diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/srem-seteq-vec-nonsplat.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/srem-seteq-vec-nonsplat.ll | 968 |
1 files changed, 215 insertions, 753 deletions
diff --git a/llvm/test/CodeGen/X86/srem-seteq-vec-nonsplat.ll b/llvm/test/CodeGen/X86/srem-seteq-vec-nonsplat.ll index 03ef755425d..134d40015d0 100644 --- a/llvm/test/CodeGen/X86/srem-seteq-vec-nonsplat.ll +++ b/llvm/test/CodeGen/X86/srem-seteq-vec-nonsplat.ll @@ -138,21 +138,10 @@ define <4 x i32> @test_srem_odd_even(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2454267027,1374389535,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 ; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq @@ -168,132 +157,55 @@ define <4 x i32> @test_srem_odd_even(<4 x i32> %X) nounwind { define <4 x i32> @test_srem_odd_allones_eq(<4 x i32> %X) nounwind { ; CHECK-SSE2-LABEL: test_srem_odd_allones_eq: ; CHECK-SSE2: # %bb.0: -; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1 -; CHECK-SSE2-NEXT: pxor %xmm2, %xmm2 -; CHECK-SSE2-NEXT: pcmpgtd %xmm0, %xmm2 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [1717986919,1717986919,0,1717986919] -; CHECK-SSE2-NEXT: pand %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm4 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,3,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1] -; CHECK-SSE2-NEXT: psubd %xmm2, %xmm3 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = <0,u,4294967295,u> -; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm2 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: paddd %xmm3, %xmm2 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3 -; CHECK-SSE2-NEXT: psrad $1, %xmm3 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm4 -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm3[3,0] -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm4[0,2] -; CHECK-SSE2-NEXT: psrld $31, %xmm2 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: paddd %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1] -; CHECK-SSE2-NEXT: psubd %xmm2, %xmm0 -; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0 -; CHECK-SSE2-NEXT: psrld $31, %xmm0 +; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] +; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] +; CHECK-SSE2-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0 ; CHECK-SSE2-NEXT: retq ; ; CHECK-SSE41-LABEL: test_srem_odd_allones_eq: ; CHECK-SSE41: # %bb.0: -; CHECK-SSE41-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-SSE41-NEXT: movd %eax, %xmm1 -; CHECK-SSE41-NEXT: pmuldq %xmm0, %xmm1 -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] -; CHECK-SSE41-NEXT: pmuldq {{.*}}(%rip), %xmm2 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [0,0,4294967295,0] -; CHECK-SSE41-NEXT: pmulld %xmm0, %xmm1 -; CHECK-SSE41-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2 -; CHECK-SSE41-NEXT: psrad $1, %xmm2 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: psrld $31, %xmm1 -; CHECK-SSE41-NEXT: pxor %xmm3, %xmm3 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7] -; CHECK-SSE41-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1 -; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0 -; CHECK-SSE41-NEXT: pcmpeqd %xmm3, %xmm0 +; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993458,858993458,4294967295,858993458] +; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1 +; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0 ; CHECK-SSE41-NEXT: psrld $31, %xmm0 ; CHECK-SSE41-NEXT: retq ; ; CHECK-AVX1-LABEL: test_srem_odd_allones_eq: ; CHECK-AVX1: # %bb.0: -; CHECK-AVX1-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX1-NEXT: vmovd %eax, %xmm1 -; CHECK-AVX1-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] -; CHECK-AVX1-NEXT: vpmuldq {{.*}}(%rip), %xmm2, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX1-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsrad $1, %xmm1, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpsrld $31, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7] -; CHECK-AVX1-NEXT: vpaddd %xmm1, %xmm2, %xmm1 -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX1-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: retq ; ; CHECK-AVX2-LABEL: test_srem_odd_allones_eq: ; CHECK-AVX2: # %bb.0: -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX2-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX2-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX2-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX2-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [429496729,429496729,429496729,429496729] +; CHECK-AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: retq ; ; CHECK-AVX512VL-LABEL: test_srem_odd_allones_eq: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX512VL-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 5, i32 4294967295, i32 5> @@ -304,133 +216,56 @@ define <4 x i32> @test_srem_odd_allones_eq(<4 x i32> %X) nounwind { define <4 x i32> @test_srem_odd_allones_ne(<4 x i32> %X) nounwind { ; CHECK-SSE2-LABEL: test_srem_odd_allones_ne: ; CHECK-SSE2: # %bb.0: -; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1 -; CHECK-SSE2-NEXT: pxor %xmm2, %xmm2 -; CHECK-SSE2-NEXT: pcmpgtd %xmm0, %xmm2 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [1717986919,1717986919,0,1717986919] -; CHECK-SSE2-NEXT: pand %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm4 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,3,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1] -; CHECK-SSE2-NEXT: psubd %xmm2, %xmm3 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = <0,u,4294967295,u> -; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm2 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: paddd %xmm3, %xmm2 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3 -; CHECK-SSE2-NEXT: psrad $1, %xmm3 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm4 -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm3[3,0] -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm4[0,2] -; CHECK-SSE2-NEXT: psrld $31, %xmm2 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: paddd %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1] -; CHECK-SSE2-NEXT: psubd %xmm2, %xmm0 -; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0 -; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] +; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] +; CHECK-SSE2-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: psrld $31, %xmm0 ; CHECK-SSE2-NEXT: retq ; ; CHECK-SSE41-LABEL: test_srem_odd_allones_ne: ; CHECK-SSE41: # %bb.0: -; CHECK-SSE41-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-SSE41-NEXT: movd %eax, %xmm1 -; CHECK-SSE41-NEXT: pmuldq %xmm0, %xmm1 -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] -; CHECK-SSE41-NEXT: pmuldq {{.*}}(%rip), %xmm2 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [0,0,4294967295,0] -; CHECK-SSE41-NEXT: pmulld %xmm0, %xmm1 -; CHECK-SSE41-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2 -; CHECK-SSE41-NEXT: psrad $1, %xmm2 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: psrld $31, %xmm1 -; CHECK-SSE41-NEXT: pxor %xmm3, %xmm3 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7] -; CHECK-SSE41-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1 -; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0 -; CHECK-SSE41-NEXT: pcmpeqd %xmm3, %xmm0 +; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993458,858993458,4294967295,858993458] +; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1 +; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0 ; CHECK-SSE41-NEXT: pandn {{.*}}(%rip), %xmm0 ; CHECK-SSE41-NEXT: retq ; ; CHECK-AVX1-LABEL: test_srem_odd_allones_ne: ; CHECK-AVX1: # %bb.0: -; CHECK-AVX1-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX1-NEXT: vmovd %eax, %xmm1 -; CHECK-AVX1-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] -; CHECK-AVX1-NEXT: vpmuldq {{.*}}(%rip), %xmm2, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX1-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsrad $1, %xmm1, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpsrld $31, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7] -; CHECK-AVX1-NEXT: vpaddd %xmm1, %xmm2, %xmm1 -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX1-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: vpandn {{.*}}(%rip), %xmm0, %xmm0 ; CHECK-AVX1-NEXT: retq ; ; CHECK-AVX2-LABEL: test_srem_odd_allones_ne: ; CHECK-AVX2: # %bb.0: -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX2-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX2-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX2-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX2-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [429496729,429496729,429496729,429496729] +; CHECK-AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [1,1,1,1] ; CHECK-AVX2-NEXT: vpandn %xmm1, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: retq ; ; CHECK-AVX512VL-LABEL: test_srem_odd_allones_ne: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX512VL-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpandnd {{.*}}(%rip){1to4}, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 5, i32 4294967295, i32 5> @@ -559,24 +394,11 @@ define <4 x i32> @test_srem_even_allones_eq(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_allones_eq: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [2454267027,2454267027,2454267027,2454267027] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: movl $-1840700269, %eax # imm = 0x92492493 -; CHECK-AVX512VL-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprord $1, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 14, i32 4294967295, i32 14> @@ -705,24 +527,11 @@ define <4 x i32> @test_srem_even_allones_ne(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_allones_ne: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [2454267027,2454267027,2454267027,2454267027] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: movl $-1840700269, %eax # imm = 0x92492493 -; CHECK-AVX512VL-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprord $1, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpandnd {{.*}}(%rip){1to4}, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 14, i32 4294967295, i32 14> @@ -869,23 +678,11 @@ define <4 x i32> @test_srem_odd_even_allones_eq(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_allones_eq: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2454267027,0,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 14, i32 4294967295, i32 100> @@ -1031,23 +828,11 @@ define <4 x i32> @test_srem_odd_even_allones_ne(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_allones_ne: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2454267027,0,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpandnd {{.*}}(%rip){1to4}, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 14, i32 4294967295, i32 100> @@ -1168,20 +953,10 @@ define <4 x i32> @test_srem_odd_poweroftwo(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_poweroftwo: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmuldq {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 ; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq @@ -1283,19 +1058,10 @@ define <4 x i32> @test_srem_even_poweroftwo(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_poweroftwo: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [2454267027,2454267027,2454267027,2454267027] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmuldq {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpaddd %xmm0, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpsrad $3, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 ; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq @@ -1438,21 +1204,10 @@ define <4 x i32> @test_srem_odd_even_poweroftwo(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_poweroftwo: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2454267027,2147483649,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 ; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq @@ -1468,132 +1223,55 @@ define <4 x i32> @test_srem_odd_even_poweroftwo(<4 x i32> %X) nounwind { define <4 x i32> @test_srem_odd_one(<4 x i32> %X) nounwind { ; CHECK-SSE2-LABEL: test_srem_odd_one: ; CHECK-SSE2: # %bb.0: -; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1 -; CHECK-SSE2-NEXT: pxor %xmm2, %xmm2 -; CHECK-SSE2-NEXT: pcmpgtd %xmm0, %xmm2 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [1717986919,1717986919,0,1717986919] -; CHECK-SSE2-NEXT: pand %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[1,3,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm4 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[1,3,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1] -; CHECK-SSE2-NEXT: psubd %xmm2, %xmm3 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = <0,u,1,u> -; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm2 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: paddd %xmm3, %xmm2 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3 -; CHECK-SSE2-NEXT: psrad $1, %xmm3 -; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm4 -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm3[3,0] -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,1],xmm4[0,2] -; CHECK-SSE2-NEXT: psrld $31, %xmm2 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: paddd %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3] -; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1] -; CHECK-SSE2-NEXT: psubd %xmm2, %xmm0 -; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0 -; CHECK-SSE2-NEXT: psrld $31, %xmm0 +; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] +; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] +; CHECK-SSE2-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0 ; CHECK-SSE2-NEXT: retq ; ; CHECK-SSE41-LABEL: test_srem_odd_one: ; CHECK-SSE41: # %bb.0: -; CHECK-SSE41-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-SSE41-NEXT: movd %eax, %xmm1 -; CHECK-SSE41-NEXT: pmuldq %xmm0, %xmm1 -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] -; CHECK-SSE41-NEXT: pmuldq {{.*}}(%rip), %xmm2 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [0,0,1,0] -; CHECK-SSE41-NEXT: pmulld %xmm0, %xmm1 -; CHECK-SSE41-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2 -; CHECK-SSE41-NEXT: psrad $1, %xmm2 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: psrld $31, %xmm1 -; CHECK-SSE41-NEXT: pxor %xmm3, %xmm3 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7] -; CHECK-SSE41-NEXT: paddd %xmm2, %xmm1 -; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1 -; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0 -; CHECK-SSE41-NEXT: pcmpeqd %xmm3, %xmm0 +; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993458,858993458,4294967295,858993458] +; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1 +; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0 ; CHECK-SSE41-NEXT: psrld $31, %xmm0 ; CHECK-SSE41-NEXT: retq ; ; CHECK-AVX1-LABEL: test_srem_odd_one: ; CHECK-AVX1: # %bb.0: -; CHECK-AVX1-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX1-NEXT: vmovd %eax, %xmm1 -; CHECK-AVX1-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] -; CHECK-AVX1-NEXT: vpmuldq {{.*}}(%rip), %xmm2, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX1-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsrad $1, %xmm1, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpsrld $31, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7] -; CHECK-AVX1-NEXT: vpaddd %xmm1, %xmm2, %xmm1 -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX1-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: retq ; ; CHECK-AVX2-LABEL: test_srem_odd_one: ; CHECK-AVX2: # %bb.0: -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX2-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX2-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX2-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX2-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [429496729,429496729,429496729,429496729] +; CHECK-AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: retq ; ; CHECK-AVX512VL-LABEL: test_srem_odd_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1717986919,1717986919,1717986919,1717986919] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: movl $1717986919, %eax # imm = 0x66666667 -; CHECK-AVX512VL-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 5, i32 1, i32 5> @@ -1711,23 +1389,11 @@ define <4 x i32> @test_srem_even_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpbroadcastd {{.*#+}} xmm2 = [2454267027,2454267027,2454267027,2454267027] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: movl $-1840700269, %eax # imm = 0x92492493 -; CHECK-AVX512VL-NEXT: vmovd %eax, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3] -; CHECK-AVX512VL-NEXT: vpaddd %xmm0, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprord $1, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 14, i32 1, i32 14> @@ -1874,23 +1540,11 @@ define <4 x i32> @test_srem_odd_even_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2454267027,0,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 14, i32 1, i32 100> @@ -2465,23 +2119,11 @@ define <4 x i32> @test_srem_odd_allones_and_poweroftwo(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_allones_and_poweroftwo: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,0,2147483649,1717986919] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 4294967295, i32 16, i32 5> @@ -2614,23 +2256,11 @@ define <4 x i32> @test_srem_even_allones_and_poweroftwo(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_allones_and_poweroftwo: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,0,2147483649,2454267027] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 4294967295, i32 16, i32 14> @@ -2776,23 +2406,11 @@ define <4 x i32> @test_srem_odd_even_allones_and_poweroftwo(<4 x i32> %X) nounwi ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_allones_and_poweroftwo: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,0,2147483649,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 4294967295, i32 16, i32 100> @@ -2807,138 +2425,55 @@ define <4 x i32> @test_srem_odd_even_allones_and_poweroftwo(<4 x i32> %X) nounwi define <4 x i32> @test_srem_odd_allones_and_one(<4 x i32> %X) nounwind { ; CHECK-SSE2-LABEL: test_srem_odd_allones_and_one: ; CHECK-SSE2: # %bb.0: -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [0,4294967295,1,0] -; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2 +; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] +; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] ; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1] -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [1717986919,0,0,1717986919] -; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1 -; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm1 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm5 = xmm1[1,3,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[2,2,3,3] -; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm5 = xmm5[0],xmm1[0],xmm5[1],xmm1[1] -; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1 -; CHECK-SSE2-NEXT: pxor %xmm3, %xmm3 -; CHECK-SSE2-NEXT: pcmpgtd %xmm0, %xmm3 -; CHECK-SSE2-NEXT: pand %xmm4, %xmm3 -; CHECK-SSE2-NEXT: psubd %xmm3, %xmm5 -; CHECK-SSE2-NEXT: paddd %xmm2, %xmm5 -; CHECK-SSE2-NEXT: movdqa %xmm5, %xmm2 -; CHECK-SSE2-NEXT: psrad $1, %xmm2 -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm5[1,2] -; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,2,3,1] -; CHECK-SSE2-NEXT: psrld $31, %xmm5 -; CHECK-SSE2-NEXT: pand {{.*}}(%rip), %xmm5 -; CHECK-SSE2-NEXT: paddd %xmm2, %xmm5 -; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [5,4294967295,1,5] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm5[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm5 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm5[0,2,2,3] -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] -; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm2 -; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3] -; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1] -; CHECK-SSE2-NEXT: psubd %xmm4, %xmm0 -; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0 -; CHECK-SSE2-NEXT: psrld $31, %xmm0 +; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] +; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] +; CHECK-SSE2-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pxor {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0 +; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0 ; CHECK-SSE2-NEXT: retq ; ; CHECK-SSE41-LABEL: test_srem_odd_allones_and_one: ; CHECK-SSE41: # %bb.0: -; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [1717986919,0,0,1717986919] -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[2,2,3,3] -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-SSE41-NEXT: pmuldq %xmm2, %xmm3 -; CHECK-SSE41-NEXT: pmuldq %xmm0, %xmm1 -; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7] -; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [0,4294967295,1,0] -; CHECK-SSE41-NEXT: pmulld %xmm0, %xmm2 -; CHECK-SSE41-NEXT: paddd %xmm1, %xmm2 -; CHECK-SSE41-NEXT: movdqa %xmm2, %xmm1 -; CHECK-SSE41-NEXT: psrad $1, %xmm1 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3,4,5],xmm1[6,7] -; CHECK-SSE41-NEXT: psrld $31, %xmm2 -; CHECK-SSE41-NEXT: pxor %xmm3, %xmm3 -; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3,4,5],xmm2[6,7] -; CHECK-SSE41-NEXT: paddd %xmm1, %xmm2 -; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2 -; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0 -; CHECK-SSE41-NEXT: pcmpeqd %xmm3, %xmm0 +; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: paddd {{.*}}(%rip), %xmm0 +; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993458,4294967295,4294967295,858993458] +; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1 +; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0 ; CHECK-SSE41-NEXT: psrld $31, %xmm0 ; CHECK-SSE41-NEXT: retq ; ; CHECK-AVX1-LABEL: test_srem_odd_allones_and_one: ; CHECK-AVX1: # %bb.0: -; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,0,0,1717986919] -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,2,3,3] -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX1-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX1-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX1-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsrad $1, %xmm1, %xmm2 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3,4,5],xmm2[6,7] -; CHECK-AVX1-NEXT: vpsrld $31, %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3,4,5],xmm1[6,7] -; CHECK-AVX1-NEXT: vpaddd %xmm1, %xmm2, %xmm1 -; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX1-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX1-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX1-NEXT: retq ; ; CHECK-AVX2-LABEL: test_srem_odd_allones_and_one: ; CHECK-AVX2: # %bb.0: -; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,0,0,1717986919] -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,2,3,3] -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX2-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX2-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX2-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1,2],xmm2[3] -; CHECK-AVX2-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX2-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837] +; CHECK-AVX2-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [429496729,429496729,429496729,429496729] +; CHECK-AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm0 +; CHECK-AVX2-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX2-NEXT: retq ; ; CHECK-AVX512VL-LABEL: test_srem_odd_allones_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,0,0,1717986919] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,2,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1,2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 4294967295, i32 1, i32 5> @@ -3070,23 +2605,11 @@ define <4 x i32> @test_srem_even_allones_and_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_allones_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,0,0,2454267027] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,2,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1,2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip){1to4}, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprord $1, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 4294967295, i32 1, i32 14> @@ -3225,23 +2748,11 @@ define <4 x i32> @test_srem_odd_even_allones_and_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_allones_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,0,0,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,2,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1,2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 4294967295, i32 1, i32 100> @@ -3385,23 +2896,11 @@ define <4 x i32> @test_srem_odd_poweroftwo_and_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_poweroftwo_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2147483649,0,1717986919] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 16, i32 1, i32 5> @@ -3522,22 +3021,11 @@ define <4 x i32> @test_srem_even_poweroftwo_and_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_even_poweroftwo_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,2147483649,0,2454267027] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpaddd %xmm0, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 16, i32 1, i32 14> @@ -3684,23 +3172,11 @@ define <4 x i32> @test_srem_odd_even_poweroftwo_and_one(<4 x i32> %X) nounwind { ; ; CHECK-AVX512VL-LABEL: test_srem_odd_even_poweroftwo_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vmovdqa {{.*#+}} xmm1 = [1717986919,2147483649,0,1374389535] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpmuldq %xmm2, %xmm3, %xmm2 -; CHECK-AVX512VL-NEXT: vpmuldq %xmm1, %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2],xmm2[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 16, i32 1, i32 100> @@ -3821,18 +3297,11 @@ define <4 x i32> @test_srem_odd_allones_and_poweroftwo_and_one(<4 x i32> %X) nou ; ; CHECK-AVX512VL-LABEL: test_srem_odd_allones_and_poweroftwo_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpmuldq {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpsrlq $32, %xmm2, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm1, %xmm2, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2],xmm3[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 5, i32 4294967295, i32 16, i32 1> @@ -3945,18 +3414,11 @@ define <4 x i32> @test_srem_even_allones_and_poweroftwo_and_one(<4 x i32> %X) no ; ; CHECK-AVX512VL-LABEL: test_srem_even_allones_and_poweroftwo_and_one: ; CHECK-AVX512VL: # %bb.0: -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm1 -; CHECK-AVX512VL-NEXT: vpmuldq {{.*}}(%rip), %xmm0, %xmm2 -; CHECK-AVX512VL-NEXT: vpsrlq $32, %xmm2, %xmm2 -; CHECK-AVX512VL-NEXT: vpaddd %xmm1, %xmm2, %xmm1 -; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm1, %xmm2 -; CHECK-AVX512VL-NEXT: vpxor %xmm3, %xmm3, %xmm3 -; CHECK-AVX512VL-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2],xmm3[3] -; CHECK-AVX512VL-NEXT: vpsravd {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpaddd %xmm2, %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1 -; CHECK-AVX512VL-NEXT: vpsubd %xmm1, %xmm0, %xmm0 -; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm3, %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpaddd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0 +; CHECK-AVX512VL-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1 +; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0 ; CHECK-AVX512VL-NEXT: retq %srem = srem <4 x i32> %X, <i32 14, i32 4294967295, i32 16, i32 1> |
