[X86][AVX2] bugzilla bug 21281 Performance regression in vector interleave in AVX2

This is a patch for an on-going bugzilla bug 21281 on the generated X86 code for a matrix transpose8x8 subroutine which requires vector interleaving. The generated code in AVX2 is currently non-optimal and requires 60 instructions as opposed to only 40 instructions generated for AVX1.
 The patch includes a fix for the AVX2 case where vector unpack instructions use less operations than the vector blend operations available in AVX2.
 In this case using vector unpack instructions is more efficient.

Reviewers:
zvi  
delena  
igorb  
craig.topper  
guyblank  
eladcohen  
m_zuckerman  
aymanmus  
RKSimon 

llvm-svn: 298840

1 files changed, 25 insertions, 37 deletions

diff --git a/llvm/test/CodeGen/X86/vector-interleave.ll b/llvm/test/CodeGen/X86/vector-interleave.ll
index 4f9dbb03fb1..1265ea10897 100644
--- a/llvm/test/CodeGen/X86/vector-interleave.ll
+++ b/llvm/test/CodeGen/X86/vector-interleave.ll
@@ -93,44 +93,32 @@ define <64 x i16> @interleave8x8(<8 x i16> %a, <8 x i16> %b, <8 x i16> %c, <8 x
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm2, %ymm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,4,u,5,u,6,u,7>
-; AVX2-NEXT:    vpermd %ymm1, %ymm9, %ymm3
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm0[0],ymm3[1],ymm0[2],ymm3[3],ymm0[4],ymm3[5],ymm0[6],ymm3[7]
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,0,u,1,u,2,u,3>
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm8 = xmm8[0],zero,xmm8[1],zero,xmm8[2],zero,xmm8[3],zero
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0],ymm1[1],ymm8[2],ymm1[3],ymm8[4],ymm1[5],ymm8[6],ymm1[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm8[2],xmm1[2],xmm8[3],xmm1[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm8[0],xmm1[0],xmm8[1],xmm1[1]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm5[4],xmm4[5],xmm5[5],xmm4[6],xmm5[6],xmm4[7],xmm5[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm6, %ymm5
-; AVX2-NEXT:    vpermd %ymm5, %ymm9, %ymm6
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm6[1],ymm4[2],ymm6[3],ymm4[4],ymm6[5],ymm4[6],ymm6[7]
-; AVX2-NEXT:    vpermd %ymm5, %ymm0, %ymm0
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4],ymm0[5],ymm2[6],ymm0[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm5, %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm2
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm1, %ymm1
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
-; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm3
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm3, %ymm3
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm4[2],xmm6[2],xmm4[3],xmm6[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm1[2],xmm5[2],xmm1[3],xmm5[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm3[4],xmm6[4],xmm3[5],xmm6[5],xmm3[6],xmm6[6],xmm3[7],xmm6[7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm6[0],xmm3[1],xmm6[1],xmm3[2],xmm6[2],xmm3[3],xmm6[3]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm9[4],xmm7[4],xmm9[5],xmm7[5],xmm9[6],xmm7[6],xmm9[7],xmm7[7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm9[0],xmm7[0],xmm9[1],xmm7[1],xmm9[2],xmm7[2],xmm9[3],xmm7[3]
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
 ; AVX2-NEXT:    retq
   %ab = shufflevector <8 x i16> %a, <8 x i16> %b, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>
   %cd = shufflevector <8 x i16> %c, <8 x i16> %d, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>