[llvm] aa3fcb9 - [X86][AVX] Add extra vperm2f128+vpermilvar combine coverage

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2020-09-07T10:58:53+01:00
New Revision: aa3fcb967110f2d448d241358cadc048954e6134

URL: https://github.com/llvm/llvm-project/commit/aa3fcb967110f2d448d241358cadc048954e6134
DIFF: https://github.com/llvm/llvm-project/commit/aa3fcb967110f2d448d241358cadc048954e6134.diff

LOG: [X86][AVX] Add extra vperm2f128+vpermilvar combine coverage

The existing test /should/ reduce to a vmovaps (concat xmm with zero upper).

Added: 
    

Modified: 
    llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
index 6ffbe095c39b..d4ef76a2a9cf 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
@@ -142,8 +142,19 @@ define <8 x float> @combine_vpermilvar_vperm2f128_zero_8f32(<8 x float> %a0) {
   ret <8 x float> %3
 }
 
-define <4 x double> @combine_vperm2f128_vpermilvar_as_vpblendpd(<4 x double> %a0) {
-; CHECK-LABEL: combine_vperm2f128_vpermilvar_as_vpblendpd:
+define <4 x double> @combine_vperm2f128_vpermilvar_as_vperm2f128(<4 x double> %a0) {
+; CHECK-LABEL: combine_vperm2f128_vpermilvar_as_vperm2f128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]
+; CHECK-NEXT:    ret{{[l|q]}}
+  %1 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)
+  %2 = shufflevector <4 x double> %1, <4 x double> zeroinitializer, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
+  %3 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %2, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)
+  ret <4 x double> %3
+}
+
+define <4 x double> @combine_vperm2f128_vpermilvar_as_vmovaps(<4 x double> %a0) {
+; CHECK-LABEL: combine_vperm2f128_vpermilvar_as_vmovaps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
 ; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]