Index: include/llvm/IR/IntrinsicsX86.td
===================================================================
--- include/llvm/IR/IntrinsicsX86.td
+++ include/llvm/IR/IntrinsicsX86.td
@@ -1439,80 +1439,6 @@
def int_x86_avx_ptestnzc_256 : GCCBuiltin<"__builtin_ia32_ptestnzc256">,
Intrinsic<[llvm_i32_ty], [llvm_v4i64_ty,
llvm_v4i64_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_d_512 : GCCBuiltin<"__builtin_ia32_ptestmd512">,
- Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
- llvm_i16_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_q_512 : GCCBuiltin<"__builtin_ia32_ptestmq512">,
- Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
- llvm_i8_ty], [IntrNoMem]>;
-
- def int_x86_avx512_ptestm_b_128 : GCCBuiltin<"__builtin_ia32_ptestmb128">,
- Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty,
- llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_b_256 : GCCBuiltin<"__builtin_ia32_ptestmb256">,
- Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty,
- llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_b_512 : GCCBuiltin<"__builtin_ia32_ptestmb512">,
- Intrinsic<[llvm_i64_ty], [llvm_v64i8_ty,
- llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_d_128 : GCCBuiltin<"__builtin_ia32_ptestmd128">,
- Intrinsic<[llvm_i8_ty], [llvm_v4i32_ty,
- llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_d_256 : GCCBuiltin<"__builtin_ia32_ptestmd256">,
- Intrinsic<[llvm_i8_ty], [llvm_v8i32_ty,
- llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_q_128 : GCCBuiltin<"__builtin_ia32_ptestmq128">,
- Intrinsic<[llvm_i8_ty], [llvm_v2i64_ty,
- llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_q_256 : GCCBuiltin<"__builtin_ia32_ptestmq256">,
- Intrinsic<[llvm_i8_ty], [llvm_v4i64_ty,
- llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_w_128 : GCCBuiltin<"__builtin_ia32_ptestmw128">,
- Intrinsic<[llvm_i8_ty], [llvm_v8i16_ty,
- llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_w_256 : GCCBuiltin<"__builtin_ia32_ptestmw256">,
- Intrinsic<[llvm_i16_ty], [llvm_v16i16_ty,
- llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestm_w_512 : GCCBuiltin<"__builtin_ia32_ptestmw512">,
- Intrinsic<[llvm_i32_ty], [llvm_v32i16_ty,
- llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>;
-
- def int_x86_avx512_ptestnm_b_128 : GCCBuiltin<"__builtin_ia32_ptestnmb128">,
- Intrinsic<[llvm_i16_ty], [llvm_v16i8_ty,
- llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_b_256 : GCCBuiltin<"__builtin_ia32_ptestnmb256">,
- Intrinsic<[llvm_i32_ty], [llvm_v32i8_ty,
- llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_b_512 : GCCBuiltin<"__builtin_ia32_ptestnmb512">,
- Intrinsic<[llvm_i64_ty], [llvm_v64i8_ty,
- llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_d_128 : GCCBuiltin<"__builtin_ia32_ptestnmd128">,
- Intrinsic<[llvm_i8_ty], [llvm_v4i32_ty,
- llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_d_256 : GCCBuiltin<"__builtin_ia32_ptestnmd256">,
- Intrinsic<[llvm_i8_ty], [llvm_v8i32_ty,
- llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_d_512 : GCCBuiltin<"__builtin_ia32_ptestnmd512">,
- Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty,
- llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_q_128 : GCCBuiltin<"__builtin_ia32_ptestnmq128">,
- Intrinsic<[llvm_i8_ty], [llvm_v2i64_ty,
- llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_q_256 : GCCBuiltin<"__builtin_ia32_ptestnmq256">,
- Intrinsic<[llvm_i8_ty], [llvm_v4i64_ty,
- llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_q_512 : GCCBuiltin<"__builtin_ia32_ptestnmq512">,
- Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty,
- llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_w_128 : GCCBuiltin<"__builtin_ia32_ptestnmw128">,
- Intrinsic<[llvm_i8_ty], [llvm_v8i16_ty,
- llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_w_256 : GCCBuiltin<"__builtin_ia32_ptestnmw256">,
- Intrinsic<[llvm_i16_ty], [llvm_v16i16_ty,
- llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
- def int_x86_avx512_ptestnm_w_512 : GCCBuiltin<"__builtin_ia32_ptestnmw512">,
- Intrinsic<[llvm_i32_ty], [llvm_v32i16_ty,
- llvm_v32i16_ty, llvm_i32_ty], [IntrNoMem]>;
def int_x86_avx512_mask_fpclass_pd_128 :
GCCBuiltin<"__builtin_ia32_fpclasspd128_mask">,
Index: lib/IR/AutoUpgrade.cpp
===================================================================
--- lib/IR/AutoUpgrade.cpp
+++ lib/IR/AutoUpgrade.cpp
@@ -256,6 +256,8 @@
Name.startswith("avx512.cvtmask2") || // Added in 5.0
(Name.startswith("xop.vpcom") && // Added in 3.2
F->arg_size() == 2) ||
+ Name.startswith("avx512.ptestm") || //Added in 6.0
+ Name.startswith("avx512.ptestnm") || //Added in 6.0
Name.startswith("sse2.pavg") || // Added in 6.0
Name.startswith("avx2.pavg") || // Added in 6.0
Name.startswith("avx512.mask.pavg")) // Added in 6.0
@@ -823,6 +825,26 @@
return Res;
}
+// Applying mask on vector of i1's and make sure result is atleast 8 bits wide.
+static Value *ApplyX86MaskOn1BitsVec(IRBuilder<> &Builder,Value *Vec, Value *Mask,
+ unsigned NumElts) {
+ const auto *C = dyn_cast<Constant>(Mask);
+ if (!C || !C->isAllOnesValue())
+ Vec = Builder.CreateAnd(Vec, getX86MaskVec(Builder, Mask, NumElts));
+
+ if (NumElts < 8) {
+ uint32_t Indices[8];
+ for (unsigned i = 0; i != NumElts; ++i)
+ Indices[i] = i;
+ for (unsigned i = NumElts; i != 8; ++i)
+ Indices[i] = NumElts + i % NumElts;
+ Vec = Builder.CreateShuffleVector(Vec,
+ Constant::getNullValue(Vec->getType()),
+ Indices);
+ }
+ return Vec;
+}
+
static Value *upgradeMaskedCompare(IRBuilder<> &Builder, CallInst &CI,
unsigned CC, bool Signed) {
Value *Op0 = CI.getArgOperand(0);
@@ -848,20 +870,8 @@
}
Value *Mask = CI.getArgOperand(CI.getNumArgOperands() - 1);
- const auto *C = dyn_cast<Constant>(Mask);
- if (!C || !C->isAllOnesValue())
- Cmp = Builder.CreateAnd(Cmp, getX86MaskVec(Builder, Mask, NumElts));
+ Cmp = ApplyX86MaskOn1BitsVec(Builder, Cmp, Mask, NumElts);
- if (NumElts < 8) {
- uint32_t Indices[8];
- for (unsigned i = 0; i != NumElts; ++i)
- Indices[i] = i;
- for (unsigned i = NumElts; i != 8; ++i)
- Indices[i] = NumElts + i % NumElts;
- Cmp = Builder.CreateShuffleVector(Cmp,
- Constant::getNullValue(Cmp->getType()),
- Indices);
- }
return Builder.CreateBitCast(Cmp, IntegerType::get(CI.getContext(),
std::max(NumElts, 8U)));
}
@@ -1027,6 +1037,21 @@
Rep = Builder.CreateICmp(CmpEq ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_SGT,
CI->getArgOperand(0), CI->getArgOperand(1));
Rep = Builder.CreateSExt(Rep, CI->getType(), "");
+ } else if (IsX86 && (Name.startswith("avx512.ptestm") ||
+ Name.startswith("avx512.ptestnm"))) {
+ Value *Op0 = CI->getArgOperand(0);
+ Value *Op1 = CI->getArgOperand(1);
+ Value *Mask = CI->getArgOperand(2);
+ Rep = Builder.CreateAnd(Op0, Op1);
+ llvm::Type *Ty = Op0->getType();
+ Value *Zero = llvm::Constant::getNullValue(Ty);
+ ICmpInst::Predicate Pred =
+ Name.startswith("avx512.ptestm") ? ICmpInst::ICMP_NE : ICmpInst::ICMP_EQ;
+ Rep = Builder.CreateICmp(Pred , Rep, Zero);
+ unsigned NumElts = Op0->getType()->getVectorNumElements();
+ Rep = ApplyX86MaskOn1BitsVec(Builder, Rep, Mask, NumElts);
+ Rep = Builder.CreateBitCast(Rep, IntegerType::get(CI->getContext(),
+ std::max(NumElts, 8U)));
} else if (IsX86 && (Name.startswith("avx512.mask.pbroadcast"))){
unsigned NumElts =
CI->getArgOperand(1)->getType()->getVectorNumElements();
Index: lib/Target/X86/X86IntrinsicsInfo.h
===================================================================
--- lib/Target/X86/X86IntrinsicsInfo.h
+++ lib/Target/X86/X86IntrinsicsInfo.h
@@ -1404,30 +1404,6 @@
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_ptestm_b_128, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_b_256, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_b_512, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_d_128, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_d_256, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_d_512, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_q_128, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_q_256, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_q_512, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_w_128, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_w_256, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestm_w_512, CMP_MASK, X86ISD::TESTM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_b_128, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_b_256, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_b_512, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_d_128, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_d_256, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_d_512, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_q_128, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_q_256, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_q_512, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_w_128, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_w_256, CMP_MASK, X86ISD::TESTNM, 0),
- X86_INTRINSIC_DATA(avx512_ptestnm_w_512, CMP_MASK, X86ISD::TESTNM, 0),
X86_INTRINSIC_DATA(avx512_rcp14_pd_128, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
X86_INTRINSIC_DATA(avx512_rcp14_pd_256, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
X86_INTRINSIC_DATA(avx512_rcp14_pd_512, INTR_TYPE_1OP_MASK, X86ISD::FRCP, 0),
Index: test/CodeGen/X86/avx512-intrinsics-upgrade.ll
===================================================================
--- test/CodeGen/X86/avx512-intrinsics-upgrade.ll
+++ test/CodeGen/X86/avx512-intrinsics-upgrade.ll
@@ -3654,3 +3654,77 @@
ret <8 x i64> %res2
}
+define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1, i8 %m) {
+; CHECK-LABEL: test_vptestmq:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmq %zmm1, %zmm0, %k0
+; CHECK-NEXT: kmovw %k0, %ecx
+; CHECK-NEXT: kmovw %edi, %k1
+; CHECK-NEXT: vptestmq %zmm1, %zmm0, %k0 {%k1}
+; CHECK-NEXT: kmovw %k0, %eax
+; CHECK-NEXT: addb %cl, %al
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq
+ %res = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
+ %res1 = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 %m)
+ %res2 = add i8 %res1, %res
+ ret i8 %res2
+}
+declare i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64>, <8 x i64>, i8)
+
+define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1, i16 %m) {
+; CHECK-LABEL: test_vptestmd:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmd %zmm1, %zmm0, %k0
+; CHECK-NEXT: kmovw %k0, %ecx
+; CHECK-NEXT: kmovw %edi, %k1
+; CHECK-NEXT: vptestmd %zmm1, %zmm0, %k0 {%k1}
+; CHECK-NEXT: kmovw %k0, %eax
+; CHECK-NEXT: addl %ecx, %eax
+; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
+; CHECK-NEXT: retq
+ %res = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
+ %res1 = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 %m)
+ %res2 = add i16 %res1, %res
+ ret i16 %res2
+}
+declare i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32>, <16 x i32>, i16)
+
+declare i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32>, <16 x i32>, i16 %x2)
+
+define i16@test_int_x86_avx512_ptestnm_d_512(<16 x i32> %x0, <16 x i32> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_512:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmd %zmm1, %zmm0, %k0
+; CHECK-NEXT: kmovw %edi, %k1
+; CHECK-NEXT: vptestnmd %zmm1, %zmm0, %k1 {%k1}
+; CHECK-NEXT: kmovw %k1, %ecx
+; CHECK-NEXT: kmovw %k0, %eax
+; CHECK-NEXT: addl %ecx, %eax
+; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
+; CHECK-NEXT: retq
+ %res = call i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32> %x0, <16 x i32> %x1, i16 %x2)
+ %res1 = call i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32> %x0, <16 x i32> %x1, i16-1)
+ %res2 = add i16 %res, %res1
+ ret i16 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64>, <8 x i64>, i8 %x2)
+
+define i8@test_int_x86_avx512_ptestnm_q_512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_512:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmq %zmm1, %zmm0, %k0
+; CHECK-NEXT: kmovw %edi, %k1
+; CHECK-NEXT: vptestnmq %zmm1, %zmm0, %k1 {%k1}
+; CHECK-NEXT: kmovw %k1, %ecx
+; CHECK-NEXT: kmovw %k0, %eax
+; CHECK-NEXT: addb %cl, %al
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq
+ %res = call i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64> %x0, <8 x i64> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
Index: test/CodeGen/X86/avx512-intrinsics.ll
===================================================================
--- test/CodeGen/X86/avx512-intrinsics.ll
+++ test/CodeGen/X86/avx512-intrinsics.ll
@@ -766,42 +766,6 @@
declare <8 x double> @llvm.x86.avx512.mask.min.pd.512(<8 x double>, <8 x double>,
<8 x double>, i8, i32)
-define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1, i8 %m) {
-; CHECK-LABEL: test_vptestmq:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1
-; CHECK-NEXT: vptestmq %zmm1, %zmm0, %k0
-; CHECK-NEXT: kmovw %k0, %ecx
-; CHECK-NEXT: vptestmq %zmm1, %zmm0, %k0 {%k1}
-; CHECK-NEXT: kmovw %k0, %eax
-; CHECK-NEXT: addb %cl, %al
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq
- %res = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
- %res1 = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 %m)
- %res2 = add i8 %res1, %res
- ret i8 %res2
-}
-declare i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64>, <8 x i64>, i8)
-
-define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1, i16 %m) {
-; CHECK-LABEL: test_vptestmd:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1
-; CHECK-NEXT: vptestmd %zmm1, %zmm0, %k0
-; CHECK-NEXT: kmovw %k0, %ecx
-; CHECK-NEXT: vptestmd %zmm1, %zmm0, %k0 {%k1}
-; CHECK-NEXT: kmovw %k0, %eax
-; CHECK-NEXT: addl %ecx, %eax
-; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
-; CHECK-NEXT: retq
- %res = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
- %res1 = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 %m)
- %res2 = add i16 %res1, %res
- ret i16 %res2
-}
-declare i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32>, <16 x i32>, i16)
-
define void @test_mask_store_ss(i8* %ptr, <4 x float> %data, i8 %mask) {
; CHECK-LABEL: test_mask_store_ss:
; CHECK: ## BB#0:
@@ -4133,47 +4097,6 @@
ret <2 x double> %res4
}
-declare i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32>, <16 x i32>, i16 %x2)
-
-define i16@test_int_x86_avx512_ptestnm_d_512(<16 x i32> %x0, <16 x i32> %x1, i16 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_512:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1
-; CHECK-NEXT: vptestnmd %zmm1, %zmm0, %k0
-; CHECK-NEXT: kmovw %k0, %ecx
-; CHECK-NEXT: vptestnmd %zmm1, %zmm0, %k0 {%k1}
-; CHECK-NEXT: kmovw %k0, %eax
-; CHECK-NEXT: addl %ecx, %eax
-; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
-; CHECK-NEXT: retq
- %res = call i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32> %x0, <16 x i32> %x1, i16 %x2)
- %res1 = call i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32> %x0, <16 x i32> %x1, i16-1)
- %res2 = add i16 %res, %res1
- ret i16 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64>, <8 x i64>, i8 %x2)
-
-define i8@test_int_x86_avx512_ptestnm_q_512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_512:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1
-; CHECK-NEXT: vptestnmq %zmm1, %zmm0, %k0
-; CHECK-NEXT: kmovw %k0, %ecx
-; CHECK-NEXT: vptestnmq %zmm1, %zmm0, %k0 {%k1}
-; CHECK-NEXT: kmovw %k0, %eax
-; CHECK-NEXT: addb %cl, %al
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq
- %res = call i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64> %x0, <8 x i64> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-
-
-
declare <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)
define <2 x double>@test_int_x86_avx512_mask_vfmadd_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3,i32 %x4 ){
Index: test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
===================================================================
--- test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
+++ test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll
@@ -3795,3 +3795,135 @@
ret <64 x i8> %res2
}
+declare i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8>, <64 x i8>, i64)
+
+define i64@test_int_x86_avx512_ptestm_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) {
+; AVX512BW-LABEL: test_int_x86_avx512_ptestm_b_512:
+; AVX512BW: ## BB#0:
+; AVX512BW-NEXT: vptestmb %zmm1, %zmm0, %k0
+; AVX512BW-NEXT: kmovq %rdi, %k1
+; AVX512BW-NEXT: vptestmb %zmm1, %zmm0, %k1 {%k1}
+; AVX512BW-NEXT: kmovq %k1, %rcx
+; AVX512BW-NEXT: kmovq %k0, %rax
+; AVX512BW-NEXT: addq %rcx, %rax
+; AVX512BW-NEXT: vzeroupper
+; AVX512BW-NEXT: retq
+;
+; AVX512F-32-LABEL: test_int_x86_avx512_ptestm_b_512:
+; AVX512F-32: # BB#0:
+; AVX512F-32-NEXT: subl $20, %esp
+; AVX512F-32-NEXT: .cfi_def_cfa_offset 24
+; AVX512F-32-NEXT: vptestmb %zmm1, %zmm0, %k0
+; AVX512F-32-NEXT: kmovq {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT: vptestmb %zmm1, %zmm0, %k1 {%k1}
+; AVX512F-32-NEXT: kmovq %k1, (%esp)
+; AVX512F-32-NEXT: movl (%esp), %eax
+; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
+; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx
+; AVX512F-32-NEXT: addl $20, %esp
+; AVX512F-32-NEXT: vzeroupper
+; AVX512F-32-NEXT: retl
+ %res = call i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
+ %res1 = call i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64-1)
+ %res2 = add i64 %res, %res1
+ ret i64 %res2
+}
+
+declare i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16>, <32 x i16>, i32)
+
+define i32@test_int_x86_avx512_ptestm_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) {
+; AVX512BW-LABEL: test_int_x86_avx512_ptestm_w_512:
+; AVX512BW: ## BB#0:
+; AVX512BW-NEXT: vptestmw %zmm1, %zmm0, %k0
+; AVX512BW-NEXT: kmovd %edi, %k1
+; AVX512BW-NEXT: vptestmw %zmm1, %zmm0, %k1 {%k1}
+; AVX512BW-NEXT: kmovd %k1, %ecx
+; AVX512BW-NEXT: kmovd %k0, %eax
+; AVX512BW-NEXT: addl %ecx, %eax
+; AVX512BW-NEXT: vzeroupper
+; AVX512BW-NEXT: retq
+;
+; AVX512F-32-LABEL: test_int_x86_avx512_ptestm_w_512:
+; AVX512F-32: # BB#0:
+; AVX512F-32-NEXT: vptestmw %zmm1, %zmm0, %k0
+; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT: vptestmw %zmm1, %zmm0, %k1 {%k1}
+; AVX512F-32-NEXT: kmovd %k1, %ecx
+; AVX512F-32-NEXT: kmovd %k0, %eax
+; AVX512F-32-NEXT: addl %ecx, %eax
+; AVX512F-32-NEXT: vzeroupper
+; AVX512F-32-NEXT: retl
+ %res = call i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
+ %res1 = call i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32-1)
+ %res2 = add i32 %res, %res1
+ ret i32 %res2
+}
+
+declare i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8>, <64 x i8>, i64 %x2)
+
+define i64@test_int_x86_avx512_ptestnm_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) {
+; AVX512BW-LABEL: test_int_x86_avx512_ptestnm_b_512:
+; AVX512BW: ## BB#0:
+; AVX512BW-NEXT: vptestnmb %zmm1, %zmm0, %k0
+; AVX512BW-NEXT: kmovq %rdi, %k1
+; AVX512BW-NEXT: vptestnmb %zmm1, %zmm0, %k1 {%k1}
+; AVX512BW-NEXT: kmovq %k1, %rcx
+; AVX512BW-NEXT: kmovq %k0, %rax
+; AVX512BW-NEXT: addq %rcx, %rax
+; AVX512BW-NEXT: vzeroupper
+; AVX512BW-NEXT: retq
+;
+; AVX512F-32-LABEL: test_int_x86_avx512_ptestnm_b_512:
+; AVX512F-32: # BB#0:
+; AVX512F-32-NEXT: subl $20, %esp
+; AVX512F-32-NEXT: .cfi_def_cfa_offset 24
+; AVX512F-32-NEXT: vptestnmb %zmm1, %zmm0, %k0
+; AVX512F-32-NEXT: kmovq {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT: vptestnmb %zmm1, %zmm0, %k1 {%k1}
+; AVX512F-32-NEXT: kmovq %k1, (%esp)
+; AVX512F-32-NEXT: movl (%esp), %eax
+; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %edx
+; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
+; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx
+; AVX512F-32-NEXT: addl $20, %esp
+; AVX512F-32-NEXT: vzeroupper
+; AVX512F-32-NEXT: retl
+ %res = call i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
+ %res1 = call i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64-1)
+ %res2 = add i64 %res, %res1
+ ret i64 %res2
+}
+
+declare i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16>, <32 x i16>, i32 %x2)
+
+define i32@test_int_x86_avx512_ptestnm_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) {
+; AVX512BW-LABEL: test_int_x86_avx512_ptestnm_w_512:
+; AVX512BW: ## BB#0:
+; AVX512BW-NEXT: vptestnmw %zmm1, %zmm0, %k0
+; AVX512BW-NEXT: kmovd %edi, %k1
+; AVX512BW-NEXT: vptestnmw %zmm1, %zmm0, %k1 {%k1}
+; AVX512BW-NEXT: kmovd %k1, %ecx
+; AVX512BW-NEXT: kmovd %k0, %eax
+; AVX512BW-NEXT: addl %ecx, %eax
+; AVX512BW-NEXT: vzeroupper
+; AVX512BW-NEXT: retq
+;
+; AVX512F-32-LABEL: test_int_x86_avx512_ptestnm_w_512:
+; AVX512F-32: # BB#0:
+; AVX512F-32-NEXT: vptestnmw %zmm1, %zmm0, %k0
+; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT: vptestnmw %zmm1, %zmm0, %k1 {%k1}
+; AVX512F-32-NEXT: kmovd %k1, %ecx
+; AVX512F-32-NEXT: kmovd %k0, %eax
+; AVX512F-32-NEXT: addl %ecx, %eax
+; AVX512F-32-NEXT: vzeroupper
+; AVX512F-32-NEXT: retl
+ %res = call i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
+ %res1 = call i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32-1)
+ %res2 = add i32 %res, %res1
+ ret i32 %res2
+}
+
Index: test/CodeGen/X86/avx512bw-intrinsics.ll
===================================================================
--- test/CodeGen/X86/avx512bw-intrinsics.ll
+++ test/CodeGen/X86/avx512bw-intrinsics.ll
@@ -1683,134 +1683,6 @@
ret <32 x i16> %res4
}
-declare i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8>, <64 x i8>, i64)
-
-define i64@test_int_x86_avx512_ptestm_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) {
-; AVX512BW-LABEL: test_int_x86_avx512_ptestm_b_512:
-; AVX512BW: ## BB#0:
-; AVX512BW-NEXT: kmovq %rdi, %k1
-; AVX512BW-NEXT: vptestmb %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT: kmovq %k0, %rcx
-; AVX512BW-NEXT: vptestmb %zmm1, %zmm0, %k0
-; AVX512BW-NEXT: kmovq %k0, %rax
-; AVX512BW-NEXT: addq %rcx, %rax
-; AVX512BW-NEXT: retq
-;
-; AVX512F-32-LABEL: test_int_x86_avx512_ptestm_b_512:
-; AVX512F-32: # BB#0:
-; AVX512F-32-NEXT: subl $20, %esp
-; AVX512F-32-NEXT: .cfi_def_cfa_offset 24
-; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k0
-; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
-; AVX512F-32-NEXT: kunpckdq %k0, %k1, %k1
-; AVX512F-32-NEXT: vptestmb %zmm1, %zmm0, %k0 {%k1}
-; AVX512F-32-NEXT: kmovq %k0, (%esp)
-; AVX512F-32-NEXT: vptestmb %zmm1, %zmm0, %k0
-; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
-; AVX512F-32-NEXT: movl (%esp), %eax
-; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %edx
-; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
-; AVX512F-32-NEXT: adcxl {{[0-9]+}}(%esp), %edx
-; AVX512F-32-NEXT: addl $20, %esp
-; AVX512F-32-NEXT: retl
- %res = call i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
- %res1 = call i64 @llvm.x86.avx512.ptestm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64-1)
- %res2 = add i64 %res, %res1
- ret i64 %res2
-}
-
-declare i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16>, <32 x i16>, i32)
-
-define i32@test_int_x86_avx512_ptestm_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) {
-; AVX512BW-LABEL: test_int_x86_avx512_ptestm_w_512:
-; AVX512BW: ## BB#0:
-; AVX512BW-NEXT: kmovd %edi, %k1
-; AVX512BW-NEXT: vptestmw %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT: kmovd %k0, %ecx
-; AVX512BW-NEXT: vptestmw %zmm1, %zmm0, %k0
-; AVX512BW-NEXT: kmovd %k0, %eax
-; AVX512BW-NEXT: addl %ecx, %eax
-; AVX512BW-NEXT: retq
-;
-; AVX512F-32-LABEL: test_int_x86_avx512_ptestm_w_512:
-; AVX512F-32: # BB#0:
-; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
-; AVX512F-32-NEXT: vptestmw %zmm1, %zmm0, %k0 {%k1}
-; AVX512F-32-NEXT: kmovd %k0, %ecx
-; AVX512F-32-NEXT: vptestmw %zmm1, %zmm0, %k0
-; AVX512F-32-NEXT: kmovd %k0, %eax
-; AVX512F-32-NEXT: addl %ecx, %eax
-; AVX512F-32-NEXT: retl
- %res = call i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
- %res1 = call i32 @llvm.x86.avx512.ptestm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32-1)
- %res2 = add i32 %res, %res1
- ret i32 %res2
-}
-
-declare i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8>, <64 x i8>, i64 %x2)
-
-define i64@test_int_x86_avx512_ptestnm_b_512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2) {
-; AVX512BW-LABEL: test_int_x86_avx512_ptestnm_b_512:
-; AVX512BW: ## BB#0:
-; AVX512BW-NEXT: kmovq %rdi, %k1
-; AVX512BW-NEXT: vptestnmb %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT: kmovq %k0, %rcx
-; AVX512BW-NEXT: vptestnmb %zmm1, %zmm0, %k0
-; AVX512BW-NEXT: kmovq %k0, %rax
-; AVX512BW-NEXT: addq %rcx, %rax
-; AVX512BW-NEXT: retq
-;
-; AVX512F-32-LABEL: test_int_x86_avx512_ptestnm_b_512:
-; AVX512F-32: # BB#0:
-; AVX512F-32-NEXT: subl $20, %esp
-; AVX512F-32-NEXT: .cfi_def_cfa_offset 24
-; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k0
-; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
-; AVX512F-32-NEXT: kunpckdq %k0, %k1, %k1
-; AVX512F-32-NEXT: vptestnmb %zmm1, %zmm0, %k0 {%k1}
-; AVX512F-32-NEXT: kmovq %k0, (%esp)
-; AVX512F-32-NEXT: vptestnmb %zmm1, %zmm0, %k0
-; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
-; AVX512F-32-NEXT: movl (%esp), %eax
-; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %edx
-; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
-; AVX512F-32-NEXT: adcxl {{[0-9]+}}(%esp), %edx
-; AVX512F-32-NEXT: addl $20, %esp
-; AVX512F-32-NEXT: retl
- %res = call i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64 %x2)
- %res1 = call i64 @llvm.x86.avx512.ptestnm.b.512(<64 x i8> %x0, <64 x i8> %x1, i64-1)
- %res2 = add i64 %res, %res1
- ret i64 %res2
-}
-
-declare i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16>, <32 x i16>, i32 %x2)
-
-define i32@test_int_x86_avx512_ptestnm_w_512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2) {
-; AVX512BW-LABEL: test_int_x86_avx512_ptestnm_w_512:
-; AVX512BW: ## BB#0:
-; AVX512BW-NEXT: kmovd %edi, %k1
-; AVX512BW-NEXT: vptestnmw %zmm1, %zmm0, %k0 {%k1}
-; AVX512BW-NEXT: kmovd %k0, %ecx
-; AVX512BW-NEXT: vptestnmw %zmm1, %zmm0, %k0
-; AVX512BW-NEXT: kmovd %k0, %eax
-; AVX512BW-NEXT: addl %ecx, %eax
-; AVX512BW-NEXT: retq
-;
-; AVX512F-32-LABEL: test_int_x86_avx512_ptestnm_w_512:
-; AVX512F-32: # BB#0:
-; AVX512F-32-NEXT: kmovd {{[0-9]+}}(%esp), %k1
-; AVX512F-32-NEXT: vptestnmw %zmm1, %zmm0, %k0 {%k1}
-; AVX512F-32-NEXT: kmovd %k0, %ecx
-; AVX512F-32-NEXT: vptestnmw %zmm1, %zmm0, %k0
-; AVX512F-32-NEXT: kmovd %k0, %eax
-; AVX512F-32-NEXT: addl %ecx, %eax
-; AVX512F-32-NEXT: retl
- %res = call i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32 %x2)
- %res1 = call i32 @llvm.x86.avx512.ptestnm.w.512(<32 x i16> %x0, <32 x i16> %x1, i32-1)
- %res2 = add i32 %res, %res1
- ret i32 %res2
-}
-
define <32 x i16> @test_x86_avx512_psll_w_512(<32 x i16> %a0, <8 x i16> %a1) {
; AVX512BW-LABEL: test_x86_avx512_psll_w_512:
; AVX512BW: ## BB#0:
Index: test/CodeGen/X86/avx512bwvl-intrinsics-upgrade.ll
===================================================================
--- test/CodeGen/X86/avx512bwvl-intrinsics-upgrade.ll
+++ test/CodeGen/X86/avx512bwvl-intrinsics-upgrade.ll
@@ -3672,3 +3672,157 @@
declare <16 x i16> @llvm.x86.avx512.mask.pavg.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
+declare i16 @llvm.x86.avx512.ptestm.b.128(<16 x i8>, <16 x i8>, i16)
+
+define i16@test_int_x86_avx512_ptestm_b_128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_b_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmb %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x08,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestmb %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
+; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i16 @llvm.x86.avx512.ptestm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2)
+ %res1 = call i16 @llvm.x86.avx512.ptestm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16-1)
+ %res2 = add i16 %res, %res1
+ ret i16 %res2
+}
+
+declare i32 @llvm.x86.avx512.ptestm.b.256(<32 x i8>, <32 x i8>, i32)
+
+define i32@test_int_x86_avx512_ptestm_b_256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_b_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmb %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x28,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestmb %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
+; CHECK-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i32 @llvm.x86.avx512.ptestm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2)
+ %res1 = call i32 @llvm.x86.avx512.ptestm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32-1)
+ %res2 = add i32 %res, %res1
+ ret i32 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestm.w.128(<8 x i16>, <8 x i16>, i8)
+
+define i8@test_int_x86_avx512_ptestm_w_128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_w_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmw %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x08,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestmw %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i16 @llvm.x86.avx512.ptestm.w.256(<16 x i16>, <16 x i16>, i16)
+
+define i16@test_int_x86_avx512_ptestm_w_256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_w_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmw %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x28,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestmw %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
+; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
+; CHECK-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i16 @llvm.x86.avx512.ptestm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2)
+ %res1 = call i16 @llvm.x86.avx512.ptestm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16-1)
+ %res2 = add i16 %res, %res1
+ ret i16 %res2
+}
+
+declare i16 @llvm.x86.avx512.ptestnm.b.128(<16 x i8>, <16 x i8>, i16)
+
+define i16@test_int_x86_avx512_ptestnm_b_128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_b_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmb %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x08,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestnmb %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7e,0x09,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
+; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i16 @llvm.x86.avx512.ptestnm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2)
+ %res1 = call i16 @llvm.x86.avx512.ptestnm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16-1)
+ %res2 = add i16 %res, %res1
+ ret i16 %res2
+}
+
+declare i32 @llvm.x86.avx512.ptestnm.b.256(<32 x i8>, <32 x i8>, i32)
+
+define i32@test_int_x86_avx512_ptestnm_b_256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_b_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmb %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x28,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestnmb %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7e,0x29,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
+; CHECK-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i32 @llvm.x86.avx512.ptestnm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2)
+ %res1 = call i32 @llvm.x86.avx512.ptestnm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32-1)
+ %res2 = add i32 %res, %res1
+ ret i32 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestnm.w.128(<8 x i16>, <8 x i16>, i8 %x2)
+
+define i8@test_int_x86_avx512_ptestnm_w_128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_w_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmw %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x08,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestnmw %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfe,0x09,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestnm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestnm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i16 @llvm.x86.avx512.ptestnm.w.256(<16 x i16>, <16 x i16>, i16 %x2)
+
+define i16@test_int_x86_avx512_ptestnm_w_256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_w_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmw %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x28,0x26,0xc1]
+; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT: vptestnmw %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfe,0x29,0x26,0xc9]
+; CHECK-NEXT: kmovd %k1, %ecx ## encoding: [0xc5,0xfb,0x93,0xc9]
+; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
+; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
+; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
+; CHECK-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77]
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i16 @llvm.x86.avx512.ptestnm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2)
+ %res1 = call i16 @llvm.x86.avx512.ptestnm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16-1)
+ %res2 = add i16 %res, %res1
+ ret i16 %res2
+}
+
Index: test/CodeGen/X86/avx512bwvl-intrinsics.ll
===================================================================
--- test/CodeGen/X86/avx512bwvl-intrinsics.ll
+++ test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -2515,154 +2515,3 @@
ret <16 x i16> %res4
}
-declare i16 @llvm.x86.avx512.ptestm.b.128(<16 x i8>, <16 x i8>, i16)
-
-define i16@test_int_x86_avx512_ptestm_b_128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_b_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestmb %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x08,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestmb %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
-; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i16 @llvm.x86.avx512.ptestm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2)
- %res1 = call i16 @llvm.x86.avx512.ptestm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16-1)
- %res2 = add i16 %res, %res1
- ret i16 %res2
-}
-
-declare i32 @llvm.x86.avx512.ptestm.b.256(<32 x i8>, <32 x i8>, i32)
-
-define i32@test_int_x86_avx512_ptestm_b_256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_b_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestmb %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestmb %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x28,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i32 @llvm.x86.avx512.ptestm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2)
- %res1 = call i32 @llvm.x86.avx512.ptestm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32-1)
- %res2 = add i32 %res, %res1
- ret i32 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestm.w.128(<8 x i16>, <8 x i16>, i8)
-
-define i8@test_int_x86_avx512_ptestm_w_128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_w_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestmw %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x08,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestmw %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i16 @llvm.x86.avx512.ptestm.w.256(<16 x i16>, <16 x i16>, i16)
-
-define i16@test_int_x86_avx512_ptestm_w_256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_w_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestmw %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x28,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestmw %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
-; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i16 @llvm.x86.avx512.ptestm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2)
- %res1 = call i16 @llvm.x86.avx512.ptestm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16-1)
- %res2 = add i16 %res, %res1
- ret i16 %res2
-}
-
-declare i16 @llvm.x86.avx512.ptestnm.b.128(<16 x i8>, <16 x i8>, i16)
-
-define i16@test_int_x86_avx512_ptestnm_b_128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_b_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestnmb %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x08,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestnmb %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7e,0x09,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
-; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i16 @llvm.x86.avx512.ptestnm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16 %x2)
- %res1 = call i16 @llvm.x86.avx512.ptestnm.b.128(<16 x i8> %x0, <16 x i8> %x1, i16-1)
- %res2 = add i16 %res, %res1
- ret i16 %res2
-}
-
-declare i32 @llvm.x86.avx512.ptestnm.b.256(<32 x i8>, <32 x i8>, i32)
-
-define i32@test_int_x86_avx512_ptestnm_b_256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_b_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestnmb %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7e,0x29,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestnmb %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x28,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i32 @llvm.x86.avx512.ptestnm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32 %x2)
- %res1 = call i32 @llvm.x86.avx512.ptestnm.b.256(<32 x i8> %x0, <32 x i8> %x1, i32-1)
- %res2 = add i32 %res, %res1
- ret i32 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestnm.w.128(<8 x i16>, <8 x i16>, i8 %x2)
-
-define i8@test_int_x86_avx512_ptestnm_w_128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_w_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestnmw %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x08,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestnmw %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfe,0x09,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestnm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestnm.w.128(<8 x i16> %x0, <8 x i16> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i16 @llvm.x86.avx512.ptestnm.w.256(<16 x i16>, <16 x i16>, i16 %x2)
-
-define i16@test_int_x86_avx512_ptestnm_w_256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_w_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
-; CHECK-NEXT: vptestnmw %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x28,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %ecx ## encoding: [0xc5,0xfb,0x93,0xc8]
-; CHECK-NEXT: vptestnmw %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfe,0x29,0x26,0xc1]
-; CHECK-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0]
-; CHECK-NEXT: addl %ecx, %eax ## encoding: [0x01,0xc8]
-; CHECK-NEXT: ## kill: %AX<def> %AX<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i16 @llvm.x86.avx512.ptestnm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16 %x2)
- %res1 = call i16 @llvm.x86.avx512.ptestnm.w.256(<16 x i16> %x0, <16 x i16> %x1, i16-1)
- %res2 = add i16 %res, %res1
- ret i16 %res2
-}
-
-
Index: test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
===================================================================
--- test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
+++ test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
@@ -5916,3 +5916,155 @@
ret <8 x i32> %res2
}
+declare i8 @llvm.x86.avx512.ptestm.d.128(<4 x i32>, <4 x i32>,i8)
+
+define i8@test_int_x86_avx512_ptestm_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_d_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmd %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x08,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestmd %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestm.d.256(<8 x i32>, <8 x i32>, i8)
+
+define i8@test_int_x86_avx512_ptestm_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_d_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmd %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x28,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestmd %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestm.q.128(<2 x i64>, <2 x i64>, i8)
+
+define i8@test_int_x86_avx512_ptestm_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_q_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmq %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x08,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestmq %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestm.q.256(<4 x i64>, <4 x i64>, i8)
+
+define i8@test_int_x86_avx512_ptestm_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestm_q_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestmq %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x28,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestmq %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestnm.d.128(<4 x i32>, <4 x i32>, i8 %x2)
+
+define i8@test_int_x86_avx512_ptestnm_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmd %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x08,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestnmd %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7e,0x09,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestnm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestnm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestnm.d.256(<8 x i32>, <8 x i32>, i8 %x2)
+
+define i8@test_int_x86_avx512_ptestnm_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmd %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x28,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestnmd %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0x7e,0x29,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestnm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestnm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestnm.q.128(<2 x i64>, <2 x i64>, i8 %x2)
+
+define i8@test_int_x86_avx512_ptestnm_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_128:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmq %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x08,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestnmq %xmm1, %xmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfe,0x09,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestnm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestnm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
+declare i8 @llvm.x86.avx512.ptestnm.q.256(<4 x i64>, <4 x i64>, i8 %x2)
+
+define i8@test_int_x86_avx512_ptestnm_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_256:
+; CHECK: ## BB#0:
+; CHECK-NEXT: vptestnmq %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x28,0x27,0xc1]
+; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT: vptestnmq %ymm1, %ymm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfe,0x29,0x27,0xc9]
+; CHECK-NEXT: kmovw %k1, %ecx ## encoding: [0xc5,0xf8,0x93,0xc9]
+; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
+; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
+; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
+; CHECK-NEXT: retq ## encoding: [0xc3]
+ %res = call i8 @llvm.x86.avx512.ptestnm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2)
+ %res1 = call i8 @llvm.x86.avx512.ptestnm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8-1)
+ %res2 = add i8 %res, %res1
+ ret i8 %res2
+}
+
Index: test/CodeGen/X86/avx512vl-intrinsics.ll
===================================================================
--- test/CodeGen/X86/avx512vl-intrinsics.ll
+++ test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -3941,160 +3941,6 @@
ret <8 x float> %res4
}
-declare i8 @llvm.x86.avx512.ptestm.d.128(<4 x i32>, <4 x i32>,i8)
-
-define i8@test_int_x86_avx512_ptestm_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_d_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestmd %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestmd %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x08,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestm.d.256(<8 x i32>, <8 x i32>, i8)
-
-define i8@test_int_x86_avx512_ptestm_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_d_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestmd %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x28,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestmd %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestm.q.128(<2 x i64>, <2 x i64>, i8)
-
-define i8@test_int_x86_avx512_ptestm_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_q_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestmq %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestmq %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x08,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestm.q.256(<4 x i64>, <4 x i64>, i8)
-
-define i8@test_int_x86_avx512_ptestm_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestm_q_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestmq %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestmq %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x28,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestnm.d.128(<4 x i32>, <4 x i32>, i8 %x2)
-
-define i8@test_int_x86_avx512_ptestnm_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestnmd %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7e,0x09,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestnmd %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x08,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestnm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestnm.d.128(<4 x i32> %x0, <4 x i32> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestnm.d.256(<8 x i32>, <8 x i32>, i8 %x2)
-
-define i8@test_int_x86_avx512_ptestnm_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestnmd %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0x7e,0x28,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestnmd %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0x7e,0x29,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestnm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestnm.d.256(<8 x i32> %x0, <8 x i32> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestnm.q.128(<2 x i64>, <2 x i64>, i8 %x2)
-
-define i8@test_int_x86_avx512_ptestnm_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_128:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestnmq %xmm1, %xmm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfe,0x09,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestnmq %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x08,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestnm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestnm.q.128(<2 x i64> %x0, <2 x i64> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-declare i8 @llvm.x86.avx512.ptestnm.q.256(<4 x i64>, <4 x i64>, i8 %x2)
-
-define i8@test_int_x86_avx512_ptestnm_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
-; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_256:
-; CHECK: ## BB#0:
-; CHECK-NEXT: kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
-; CHECK-NEXT: vptestnmq %ymm1, %ymm0, %k0 {%k1} ## encoding: [0x62,0xf2,0xfe,0x29,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8]
-; CHECK-NEXT: vptestnmq %ymm1, %ymm0, %k0 ## encoding: [0x62,0xf2,0xfe,0x28,0x27,0xc1]
-; CHECK-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0]
-; CHECK-NEXT: addb %cl, %al ## encoding: [0x00,0xc8]
-; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill>
-; CHECK-NEXT: retq ## encoding: [0xc3]
- %res = call i8 @llvm.x86.avx512.ptestnm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2)
- %res1 = call i8 @llvm.x86.avx512.ptestnm.q.256(<4 x i64> %x0, <4 x i64> %x1, i8-1)
- %res2 = add i8 %res, %res1
- ret i8 %res2
-}
-
-
-
define <2 x i64> @test_x86_avx512_psra_q_128(<2 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psra_q_128:
; CHECK: ## BB#0: