Mercurial > hg > CbC > CbC_llvm

diff test/CodeGen/X86/avx512-fma.ll @ 100:7d135dc70f03 LLVM 3.9

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 3.9
author Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
date Tue, 26 Jan 2016 22:53:40 +0900
parents afa8332a0e37
children 1172e4bd9c6f
line wrap: on
line diff
--- a/test/CodeGen/X86/avx512-fma.ll	Tue Oct 13 17:49:56 2015 +0900
+++ b/test/CodeGen/X86/avx512-fma.ll	Tue Jan 26 22:53:40 2016 +0900
@@ -1,81 +1,93 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f -fp-contract=fast | FileCheck %s
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx -fp-contract=fast | FileCheck %s --check-prefix=SKX
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f -fp-contract=fast | FileCheck %s --check-prefix=ALL --check-prefix=KNL
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx -fp-contract=fast | FileCheck %s --check-prefix=ALL --check-prefix=SKX
 
-; CHECK-LABEL: test_x86_fmadd_ps_z
-; CHECK: vfmadd213ps     %zmm2, %zmm1, %zmm0
-; CHECK: ret
 define <16 x float> @test_x86_fmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
+; ALL-LABEL: test_x86_fmadd_ps_z:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
   %res = fadd <16 x float> %x, %a2
   ret <16 x float> %res
 }
 
-; CHECK-LABEL: test_x86_fmsub_ps_z
-; CHECK: vfmsub213ps     %zmm2, %zmm1, %zmm0
-; CHECK: ret
 define <16 x float> @test_x86_fmsub_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
+; ALL-LABEL: test_x86_fmsub_ps_z:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmsub213ps %zmm2, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
   %res = fsub <16 x float> %x, %a2
   ret <16 x float> %res
 }
 
-; CHECK-LABEL: test_x86_fnmadd_ps_z
-; CHECK: vfnmadd213ps     %zmm2, %zmm1, %zmm0
-; CHECK: ret
 define <16 x float> @test_x86_fnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
+; ALL-LABEL: test_x86_fnmadd_ps_z:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfnmadd213ps %zmm2, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
   %res = fsub <16 x float> %a2, %x
   ret <16 x float> %res
 }
 
-; CHECK-LABEL: test_x86_fnmsub_ps_z
-; CHECK: vfnmsub213ps     %zmm2, %zmm1, %zmm0
-; CHECK: ret
 define <16 x float> @test_x86_fnmsub_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
+; ALL-LABEL: test_x86_fnmsub_ps_z:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfnmsub213ps %zmm2, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul <16 x float> %a0, %a1
-  %y = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, 
+  %y = 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>, %x
+                          float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00,
+                          float -0.000000e+00>, %x
   %res = fsub <16 x float> %y, %a2
   ret <16 x float> %res
 }
 
-; CHECK-LABEL: test_x86_fmadd_pd_z
-; CHECK: vfmadd213pd     %zmm2, %zmm1, %zmm0
-; CHECK: ret
 define <8 x double> @test_x86_fmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
+; ALL-LABEL: test_x86_fmadd_pd_z:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul <8 x double> %a0, %a1
   %res = fadd <8 x double> %x, %a2
   ret <8 x double> %res
 }
 
-; CHECK-LABEL: test_x86_fmsub_pd_z
-; CHECK: vfmsub213pd     %zmm2, %zmm1, %zmm0
-; CHECK: ret
 define <8 x double> @test_x86_fmsub_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
+; ALL-LABEL: test_x86_fmsub_pd_z:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmsub213pd %zmm2, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul <8 x double> %a0, %a1
   %res = fsub <8 x double> %x, %a2
   ret <8 x double> %res
 }
 
 define double @test_x86_fmsub_213(double %a0, double %a1, double %a2) {
-; CHECK-LABEL: test_x86_fmsub_213:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vfmsub213sd %xmm2, %xmm0, %xmm1
-; CHECK-NEXT:    vmovaps %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; ALL-LABEL: test_x86_fmsub_213:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmsub213sd %xmm2, %xmm0, %xmm1
+; ALL-NEXT:    vmovaps %zmm1, %zmm0
+; ALL-NEXT:    retq
   %x = fmul double %a0, %a1
   %res = fsub double %x, %a2
   ret double %res
 }
 
 define double @test_x86_fmsub_213_m(double %a0, double %a1, double * %a2_ptr) {
-; CHECK-LABEL: test_x86_fmsub_213_m:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vfmsub213sd (%rdi), %xmm0, %xmm1
-; CHECK-NEXT:    vmovaps %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; KNL-LABEL: test_x86_fmsub_213_m:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vfmsub213sd (%rdi), %xmm0, %xmm1
+; KNL-NEXT:    vmovaps %zmm1, %zmm0
+; KNL-NEXT:    retq
+;
+; SKX-LABEL: test_x86_fmsub_213_m:
+; SKX:       ## BB#0:
+; SKX-NEXT:    vfmsub213sd (%rdi), %xmm1, %xmm0
+; SKX-NEXT:    retq
   %a2 = load double , double *%a2_ptr
   %x = fmul double %a0, %a1
   %res = fsub double %x, %a2
@@ -83,11 +95,11 @@
 }
 
 define double @test_x86_fmsub_231_m(double %a0, double %a1, double * %a2_ptr) {
-; CHECK-LABEL: test_x86_fmsub_231_m:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vfmsub231sd (%rdi), %xmm0, %xmm1
-; CHECK-NEXT:    vmovaps %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; ALL-LABEL: test_x86_fmsub_231_m:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmsub231sd (%rdi), %xmm0, %xmm1
+; ALL-NEXT:    vmovaps %zmm1, %zmm0
+; ALL-NEXT:    retq
   %a2 = load double , double *%a2_ptr
   %x = fmul double %a0, %a2
   %res = fsub double %x, %a1
@@ -95,21 +107,21 @@
 }
 
 define <16 x float> @test231_br(<16 x float> %a1, <16 x float> %a2) nounwind {
-; CHECK-LABEL: test231_br:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vfmadd231ps {{.*}}(%rip){1to16}, %zmm0, %zmm1
-; CHECK-NEXT:    vmovaps %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; ALL-LABEL: test231_br:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmadd231ps {{.*}}(%rip){1to16}, %zmm0, %zmm1
+; ALL-NEXT:    vmovaps %zmm1, %zmm0
+; ALL-NEXT:    retq
   %b1 = fmul <16 x float> %a1, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
   %b2 = fadd <16 x float> %b1, %a2
   ret <16 x float> %b2
 }
 
 define <16 x float> @test213_br(<16 x float> %a1, <16 x float> %a2) nounwind {
-; CHECK-LABEL: test213_br:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vfmadd213ps {{.*}}(%rip){1to16}, %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; ALL-LABEL: test213_br:
+; ALL:       ## BB#0:
+; ALL-NEXT:    vfmadd213ps {{.*}}(%rip){1to16}, %zmm1, %zmm0
+; ALL-NEXT:    retq
   %b1 = fmul <16 x float> %a1, %a2
   %b2 = fadd <16 x float> %b1, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
   ret <16 x float> %b2
@@ -117,16 +129,17 @@
 
 ;mask (a*c+b , a)
 define <16 x float> @test_x86_fmadd132_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> *%a2_ptrt, <16 x i1> %mask) {
-; CHECK-LABEL: test_x86_fmadd132_ps:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpmovsxbd %xmm2, %zmm2
-; CHECK-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; CHECK-NEXT:    vptestmd %zmm2, %zmm2, %k1
-; CHECK-NEXT:    vfmadd132ps (%rdi), %zmm1, %zmm0 {%k1}
-; CHECK-NEXT:    retq
+; KNL-LABEL: test_x86_fmadd132_ps:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; KNL-NEXT:    vfmadd132ps (%rdi), %zmm1, %zmm0 {%k1}
+; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_x86_fmadd132_ps:
 ; SKX:       ## BB#0:
+; SKX-NEXT:    vpsllw $7, %xmm2, %xmm2
 ; SKX-NEXT:    vpmovb2m %xmm2, %k1
 ; SKX-NEXT:    vfmadd132ps (%rdi), %zmm1, %zmm0 {%k1}
 ; SKX-NEXT:    retq
@@ -139,17 +152,18 @@
 
 ;mask (a*c+b , b)
 define <16 x float> @test_x86_fmadd231_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> *%a2_ptrt, <16 x i1> %mask) {
-; CHECK-LABEL: test_x86_fmadd231_ps:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpmovsxbd %xmm2, %zmm2
-; CHECK-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; CHECK-NEXT:    vptestmd %zmm2, %zmm2, %k1
-; CHECK-NEXT:    vfmadd231ps (%rdi), %zmm0, %zmm1 {%k1}
-; CHECK-NEXT:    vmovaps %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; KNL-LABEL: test_x86_fmadd231_ps:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; KNL-NEXT:    vfmadd231ps (%rdi), %zmm0, %zmm1 {%k1}
+; KNL-NEXT:    vmovaps %zmm1, %zmm0
+; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_x86_fmadd231_ps:
 ; SKX:       ## BB#0:
+; SKX-NEXT:    vpsllw $7, %xmm2, %xmm2
 ; SKX-NEXT:    vpmovb2m %xmm2, %k1
 ; SKX-NEXT:    vfmadd231ps (%rdi), %zmm0, %zmm1 {%k1}
 ; SKX-NEXT:    vmovaps %zmm1, %zmm0
@@ -163,17 +177,18 @@
 
 ;mask (b*a+c , b)
 define <16 x float> @test_x86_fmadd213_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> *%a2_ptrt, <16 x i1> %mask) {
-; CHECK-LABEL: test_x86_fmadd213_ps:
-; CHECK:       ## BB#0:
-; CHECK-NEXT:    vpmovsxbd %xmm2, %zmm2
-; CHECK-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
-; CHECK-NEXT:    vptestmd %zmm2, %zmm2, %k1
-; CHECK-NEXT:    vfmadd213ps (%rdi), %zmm0, %zmm1 {%k1}
-; CHECK-NEXT:    vmovaps %zmm1, %zmm0
-; CHECK-NEXT:    retq
+; KNL-LABEL: test_x86_fmadd213_ps:
+; KNL:       ## BB#0:
+; KNL-NEXT:    vpmovsxbd %xmm2, %zmm2
+; KNL-NEXT:    vpslld $31, %zmm2, %zmm2
+; KNL-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; KNL-NEXT:    vfmadd213ps (%rdi), %zmm0, %zmm1 {%k1}
+; KNL-NEXT:    vmovaps %zmm1, %zmm0
+; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_x86_fmadd213_ps:
 ; SKX:       ## BB#0:
+; SKX-NEXT:    vpsllw $7, %xmm2, %xmm2
 ; SKX-NEXT:    vpmovb2m %xmm2, %k1
 ; SKX-NEXT:    vfmadd213ps (%rdi), %zmm0, %zmm1 {%k1}
 ; SKX-NEXT:    vmovaps %zmm1, %zmm0