From bce1f6b49196e174f299e577a42151e2adbf49d8 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sun, 2 Oct 2016 20:43:02 +0000
Subject: [PATCH] [X86][AVX2] Missed opportunities to combine to VPERMD/VPERMPS

llvm-svn: 283077
---
 .../CodeGen/X86/vector-shuffle-combining-avx2.ll   | 48 ++++++++++++++++++++++
 1 file changed, 48 insertions(+)

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index bea9c4a..72eff32 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -89,6 +89,54 @@ define <4 x i64> @combine_permq_pshufb_as_vperm2i128(<4 x i64> %a0) {
   ret <4 x i64> %5
 }
 
+define <8 x i32> @combine_as_vpermd(<8 x i32> %a0) {
+; X32-LABEL: combine_as_vpermd:
+; X32:       # BB#0:
+; X32-NEXT:    vmovdqa {{.*#+}} ymm1 = <4,u,u,5,u,u,0,7>
+; X32-NEXT:    vpermd %ymm0, %ymm1, %ymm1
+; X32-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
+; X32-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4,5],ymm1[6,7]
+; X32-NEXT:    retl
+;
+; X64-LABEL: combine_as_vpermd:
+; X64:       # BB#0:
+; X64-NEXT:    vmovdqa {{.*#+}} ymm1 = <4,u,u,5,u,u,0,7>
+; X64-NEXT:    vpermd %ymm0, %ymm1, %ymm1
+; X64-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
+; X64-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4,5],ymm1[6,7]
+; X64-NEXT:    retq
+  %1 = shufflevector <8 x i32> %a0, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+  %2 = tail call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %a0, <8 x i32> <i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 7, i32 6>)
+  %3 = shufflevector <8 x i32> %1, <8 x i32> %2, <8 x i32> <i32 0, i32 8, i32 9, i32 1, i32 15, i32 14, i32 4, i32 3>
+  ret <8 x i32> %3
+}
+
+define <8 x float> @combine_as_vpermps(<8 x float> %a0) {
+; X32-LABEL: combine_as_vpermps:
+; X32:       # BB#0:
+; X32-NEXT:    vpermilps {{.*#+}} ymm1 = ymm0[1,0,3,2,5,4,7,6]
+; X32-NEXT:    vmovaps {{.*#+}} ymm2 = <u,4,u,5,u,u,4,7>
+; X32-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; X32-NEXT:    vmovaps {{.*#+}} ymm2 = <7,u,6,u,0,1,u,u>
+; X32-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; X32-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6,7]
+; X32-NEXT:    retl
+;
+; X64-LABEL: combine_as_vpermps:
+; X64:       # BB#0:
+; X64-NEXT:    vpermilps {{.*#+}} ymm1 = ymm0[1,0,3,2,5,4,7,6]
+; X64-NEXT:    vmovaps {{.*#+}} ymm2 = <u,4,u,5,u,u,4,7>
+; X64-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; X64-NEXT:    vmovaps {{.*#+}} ymm2 = <7,u,6,u,0,1,u,u>
+; X64-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; X64-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6,7]
+; X64-NEXT:    retq
+  %1 = shufflevector <8 x float> %a0, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
+  %2 = tail call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a0, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>)
+  %3 = shufflevector <8 x float> %1, <8 x float> %2, <8 x i32> <i32 15, i32 0, i32 14, i32 1, i32 8, i32 9, i32 4, i32 3>
+  ret <8 x float> %3
+}
+
 define <32 x i8> @combine_permq_pshufb_as_vpblendd(<4 x i64> %a0) {
 ; X32-LABEL: combine_permq_pshufb_as_vpblendd:
 ; X32:       # BB#0:
-- 
2.7.4