[llvm] r290559 - [AVX-512][InstCombine] Teach InstCombine to turn packed add/sub/mul/div with rounding intrinsics into normal IR operations if the rounding mode is CUR_DIRECTION.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Mon Dec 26 16:23:17 PST 2016
Author: ctopper
Date: Mon Dec 26 18:23:16 2016
New Revision: 290559
URL: http://llvm.org/viewvc/llvm-project?rev=290559&view=rev
Log:
[AVX-512][InstCombine] Teach InstCombine to turn packed add/sub/mul/div with rounding intrinsics into normal IR operations if the rounding mode is CUR_DIRECTION.
Modified:
llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp
llvm/trunk/test/Transforms/InstCombine/x86-avx512.ll
Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp?rev=290559&r1=290558&r2=290559&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp (original)
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp Mon Dec 26 18:23:16 2016
@@ -1789,6 +1789,50 @@ Instruction *InstCombiner::visitCallInst
break;
}
+ case Intrinsic::x86_avx512_mask_add_ps_512:
+ case Intrinsic::x86_avx512_mask_div_ps_512:
+ case Intrinsic::x86_avx512_mask_mul_ps_512:
+ case Intrinsic::x86_avx512_mask_sub_ps_512:
+ case Intrinsic::x86_avx512_mask_add_pd_512:
+ case Intrinsic::x86_avx512_mask_div_pd_512:
+ case Intrinsic::x86_avx512_mask_mul_pd_512:
+ case Intrinsic::x86_avx512_mask_sub_pd_512:
+ // If the rounding mode is CUR_DIRECTION(4) we can turn these into regular
+ // IR operations.
+ if (auto *R = dyn_cast<ConstantInt>(II->getArgOperand(4))) {
+ if (R->getValue() == 4) {
+ Value *Arg0 = II->getArgOperand(0);
+ Value *Arg1 = II->getArgOperand(1);
+
+ Value *V;
+ switch (II->getIntrinsicID()) {
+ default: llvm_unreachable("Case stmts out of sync!");
+ case Intrinsic::x86_avx512_mask_add_ps_512:
+ case Intrinsic::x86_avx512_mask_add_pd_512:
+ V = Builder->CreateFAdd(Arg0, Arg1);
+ break;
+ case Intrinsic::x86_avx512_mask_sub_ps_512:
+ case Intrinsic::x86_avx512_mask_sub_pd_512:
+ V = Builder->CreateFSub(Arg0, Arg1);
+ break;
+ case Intrinsic::x86_avx512_mask_mul_ps_512:
+ case Intrinsic::x86_avx512_mask_mul_pd_512:
+ V = Builder->CreateFMul(Arg0, Arg1);
+ break;
+ case Intrinsic::x86_avx512_mask_div_ps_512:
+ case Intrinsic::x86_avx512_mask_div_pd_512:
+ V = Builder->CreateFDiv(Arg0, Arg1);
+ break;
+ }
+
+ // Create a select for the masking.
+ V = emitX86MaskSelect(II->getArgOperand(3), V, II->getArgOperand(2),
+ *Builder);
+ return replaceInstUsesWith(*II, V);
+ }
+ }
+ break;
+
case Intrinsic::x86_avx512_mask_add_ss_round:
case Intrinsic::x86_avx512_mask_div_ss_round:
case Intrinsic::x86_avx512_mask_mul_ss_round:
Modified: llvm/trunk/test/Transforms/InstCombine/x86-avx512.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/x86-avx512.ll?rev=290559&r1=290558&r2=290559&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/x86-avx512.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/x86-avx512.ll Mon Dec 26 18:23:16 2016
@@ -2289,3 +2289,323 @@ define <64 x i8> @undef_test_permvar_qi_
%a = tail call <64 x i8> @llvm.x86.avx512.mask.permvar.qi.512(<64 x i8> %a0, <64 x i8> <i8 undef, i8 62, i8 61, i8 60, i8 59, i8 58, i8 57, i8 56, i8 55, i8 54, i8 53, i8 52, i8 51, i8 50, i8 49, i8 48, i8 47, i8 46, i8 45, i8 44, i8 43, i8 42, i8 41, i8 40, i8 39, i8 38, i8 37, i8 36, i8 35, i8 34, i8 33, i8 32, i8 31, i8 30, i8 29, i8 28, i8 27, i8 26, i8 25, i8 24, i8 23, i8 22, i8 21, i8 20, i8 19, i8 18, i8 17, i8 16, i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>, <64 x i8> %passthru, i64 %mask)
ret <64 x i8> %a
}
+
+declare <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
+
+define <16 x float> @test_add_ps(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_add_ps(
+; CHECK-NEXT: [[TMP1:%.*]] = fadd <16 x float> %a, %b
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_add_ps_round(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_add_ps_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_add_ps_mask(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_add_ps_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fadd <16 x float> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 %mask to <16 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[TMP2]], <16 x float> [[TMP1]], <16 x float> %c
+; CHECK-NEXT: ret <16 x float> [[TMP3]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_add_ps_mask_round(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_add_ps_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+ ret <16 x float> %1
+}
+
+declare <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
+
+define <8 x double> @test_add_pd(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_add_pd(
+; CHECK-NEXT: [[TMP1:%.*]] = fadd <8 x double> %a, %b
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_add_pd_round(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_add_pd_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_add_pd_mask(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_add_pd_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fadd <8 x double> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 %mask to <8 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x double> [[TMP1]], <8 x double> %c
+; CHECK-NEXT: ret <8 x double> [[TMP3]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_add_pd_mask_round(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_add_pd_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.add.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+ ret <8 x double> %1
+}
+
+declare <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
+
+define <16 x float> @test_sub_ps(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_sub_ps(
+; CHECK-NEXT: [[TMP1:%.*]] = fsub <16 x float> %a, %b
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_sub_ps_round(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_sub_ps_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_sub_ps_mask(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_sub_ps_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fsub <16 x float> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 %mask to <16 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[TMP2]], <16 x float> [[TMP1]], <16 x float> %c
+; CHECK-NEXT: ret <16 x float> [[TMP3]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_sub_ps_mask_round(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_sub_ps_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+ ret <16 x float> %1
+}
+
+declare <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
+
+define <8 x double> @test_sub_pd(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_sub_pd(
+; CHECK-NEXT: [[TMP1:%.*]] = fsub <8 x double> %a, %b
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_sub_pd_round(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_sub_pd_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_sub_pd_mask(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_sub_pd_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fsub <8 x double> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 %mask to <8 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x double> [[TMP1]], <8 x double> %c
+; CHECK-NEXT: ret <8 x double> [[TMP3]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_sub_pd_mask_round(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_sub_pd_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.sub.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+ ret <8 x double> %1
+}
+
+declare <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
+
+define <16 x float> @test_mul_ps(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_mul_ps(
+; CHECK-NEXT: [[TMP1:%.*]] = fmul <16 x float> %a, %b
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_mul_ps_round(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_mul_ps_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_mul_ps_mask(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_mul_ps_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fmul <16 x float> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 %mask to <16 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[TMP2]], <16 x float> [[TMP1]], <16 x float> %c
+; CHECK-NEXT: ret <16 x float> [[TMP3]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_mul_ps_mask_round(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_mul_ps_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+ ret <16 x float> %1
+}
+
+declare <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
+
+define <8 x double> @test_mul_pd(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_mul_pd(
+; CHECK-NEXT: [[TMP1:%.*]] = fmul <8 x double> %a, %b
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_mul_pd_round(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_mul_pd_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_mul_pd_mask(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_mul_pd_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fmul <8 x double> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 %mask to <8 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x double> [[TMP1]], <8 x double> %c
+; CHECK-NEXT: ret <8 x double> [[TMP3]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_mul_pd_mask_round(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_mul_pd_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+ ret <8 x double> %1
+}
+
+declare <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
+
+define <16 x float> @test_div_ps(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_div_ps(
+; CHECK-NEXT: [[TMP1:%.*]] = fdiv <16 x float> %a, %b
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_div_ps_round(<16 x float> %a, <16 x float> %b) {
+; CHECK-LABEL: @test_div_ps_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> undef, i16 -1, i32 8)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_div_ps_mask(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_div_ps_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fdiv <16 x float> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i16 %mask to <16 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[TMP2]], <16 x float> [[TMP1]], <16 x float> %c
+; CHECK-NEXT: ret <16 x float> [[TMP3]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 4)
+ ret <16 x float> %1
+}
+
+define <16 x float> @test_div_ps_mask_round(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
+; CHECK-LABEL: @test_div_ps_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+; CHECK-NEXT: ret <16 x float> [[TMP1]]
+;
+ %1 = tail call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask, i32 8)
+ ret <16 x float> %1
+}
+
+declare <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
+
+define <8 x double> @test_div_pd(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_div_pd(
+; CHECK-NEXT: [[TMP1:%.*]] = fdiv <8 x double> %a, %b
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_div_pd_round(<8 x double> %a, <8 x double> %b) {
+; CHECK-LABEL: @test_div_pd_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> undef, i8 -1, i32 8)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_div_pd_mask(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_div_pd_mask(
+; CHECK-NEXT: [[TMP1:%.*]] = fdiv <8 x double> %a, %b
+; CHECK-NEXT: [[TMP2:%.*]] = bitcast i8 %mask to <8 x i1>
+; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x double> [[TMP1]], <8 x double> %c
+; CHECK-NEXT: ret <8 x double> [[TMP3]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 4)
+ ret <8 x double> %1
+}
+
+define <8 x double> @test_div_pd_mask_round(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask) {
+; CHECK-LABEL: @test_div_pd_mask_round(
+; CHECK-NEXT: [[TMP1:%.*]] = tail call <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+; CHECK-NEXT: ret <8 x double> [[TMP1]]
+;
+ %1 = tail call <8 x double> @llvm.x86.avx512.mask.div.pd.512(<8 x double> %a, <8 x double> %b, <8 x double> %c, i8 %mask, i32 8)
+ ret <8 x double> %1
+}
More information about the llvm-commits
mailing list