summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen')
-rw-r--r--llvm/test/CodeGen/X86/fma-commute-x86.ll506
-rw-r--r--llvm/test/CodeGen/X86/fma_patterns.ll21
2 files changed, 515 insertions, 12 deletions
diff --git a/llvm/test/CodeGen/X86/fma-commute-x86.ll b/llvm/test/CodeGen/X86/fma-commute-x86.ll
new file mode 100644
index 00000000000..7b0f13e5b03
--- /dev/null
+++ b/llvm/test/CodeGen/X86/fma-commute-x86.ll
@@ -0,0 +1,506 @@
+; RUN: llc < %s -mtriple=x86_64-pc-win32 -mcpu=core-avx2 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-pc-win32 -mattr=+fma,+fma4 | FileCheck %s
+; RUN: llc < %s -mcpu=bdver2 -mtriple=x86_64-pc-win32 -mattr=-fma4 | FileCheck %s
+
+attributes #0 = { nounwind }
+
+declare <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
+define <4 x float> @test_x86_fmadd_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_baa_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfmadd132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fmadd_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_aba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfmadd231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fmadd_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_bba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %xmm0
+; CHECK-NEXT: vfmadd213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>) nounwind readnone
+define <8 x float> @test_x86_fmadd_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_baa_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfmadd132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fmadd_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_aba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfmadd231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fmadd_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_bba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %ymm0
+; CHECK-NEXT: vfmadd213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+declare <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
+define <2 x double> @test_x86_fmadd_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_baa_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfmadd132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fmadd_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_aba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfmadd231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fmadd_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_bba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %xmm0
+; CHECK-NEXT: vfmadd213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
+define <4 x double> @test_x86_fmadd_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_baa_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfmadd132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fmadd_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_aba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfmadd231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fmadd_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmadd_bba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %ymm0
+; CHECK-NEXT: vfmadd213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+
+
+declare <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
+define <4 x float> @test_x86_fnmadd_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_baa_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfnmadd132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fnmadd_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_aba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfnmadd231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fnmadd_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_bba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %xmm0
+; CHECK-NEXT: vfnmadd213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x float>) nounwind readnone
+define <8 x float> @test_x86_fnmadd_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_baa_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfnmadd132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fnmadd_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_aba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfnmadd231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fnmadd_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_bba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %ymm0
+; CHECK-NEXT: vfnmadd213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+declare <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
+define <2 x double> @test_x86_fnmadd_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_baa_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfnmadd132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fnmadd_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_aba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfnmadd231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fnmadd_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_bba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %xmm0
+; CHECK-NEXT: vfnmadd213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
+define <4 x double> @test_x86_fnmadd_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_baa_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfnmadd132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fnmadd_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_aba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfnmadd231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fnmadd_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmadd_bba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %ymm0
+; CHECK-NEXT: vfnmadd213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+
+declare <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
+define <4 x float> @test_x86_fmsub_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_baa_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfmsub132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fmsub_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_aba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfmsub231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fmsub_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_bba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %xmm0
+; CHECK-NEXT: vfmsub213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>) nounwind readnone
+define <8 x float> @test_x86_fmsub_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_baa_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfmsub132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fmsub_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_aba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfmsub231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fmsub_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_bba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %ymm0
+; CHECK-NEXT: vfmsub213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+declare <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
+define <2 x double> @test_x86_fmsub_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_baa_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfmsub132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fmsub_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_aba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfmsub231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fmsub_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_bba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %xmm0
+; CHECK-NEXT: vfmsub213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
+define <4 x double> @test_x86_fmsub_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_baa_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfmsub132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fmsub_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_aba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfmsub231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fmsub_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fmsub_bba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %ymm0
+; CHECK-NEXT: vfmsub213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+
+declare <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
+define <4 x float> @test_x86_fnmsub_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_baa_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfnmsub132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fnmsub_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_aba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %xmm0
+; CHECK-NEXT: vfnmsub231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+define <4 x float> @test_x86_fnmsub_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_bba_ps:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %xmm0
+; CHECK-NEXT: vfnmsub213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
+ ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>) nounwind readnone
+define <8 x float> @test_x86_fnmsub_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_baa_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfnmsub132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fnmsub_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_aba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rcx), %ymm0
+; CHECK-NEXT: vfnmsub231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+define <8 x float> @test_x86_fnmsub_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_bba_ps_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovaps (%rdx), %ymm0
+; CHECK-NEXT: vfnmsub213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
+ ret <8 x float> %res
+}
+
+declare <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
+define <2 x double> @test_x86_fnmsub_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_baa_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfnmsub132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fnmsub_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_aba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %xmm0
+; CHECK-NEXT: vfnmsub231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+define <2 x double> @test_x86_fnmsub_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_bba_pd:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %xmm0
+; CHECK-NEXT: vfnmsub213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT: retq
+ %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
+ ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
+define <4 x double> @test_x86_fnmsub_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_baa_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfnmsub132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fnmsub_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_aba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rcx), %ymm0
+; CHECK-NEXT: vfnmsub231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
+define <4 x double> @test_x86_fnmsub_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
+; CHECK-LABEL: test_x86_fnmsub_bba_pd_y:
+; CHECK: # BB#0:
+; CHECK-NEXT: vmovapd (%rdx), %ymm0
+; CHECK-NEXT: vfnmsub213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT: retq
+ %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
+ ret <4 x double> %res
+}
+
diff --git a/llvm/test/CodeGen/X86/fma_patterns.ll b/llvm/test/CodeGen/X86/fma_patterns.ll
index 383ecfcd5e3..6a42744543a 100644
--- a/llvm/test/CodeGen/X86/fma_patterns.ll
+++ b/llvm/test/CodeGen/X86/fma_patterns.ll
@@ -235,11 +235,10 @@ define float @test_x86_fnmsub_ss(float %a0, float %a1, float %a2) {
}
define <4 x float> @test_x86_fmadd_ps_load(<4 x float>* %a0, <4 x float> %a1, <4 x float> %a2) {
-; CHECK_FMA-LABEL: test_x86_fmadd_ps_load:
-; CHECK_FMA: # BB#0:
-; CHECK_FMA-NEXT: vmovaps (%rdi), %xmm2
-; CHECK_FMA-NEXT: vfmadd213ps %xmm1, %xmm2, %xmm0
-; CHECK_FMA-NEXT: retq
+; CHECK-FMA-LABEL: test_x86_fmadd_ps_load:
+; CHECK-FMA: # BB#0:
+; CHECK-FMA-NEXT: vfmadd132ps (%rdi), %xmm1, %xmm0
+; CHECK-FMA-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmadd_ps_load:
; CHECK_FMA4: # BB#0:
@@ -252,11 +251,10 @@ define <4 x float> @test_x86_fmadd_ps_load(<4 x float>* %a0, <4 x float> %a1, <4
}
define <4 x float> @test_x86_fmsub_ps_load(<4 x float>* %a0, <4 x float> %a1, <4 x float> %a2) {
-; CHECK_FMA-LABEL: test_x86_fmsub_ps_load:
-; CHECK_FMA: # BB#0:
-; CHECK_FMA-NEXT: vmovaps (%rdi), %xmm2
-; CHECK_FMA-NEXT: vfmsub213ps %xmm1, %xmm2, %xmm0
-; CHECK_FMA-NEXT: retq
+; CHECK-FMA-LABEL: test_x86_fmsub_ps_load:
+; CHECK-FMA: # BB#0:
+; CHECK-FMA-NEXT: vfmsub132ps (%rdi), %xmm1, %xmm0
+; CHECK-FMA-NEXT: retq
;
; CHECK_FMA4-LABEL: test_x86_fmsub_ps_load:
; CHECK_FMA4: # BB#0:
@@ -588,8 +586,7 @@ define <4 x float> @test_v4f32_fma_x_c1_fmul_x_c2(<4 x float> %x) #0 {
define <4 x float> @test_v4f32_fma_fmul_x_c1_c2_y(<4 x float> %x, <4 x float> %y) #0 {
; CHECK_FMA-LABEL: test_v4f32_fma_fmul_x_c1_c2_y:
; CHECK_FMA: # BB#0:
-; CHECK_FMA-NEXT: vmovaps {{.*#+}} xmm2 = [4.000000e+00,6.000000e+00,6.000000e+00,4.000000e+00]
-; CHECK_FMA-NEXT: vfmadd213ps %xmm1, %xmm2, %xmm0
+; CHECK_FMA-NEXT: vfmadd132ps {{.*}}(%rip), %xmm1, %xmm0
; CHECK_FMA-NEXT: retq
;
; CHECK_FMA4-LABEL: test_v4f32_fma_fmul_x_c1_c2_y:
OpenPOWER on IntegriCloud