From 3def9e11e2652aff5e084c764320981d9c5e786a Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Mon, 28 Nov 2016 15:50:39 +0000
Subject: [PATCH] [X86][SSE] Added tests showing missed combines of shifts with
 shuffles.

llvm-svn: 288037
---
 .../CodeGen/X86/vector-shuffle-combining-avx2.ll   | 54 ++++++++++++++++++++++
 .../CodeGen/X86/vector-shuffle-combining-ssse3.ll  | 54 ++++++++++++++++++++++
 2 files changed, 108 insertions(+)

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index 89194c3..2be4a06 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -596,6 +596,60 @@ define <32 x i8> @combine_pshufb_not_as_pshufw(<32 x i8> %a0) {
   ret <32 x i8> %res1
 }
 
+define <32 x i8> @combine_psrlw_pshufb(<16 x i16> %a0) {
+; X32-LABEL: combine_psrlw_pshufb:
+; X32:       # BB#0:
+; X32-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X32-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14,17,16,19,18,21,20,23,22,25,24,27,26,29,28,31,30]
+; X32-NEXT:    retl
+;
+; X64-LABEL: combine_psrlw_pshufb:
+; X64:       # BB#0:
+; X64-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X64-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14,17,16,19,18,21,20,23,22,25,24,27,26,29,28,31,30]
+; X64-NEXT:    retq
+  %1 = lshr <16 x i16> %a0, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8>
+  %2 = bitcast <16 x i16> %1 to <32 x i8>
+  %3 = tail call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %2, <32 x i8> <i8 1, i8 0, i8 3, i8 2, i8 5, i8 4, i8 7, i8 6, i8 9, i8 8, i8 11, i8 10, i8 13, i8 12, i8 15, i8 14, i8 17, i8 16, i8 19, i8 18, i8 21, i8 20, i8 23, i8 22, i8 25, i8 24, i8 27, i8 26, i8 29, i8 28, i8 31, i8 30>)
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @combine_pslld_pshufb(<8 x i32> %a0) {
+; X32-LABEL: combine_pslld_pshufb:
+; X32:       # BB#0:
+; X32-NEXT:    vpslld $24, %ymm0, %ymm0
+; X32-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12,19,18,17,16,23,22,21,20,27,26,25,24,31,30,29,28]
+; X32-NEXT:    retl
+;
+; X64-LABEL: combine_pslld_pshufb:
+; X64:       # BB#0:
+; X64-NEXT:    vpslld $24, %ymm0, %ymm0
+; X64-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12,19,18,17,16,23,22,21,20,27,26,25,24,31,30,29,28]
+; X64-NEXT:    retq
+  %1 = shl <8 x i32> %a0, <i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24>
+  %2 = bitcast <8 x i32> %1 to <32 x i8>
+  %3 = tail call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %2, <32 x i8> <i8 3, i8 2, i8 1, i8 0, i8 7, i8 6, i8 5, i8 4, i8 11, i8 10, i8 9, i8 8, i8 15, i8 14, i8 13, i8 12, i8 19, i8 18, i8 17, i8 16, i8 23, i8 22, i8 21, i8 20, i8 27, i8 26, i8 25, i8 24, i8 31, i8 30, i8 29, i8 28>)
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @combine_psrlq_pshufb(<4 x i64> %a0) {
+; X32-LABEL: combine_psrlq_pshufb:
+; X32:       # BB#0:
+; X32-NEXT:    vpsrlq $32, %ymm0, %ymm0
+; X32-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,23,22,21,20,19,18,17,31,30,29,28,27,26,25,24,23]
+; X32-NEXT:    retl
+;
+; X64-LABEL: combine_psrlq_pshufb:
+; X64:       # BB#0:
+; X64-NEXT:    vpsrlq $32, %ymm0, %ymm0
+; X64-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,23,22,21,20,19,18,17,31,30,29,28,27,26,25,24,23]
+; X64-NEXT:    retq
+  %1 = lshr <4 x i64> %a0, <i64 32, i64 32, i64 32, i64 32>
+  %2 = bitcast <4 x i64> %1 to <32 x i8>
+  %3 = tail call <32 x i8> @llvm.x86.avx2.pshuf.b(<32 x i8> %2, <32 x i8> <i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 23, i8 22, i8 21, i8 20, i8 19, i8 18, i8 17, i8 31, i8 30, i8 29, i8 28, i8 27, i8 26, i8 25, i8 24, i8 23>)
+  ret <32 x i8> %3
+}
+
 define <8 x i32> @constant_fold_permd() {
 ; X32-LABEL: constant_fold_permd:
 ; X32:       # BB#0:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
index 7676e83..277d541 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-ssse3.ll
@@ -409,6 +409,60 @@ define <16 x i8> @combine_pshufb_as_unary_unpckhwd(<16 x i8> %a0) {
   ret <16 x i8> %1
 }
 
+define <16 x i8> @combine_psrlw_pshufb(<8 x i16> %a0) {
+; SSE-LABEL: combine_psrlw_pshufb:
+; SSE:       # BB#0:
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[0],zero,zero,zero,xmm0[0],zero,zero,zero,xmm0[0],zero,zero,zero
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_psrlw_pshufb:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; AVX-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[0],zero,zero,zero,xmm0[0],zero,zero,zero,xmm0[0],zero,zero,zero
+; AVX-NEXT:    retq
+  %1 = lshr <8 x i16> %a0, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8>
+  %2 = bitcast <8 x i16> %1 to <16 x i8>
+  %3 = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %2, <16 x i8> <i8 0, i8 -1, i8 -1, i8 -1, i8 0, i8 -1, i8 -1, i8 -1, i8 0, i8 -1, i8 -1, i8 -1, i8 0, i8 -1, i8 -1, i8 -1>)
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @combine_pslld_pshufb(<4 x i32> %a0) {
+; SSE-LABEL: combine_pslld_pshufb:
+; SSE:       # BB#0:
+; SSE-NEXT:    pslld $8, %xmm0
+; SSE-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_pslld_pshufb:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpslld $8, %xmm0, %xmm0
+; AVX-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
+; AVX-NEXT:    retq
+  %1 = shl <4 x i32> %a0, <i32 8, i32 8, i32 8, i32 8>
+  %2 = bitcast <4 x i32> %1 to <16 x i8>
+  %3 = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %2, <16 x i8> <i8 3, i8 2, i8 1, i8 0, i8 7, i8 6, i8 5, i8 4, i8 11, i8 10, i8 9, i8 8, i8 15, i8 14, i8 13, i8 12>)
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @combine_psrlq_pshufb(<2 x i64> %a0) {
+; SSE-LABEL: combine_psrlq_pshufb:
+; SSE:       # BB#0:
+; SSE-NEXT:    psrlq $48, %xmm0
+; SSE-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_psrlq_pshufb:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpsrlq $48, %xmm0, %xmm0
+; AVX-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8]
+; AVX-NEXT:    retq
+  %1 = lshr <2 x i64> %a0, <i64 48, i64 48>
+  %2 = bitcast <2 x i64> %1 to <16 x i8>
+  %3 = tail call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %2, <16 x i8> <i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8>)
+  ret <16 x i8> %3
+}
+
 define <16 x i8> @combine_unpckl_arg0_pshufb(<16 x i8> %a0, <16 x i8> %a1) {
 ; SSE-LABEL: combine_unpckl_arg0_pshufb:
 ; SSE:       # BB#0:
-- 
2.7.4