x1d,0x31,0xab]
+          vcvtps2ph $0xab, %ymm30, (%rcx) {%k2}
+
+// CHECK: vcvtps2ph $123, %ymm30, (%rcx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0x31,0x7b]
+          vcvtps2ph $0x7b, %ymm30, (%rcx)
+
+// CHECK: vcvtps2ph $123, %ymm30, 291(%rax,%r14,8)
+// CHECK:  encoding: [0x62,0x23,0x7d,0x28,0x1d,0xb4,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vcvtps2ph $0x7b, %ymm30, 291(%rax,%r14,8)
+
+// CHECK: vcvtps2ph $123, %ymm30, 2032(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0x72,0x7f,0x7b]
+          vcvtps2ph $0x7b, %ymm30, 2032(%rdx)
+
+// CHECK: vcvtps2ph $123, %ymm30, 2048(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0xb2,0x00,0x08,0x00,0x00,0x7b]
+          vcvtps2ph $0x7b, %ymm30, 2048(%rdx)
+
+// CHECK: vcvtps2ph $123, %ymm30, -2048(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0x72,0x80,0x7b]
+          vcvtps2ph $0x7b, %ymm30, -2048(%rdx)
+
+// CHECK: vcvtps2ph $123, %ymm30, -2064(%rdx)
+// CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0xb2,0xf0,0xf7,0xff,0xff,0x7b]
+          vcvtps2ph $0x7b, %ymm30, -2064(%rdx)