summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--llvm/include/llvm/IR/IntrinsicsX86.td24
-rw-r--r--llvm/lib/IR/AutoUpgrade.cpp18
-rw-r--r--llvm/lib/Target/X86/X86IntrinsicsInfo.h8
-rw-r--r--llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp46
-rw-r--r--llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll396
-rw-r--r--llvm/test/CodeGen/X86/avx512bw-intrinsics.ll100
6 files changed, 485 insertions, 107 deletions
diff --git a/llvm/include/llvm/IR/IntrinsicsX86.td b/llvm/include/llvm/IR/IntrinsicsX86.td
index 71321ff612b..9d723a710d1 100644
--- a/llvm/include/llvm/IR/IntrinsicsX86.td
+++ b/llvm/include/llvm/IR/IntrinsicsX86.td
@@ -3653,18 +3653,18 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
}
// Integer arithmetic ops
let TargetPrefix = "x86" in {
- def int_x86_avx512_mask_padds_b_512 : GCCBuiltin<"__builtin_ia32_paddsb512_mask">,
- Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty,
- llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>;
- def int_x86_avx512_mask_padds_w_512 : GCCBuiltin<"__builtin_ia32_paddsw512_mask">,
- Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty,
- llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>;
- def int_x86_avx512_mask_psubs_b_512 : GCCBuiltin<"__builtin_ia32_psubsb512_mask">,
- Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty,
- llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>;
- def int_x86_avx512_mask_psubs_w_512 : GCCBuiltin<"__builtin_ia32_psubsw512_mask">,
- Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty,
- llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>;
+ def int_x86_avx512_padds_b_512 : GCCBuiltin<"__builtin_ia32_paddsb512">,
+ Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty],
+ [IntrNoMem]>;
+ def int_x86_avx512_padds_w_512 : GCCBuiltin<"__builtin_ia32_paddsw512">,
+ Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty],
+ [IntrNoMem]>;
+ def int_x86_avx512_psubs_b_512 : GCCBuiltin<"__builtin_ia32_psubsb512">,
+ Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty],
+ [IntrNoMem]>;
+ def int_x86_avx512_psubs_w_512 : GCCBuiltin<"__builtin_ia32_psubsw512">,
+ Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty],
+ [IntrNoMem]>;
def int_x86_avx512_pmulhu_w_512 : GCCBuiltin<"__builtin_ia32_pmulhuw512">,
Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty,
llvm_v32i16_ty], [IntrNoMem, Commutative]>;
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index ae88a285a0e..1864ad29b30 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -282,14 +282,8 @@ static bool ShouldUpgradeX86Intrinsic(Function *F, StringRef Name) {
Name.startswith("avx512.mask.pror.") || // Added in 7.0
Name.startswith("avx512.mask.prolv.") || // Added in 7.0
Name.startswith("avx512.mask.prol.") || // Added in 7.0
- Name == "avx512.mask.padds.b.128" || // Added in 8.0
- Name == "avx512.mask.padds.b.256" || // Added in 8.0
- Name == "avx512.mask.padds.w.128" || // Added in 8.0
- Name == "avx512.mask.padds.w.256" || // Added in 8.0
- Name == "avx512.mask.psubs.b.128" || // Added in 8.0
- Name == "avx512.mask.psubs.b.256" || // Added in 8.0
- Name == "avx512.mask.psubs.w.128" || // Added in 8.0
- Name == "avx512.mask.psubs.w.256" || // Added in 8.0
+ Name.startswith("avx512.mask.padds.") || // Added in 8.0
+ Name.startswith("avx512.mask.psubs.") || // Added in 8.0
Name == "sse.cvtsi2ss" || // Added in 7.0
Name == "sse.cvtsi642ss" || // Added in 7.0
Name == "sse2.cvtsi2sd" || // Added in 7.0
@@ -1417,10 +1411,14 @@ static bool upgradeAVX512MaskToSelect(StringRef Name, IRBuilder<> &Builder,
IID = Intrinsic::x86_sse2_padds_b;
else if (VecWidth == 256 && EltWidth == 8)
IID = Intrinsic::x86_avx2_padds_b;
+ else if (VecWidth == 512 && EltWidth == 8)
+ IID = Intrinsic::x86_avx512_padds_b_512;
else if (VecWidth == 128 && EltWidth == 16)
IID = Intrinsic::x86_sse2_padds_w;
else if (VecWidth == 256 && EltWidth == 16)
IID = Intrinsic::x86_avx2_padds_w;
+ else if (VecWidth == 512 && EltWidth == 16)
+ IID = Intrinsic::x86_avx512_padds_w_512;
else
llvm_unreachable("Unexpected intrinsic");
} else if (Name.startswith("psubs.")) {
@@ -1428,10 +1426,14 @@ static bool upgradeAVX512MaskToSelect(StringRef Name, IRBuilder<> &Builder,
IID = Intrinsic::x86_sse2_psubs_b;
else if (VecWidth == 256 && EltWidth == 8)
IID = Intrinsic::x86_avx2_psubs_b;
+ else if (VecWidth == 512 && EltWidth == 8)
+ IID = Intrinsic::x86_avx512_psubs_b_512;
else if (VecWidth == 128 && EltWidth == 16)
IID = Intrinsic::x86_sse2_psubs_w;
else if (VecWidth == 256 && EltWidth == 16)
IID = Intrinsic::x86_avx2_psubs_w;
+ else if (VecWidth == 512 && EltWidth == 16)
+ IID = Intrinsic::x86_avx512_psubs_w_512;
else
llvm_unreachable("Unexpected intrinsic");
} else
diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
index 95f17f76298..a9ee3182aaa 100644
--- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
@@ -673,8 +673,6 @@ static const IntrinsicData IntrinsicsWithoutChain[] = {
X86ISD::FMULS_RND, 0),
X86_INTRINSIC_DATA(avx512_mask_mul_ss_round, INTR_TYPE_SCALAR_MASK_RM,
X86ISD::FMULS_RND, 0),
- X86_INTRINSIC_DATA(avx512_mask_padds_b_512, INTR_TYPE_2OP_MASK, X86ISD::ADDS, 0),
- X86_INTRINSIC_DATA(avx512_mask_padds_w_512, INTR_TYPE_2OP_MASK, X86ISD::ADDS, 0),
X86_INTRINSIC_DATA(avx512_mask_pmov_db_128, INTR_TYPE_1OP_MASK,
X86ISD::VTRUNC, 0),
X86_INTRINSIC_DATA(avx512_mask_pmov_db_256, INTR_TYPE_1OP_MASK,
@@ -789,8 +787,6 @@ static const IntrinsicData IntrinsicsWithoutChain[] = {
X86ISD::MULTISHIFT, 0),
X86_INTRINSIC_DATA(avx512_mask_pmultishift_qb_512, INTR_TYPE_2OP_MASK,
X86ISD::MULTISHIFT, 0),
- X86_INTRINSIC_DATA(avx512_mask_psubs_b_512, INTR_TYPE_2OP_MASK, X86ISD::SUBS, 0),
- X86_INTRINSIC_DATA(avx512_mask_psubs_w_512, INTR_TYPE_2OP_MASK, X86ISD::SUBS, 0),
X86_INTRINSIC_DATA(avx512_mask_range_pd_128, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, 0),
X86_INTRINSIC_DATA(avx512_mask_range_pd_256, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, 0),
X86_INTRINSIC_DATA(avx512_mask_range_pd_512, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, X86ISD::VRANGE_RND),
@@ -926,6 +922,8 @@ static const IntrinsicData IntrinsicsWithoutChain[] = {
X86_INTRINSIC_DATA(avx512_packsswb_512, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
X86_INTRINSIC_DATA(avx512_packusdw_512, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
X86_INTRINSIC_DATA(avx512_packuswb_512, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
+ X86_INTRINSIC_DATA(avx512_padds_b_512, INTR_TYPE_2OP, X86ISD::ADDS, 0),
+ X86_INTRINSIC_DATA(avx512_padds_w_512, INTR_TYPE_2OP, X86ISD::ADDS, 0),
X86_INTRINSIC_DATA(avx512_permvar_df_256, VPERM_2OP, X86ISD::VPERMV, 0),
X86_INTRINSIC_DATA(avx512_permvar_df_512, VPERM_2OP, X86ISD::VPERMV, 0),
X86_INTRINSIC_DATA(avx512_permvar_di_256, VPERM_2OP, X86ISD::VPERMV, 0),
@@ -1008,6 +1006,8 @@ static const IntrinsicData IntrinsicsWithoutChain[] = {
X86_INTRINSIC_DATA(avx512_psrlv_w_128, INTR_TYPE_2OP, ISD::SRL, 0),
X86_INTRINSIC_DATA(avx512_psrlv_w_256, INTR_TYPE_2OP, ISD::SRL, 0),
X86_INTRINSIC_DATA(avx512_psrlv_w_512, INTR_TYPE_2OP, ISD::SRL, 0),
+ X86_INTRINSIC_DATA(avx512_psubs_b_512, INTR_TYPE_2OP, X86ISD::SUBS, 0),
+ X86_INTRINSIC_DATA(avx512_psubs_w_512, INTR_TYPE_2OP, X86ISD::SUBS, 0),
X86_INTRINSIC_DATA(avx512_pternlog_d_128, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
X86_INTRINSIC_DATA(avx512_pternlog_d_256, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
X86_INTRINSIC_DATA(avx512_pternlog_d_512, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 7773c3d234e..1d0cd11f941 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -245,8 +245,7 @@ Instruction *InstCombiner::SimplifyAnyMemSet(AnyMemSetInst *MI) {
static Value *simplifyX86AddsSubs(const IntrinsicInst &II,
InstCombiner::BuilderTy &Builder) {
- bool IsAddition = false;
- bool IsMasked = false;
+ bool IsAddition;
switch (II.getIntrinsicID()) {
default: llvm_unreachable("Unexpected intrinsic!");
@@ -254,21 +253,17 @@ static Value *simplifyX86AddsSubs(const IntrinsicInst &II,
case Intrinsic::x86_sse2_padds_w:
case Intrinsic::x86_avx2_padds_b:
case Intrinsic::x86_avx2_padds_w:
- IsAddition = true; IsMasked = false;
+ case Intrinsic::x86_avx512_padds_b_512:
+ case Intrinsic::x86_avx512_padds_w_512:
+ IsAddition = true;
break;
case Intrinsic::x86_sse2_psubs_b:
case Intrinsic::x86_sse2_psubs_w:
case Intrinsic::x86_avx2_psubs_b:
case Intrinsic::x86_avx2_psubs_w:
- IsAddition = false; IsMasked = false;
- break;
- case Intrinsic::x86_avx512_mask_padds_b_512:
- case Intrinsic::x86_avx512_mask_padds_w_512:
- IsAddition = true; IsMasked = true;
- break;
- case Intrinsic::x86_avx512_mask_psubs_b_512:
- case Intrinsic::x86_avx512_mask_psubs_w_512:
- IsAddition = false; IsMasked = true;
+ case Intrinsic::x86_avx512_psubs_b_512:
+ case Intrinsic::x86_avx512_psubs_w_512:
+ IsAddition = false;
break;
}
@@ -278,7 +273,7 @@ static Value *simplifyX86AddsSubs(const IntrinsicInst &II,
auto SVT = VT->getElementType();
unsigned NumElems = VT->getNumElements();
- if (!Arg0 || !Arg1 || (IsMasked && !isa<Constant>(II.getOperand(2))))
+ if (!Arg0 || !Arg1)
return nullptr;
SmallVector<Constant *, 64> Result;
@@ -306,21 +301,7 @@ static Value *simplifyX86AddsSubs(const IntrinsicInst &II,
Result.push_back(Constant::getIntegerValue(SVT, ResultElem));
}
- Value *ResultVec = ConstantVector::get(Result);
-
- if (II.getNumArgOperands() == 4) { // For masked intrinsics.
- Value *Src = II.getOperand(2);
- auto Mask = II.getOperand(3);
- if (auto *C = dyn_cast<Constant>(Mask))
- if (C->isAllOnesValue())
- return ResultVec;
- auto *MaskTy = VectorType::get(
- Builder.getInt1Ty(), cast<IntegerType>(Mask->getType())->getBitWidth());
- Mask = Builder.CreateBitCast(Mask, MaskTy);
- ResultVec = Builder.CreateSelect(Mask, ResultVec, Src);
- }
-
- return ResultVec;
+ return ConstantVector::get(Result);
}
static Value *simplifyX86immShift(const IntrinsicInst &II,
@@ -2614,14 +2595,13 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
case Intrinsic::x86_avx2_padds_w:
case Intrinsic::x86_avx2_psubs_b:
case Intrinsic::x86_avx2_psubs_w:
- case Intrinsic::x86_avx512_mask_padds_b_512:
- case Intrinsic::x86_avx512_mask_padds_w_512:
- case Intrinsic::x86_avx512_mask_psubs_b_512:
- case Intrinsic::x86_avx512_mask_psubs_w_512:
+ case Intrinsic::x86_avx512_padds_b_512:
+ case Intrinsic::x86_avx512_padds_w_512:
+ case Intrinsic::x86_avx512_psubs_b_512:
+ case Intrinsic::x86_avx512_psubs_w_512:
if (Value *V = simplifyX86AddsSubs(*II, Builder))
return replaceInstUsesWith(*II, V);
break;
-
// Constant fold ashr( <A x Bi>, Ci ).
// Constant fold lshr( <A x Bi>, Ci ).
diff --git a/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
index 31a557aa4ae..c623d8bbf2d 100644
--- a/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
@@ -3284,3 +3284,399 @@ define <64 x i8> @test_mask_subs_epu8_rmkz_512(<64 x i8> %a, <64 x i8>* %ptr_b,
}
declare <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <32 x i16> @test_mask_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
+; CHECK-LABEL: test_mask_adds_epi16_rr_512:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vpaddsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0xc1]
+; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
+ %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_mask_adds_epi16_rrk_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
+; X86-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi16_rrk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT: vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
+; X64-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
+; X86-LABEL: test_mask_adds_epi16_rrkz_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi16_rrkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT: vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
+; X86-LABEL: test_mask_adds_epi16_rm_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: vpaddsw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi16_rm_512:
+; X64: # %bb.0:
+; X64-NEXT: vpaddsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <32 x i16>, <32 x i16>* %ptr_b
+ %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_mask_adds_epi16_rmk_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpaddsw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0x08]
+; X86-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi16_rmk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT: vpaddsw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0x0f]
+; X64-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <32 x i16>, <32 x i16>* %ptr_b
+ %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_adds_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
+; X86-LABEL: test_mask_adds_epi16_rmkz_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpaddsw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi16_rmkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT: vpaddsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <32 x i16>, <32 x i16>* %ptr_b
+ %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+ ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <32 x i16> @test_mask_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
+; CHECK-LABEL: test_mask_subs_epi16_rr_512:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vpsubsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0xc1]
+; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
+ %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_mask_subs_epi16_rrk_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
+; X86-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi16_rrk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT: vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
+; X64-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
+; X86-LABEL: test_mask_subs_epi16_rrkz_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi16_rrkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT: vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
+; X86-LABEL: test_mask_subs_epi16_rm_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: vpsubsw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi16_rm_512:
+; X64: # %bb.0:
+; X64-NEXT: vpsubsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <32 x i16>, <32 x i16>* %ptr_b
+ %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_mask_subs_epi16_rmk_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpsubsw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0x08]
+; X86-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi16_rmk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT: vpsubsw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0x0f]
+; X64-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <32 x i16>, <32 x i16>* %ptr_b
+ %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
+ ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
+; X86-LABEL: test_mask_subs_epi16_rmkz_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpsubsw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi16_rmkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT: vpsubsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <32 x i16>, <32 x i16>* %ptr_b
+ %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
+ ret <32 x i16> %res
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+define <64 x i8> @test_mask_adds_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) {
+; CHECK-LABEL: test_mask_adds_epi8_rr_512:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vpaddsb %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xec,0xc1]
+; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
+ %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) {
+; X86-LABEL: test_mask_adds_epi8_rrk_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpaddsb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xec,0xd1]
+; X86-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi8_rrk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
+; X64-NEXT: vpaddsb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xec,0xd1]
+; X64-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) {
+; X86-LABEL: test_mask_adds_epi8_rrkz_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpaddsb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xec,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi8_rrkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
+; X64-NEXT: vpaddsb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xec,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rm_512(<64 x i8> %a, <64 x i8>* %ptr_b) {
+; X86-LABEL: test_mask_adds_epi8_rm_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: vpaddsb (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xec,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi8_rm_512:
+; X64: # %bb.0:
+; X64-NEXT: vpaddsb (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xec,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <64 x i8>, <64 x i8>* %ptr_b
+ %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rmk_512(<64 x i8> %a, <64 x i8>* %ptr_b, <64 x i8> %passThru, i64 %mask) {
+; X86-LABEL: test_mask_adds_epi8_rmk_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpaddsb (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xec,0x08]
+; X86-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi8_rmk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
+; X64-NEXT: vpaddsb (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xec,0x0f]
+; X64-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <64 x i8>, <64 x i8>* %ptr_b
+ %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_adds_epi8_rmkz_512(<64 x i8> %a, <64 x i8>* %ptr_b, i64 %mask) {
+; X86-LABEL: test_mask_adds_epi8_rmkz_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpaddsb (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xec,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_adds_epi8_rmkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
+; X64-NEXT: vpaddsb (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xec,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <64 x i8>, <64 x i8>* %ptr_b
+ %res = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+ ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+define <64 x i8> @test_mask_subs_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) {
+; CHECK-LABEL: test_mask_subs_epi8_rr_512:
+; CHECK: # %bb.0:
+; CHECK-NEXT: vpsubsb %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe8,0xc1]
+; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
+ %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) {
+; X86-LABEL: test_mask_subs_epi8_rrk_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpsubsb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe8,0xd1]
+; X86-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi8_rrk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
+; X64-NEXT: vpsubsb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe8,0xd1]
+; X64-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) {
+; X86-LABEL: test_mask_subs_epi8_rrkz_512:
+; X86: # %bb.0:
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT: vpsubsb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe8,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi8_rrkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
+; X64-NEXT: vpsubsb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe8,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rm_512(<64 x i8> %a, <64 x i8>* %ptr_b) {
+; X86-LABEL: test_mask_subs_epi8_rm_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: vpsubsb (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe8,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi8_rm_512:
+; X64: # %bb.0:
+; X64-NEXT: vpsubsb (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe8,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <64 x i8>, <64 x i8>* %ptr_b
+ %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 -1)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rmk_512(<64 x i8> %a, <64 x i8>* %ptr_b, <64 x i8> %passThru, i64 %mask) {
+; X86-LABEL: test_mask_subs_epi8_rmk_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpsubsb (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe8,0x08]
+; X86-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi8_rmk_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
+; X64-NEXT: vpsubsb (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe8,0x0f]
+; X64-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <64 x i8>, <64 x i8>* %ptr_b
+ %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask)
+ ret <64 x i8> %res
+}
+
+define <64 x i8> @test_mask_subs_epi8_rmkz_512(<64 x i8> %a, <64 x i8>* %ptr_b, i64 %mask) {
+; X86-LABEL: test_mask_subs_epi8_rmkz_512:
+; X86: # %bb.0:
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT: vpsubsb (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe8,0x00]
+; X86-NEXT: retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_subs_epi8_rmkz_512:
+; X64: # %bb.0:
+; X64-NEXT: kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
+; X64-NEXT: vpsubsb (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe8,0x07]
+; X64-NEXT: retq # encoding: [0xc3]
+ %b = load <64 x i8>, <64 x i8>* %ptr_b
+ %res = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
+ ret <64 x i8> %res
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
diff --git a/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
index 4e59ade192c..87e2e7c037e 100644
--- a/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bw-intrinsics.ll
@@ -564,8 +564,8 @@ define <32 x i16> @test_mask_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
; CHECK: # %bb.0:
; CHECK-NEXT: vpaddsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0xc1]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
- %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+ ret <32 x i16> %1
}
define <32 x i16> @test_mask_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
@@ -582,8 +582,10 @@ define <32 x i16> @test_mask_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <3
; X64-NEXT: vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
; X64-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT: retq # encoding: [0xc3]
- %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+ ret <32 x i16> %3
}
define <32 x i16> @test_mask_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
@@ -598,8 +600,10 @@ define <32 x i16> @test_mask_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
- %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+ ret <32 x i16> %3
}
define <32 x i16> @test_mask_adds_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
@@ -614,8 +618,8 @@ define <32 x i16> @test_mask_adds_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b
; X64-NEXT: vpaddsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%b = load <32 x i16>, <32 x i16>* %ptr_b
- %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+ ret <32 x i16> %1
}
define <32 x i16> @test_mask_adds_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
@@ -634,8 +638,10 @@ define <32 x i16> @test_mask_adds_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_
; X64-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
%b = load <32 x i16>, <32 x i16>* %ptr_b
- %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+ ret <32 x i16> %3
}
define <32 x i16> @test_mask_adds_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
@@ -652,19 +658,21 @@ define <32 x i16> @test_mask_adds_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr
; X64-NEXT: vpaddsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%b = load <32 x i16>, <32 x i16>* %ptr_b
- %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+ ret <32 x i16> %3
}
-declare <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+declare <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16>, <32 x i16>)
define <32 x i16> @test_mask_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
; CHECK-LABEL: test_mask_subs_epi16_rr_512:
; CHECK: # %bb.0:
; CHECK-NEXT: vpsubsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0xc1]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
- %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+ ret <32 x i16> %1
}
define <32 x i16> @test_mask_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
@@ -681,8 +689,10 @@ define <32 x i16> @test_mask_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <3
; X64-NEXT: vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
; X64-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT: retq # encoding: [0xc3]
- %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+ ret <32 x i16> %3
}
define <32 x i16> @test_mask_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
@@ -697,8 +707,10 @@ define <32 x i16> @test_mask_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
- %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+ ret <32 x i16> %3
}
define <32 x i16> @test_mask_subs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
@@ -713,8 +725,8 @@ define <32 x i16> @test_mask_subs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b
; X64-NEXT: vpsubsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%b = load <32 x i16>, <32 x i16>* %ptr_b
- %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+ ret <32 x i16> %1
}
define <32 x i16> @test_mask_subs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
@@ -733,8 +745,10 @@ define <32 x i16> @test_mask_subs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_
; X64-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
%b = load <32 x i16>, <32 x i16>* %ptr_b
- %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+ ret <32 x i16> %3
}
define <32 x i16> @test_mask_subs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
@@ -751,11 +765,13 @@ define <32 x i16> @test_mask_subs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr
; X64-NEXT: vpsubsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%b = load <32 x i16>, <32 x i16>* %ptr_b
- %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
- ret <32 x i16> %res
+ %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+ %2 = bitcast i32 %mask to <32 x i1>
+ %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+ ret <32 x i16> %3
}
-declare <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+declare <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16>, <32 x i16>)
define <64 x i8> @test_mask_adds_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) {
; CHECK-LABEL: test_mask_adds_epi8_rr_512:
@@ -769,9 +785,7 @@ define <64 x i8> @test_mask_adds_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) {
define <64 x i8> @test_mask_adds_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) {
; X86-LABEL: test_mask_adds_epi8_rrk_512:
; X86: # %bb.0:
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpaddsb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xec,0xd1]
; X86-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT: retl # encoding: [0xc3]
@@ -789,9 +803,7 @@ define <64 x i8> @test_mask_adds_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x
define <64 x i8> @test_mask_adds_epi8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) {
; X86-LABEL: test_mask_adds_epi8_rrkz_512:
; X86: # %bb.0:
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpaddsb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xec,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
@@ -824,9 +836,7 @@ define <64 x i8> @test_mask_adds_epi8_rmk_512(<64 x i8> %a, <64 x i8>* %ptr_b, <
; X86-LABEL: test_mask_adds_epi8_rmk_512:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x08]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x0c]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpaddsb (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xec,0x08]
; X86-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
@@ -846,9 +856,7 @@ define <64 x i8> @test_mask_adds_epi8_rmkz_512(<64 x i8> %a, <64 x i8>* %ptr_b,
; X86-LABEL: test_mask_adds_epi8_rmkz_512:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x08]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x0c]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpaddsb (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xec,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
@@ -876,9 +884,7 @@ define <64 x i8> @test_mask_subs_epi8_rr_512(<64 x i8> %a, <64 x i8> %b) {
define <64 x i8> @test_mask_subs_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x i8> %passThru, i64 %mask) {
; X86-LABEL: test_mask_subs_epi8_rrk_512:
; X86: # %bb.0:
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpsubsb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe8,0xd1]
; X86-NEXT: vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT: retl # encoding: [0xc3]
@@ -896,9 +902,7 @@ define <64 x i8> @test_mask_subs_epi8_rrk_512(<64 x i8> %a, <64 x i8> %b, <64 x
define <64 x i8> @test_mask_subs_epi8_rrkz_512(<64 x i8> %a, <64 x i8> %b, i64 %mask) {
; X86-LABEL: test_mask_subs_epi8_rrkz_512:
; X86: # %bb.0:
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpsubsb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe8,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
@@ -931,9 +935,7 @@ define <64 x i8> @test_mask_subs_epi8_rmk_512(<64 x i8> %a, <64 x i8>* %ptr_b, <
; X86-LABEL: test_mask_subs_epi8_rmk_512:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x08]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x0c]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpsubsb (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe8,0x08]
; X86-NEXT: vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
@@ -953,9 +955,7 @@ define <64 x i8> @test_mask_subs_epi8_rmkz_512(<64 x i8> %a, <64 x i8>* %ptr_b,
; X86-LABEL: test_mask_subs_epi8_rmkz_512:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k0 # encoding: [0xc4,0xe1,0xf9,0x90,0x44,0x24,0x08]
-; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x0c]
-; X86-NEXT: kunpckdq %k0, %k1, %k1 # encoding: [0xc4,0xe1,0xf4,0x4b,0xc8]
+; X86-NEXT: kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpsubsb (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe8,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
OpenPOWER on IntegriCloud