[llvm] r363215 - [NFC][CodeGen] Add unary FNeg tests to X86/avx512-intrinsics-fast-isel.ll
Cameron McInally via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 12 15:50:44 PDT 2019
Author: mcinally
Date: Wed Jun 12 15:50:44 2019
New Revision: 363215
URL: http://llvm.org/viewvc/llvm-project?rev=363215&view=rev
Log:
[NFC][CodeGen] Add unary FNeg tests to X86/avx512-intrinsics-fast-isel.ll
Patch 1 of n.
Modified:
llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
Modified: llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll?rev=363215&r1=363214&r2=363215&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll Wed Jun 12 15:50:44 2019
@@ -3099,6 +3099,27 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_mask_fmsub_round_pd_unary_fneg(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_mask_fmsub_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg, 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) {
; X86-LABEL: test_mm512_maskz_fmsub_round_pd:
; X86: # %bb.0: # %entry
@@ -3120,6 +3141,27 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_maskz_fmsub_round_pd_unary_fneg(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_maskz_fmsub_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg, 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) {
; X86-LABEL: test_mm512_fnmadd_round_pd:
; X86: # %bb.0: # %entry
@@ -3138,6 +3180,24 @@ entry:
ret <8 x double> %0
}
+define <8 x double> @test_mm512_fnmadd_round_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_fnmadd_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm0, %zmm0
+; X86-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fnmadd_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0
+; X64-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__A
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %neg, <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) {
; X86-LABEL: test_mm512_mask3_fnmadd_round_pd:
; X86: # %bb.0: # %entry
@@ -3161,6 +3221,29 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_mask3_fnmadd_round_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm512_mask3_fnmadd_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X86-NEXT: vmovapd %zmm2, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_round_pd_unary_fneg:
+; 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:
+ %neg = fneg <8 x double> %__A
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %neg, <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) {
; X86-LABEL: test_mm512_maskz_fnmadd_round_pd:
; X86: # %bb.0: # %entry
@@ -3182,6 +3265,27 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_maskz_fnmadd_round_pd_unary_fneg(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmadd_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__A
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %neg, <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) {
; CHECK-LABEL: test_mm512_fnmsub_round_pd:
; CHECK: # %bb.0: # %entry
@@ -3197,6 +3301,21 @@ entry:
ret <8 x double> %0
}
+define <8 x double> @test_mm512_fnmsub_round_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_round_pd_unary_fneg:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vpbroadcastq {{.*#+}} zmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
+; CHECK-NEXT: vpxorq %zmm3, %zmm0, %zmm4
+; CHECK-NEXT: vpxorq %zmm3, %zmm2, %zmm0
+; CHECK-NEXT: vfmadd231pd {rn-sae}, %zmm4, %zmm1, %zmm0
+; CHECK-NEXT: ret{{[l|q]}}
+entry:
+ %neg = fneg <8 x double> %__A
+ %neg1 = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %neg, <8 x double> %__B, <8 x double> %neg1, 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) {
; X86-LABEL: test_mm512_maskz_fnmsub_round_pd:
; X86: # %bb.0: # %entry
@@ -3219,6 +3338,28 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_maskz_fnmsub_round_pd_unary_fneg(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmsub_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__A
+ %neg1 = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %neg, <8 x double> %__B, <8 x double> %neg1, 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) {
; CHECK-LABEL: test_mm512_fmadd_pd:
; CHECK: # %bb.0: # %entry
@@ -3309,6 +3450,24 @@ entry:
ret <8 x double> %0
}
+define <8 x double> @test_mm512_fmsub_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_fmsub_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2
+; X86-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fmsub_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; X64-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg.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) {
; X86-LABEL: test_mm512_mask_fmsub_pd:
; X86: # %bb.0: # %entry
@@ -3330,6 +3489,27 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_mask_fmsub_pd_unary_fneg(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_mask_fmsub_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub132pd {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg.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) {
; X86-LABEL: test_mm512_maskz_fmsub_pd:
; X86: # %bb.0: # %entry
@@ -3351,6 +3531,27 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_maskz_fmsub_pd_unary_fneg(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_maskz_fmsub_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub213pd {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg.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) {
; X86-LABEL: test_mm512_fnmadd_pd:
; X86: # %bb.0: # %entry
@@ -3369,6 +3570,24 @@ entry:
ret <8 x double> %0
}
+define <8 x double> @test_mm512_fnmadd_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_fnmadd_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm0, %zmm0
+; X86-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fnmadd_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm0, %zmm0
+; X64-NEXT: vfmadd213pd {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <8 x double> %__A
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %neg.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) {
; X86-LABEL: test_mm512_mask3_fnmadd_pd:
; X86: # %bb.0: # %entry
@@ -3392,6 +3611,29 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_mask3_fnmadd_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C, i8 zeroext %__U) {
+; X86-LABEL: test_mm512_mask3_fnmadd_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd231pd {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2
+; X86-NEXT: vmovapd %zmm2, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_pd_unary_fneg:
+; 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:
+ %neg.i = fneg <8 x double> %__A
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %neg.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) {
; X86-LABEL: test_mm512_maskz_fnmadd_pd:
; X86: # %bb.0: # %entry
@@ -3413,6 +3655,27 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_maskz_fnmadd_pd_unary_fneg(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmadd_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmadd213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <8 x double> %__A
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %neg.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) {
; CHECK-LABEL: test_mm512_fnmsub_pd:
; CHECK: # %bb.0: # %entry
@@ -3428,6 +3691,21 @@ entry:
ret <8 x double> %0
}
+define <8 x double> @test_mm512_fnmsub_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_pd_unary_fneg:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vpbroadcastq {{.*#+}} zmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
+; CHECK-NEXT: vpxorq %zmm3, %zmm0, %zmm4
+; CHECK-NEXT: vpxorq %zmm3, %zmm2, %zmm0
+; CHECK-NEXT: vfmadd231pd {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0
+; CHECK-NEXT: ret{{[l|q]}}
+entry:
+ %neg.i = fneg <8 x double> %__A
+ %neg1.i = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %neg.i, <8 x double> %__B, <8 x double> %neg1.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) {
; X86-LABEL: test_mm512_maskz_fnmsub_pd:
; X86: # %bb.0: # %entry
@@ -3450,6 +3728,28 @@ entry:
ret <8 x double> %2
}
+define <8 x double> @test_mm512_maskz_fnmsub_pd_unary_fneg(i8 zeroext %__U, <8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmsub_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmsub213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmsub213pd {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <8 x double> %__A
+ %neg1.i = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.fma.v8f64(<8 x double> %neg.i, <8 x double> %__B, <8 x double> %neg1.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) {
; CHECK-LABEL: test_mm512_fmadd_round_ps:
; CHECK: # %bb.0: # %entry
@@ -3542,6 +3842,24 @@ entry:
ret <16 x float> %0
}
+define <16 x float> @test_mm512_fmsub_round_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_fmsub_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2
+; X86-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fmsub_round_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; X64-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT: retq
+entry:
+ %neg = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %neg, 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) {
; X86-LABEL: test_mm512_mask_fmsub_round_ps:
; X86: # %bb.0: # %entry
@@ -3563,6 +3881,27 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_mask_fmsub_round_ps_unary_fneg(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_mask_fmsub_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_round_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %neg, 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) {
; X86-LABEL: test_mm512_maskz_fmsub_round_ps:
; X86: # %bb.0: # %entry
@@ -3584,6 +3923,27 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_maskz_fmsub_round_ps_unary_fneg(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_maskz_fmsub_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_round_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %__A, <16 x float> %__B, <16 x float> %neg, 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) {
; X86-LABEL: test_mm512_fnmadd_round_ps:
; X86: # %bb.0: # %entry
@@ -3602,6 +3962,24 @@ entry:
ret <16 x float> %0
}
+define <16 x float> @test_mm512_fnmadd_round_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_fnmadd_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm0, %zmm0
+; X86-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fnmadd_round_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm0
+; X64-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT: retq
+entry:
+ %neg = fneg <16 x float> %__A
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %neg, <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) {
; X86-LABEL: test_mm512_mask3_fnmadd_round_ps:
; X86: # %bb.0: # %entry
@@ -3625,6 +4003,29 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_mask3_fnmadd_round_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X86-LABEL: test_mm512_mask3_fnmadd_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
+; X86-NEXT: vmovaps %zmm2, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_round_ps_unary_fneg:
+; 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:
+ %neg = fneg <16 x float> %__A
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %neg, <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) {
; X86-LABEL: test_mm512_maskz_fnmadd_round_ps:
; X86: # %bb.0: # %entry
@@ -3646,6 +4047,27 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_maskz_fnmadd_round_ps_unary_fneg(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmadd_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_round_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <16 x float> %__A
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %neg, <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) {
; CHECK-LABEL: test_mm512_fnmsub_round_ps:
; CHECK: # %bb.0: # %entry
@@ -3661,6 +4083,21 @@ entry:
ret <16 x float> %0
}
+define <16 x float> @test_mm512_fnmsub_round_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_round_ps_unary_fneg:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vpbroadcastd {{.*#+}} zmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
+; CHECK-NEXT: vpxord %zmm3, %zmm0, %zmm4
+; CHECK-NEXT: vpxord %zmm3, %zmm2, %zmm0
+; CHECK-NEXT: vfmadd231ps {rn-sae}, %zmm4, %zmm1, %zmm0
+; CHECK-NEXT: ret{{[l|q]}}
+entry:
+ %neg = fneg <16 x float> %__A
+ %neg1 = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %neg, <16 x float> %__B, <16 x float> %neg1, 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) {
; X86-LABEL: test_mm512_maskz_fnmsub_round_ps:
; X86: # %bb.0: # %entry
@@ -3683,6 +4120,28 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_maskz_fnmsub_round_ps_unary_fneg(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmsub_round_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_round_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmsub213ps {rn-sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <16 x float> %__A
+ %neg1 = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %neg, <16 x float> %__B, <16 x float> %neg1, 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) {
; CHECK-LABEL: test_mm512_fmadd_ps:
; CHECK: # %bb.0: # %entry
@@ -3773,6 +4232,24 @@ entry:
ret <16 x float> %0
}
+define <16 x float> @test_mm512_fmsub_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_fmsub_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm2, %zmm2
+; X86-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fmsub_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; X64-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %neg.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) {
; X86-LABEL: test_mm512_mask_fmsub_ps:
; X86: # %bb.0: # %entry
@@ -3794,6 +4271,27 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_mask_fmsub_ps_unary_fneg(<16 x float> %__A, i16 zeroext %__U, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_mask_fmsub_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask_fmsub_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub132ps {{.*#+}} zmm0 = (zmm0 * zmm1) - zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %neg.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) {
; X86-LABEL: test_mm512_maskz_fmsub_ps:
; X86: # %bb.0: # %entry
@@ -3815,6 +4313,27 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_maskz_fmsub_ps_unary_fneg(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_maskz_fmsub_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fmsub_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsub213ps {{.*#+}} zmm0 = (zmm1 * zmm0) - zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %__A, <16 x float> %__B, <16 x float> %neg.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) {
; X86-LABEL: test_mm512_fnmadd_ps:
; X86: # %bb.0: # %entry
@@ -3833,6 +4352,24 @@ entry:
ret <16 x float> %0
}
+define <16 x float> @test_mm512_fnmadd_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_fnmadd_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxord {{\.LCPI.*}}{1to16}, %zmm0, %zmm0
+; X86-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fnmadd_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxord {{.*}}(%rip){1to16}, %zmm0, %zmm0
+; X64-NEXT: vfmadd213ps {{.*#+}} zmm0 = (zmm1 * zmm0) + zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <16 x float> %__A
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %neg.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) {
; X86-LABEL: test_mm512_mask3_fnmadd_ps:
; X86: # %bb.0: # %entry
@@ -3856,6 +4393,29 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_mask3_fnmadd_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C, i16 zeroext %__U) {
+; X86-LABEL: test_mm512_mask3_fnmadd_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd231ps {{.*#+}} zmm2 = -(zmm0 * zmm1) + zmm2
+; X86-NEXT: vmovaps %zmm2, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask3_fnmadd_ps_unary_fneg:
+; 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:
+ %neg.i = fneg <16 x float> %__A
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %neg.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) {
; X86-LABEL: test_mm512_maskz_fnmadd_ps:
; X86: # %bb.0: # %entry
@@ -3877,6 +4437,27 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_maskz_fnmadd_ps_unary_fneg(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmadd_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmadd213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmadd_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmadd213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) + zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <16 x float> %__A
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %neg.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) {
; CHECK-LABEL: test_mm512_fnmsub_ps:
; CHECK: # %bb.0: # %entry
@@ -3892,6 +4473,21 @@ entry:
ret <16 x float> %0
}
+define <16 x float> @test_mm512_fnmsub_ps_unary_fneg(<16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; CHECK-LABEL: test_mm512_fnmsub_ps_unary_fneg:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vpbroadcastd {{.*#+}} zmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]
+; CHECK-NEXT: vpxord %zmm3, %zmm0, %zmm4
+; CHECK-NEXT: vpxord %zmm3, %zmm2, %zmm0
+; CHECK-NEXT: vfmadd231ps {{.*#+}} zmm0 = (zmm1 * zmm4) + zmm0
+; CHECK-NEXT: ret{{[l|q]}}
+entry:
+ %neg.i = fneg <16 x float> %__A
+ %neg1.i = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %neg.i, <16 x float> %__B, <16 x float> %neg1.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) {
; X86-LABEL: test_mm512_maskz_fnmsub_ps:
; X86: # %bb.0: # %entry
@@ -3914,6 +4510,28 @@ entry:
ret <16 x float> %2
}
+define <16 x float> @test_mm512_maskz_fnmsub_ps_unary_fneg(i16 zeroext %__U, <16 x float> %__A, <16 x float> %__B, <16 x float> %__C) {
+; X86-LABEL: test_mm512_maskz_fnmsub_ps_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfnmsub213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_maskz_fnmsub_ps_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfnmsub213ps {{.*#+}} zmm0 = -(zmm1 * zmm0) - zmm2
+; X64-NEXT: retq
+entry:
+ %neg.i = fneg <16 x float> %__A
+ %neg1.i = fneg <16 x float> %__C
+ %0 = tail call <16 x float> @llvm.fma.v16f32(<16 x float> %neg.i, <16 x float> %__B, <16 x float> %neg1.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) {
; CHECK-LABEL: test_mm512_fmaddsub_round_pd:
; CHECK: # %bb.0: # %entry
@@ -4006,6 +4624,24 @@ entry:
ret <8 x double> %0
}
+define <8 x double> @test_mm512_fmsubadd_round_pd_unary_fneg(<8 x double> %__A, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_fmsubadd_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: vpxorq {{\.LCPI.*}}{1to8}, %zmm2, %zmm2
+; X86-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_fmsubadd_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: vpxorq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; X64-NEXT: vfmaddsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg, 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) {
; X86-LABEL: test_mm512_mask_fmsubadd_round_pd:
; X86: # %bb.0: # %entry
@@ -4025,6 +4661,27 @@ entry:
%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_fmsubadd_round_pd_unary_fneg(<8 x double> %__A, i8 zeroext %__U, <8 x double> %__B, <8 x double> %__C) {
+; X86-LABEL: test_mm512_mask_fmsubadd_round_pd_unary_fneg:
+; X86: # %bb.0: # %entry
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: kmovw %eax, %k1
+; X86-NEXT: vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X86-NEXT: retl
+;
+; X64-LABEL: test_mm512_mask_fmsubadd_round_pd_unary_fneg:
+; X64: # %bb.0: # %entry
+; X64-NEXT: kmovw %edi, %k1
+; X64-NEXT: vfmsubadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1}
+; X64-NEXT: retq
+entry:
+ %neg = fneg <8 x double> %__C
+ %0 = tail call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %__A, <8 x double> %__B, <8 x double> %neg, 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) {
More information about the llvm-commits
mailing list