diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/vector-compare-results.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/vector-compare-results.ll | 260 |
1 files changed, 130 insertions, 130 deletions
diff --git a/llvm/test/CodeGen/X86/vector-compare-results.ll b/llvm/test/CodeGen/X86/vector-compare-results.ll index 473d54d28a6..a9d3152fd6b 100644 --- a/llvm/test/CodeGen/X86/vector-compare-results.ll +++ b/llvm/test/CodeGen/X86/vector-compare-results.ll @@ -7895,89 +7895,89 @@ define <32 x i1> @test_cmp_v32f64(<32 x double> %a0, <32 x double> %a1) nounwind ; SSE2-LABEL: test_cmp_v32f64: ; SSE2: # BB#0: ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm8 -; SSE2-NEXT: cmpltpd %xmm7, %xmm8 -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm7 -; SSE2-NEXT: cmpltpd %xmm6, %xmm7 -; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm8 = xmm8[0,1,0,2,4,5,6,7] -; SSE2-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm6 = xmm7[0,1,0,2,4,5,6,7] -; SSE2-NEXT: punpckldq {{.*#+}} xmm6 = xmm6[0],xmm8[0],xmm6[1],xmm8[1] -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm7 -; SSE2-NEXT: cmpltpd %xmm5, %xmm7 +; SSE2-NEXT: cmpltpd %xmm5, %xmm8 ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm5 ; SSE2-NEXT: cmpltpd %xmm4, %xmm5 -; SSE2-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm7[0,2,2,3,4,5,6,7] +; SSE2-NEXT: shufps {{.*#+}} xmm8 = xmm8[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm8[0,2,2,3,4,5,6,7] ; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3] ; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7] ; SSE2-NEXT: punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1] -; SSE2-NEXT: movsd {{.*#+}} xmm6 = xmm5[0],xmm6[1] ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm4 -; SSE2-NEXT: cmpltpd %xmm3, %xmm4 -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm5 -; SSE2-NEXT: cmpltpd %xmm2, %xmm5 -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm3 +; SSE2-NEXT: cmpltpd %xmm7, %xmm4 +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm7 +; SSE2-NEXT: cmpltpd %xmm6, %xmm7 ; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,2,4,5,6,7] -; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm5[0,1,0,2,4,5,6,7] -; SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1] -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm4 -; SSE2-NEXT: cmpltpd %xmm1, %xmm4 -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm1 -; SSE2-NEXT: cmpltpd %xmm0, %xmm1 +; SSE2-NEXT: pshuflw {{.*#+}} xmm6 = xmm4[0,1,0,2,4,5,6,7] +; SSE2-NEXT: shufps {{.*#+}} xmm7 = xmm7[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm7[0,1,0,2,4,5,6,7] +; SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1] +; SSE2-NEXT: movsd {{.*#+}} xmm4 = xmm5[0],xmm4[1] ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm5 -; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7] -; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7] -; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1] -; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm4 -; SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1] -; SSE2-NEXT: movapd {{.*#+}} xmm0 = [255,255,255,255,255,255,255,255] -; SSE2-NEXT: andpd %xmm0, %xmm6 -; SSE2-NEXT: andpd %xmm0, %xmm2 -; SSE2-NEXT: packuswb %xmm6, %xmm2 +; SSE2-NEXT: cmpltpd %xmm1, %xmm5 +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm6 +; SSE2-NEXT: cmpltpd %xmm0, %xmm6 ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm1 -; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm1 -; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm6 = xmm1[0,1,0,2,4,5,6,7] -; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm4 -; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm4[0,1,0,2,4,5,6,7] -; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1] +; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm5[0,2,2,3,4,5,6,7] +; SSE2-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm6[0,2,2,3,4,5,6,7] +; SSE2-NEXT: punpckldq {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1] +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm0 +; SSE2-NEXT: cmpltpd %xmm3, %xmm0 +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm3 +; SSE2-NEXT: cmpltpd %xmm2, %xmm3 +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm6 +; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm2 = xmm0[0,1,0,2,4,5,6,7] +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm3[0,1,0,2,4,5,6,7] +; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm3 +; SSE2-NEXT: movsd {{.*#+}} xmm0 = xmm5[0],xmm0[1] +; SSE2-NEXT: movapd {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255] +; SSE2-NEXT: andpd %xmm2, %xmm4 +; SSE2-NEXT: andpd %xmm2, %xmm0 +; SSE2-NEXT: packuswb %xmm4, %xmm0 ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm4 ; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm4 ; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,2,2,3] ; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7] -; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm5 -; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7] +; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm3 +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm3[0,2,2,3,4,5,6,7] ; SSE2-NEXT: punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1] +; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm3 +; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm3 +; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm3[0,1,0,2,4,5,6,7] +; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm6 +; SSE2-NEXT: shufps {{.*#+}} xmm6 = xmm6[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm6[0,1,0,2,4,5,6,7] +; SSE2-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1] ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm4 -; SSE2-NEXT: movsd {{.*#+}} xmm1 = xmm5[0],xmm1[1] +; SSE2-NEXT: movsd {{.*#+}} xmm3 = xmm5[0],xmm3[1] ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm5 ; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm5 ; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,1,0,2,4,5,6,7] +; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7] ; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm4 ; SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,1,0,2,4,5,6,7] +; SSE2-NEXT: pshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7] ; SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1] ; SSE2-NEXT: movapd {{[0-9]+}}(%rsp), %xmm5 ; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm5 ; SSE2-NEXT: shufps {{.*#+}} xmm5 = xmm5[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7] -; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm3 -; SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7] -; SSE2-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1] -; SSE2-NEXT: movsd {{.*#+}} xmm4 = xmm3[0],xmm4[1] -; SSE2-NEXT: andpd %xmm0, %xmm1 -; SSE2-NEXT: andpd %xmm0, %xmm4 -; SSE2-NEXT: packuswb %xmm1, %xmm4 -; SSE2-NEXT: movdqa %xmm4, -{{[0-9]+}}(%rsp) +; SSE2-NEXT: pshuflw {{.*#+}} xmm5 = xmm5[0,1,0,2,4,5,6,7] +; SSE2-NEXT: cmpltpd {{[0-9]+}}(%rsp), %xmm1 +; SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7] +; SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1] +; SSE2-NEXT: movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1] +; SSE2-NEXT: andpd %xmm2, %xmm3 +; SSE2-NEXT: andpd %xmm2, %xmm1 +; SSE2-NEXT: packuswb %xmm3, %xmm1 +; SSE2-NEXT: movdqa %xmm1, -{{[0-9]+}}(%rsp) ; SSE2-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-NEXT: andb $1, %al ; SSE2-NEXT: movb %al, 2(%rdi) @@ -8026,7 +8026,7 @@ define <32 x i1> @test_cmp_v32f64(<32 x double> %a0, <32 x double> %a1) nounwind ; SSE2-NEXT: movb %cl, 2(%rdi) ; SSE2-NEXT: andb $1, %al ; SSE2-NEXT: movb %al, 2(%rdi) -; SSE2-NEXT: movdqa %xmm2, -{{[0-9]+}}(%rsp) +; SSE2-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp) ; SSE2-NEXT: movb -{{[0-9]+}}(%rsp), %al ; SSE2-NEXT: andb $1, %al ; SSE2-NEXT: movb %al, (%rdi) @@ -9010,26 +9010,57 @@ define <32 x i1> @test_cmp_v32i64(<32 x i64> %a0, <32 x i64> %a1) nounwind { ; SSE2-NEXT: pcmpgtd %xmm9, %xmm11 ; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm11[0,0,2,2] ; SSE2-NEXT: pcmpeqd %xmm9, %xmm10 -; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm10[1,1,3,3] -; SSE2-NEXT: pand %xmm12, %xmm9 -; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm11[1,1,3,3] -; SSE2-NEXT: por %xmm9, %xmm10 -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 -; SSE2-NEXT: pxor %xmm8, %xmm9 +; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm10[1,1,3,3] +; SSE2-NEXT: pand %xmm12, %xmm10 +; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm11[1,1,3,3] +; SSE2-NEXT: por %xmm10, %xmm9 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10 +; SSE2-NEXT: pxor %xmm8, %xmm10 ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11 ; SSE2-NEXT: pxor %xmm8, %xmm11 ; SSE2-NEXT: movdqa %xmm11, %xmm12 -; SSE2-NEXT: pcmpgtd %xmm9, %xmm12 +; SSE2-NEXT: pcmpgtd %xmm10, %xmm12 ; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm12[0,0,2,2] -; SSE2-NEXT: pcmpeqd %xmm9, %xmm11 -; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm11[1,1,3,3] -; SSE2-NEXT: pand %xmm13, %xmm9 +; SSE2-NEXT: pcmpeqd %xmm10, %xmm11 +; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm11[1,1,3,3] +; SSE2-NEXT: pand %xmm13, %xmm11 +; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm12[1,1,3,3] +; SSE2-NEXT: por %xmm11, %xmm10 +; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm10[2,3,0,1] +; SSE2-NEXT: shufps {{.*#+}} xmm10 = xmm10[0,2],xmm9[0,2] +; SSE2-NEXT: punpcklwd {{.*#+}} xmm10 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3] +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11 +; SSE2-NEXT: pxor %xmm8, %xmm11 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm12 +; SSE2-NEXT: pxor %xmm8, %xmm12 +; SSE2-NEXT: movdqa %xmm12, %xmm13 +; SSE2-NEXT: pcmpgtd %xmm11, %xmm13 +; SSE2-NEXT: pshufd {{.*#+}} xmm14 = xmm13[0,0,2,2] +; SSE2-NEXT: pcmpeqd %xmm11, %xmm12 ; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm12[1,1,3,3] -; SSE2-NEXT: por %xmm9, %xmm11 -; SSE2-NEXT: shufps {{.*#+}} xmm11 = xmm11[0,2],xmm10[0,2] -; SSE2-NEXT: pshuflw {{.*#+}} xmm9 = xmm11[0,2,2,3,4,5,6,7] -; SSE2-NEXT: pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,4,6,6,7] -; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm9[0,1,0,2] +; SSE2-NEXT: pand %xmm14, %xmm11 +; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm13[1,1,3,3] +; SSE2-NEXT: por %xmm11, %xmm12 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm11 +; SSE2-NEXT: pxor %xmm8, %xmm11 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm13 +; SSE2-NEXT: pxor %xmm8, %xmm13 +; SSE2-NEXT: movdqa %xmm13, %xmm14 +; SSE2-NEXT: pcmpgtd %xmm11, %xmm14 +; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2] +; SSE2-NEXT: pcmpeqd %xmm11, %xmm13 +; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm13[1,1,3,3] +; SSE2-NEXT: pand %xmm15, %xmm11 +; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm14[1,1,3,3] +; SSE2-NEXT: por %xmm11, %xmm13 +; SSE2-NEXT: shufps {{.*#+}} xmm13 = xmm13[0,2],xmm12[0,2] +; SSE2-NEXT: pshuflw {{.*#+}} xmm11 = xmm13[0,2,2,3,4,5,6,7] +; SSE2-NEXT: pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,4,6,6,7] +; SSE2-NEXT: pshufd {{.*#+}} xmm11 = xmm11[0,1,0,2] +; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm9[0,2,2,3] +; SSE2-NEXT: pshuflw {{.*#+}} xmm9 = xmm9[0,1,0,2,4,5,6,7] +; SSE2-NEXT: movsd {{.*#+}} xmm11 = xmm9[0],xmm11[1] +; SSE2-NEXT: movss {{.*#+}} xmm11 = xmm10[0],xmm11[1,2,3] ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 ; SSE2-NEXT: pxor %xmm8, %xmm9 ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10 @@ -9042,9 +9073,6 @@ define <32 x i1> @test_cmp_v32i64(<32 x i64> %a0, <32 x i64> %a1) nounwind { ; SSE2-NEXT: pand %xmm13, %xmm9 ; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm12[1,1,3,3] ; SSE2-NEXT: por %xmm9, %xmm10 -; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm10[0,2,2,3] -; SSE2-NEXT: pshuflw {{.*#+}} xmm9 = xmm9[0,1,0,2,4,5,6,7] -; SSE2-NEXT: movsd {{.*#+}} xmm11 = xmm9[0],xmm11[1] ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 ; SSE2-NEXT: pxor %xmm8, %xmm9 ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm12 @@ -9060,68 +9088,40 @@ define <32 x i1> @test_cmp_v32i64(<32 x i64> %a0, <32 x i64> %a1) nounwind { ; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm12[2,3,0,1] ; SSE2-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,2],xmm10[0,2] ; SSE2-NEXT: punpcklwd {{.*#+}} xmm12 = xmm12[0],xmm9[0],xmm12[1],xmm9[1],xmm12[2],xmm9[2],xmm12[3],xmm9[3] -; SSE2-NEXT: movss {{.*#+}} xmm11 = xmm12[0],xmm11[1,2,3] ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 ; SSE2-NEXT: pxor %xmm8, %xmm9 -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10 -; SSE2-NEXT: pxor %xmm8, %xmm10 -; SSE2-NEXT: movdqa %xmm10, %xmm12 -; SSE2-NEXT: pcmpgtd %xmm9, %xmm12 -; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm12[0,0,2,2] -; SSE2-NEXT: pcmpeqd %xmm9, %xmm10 -; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm10[1,1,3,3] -; SSE2-NEXT: pand %xmm13, %xmm9 -; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm12[1,1,3,3] -; SSE2-NEXT: por %xmm9, %xmm10 +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm13 +; SSE2-NEXT: pxor %xmm8, %xmm13 +; SSE2-NEXT: movdqa %xmm13, %xmm14 +; SSE2-NEXT: pcmpgtd %xmm9, %xmm14 +; SSE2-NEXT: pshufd {{.*#+}} xmm15 = xmm14[0,0,2,2] +; SSE2-NEXT: pcmpeqd %xmm9, %xmm13 +; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm13[1,1,3,3] +; SSE2-NEXT: pand %xmm15, %xmm9 +; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm14[1,1,3,3] +; SSE2-NEXT: por %xmm9, %xmm13 ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm9 ; SSE2-NEXT: pxor %xmm8, %xmm9 -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm12 -; SSE2-NEXT: pxor %xmm8, %xmm12 -; SSE2-NEXT: movdqa %xmm12, %xmm13 -; SSE2-NEXT: pcmpgtd %xmm9, %xmm13 -; SSE2-NEXT: pshufd {{.*#+}} xmm14 = xmm13[0,0,2,2] -; SSE2-NEXT: pcmpeqd %xmm9, %xmm12 -; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm12[1,1,3,3] -; SSE2-NEXT: pand %xmm14, %xmm9 -; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm13[1,1,3,3] -; SSE2-NEXT: por %xmm9, %xmm12 -; SSE2-NEXT: shufps {{.*#+}} xmm12 = xmm12[0,2],xmm10[0,2] -; SSE2-NEXT: pshuflw {{.*#+}} xmm9 = xmm12[0,2,2,3,4,5,6,7] +; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm14 +; SSE2-NEXT: pxor %xmm8, %xmm14 +; SSE2-NEXT: movdqa %xmm14, %xmm15 +; SSE2-NEXT: pcmpgtd %xmm9, %xmm15 +; SSE2-NEXT: pcmpeqd %xmm9, %xmm14 +; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm15[0,0,2,2] +; SSE2-NEXT: pshufd {{.*#+}} xmm14 = xmm14[1,1,3,3] +; SSE2-NEXT: pand %xmm9, %xmm14 +; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm15[1,1,3,3] +; SSE2-NEXT: por %xmm14, %xmm9 +; SSE2-NEXT: shufps {{.*#+}} xmm9 = xmm9[0,2],xmm13[0,2] +; SSE2-NEXT: pshuflw {{.*#+}} xmm9 = xmm9[0,2,2,3,4,5,6,7] ; SSE2-NEXT: pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,4,6,6,7] ; SSE2-NEXT: pshufd {{.*#+}} xmm9 = xmm9[0,1,0,2] -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10 -; SSE2-NEXT: pxor %xmm8, %xmm10 -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm12 -; SSE2-NEXT: pxor %xmm8, %xmm12 -; SSE2-NEXT: movdqa %xmm12, %xmm13 -; SSE2-NEXT: pcmpgtd %xmm10, %xmm13 -; SSE2-NEXT: pshufd {{.*#+}} xmm14 = xmm13[0,0,2,2] -; SSE2-NEXT: pcmpeqd %xmm10, %xmm12 -; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm12[1,1,3,3] -; SSE2-NEXT: pand %xmm14, %xmm10 -; SSE2-NEXT: pshufd {{.*#+}} xmm12 = xmm13[1,1,3,3] -; SSE2-NEXT: por %xmm10, %xmm12 -; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm12[0,2,2,3] +; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm10[0,2,2,3] ; SSE2-NEXT: pshuflw {{.*#+}} xmm10 = xmm10[0,1,0,2,4,5,6,7] ; SSE2-NEXT: movsd {{.*#+}} xmm9 = xmm10[0],xmm9[1] -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm10 -; SSE2-NEXT: pxor %xmm8, %xmm10 -; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm13 -; SSE2-NEXT: pxor %xmm8, %xmm13 -; SSE2-NEXT: movdqa %xmm13, %xmm14 -; SSE2-NEXT: pcmpgtd %xmm10, %xmm14 -; SSE2-NEXT: pcmpeqd %xmm10, %xmm13 -; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm14[0,0,2,2] -; SSE2-NEXT: pshufd {{.*#+}} xmm13 = xmm13[1,1,3,3] -; SSE2-NEXT: pand %xmm10, %xmm13 -; SSE2-NEXT: pshufd {{.*#+}} xmm14 = xmm14[1,1,3,3] -; SSE2-NEXT: por %xmm13, %xmm14 -; SSE2-NEXT: pshufd {{.*#+}} xmm10 = xmm14[2,3,0,1] -; SSE2-NEXT: shufps {{.*#+}} xmm14 = xmm14[0,2],xmm12[0,2] -; SSE2-NEXT: punpcklwd {{.*#+}} xmm14 = xmm14[0],xmm10[0],xmm14[1],xmm10[1],xmm14[2],xmm10[2],xmm14[3],xmm10[3] ; SSE2-NEXT: movaps {{.*#+}} xmm10 = [255,255,255,255,255,255,255,255] ; SSE2-NEXT: andps %xmm10, %xmm11 -; SSE2-NEXT: movss {{.*#+}} xmm9 = xmm14[0],xmm9[1,2,3] +; SSE2-NEXT: movss {{.*#+}} xmm9 = xmm12[0],xmm9[1,2,3] ; SSE2-NEXT: andps %xmm10, %xmm9 ; SSE2-NEXT: packuswb %xmm11, %xmm9 ; SSE2-NEXT: pxor %xmm8, %xmm5 |
