summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/bswap-vector.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/bswap-vector.ll')
-rw-r--r--llvm/test/CodeGen/X86/bswap-vector.ll287
1 files changed, 92 insertions, 195 deletions
diff --git a/llvm/test/CodeGen/X86/bswap-vector.ll b/llvm/test/CodeGen/X86/bswap-vector.ll
index 0351b22a700..7d5f380c1e2 100644
--- a/llvm/test/CodeGen/X86/bswap-vector.ll
+++ b/llvm/test/CodeGen/X86/bswap-vector.ll
@@ -13,37 +13,15 @@ declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
define <8 x i16> @test1(<8 x i16> %v) {
; CHECK-NOSSSE3-LABEL: test1:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: pextrw $7, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: pextrw $3, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; CHECK-NOSSSE3-NEXT: pextrw $5, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: pextrw $1, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; CHECK-NOSSSE3-NEXT: pextrw $6, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: pextrw $2, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; CHECK-NOSSSE3-NEXT: pextrw $4, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[1,0,3,2,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,4,7,6]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,7,6]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test1:
@@ -68,25 +46,15 @@ entry:
define <4 x i32> @test2(<4 x i32> %v) {
; CHECK-NOSSSE3-LABEL: test2:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm2, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm0
+; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test2:
@@ -111,15 +79,17 @@ entry:
define <2 x i64> @test3(<2 x i64> %v) {
; CHECK-NOSSSE3-LABEL: test3:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %rax
-; CHECK-NOSSSE3-NEXT: bswapq %rax
-; CHECK-NOSSSE3-NEXT: movd %rax, %xmm1
+; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %rax
-; CHECK-NOSSSE3-NEXT: bswapq %rax
-; CHECK-NOSSSE3-NEXT: movd %rax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm0
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test3:
@@ -148,68 +118,23 @@ declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
define <16 x i16> @test4(<16 x i16> %v) {
; CHECK-NOSSSE3-LABEL: test4:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: pextrw $7, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: pextrw $3, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; CHECK-NOSSSE3-NEXT: pextrw $5, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm4
-; CHECK-NOSSSE3-NEXT: pextrw $1, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: pextrw $6, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: pextrw $2, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm4
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: pextrw $4, %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; CHECK-NOSSSE3-NEXT: pextrw $7, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: pextrw $3, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; CHECK-NOSSSE3-NEXT: pextrw $5, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm4
-; CHECK-NOSSSE3-NEXT: pextrw $1, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: pextrw $6, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: pextrw $2, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm4
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: pextrw $4, %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: movd %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: rolw $8, %ax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; CHECK-NOSSSE3-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; CHECK-NOSSSE3-NEXT: pxor %xmm2, %xmm2
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm3
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[1,0,3,2,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,4,7,6]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,4,7,6]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm0
+; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm3
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[1,0,3,2,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,4,7,6]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[1,0,3,2,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,4,7,6]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm1
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test4:
@@ -236,44 +161,23 @@ entry:
define <8 x i32> @test5(<8 x i32> %v) {
; CHECK-NOSSSE3-LABEL: test5:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm2[3,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm0
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm3, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; CHECK-NOSSSE3-NEXT: movd %xmm2, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm0
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm2, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm2, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm3, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm3
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; CHECK-NOSSSE3-NEXT: movd %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
-; CHECK-NOSSSE3-NEXT: movdqa %xmm2, %xmm1
+; CHECK-NOSSSE3-NEXT: pxor %xmm2, %xmm2
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm3
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm0
+; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm3
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm1
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test5:
@@ -300,24 +204,27 @@ entry:
define <4 x i64> @test6(<4 x i64> %v) {
; CHECK-NOSSSE3-LABEL: test6:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %rax
-; CHECK-NOSSSE3-NEXT: bswapq %rax
-; CHECK-NOSSSE3-NEXT: movd %rax, %xmm2
+; CHECK-NOSSSE3-NEXT: pxor %xmm2, %xmm2
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm3
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %rax
-; CHECK-NOSSSE3-NEXT: bswapq %rax
-; CHECK-NOSSSE3-NEXT: movd %rax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm0[0]
-; CHECK-NOSSSE3-NEXT: movd %xmm1, %rax
-; CHECK-NOSSSE3-NEXT: bswapq %rax
-; CHECK-NOSSSE3-NEXT: movd %rax, %xmm3
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %rax
-; CHECK-NOSSSE3-NEXT: bswapq %rax
-; CHECK-NOSSSE3-NEXT: movd %rax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm0[0]
-; CHECK-NOSSSE3-NEXT: movdqa %xmm2, %xmm0
-; CHECK-NOSSSE3-NEXT: movdqa %xmm3, %xmm1
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm0
+; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm3
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm3, %xmm1
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test6:
@@ -346,26 +253,16 @@ declare <4 x i16> @llvm.bswap.v4i16(<4 x i16>)
define <4 x i16> @test7(<4 x i16> %v) {
; CHECK-NOSSSE3-LABEL: test7:
; CHECK-NOSSSE3: # BB#0: # %entry
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm1, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; CHECK-NOSSSE3-NEXT: movd %xmm2, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm2
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm1
-; CHECK-NOSSSE3-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-NOSSSE3-NEXT: movd %xmm0, %eax
-; CHECK-NOSSSE3-NEXT: bswapl %eax
-; CHECK-NOSSSE3-NEXT: movd %eax, %xmm0
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; CHECK-NOSSSE3-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; CHECK-NOSSSE3-NEXT: psrld $16, %xmm1
-; CHECK-NOSSSE3-NEXT: movdqa %xmm1, %xmm0
+; CHECK-NOSSSE3-NEXT: pxor %xmm1, %xmm1
+; CHECK-NOSSSE3-NEXT: movdqa %xmm0, %xmm2
+; CHECK-NOSSSE3-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; CHECK-NOSSSE3-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
+; CHECK-NOSSSE3-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
+; CHECK-NOSSSE3-NEXT: packuswb %xmm2, %xmm0
+; CHECK-NOSSSE3-NEXT: psrld $16, %xmm0
; CHECK-NOSSSE3-NEXT: retq
;
; CHECK-SSSE3-LABEL: test7:
OpenPOWER on IntegriCloud