diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll | 3189 |
1 files changed, 3189 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll index 00e54d03dc1..3a7ac80273b 100644 --- a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll +++ b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll @@ -2811,5 +2811,3194 @@ entry: } +define <8 x double> @test_mm512_fmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + ret <8 x double> %0 +} + +declare <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) #1 + +define <8 x double> @test_mm512_mask_fmadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fmadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2 +; X32-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2 +; X64-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask_fmsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fmsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fnmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fnmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm0, %zmm0 +; X32-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; X64-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8) + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask3_fnmadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fnmadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fnmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fnmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fnmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0] +; X32-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X32-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X32-NEXT: vfmadd231pd {rn-sae}, %zmm4, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0] +; X64-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X64-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X64-NEXT: vfmadd231pd {rn-sae}, %zmm4, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %sub1, i32 8) + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_maskz_fnmsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fnmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %sub1, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask_fmadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2 +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2 +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fmadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2 +; X32-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2 +; X64-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask_fmsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fmsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fnmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fnmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm0, %zmm0 +; X32-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0 +; X64-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10 + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask3_fnmadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2 +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2 +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fnmadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fnmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fnmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fnmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0] +; X32-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X32-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X32-NEXT: vfmadd231pd {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpbroadcastq {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0] +; X64-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X64-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X64-NEXT: vfmadd231pd {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %sub1.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %sub1.i) #10 + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_maskz_fnmsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fnmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %sub1.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %sub1.i) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <16 x float> @test_mm512_fmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + ret <16 x float> %0 +} + +declare <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) #1 + +define <16 x float> @test_mm512_mask_fmadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fmadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2 +; X32-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2 +; X64-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_mask_fmsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fmsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fnmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fnmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm0, %zmm0 +; X32-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm0 +; X64-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8) + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_mask3_fnmadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fnmadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fnmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fnmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fnmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0] +; X32-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X32-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X32-NEXT: vfmadd231ps {rn-sae}, %zmm4, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0] +; X64-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X64-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X64-NEXT: vfmadd231ps {rn-sae}, %zmm4, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %sub1, i32 8) + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_maskz_fnmsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fnmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %sub1, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_mask_fmadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) + zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2 +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) + zmm2 +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fmadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2 +; X32-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2 +; X64-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_mask_fmsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fmsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fnmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fnmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm0, %zmm0 +; X32-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm0 +; X64-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10 + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_mask3_fnmadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmadd231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2 +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2 +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fnmadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fnmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmadd213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fnmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fnmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0] +; X32-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X32-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X32-NEXT: vfmadd231ps {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fnmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpbroadcastd {{.*#+}} zmm3 = [-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0,-0] +; X64-NEXT: vpxorq %zmm3, %zmm0, %zmm4 +; X64-NEXT: vpxorq %zmm3, %zmm2, %zmm0 +; X64-NEXT: vfmadd231ps {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %sub1.i) #10 + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_maskz_fnmsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fnmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmsub213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fnmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %sub1.i) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <8 x double> @test_mm512_fmaddsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmaddsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmaddsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + ret <8 x double> %0 +} + +declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) #1 + +define <8 x double> @test_mm512_mask_fmaddsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmaddsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmaddsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmaddsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fmaddsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmaddsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmaddsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmaddsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fmaddsub_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmaddsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmaddsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fmsubadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmsubadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2 +; X32-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsubadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2 +; X64-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + ret <8 x double> %0 +} + +define <8 x double> @test_mm512_mask_fmsubadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmsubadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsubadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_maskz_fmsubadd_round_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmsubadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsubadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> zeroinitializer + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_fmaddsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmaddsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmaddsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10 + %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + ret <8 x double> %3 +} + +define <8 x double> @test_mm512_mask_fmaddsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmaddsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmaddsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmaddsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10 + %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %__A + ret <8 x double> %5 +} + +define <8 x double> @test_mm512_mask3_fmaddsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmaddsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmaddsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2 +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmaddsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2 +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10 + %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %__C + ret <8 x double> %5 +} + +define <8 x double> @test_mm512_maskz_fmaddsub_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmaddsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmaddsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %2 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %1) #10 + %3 = shufflevector <8 x double> %2, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + %4 = bitcast i8 %__U to <8 x i1> + %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> zeroinitializer + ret <8 x double> %5 +} + +define <8 x double> @test_mm512_fmsubadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_fmsubadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsubadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask_fmsubadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fmsubadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsubadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsubadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd132pd {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + %3 = bitcast i8 %__U to <8 x i1> + %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__A + ret <8 x double> %4 +} + +define <8 x double> @test_mm512_maskz_fmsubadd_pd(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_maskz_fmsubadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsubadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + %3 = bitcast i8 %__U to <8 x i1> + %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> zeroinitializer + ret <8 x double> %4 +} + +define <16 x float> @test_mm512_fmaddsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmaddsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmaddsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + ret <16 x float> %0 +} + +declare <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) #1 + +define <16 x float> @test_mm512_mask_fmaddsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmaddsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmaddsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmaddsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fmaddsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmaddsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmaddsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmaddsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fmaddsub_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmaddsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmaddsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fmsubadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmsubadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2 +; X32-NEXT: vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsubadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2 +; X64-NEXT: vfmaddsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + ret <16 x float> %0 +} + +define <16 x float> @test_mm512_mask_fmsubadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmsubadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsubadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsubadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_maskz_fmsubadd_round_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmsubadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsubadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsubadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> zeroinitializer + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_fmaddsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmaddsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmaddsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10 + %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + ret <16 x float> %3 +} + +define <16 x float> @test_mm512_mask_fmaddsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmaddsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmaddsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmaddsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) +/- zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10 + %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %__A + ret <16 x float> %5 +} + +define <16 x float> @test_mm512_mask3_fmaddsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmaddsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmaddsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2 +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmaddsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) +/- zmm2 +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10 + %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %__C + ret <16 x float> %5 +} + +define <16 x float> @test_mm512_maskz_fmaddsub_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmaddsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmaddsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmaddsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) +/- zmm2 +; X64-NEXT: retq +entry: + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %2 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %1) #10 + %3 = shufflevector <16 x float> %2, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + %4 = bitcast i16 %__U to <16 x i1> + %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> zeroinitializer + ret <16 x float> %5 +} + +define <16 x float> @test_mm512_fmsubadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_fmsubadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_fmsubadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask_fmsubadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fmsubadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsubadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fmsubadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd132ps {{.*#+}} zmm0 = (zmm0 * zmm1) -/+ zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + %3 = bitcast i16 %__U to <16 x i1> + %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__A + ret <16 x float> %4 +} + +define <16 x float> @test_mm512_maskz_fmsubadd_ps(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_maskz_fmsubadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_fmsubadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) -/+ zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + %3 = bitcast i16 %__U to <16 x i1> + %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> zeroinitializer + ret <16 x float> %4 +} + +define <8 x double> @test_mm512_mask3_fmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2 +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231pd {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2 +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <16 x float> @test_mm512_mask3_fmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2 +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231ps {{.*#+}} zmm2 = (zmm0 * zmm1) - zmm2 +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <8 x double> @test_mm512_mask3_fmsubadd_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsubadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsubadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsubadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fmsubadd_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsubadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsubadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2 +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsubadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd231pd {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2 +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %sub.i) #10 + %1 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) #10 + %2 = shufflevector <8 x double> %1, <8 x double> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15> + %3 = bitcast i8 %__U to <8 x i1> + %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %__C + ret <8 x double> %4 +} + +define <16 x float> @test_mm512_mask3_fmsubadd_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsubadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsubadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsubadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fmsubadd_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fmsubadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfmsubadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2 +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fmsubadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsubadd231ps {{.*#+}} zmm2 = (zmm0 * zmm1) -/+ zmm2 +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %sub.i) #10 + %1 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) #10 + %2 = shufflevector <16 x float> %1, <16 x float> %0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31> + %3 = bitcast i16 %__U to <16 x i1> + %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %__C + ret <16 x float> %4 +} + +define <8 x double> @test_mm512_mask_fnmadd_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fnmadd_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmadd_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %sub, <8 x double> %__B, <8 x double> %__C, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask_fnmadd_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fnmadd_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmadd_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %sub.i, <8 x double> %__B, <8 x double> %__C) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <16 x float> @test_mm512_mask_fnmadd_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fnmadd_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmadd_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %sub, <16 x float> %__B, <16 x float> %__C, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask_fnmadd_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fnmadd_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmadd132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmadd_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) + zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %sub.i, <16 x float> %__B, <16 x float> %__C) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <8 x double> @test_mm512_mask_fnmsub_round_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fnmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B + %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %sub, <8 x double> %sub1, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fnmsub_round_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmsub_round_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmsub_round_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B + %sub1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %sub, <8 x double> %sub1, i32 8) + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask_fnmsub_pd(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) { +; X32-LABEL: test_mm512_mask_fnmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub132pd {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B + %sub2.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %sub.i, <8 x double> %sub2.i) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__A + ret <8 x double> %2 +} + +define <8 x double> @test_mm512_mask3_fnmsub_pd(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmsub_pd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2 +; X32-NEXT: vmovapd %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmsub_pd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2 +; X64-NEXT: vmovapd %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__B + %sub2.i = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %sub.i, <8 x double> %sub2.i) #10 + %1 = bitcast i8 %__U to <8 x i1> + %2 = select <8 x i1> %1, <8 x double> %0, <8 x double> %__C + ret <8 x double> %2 +} + +define <16 x float> @test_mm512_mask_fnmsub_round_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fnmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %sub, <16 x float> %sub1, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fnmsub_round_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmsub_round_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmsub_round_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1} +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %sub1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %sub, <16 x float> %sub1, i32 8) + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask_fnmsub_ps(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) { +; X32-LABEL: test_mm512_mask_fnmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmsub132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_fnmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub132ps {{.*#+}} zmm0 = -(zmm0 * zmm1) - zmm2 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %sub.i, <16 x float> %sub1.i) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__A + ret <16 x float> %2 +} + +define <16 x float> @test_mm512_mask3_fnmsub_ps(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) { +; X32-LABEL: test_mm512_mask3_fnmsub_ps: +; X32: # %bb.0: # %entry +; X32-NEXT: kmovw {{[0-9]+}}(%esp), %k1 +; X32-NEXT: vfnmsub231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2 +; X32-NEXT: vmovaps %zmm2, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask3_fnmsub_ps: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) - zmm2 +; X64-NEXT: vmovaps %zmm2, %zmm0 +; X64-NEXT: retq +entry: + %sub.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %sub1.i = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %sub.i, <16 x float> %sub1.i) #10 + %1 = bitcast i16 %__U to <16 x i1> + %2 = select <16 x i1> %1, <16 x float> %0, <16 x float> %__C + ret <16 x float> %2 +} + +define <4 x float> @test_mm_mask_fmadd_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fmadd_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmadd_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %1 = extractelement <4 x float> %__A, i64 0 + %2 = extractelement <4 x float> %__B, i64 0 + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <4 x float> %__W, i32 0 + %cond.i = select i1 %tobool.i, float %vecext1.i, float %3 + %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask_fmadd_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fmadd_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmadd_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %__A, <4 x float> %__B, i8 %__U, i32 4) + ret <4 x float> %0 +} + +declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1 + +define <4 x float> @test_mm_maskz_fmadd_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fmadd_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmadd_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__A, i64 0 + %1 = extractelement <4 x float> %__B, i64 0 + %2 = extractelement <4 x float> %__C, i64 0 + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3 + %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_maskz_fmadd_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fmadd_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmadd_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %__B, <4 x float> %__C, i8 %__U, i32 4) + ret <4 x float> %0 +} + +declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1 + +define <4 x float> @test_mm_mask3_fmadd_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmadd_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmadd_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %1 = extractelement <4 x float> %__X, i64 0 + %2 = extractelement <4 x float> %__Y, i64 0 + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <4 x float> %__Y, i32 0 + %cond.i = select i1 %tobool.i, float %vecext1.i, float %3 + %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask3_fmadd_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmadd_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmadd_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 %__U, i32 4) + ret <4 x float> %0 +} + +declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1 + +define <4 x float> @test_mm_mask_fmsub_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fmsub_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmsub_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %1 = extractelement <4 x float> %__A, i64 0 + %.rhs.i = extractelement <4 x float> %__B, i64 0 + %2 = fsub float -0.000000e+00, %.rhs.i + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <4 x float> %__W, i32 0 + %cond.i = select i1 %tobool.i, float %vecext1.i, float %3 + %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask_fmsub_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fmsub_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmsub_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %__A, <4 x float> %sub, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_maskz_fmsub_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fmsub_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmsub_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__A, i64 0 + %1 = extractelement <4 x float> %__B, i64 0 + %.rhs.i = extractelement <4 x float> %__C, i64 0 + %2 = fsub float -0.000000e+00, %.rhs.i + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3 + %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_maskz_fmsub_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fmsub_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmsub_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %__B, <4 x float> %sub, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_mask3_fmsub_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmsub_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmsub_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %1 = extractelement <4 x float> %__X, i64 0 + %.rhs.i = extractelement <4 x float> %__Y, i64 0 + %2 = fsub float -0.000000e+00, %.rhs.i + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <4 x float> %__Y, i32 0 + %cond.i = select i1 %tobool.i, float %vecext1.i, float %3 + %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask3_fmsub_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmsub_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmsub_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 %__U, i32 4) + ret <4 x float> %0 +} + +declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) #1 + +define <4 x float> @test_mm_mask_fnmadd_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fnmadd_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmadd_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %.rhs.i = extractelement <4 x float> %__A, i64 0 + %1 = fsub float -0.000000e+00, %.rhs.i + %2 = extractelement <4 x float> %__B, i64 0 + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <4 x float> %__W, i32 0 + %cond.i = select i1 %tobool.i, float %vecext1.i, float %3 + %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask_fnmadd_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fnmadd_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmadd_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__B, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_maskz_fnmadd_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fnmadd_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmadd_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__A, i64 0 + %.rhs.i = extractelement <4 x float> %__B, i64 0 + %1 = fsub float -0.000000e+00, %.rhs.i + %2 = extractelement <4 x float> %__C, i64 0 + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3 + %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_maskz_fnmadd_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fnmadd_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmadd_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %sub, <4 x float> %__C, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_mask3_fnmadd_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmadd_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmadd_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %.rhs.i = extractelement <4 x float> %__X, i64 0 + %1 = fsub float -0.000000e+00, %.rhs.i + %2 = extractelement <4 x float> %__Y, i64 0 + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <4 x float> %__Y, i32 0 + %cond.i = select i1 %tobool.i, float %vecext1.i, float %3 + %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask3_fnmadd_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmadd_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmadd_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231ss {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__X + %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__Y, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_mask_fnmsub_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fnmsub_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmsub_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %.rhs.i = extractelement <4 x float> %__A, i64 0 + %1 = fsub float -0.000000e+00, %.rhs.i + %.rhs7.i = extractelement <4 x float> %__B, i64 0 + %2 = fsub float -0.000000e+00, %.rhs7.i + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext2.i = extractelement <4 x float> %__W, i32 0 + %cond.i = select i1 %tobool.i, float %vecext2.i, float %3 + %vecins.i = insertelement <4 x float> %__W, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask_fnmsub_round_ss(<4 x float> %__W, i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B) { +; X32-LABEL: test_mm_mask_fnmsub_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X32-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmsub_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X64-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__A + %sub1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %0 = tail call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %sub1, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_maskz_fnmsub_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fnmsub_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmsub_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__A, i64 0 + %.rhs.i = extractelement <4 x float> %__B, i64 0 + %1 = fsub float -0.000000e+00, %.rhs.i + %.rhs5.i = extractelement <4 x float> %__C, i64 0 + %2 = fsub float -0.000000e+00, %.rhs5.i + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, float 0.000000e+00, float %3 + %vecins.i = insertelement <4 x float> %__A, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_maskz_fnmsub_round_ss(i8 zeroext %__U, <4 x float> %__A, <4 x float> %__B, <4 x float> %__C) { +; X32-LABEL: test_mm_maskz_fnmsub_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X32-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmsub_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X64-NEXT: vxorps %xmm3, %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__B + %sub1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__C + %0 = tail call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %__A, <4 x float> %sub, <4 x float> %sub1, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <4 x float> @test_mm_mask3_fnmsub_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmsub_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmsub_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub231ss {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <4 x float> %__W, i64 0 + %.rhs.i = extractelement <4 x float> %__X, i64 0 + %1 = fsub float -0.000000e+00, %.rhs.i + %.rhs7.i = extractelement <4 x float> %__Y, i64 0 + %2 = fsub float -0.000000e+00, %.rhs7.i + %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext2.i = extractelement <4 x float> %__Y, i32 0 + %cond.i = select i1 %tobool.i, float %vecext2.i, float %3 + %vecins.i = insertelement <4 x float> %__Y, float %cond.i, i32 0 + ret <4 x float> %vecins.i +} + +define <4 x float> @test_mm_mask3_fnmsub_round_ss(<4 x float> %__W, <4 x float> %__X, <4 x float> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmsub_round_ss: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X32-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovaps %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmsub_round_ss: +; X64: # %bb.0: # %entry +; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] +; X64-NEXT: vxorps %xmm3, %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231ss {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovaps %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %__X + %0 = tail call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %__W, <4 x float> %sub, <4 x float> %__Y, i8 %__U, i32 4) + ret <4 x float> %0 +} + +define <2 x double> @test_mm_mask_fmadd_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fmadd_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmadd_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %1 = extractelement <2 x double> %__A, i64 0 + %2 = extractelement <2 x double> %__B, i64 0 + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <2 x double> %__W, i32 0 + %cond.i = select i1 %tobool.i, double %vecext1.i, double %3 + %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask_fmadd_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fmadd_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmadd_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %__A, <2 x double> %__B, i8 %__U, i32 4) + ret <2 x double> %0 +} + +declare <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1 + +define <2 x double> @test_mm_maskz_fmadd_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fmadd_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmadd_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__A, i64 0 + %1 = extractelement <2 x double> %__B, i64 0 + %2 = extractelement <2 x double> %__C, i64 0 + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3 + %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_maskz_fmadd_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fmadd_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmadd_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %__B, <2 x double> %__C, i8 %__U, i32 4) + ret <2 x double> %0 +} + +declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1 + +define <2 x double> @test_mm_mask3_fmadd_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmadd_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmadd_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %1 = extractelement <2 x double> %__X, i64 0 + %2 = extractelement <2 x double> %__Y, i64 0 + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <2 x double> %__Y, i32 0 + %cond.i = select i1 %tobool.i, double %vecext1.i, double %3 + %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask3_fmadd_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmadd_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmadd_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 %__U, i32 4) + ret <2 x double> %0 +} + +declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1 + +define <2 x double> @test_mm_mask_fmsub_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fmsub_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmsub_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %1 = extractelement <2 x double> %__A, i64 0 + %.rhs.i = extractelement <2 x double> %__B, i64 0 + %2 = fsub double -0.000000e+00, %.rhs.i + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <2 x double> %__W, i32 0 + %cond.i = select i1 %tobool.i, double %vecext1.i, double %3 + %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask_fmsub_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fmsub_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vxorpd {{\.LCPI.*}}, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fmsub_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vxorpd {{.*}}(%rip), %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B + %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %__A, <2 x double> %sub, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_maskz_fmsub_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fmsub_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmsub_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__A, i64 0 + %1 = extractelement <2 x double> %__B, i64 0 + %.rhs.i = extractelement <2 x double> %__C, i64 0 + %2 = fsub double -0.000000e+00, %.rhs.i + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3 + %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_maskz_fmsub_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fmsub_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vxorpd {{\.LCPI.*}}, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fmsub_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vxorpd {{.*}}(%rip), %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %__B, <2 x double> %sub, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_mask3_fmsub_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmsub_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmsub_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %1 = extractelement <2 x double> %__X, i64 0 + %.rhs.i = extractelement <2 x double> %__Y, i64 0 + %2 = fsub double -0.000000e+00, %.rhs.i + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <2 x double> %__Y, i32 0 + %cond.i = select i1 %tobool.i, double %vecext1.i, double %3 + %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask3_fmsub_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fmsub_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fmsub_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 %__U, i32 4) + ret <2 x double> %0 +} + +declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) #1 + +define <2 x double> @test_mm_mask_fnmadd_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fnmadd_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmadd_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %.rhs.i = extractelement <2 x double> %__A, i64 0 + %1 = fsub double -0.000000e+00, %.rhs.i + %2 = extractelement <2 x double> %__B, i64 0 + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <2 x double> %__W, i32 0 + %cond.i = select i1 %tobool.i, double %vecext1.i, double %3 + %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask_fnmadd_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fnmadd_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vxorpd {{\.LCPI.*}}, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmadd_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vxorpd {{.*}}(%rip), %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A + %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__B, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_maskz_fnmadd_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fnmadd_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmadd_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__A, i64 0 + %.rhs.i = extractelement <2 x double> %__B, i64 0 + %1 = fsub double -0.000000e+00, %.rhs.i + %2 = extractelement <2 x double> %__C, i64 0 + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3 + %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_maskz_fnmadd_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fnmadd_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vxorpd {{\.LCPI.*}}, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmadd_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vxorpd {{.*}}(%rip), %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B + %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %sub, <2 x double> %__C, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_mask3_fnmadd_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmadd_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmadd231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmadd_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmadd231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %.rhs.i = extractelement <2 x double> %__X, i64 0 + %1 = fsub double -0.000000e+00, %.rhs.i + %2 = extractelement <2 x double> %__Y, i64 0 + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext1.i = extractelement <2 x double> %__Y, i32 0 + %cond.i = select i1 %tobool.i, double %vecext1.i, double %3 + %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask3_fnmadd_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmadd_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vxorpd {{\.LCPI.*}}, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmadd_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vxorpd {{.*}}(%rip), %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd231sd {{.*#+}} xmm2 = (xmm0 * xmm1) + xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__X + %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__Y, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_mask_fnmsub_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fnmsub_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmsub_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %.rhs.i = extractelement <2 x double> %__A, i64 0 + %1 = fsub double -0.000000e+00, %.rhs.i + %.rhs7.i = extractelement <2 x double> %__B, i64 0 + %2 = fsub double -0.000000e+00, %.rhs7.i + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext2.i = extractelement <2 x double> %__W, i32 0 + %cond.i = select i1 %tobool.i, double %vecext2.i, double %3 + %vecins.i = insertelement <2 x double> %__W, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask_fnmsub_round_sd(<2 x double> %__W, i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B) { +; X32-LABEL: test_mm_mask_fnmsub_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00] +; X32-NEXT: vxorpd %xmm3, %xmm1, %xmm1 +; X32-NEXT: vxorpd %xmm3, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask_fnmsub_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00] +; X64-NEXT: vxorpd %xmm3, %xmm1, %xmm1 +; X64-NEXT: vxorpd %xmm3, %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__A + %sub1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B + %0 = tail call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %sub1, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_maskz_fnmsub_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fnmsub_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmsub_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__A, i64 0 + %.rhs.i = extractelement <2 x double> %__B, i64 0 + %1 = fsub double -0.000000e+00, %.rhs.i + %.rhs5.i = extractelement <2 x double> %__C, i64 0 + %2 = fsub double -0.000000e+00, %.rhs5.i + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %cond.i = select i1 %tobool.i, double 0.000000e+00, double %3 + %vecins.i = insertelement <2 x double> %__A, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_maskz_fnmsub_round_sd(i8 zeroext %__U, <2 x double> %__A, <2 x double> %__B, <2 x double> %__C) { +; X32-LABEL: test_mm_maskz_fnmsub_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00] +; X32-NEXT: vxorpd %xmm3, %xmm1, %xmm1 +; X32-NEXT: vxorpd %xmm3, %xmm2, %xmm2 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_maskz_fnmsub_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vmovapd {{.*#+}} xmm3 = [-0.000000e+00,-0.000000e+00] +; X64-NEXT: vxorpd %xmm3, %xmm1, %xmm1 +; X64-NEXT: vxorpd %xmm3, %xmm2, %xmm2 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__B + %sub1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__C + %0 = tail call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %__A, <2 x double> %sub, <2 x double> %sub1, i8 %__U, i32 4) + ret <2 x double> %0 +} + +define <2 x double> @test_mm_mask3_fnmsub_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmsub_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfnmsub231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmsub_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfnmsub231sd {{.*#+}} xmm2 = -(xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %0 = extractelement <2 x double> %__W, i64 0 + %.rhs.i = extractelement <2 x double> %__X, i64 0 + %1 = fsub double -0.000000e+00, %.rhs.i + %.rhs7.i = extractelement <2 x double> %__Y, i64 0 + %2 = fsub double -0.000000e+00, %.rhs7.i + %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #10 + %4 = and i8 %__U, 1 + %tobool.i = icmp eq i8 %4, 0 + %vecext2.i = extractelement <2 x double> %__Y, i32 0 + %cond.i = select i1 %tobool.i, double %vecext2.i, double %3 + %vecins.i = insertelement <2 x double> %__Y, double %cond.i, i32 0 + ret <2 x double> %vecins.i +} + +define <2 x double> @test_mm_mask3_fnmsub_round_sd(<2 x double> %__W, <2 x double> %__X, <2 x double> %__Y, i8 zeroext %__U) { +; X32-LABEL: test_mm_mask3_fnmsub_round_sd: +; X32: # %bb.0: # %entry +; X32-NEXT: movb {{[0-9]+}}(%esp), %al +; X32-NEXT: vxorpd {{\.LCPI.*}}, %xmm1, %xmm1 +; X32-NEXT: kmovw %eax, %k1 +; X32-NEXT: vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X32-NEXT: vmovapd %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm_mask3_fnmsub_round_sd: +; X64: # %bb.0: # %entry +; X64-NEXT: vxorpd {{.*}}(%rip), %xmm1, %xmm1 +; X64-NEXT: kmovw %edi, %k1 +; X64-NEXT: vfmsub231sd {{.*#+}} xmm2 = (xmm0 * xmm1) - xmm2 +; X64-NEXT: vmovapd %xmm2, %xmm0 +; X64-NEXT: retq +entry: + %sub = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %__X + %0 = tail call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %__W, <2 x double> %sub, <2 x double> %__Y, i8 %__U, i32 4) + ret <2 x double> %0 +} + +declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>) #9 +declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>) #9 +declare float @llvm.fma.f32(float, float, float) #9 +declare double @llvm.fma.f64(double, double, double) #9 + !0 = !{i32 1} |
