RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00
Code Issues Projects Releases Wiki Activity
c849a98c1b
llvm-mirror/test/CodeGen/X86/urem-seteq-vec-nonsplat.ll
Simon Pilgrim 1d52f2e8fe [X86][SSE] lowerV2I64Shuffle - use undef elements in PSHUFD mask widening 
If we lower a v2i64 shuffle to PSHUFD, we currently clamp undef elements to 0, (elements 0,1 of the v4i32) which can result in the shuffle referencing more elements of the source vector than expected, affecting later shuffle combines and KnownBits/SimplifyDemanded calls.

By ensuring we widen the undef mask element we allow getV4X86ShuffleImm8 to use inline elements as the default, which are more likely to fold.
2020-07-26 16:04:22 +01:00

2616 lines
128 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SSE,CHECK-SSE2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.1 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SSE,CHECK-SSE41
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX1
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX2
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl < %s | FileCheck %s --check-prefixes=CHECK,CHECK-AVX,CHECK-AVX512VL
; Odd+Even divisors
define <4 x i32> @test_urem_odd_even(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2454267027,1374389535,1374389535]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm3
; CHECK-SSE2-NEXT: psrld $1, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3
; CHECK-SSE2-NEXT: psrld $2, %xmm3
; CHECK-SSE2-NEXT: psrld $3, %xmm2
; CHECK-SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [5,14,25,100]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: psrld $5, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,2454267027,1374389535,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3],xmm2[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: psrld $5, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $3, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,2454267027,1374389535,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm3, %xmm3
; CHECK-AVX1-NEXT: vpsrld $5, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $3, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,1374389535,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 14, i32 25, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;==============================================================================;
; One all-ones divisor in odd divisor
define <4 x i32> @test_urem_odd_allones_eq(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_allones_eq:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; 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_urem_odd_allones_eq:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993459,858993459,1,858993459]
; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX-LABEL: test_urem_odd_allones_eq:
; CHECK-AVX: # %bb.0:
; CHECK-AVX-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX-NEXT: retq
%urem = urem <4 x i32> %X, <i32 5, i32 5, i32 4294967295, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
define <4 x i32> @test_urem_odd_allones_ne(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_allones_ne:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; 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_urem_odd_allones_ne:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993460,858993460,2,858993460]
; CHECK-SSE41-NEXT: pmaxud %xmm0, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX-LABEL: test_urem_odd_allones_ne:
; CHECK-AVX: # %bb.0:
; CHECK-AVX-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX-NEXT: vpmaxud {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX-NEXT: retq
%urem = urem <4 x i32> %X, <i32 5, i32 5, i32 4294967295, i32 5>
%cmp = icmp ne <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One all-ones divisor in even divisor
define <4 x i32> @test_urem_even_allones_eq(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_allones_eq:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: psrld $31, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_allones_eq:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: psrld $31, %xmm1
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_allones_eq:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_allones_eq:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
; CHECK-AVX2-NEXT: vpmuludq %xmm3, %xmm2, %xmm2
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_allones_eq:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 14, i32 4294967295, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
define <4 x i32> @test_urem_even_allones_ne(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_allones_ne:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: psrld $31, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_allones_ne:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: psrld $31, %xmm1
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pandn {{.*}}(%rip), %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_allones_ne:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpandn {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_allones_ne:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
; CHECK-AVX2-NEXT: vpmuludq %xmm3, %xmm2, %xmm2
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %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_urem_even_allones_ne:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: vpmaxud {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: retq
%urem = urem <4 x i32> %X, <i32 14, i32 14, i32 4294967295, i32 14>
%cmp = icmp ne <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One all-ones divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_allones_eq(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_allones_eq:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm3
; CHECK-SSE2-NEXT: psrld $1, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3
; CHECK-SSE2-NEXT: psrld $2, %xmm3
; CHECK-SSE2-NEXT: psrld $31, %xmm2
; CHECK-SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [5,14,4294967295,100]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: psrld $5, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_allones_eq:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3],xmm2[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: psrld $5, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $31, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_allones_eq:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm3, %xmm3
; CHECK-AVX1-NEXT: vpsrld $5, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_allones_eq:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_allones_eq:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 14, i32 4294967295, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
define <4 x i32> @test_urem_odd_even_allones_ne(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_allones_ne:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm3
; CHECK-SSE2-NEXT: psrld $1, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3
; CHECK-SSE2-NEXT: psrld $2, %xmm3
; CHECK-SSE2-NEXT: psrld $31, %xmm2
; CHECK-SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [5,14,4294967295,100]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: psrld $5, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_allones_ne:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3],xmm2[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: psrld $5, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $31, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pandn {{.*}}(%rip), %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_allones_ne:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm3, %xmm3
; CHECK-AVX1-NEXT: vpsrld $5, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpandn {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_allones_ne:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,2147483649,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %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_urem_odd_even_allones_ne:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: vprorvd {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: vpmaxud {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX512VL-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX512VL-NEXT: retq
%urem = urem <4 x i32> %X, <i32 5, i32 14, i32 4294967295, i32 100>
%cmp = icmp ne <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
; One power-of-two divisor in odd divisor
define <4 x i32> @test_urem_odd_poweroftwo(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_poweroftwo:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = <3435973837,u,268435456,u>
; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_poweroftwo:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = <3435973837,u,268435456,u>
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_poweroftwo:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_poweroftwo:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_poweroftwo:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 5, i32 16, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One power-of-two divisor in even divisor
define <4 x i32> @test_urem_even_poweroftwo(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_poweroftwo:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_poweroftwo:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_poweroftwo:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_poweroftwo:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
; CHECK-AVX2-NEXT: vpmuludq %xmm3, %xmm2, %xmm2
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_poweroftwo:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 14, i32 16, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One power-of-two divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_poweroftwo(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_poweroftwo:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2454267027,268435456,1374389535]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm3
; CHECK-SSE2-NEXT: psrld $1, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3
; CHECK-SSE2-NEXT: psrld $2, %xmm3
; CHECK-SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [5,14,16,100]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: psrld $5, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_poweroftwo:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,2454267027,268435456,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[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 %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $5, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm3
; CHECK-SSE41-NEXT: psubd %xmm3, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_poweroftwo:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,2454267027,268435456,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $5, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_poweroftwo:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,268435456,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_poweroftwo:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 14, i32 16, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
; One one divisor in odd divisor
define <4 x i32> @test_urem_odd_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_one:
; CHECK-SSE2: # %bb.0:
; 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: pxor {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: pcmpgtd {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: pandn {{.*}}(%rip), %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993459,858993459,4294967295,858993459]
; 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_urem_odd_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%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_urem_odd_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [3435973837,3435973837,3435973837,3435973837]
; CHECK-AVX2-NEXT: vpmulld %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_urem_odd_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 5, i32 1, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One one divisor in even divisor
define <4 x i32> @test_urem_even_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm0[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
; CHECK-AVX2-NEXT: vpmuludq %xmm3, %xmm2, %xmm2
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %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: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 14, i32 1, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One one divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2454267027,0,1374389535]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm3
; CHECK-SSE2-NEXT: psrld $1, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: movaps %xmm0, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: movaps %xmm2, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,1],xmm3[0,2]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [5,14,1,100]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
; CHECK-SSE2-NEXT: psrld $5, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
; CHECK-SSE2-NEXT: psubd %xmm4, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,2454267027,0,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: psrld $5, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm1[0,1,2,3],xmm3[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm0[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm3
; CHECK-SSE41-NEXT: psubd %xmm3, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,2454267027,0,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpsrld $5, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,0,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 14, i32 1, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
; One INT_MIN divisor in odd divisor
define <4 x i32> @test_urem_odd_INT_MIN(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_INT_MIN:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = <3435973837,u,2,u>
; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_INT_MIN:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = <3435973837,u,2,u>
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_INT_MIN:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_INT_MIN:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [3435973837,3435973837,3435973837,3435973837]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_INT_MIN:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 5, i32 2147483648, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One INT_MIN divisor in even divisor
define <4 x i32> @test_urem_even_INT_MIN(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_INT_MIN:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq {{.*}}(%rip), %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_INT_MIN:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_INT_MIN:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_INT_MIN:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpbroadcastd {{.*#+}} xmm3 = [2454267027,2454267027,2454267027,2454267027]
; CHECK-AVX2-NEXT: vpmuludq %xmm3, %xmm2, %xmm2
; CHECK-AVX2-NEXT: vpmuludq {{.*}}(%rip), %xmm1, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_INT_MIN:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 14, i32 2147483648, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One INT_MIN divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_INT_MIN(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_INT_MIN:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2454267027,2,1374389535]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm3
; CHECK-SSE2-NEXT: psrld $1, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm3
; CHECK-SSE2-NEXT: psrld $2, %xmm3
; CHECK-SSE2-NEXT: movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [5,14,2147483648,100]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: psrld $5, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm4[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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_INT_MIN:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [3435973837,2454267027,2,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[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 %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $5, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm3
; CHECK-SSE41-NEXT: psubd %xmm3, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_INT_MIN:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [3435973837,2454267027,2,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $5, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_INT_MIN:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2454267027,2,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_INT_MIN:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 14, i32 2147483648, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;==============================================================================;
; One all-ones divisor and power-of-two divisor divisor in odd divisor
define <4 x i32> @test_urem_odd_allones_and_poweroftwo(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_allones_and_poweroftwo:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,3435973837]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; CHECK-SSE2-NEXT: movdqa %xmm2, %xmm1
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: psrld $31, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm2[2,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [5,4294967295,16,5]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[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 {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_allones_and_poweroftwo:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,3435973837]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE41-NEXT: pmuludq %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 %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $31, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm3
; CHECK-SSE41-NEXT: psubd %xmm3, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_allones_and_poweroftwo:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,3435973837]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX1-NEXT: vpmuludq %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: vpsrld $2, %xmm1, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_allones_and_poweroftwo:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,3435973837]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_allones_and_poweroftwo:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 4294967295, i32 16, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One all-ones divisor and power-of-two divisor divisor in even divisor
define <4 x i32> @test_urem_even_allones_and_poweroftwo(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_allones_and_poweroftwo:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2454267027,2147483649,268435456,2454267027]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[1,2]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2,3,1]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: psrld $31, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,1],xmm2[3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm1[2,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [14,4294967295,16,14]
; 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: pmuludq %xmm4, %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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_allones_and_poweroftwo:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm3 = [2454267027,2147483649,268435456,2454267027]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm4
; CHECK-SSE41-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: psrld $31, %xmm4
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm2[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm4 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm4
; CHECK-SSE41-NEXT: psubd %xmm4, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_allones_and_poweroftwo:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [2454267027,2147483649,268435456,2454267027]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm4, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %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: vpsrld $2, %xmm1, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_allones_and_poweroftwo:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,2147483649,268435456,2454267027]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_allones_and_poweroftwo:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 4294967295, i32 16, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One all-ones divisor and power-of-two divisor divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_allones_and_poweroftwo(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_allones_and_poweroftwo:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,1374389535]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE2-NEXT: movdqa %xmm3, %xmm2
; CHECK-SSE2-NEXT: psrld $5, %xmm2
; CHECK-SSE2-NEXT: psrld $31, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [5,4294967295,16,100]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm4[2,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_allones_and_poweroftwo:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE41-NEXT: movdqa %xmm3, %xmm2
; CHECK-SSE41-NEXT: psrld $5, %xmm2
; CHECK-SSE41-NEXT: psrld $31, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $2, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3],xmm2[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_allones_and_poweroftwo:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX1-NEXT: vpsrld $5, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_allones_and_poweroftwo:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_allones_and_poweroftwo:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 4294967295, i32 16, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
; One all-ones divisor and one one divisor in odd divisor
define <4 x i32> @test_urem_odd_allones_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_allones_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,4294967295,0,3435973837]
; 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: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; 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_urem_odd_allones_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [858993459,1,4294967295,858993459]
; CHECK-SSE41-NEXT: pminud %xmm0, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX-LABEL: test_urem_odd_allones_and_one:
; CHECK-AVX: # %bb.0:
; CHECK-AVX-NEXT: vpmulld {{.*}}(%rip), %xmm0, %xmm0
; CHECK-AVX-NEXT: vpminud {{.*}}(%rip), %xmm0, %xmm1
; CHECK-AVX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX-NEXT: retq
%urem = urem <4 x i32> %X, <i32 5, i32 4294967295, i32 1, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One all-ones divisor and one one divisor in even divisor
define <4 x i32> @test_urem_even_allones_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_allones_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2454267027,2147483649,0,2454267027]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[1,2]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2,3,1]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: movaps %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: psrld $31, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [14,4294967295,1,14]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_allones_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm3 = [2454267027,2147483649,0,2454267027]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm4
; CHECK-SSE41-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: psrld $31, %xmm4
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm4 = xmm1[0,1],xmm4[2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm0[4,5],xmm4[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm4
; CHECK-SSE41-NEXT: psubd %xmm4, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_allones_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [2454267027,2147483649,0,2454267027]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm4, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %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: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_allones_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,2147483649,0,2454267027]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_allones_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 4294967295, i32 1, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One all-ones divisor and one one divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_allones_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_allones_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,0,1374389535]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE2-NEXT: movdqa %xmm3, %xmm2
; CHECK-SSE2-NEXT: psrld $5, %xmm2
; CHECK-SSE2-NEXT: psrld $31, %xmm3
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [5,4294967295,1,100]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm4[0,2]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_allones_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,0,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE41-NEXT: movdqa %xmm3, %xmm2
; CHECK-SSE41-NEXT: psrld $5, %xmm2
; CHECK-SSE41-NEXT: psrld $31, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_allones_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,0,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX1-NEXT: vpsrld $5, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_allones_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,0,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_allones_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 4294967295, i32 1, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
; One power-of-two divisor divisor and one divisor in odd divisor
define <4 x i32> @test_urem_odd_poweroftwo_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_poweroftwo_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,268435456,0,3435973837]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm3
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[3,0]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[0,2]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [5,16,1,5]
; 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: 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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_poweroftwo_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,268435456,0,3435973837]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE41-NEXT: pmuludq %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: psrld $2, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_poweroftwo_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,0,3435973837]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX1-NEXT: vpmuludq %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: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_poweroftwo_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,0,3435973837]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_poweroftwo_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 16, i32 1, i32 5>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One power-of-two divisor divisor and one divisor in even divisor
define <4 x i32> @test_urem_even_poweroftwo_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_poweroftwo_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [2454267027,268435456,0,2454267027]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm4[1,3,2,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[1,2]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,2,3,1]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,3,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: movaps %xmm0, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,1],xmm1[3,3]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0,2]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [14,16,1,14]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_poweroftwo_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm3 = [2454267027,268435456,0,2454267027]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm4
; CHECK-SSE41-NEXT: pmuludq %xmm3, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5],xmm4[6,7]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_poweroftwo_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [2454267027,268435456,0,2454267027]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm4, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpmuludq %xmm3, %xmm1, %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: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_poweroftwo_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,268435456,0,2454267027]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_poweroftwo_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 16, i32 1, i32 14>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
; One power-of-two divisor divisor and one divisor in odd+even divisor
define <4 x i32> @test_urem_odd_even_poweroftwo_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_even_poweroftwo_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,268435456,0,1374389535]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE2-NEXT: movdqa %xmm3, %xmm2
; CHECK-SSE2-NEXT: psrld $5, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm3 = xmm3[1,1],xmm2[3,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm2 = [5,16,1,100]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %xmm4
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
; CHECK-SSE2-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE2-NEXT: psrld $2, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[2,0],xmm1[3,0]
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm1 = xmm1[0,1],xmm4[0,2]
; CHECK-SSE2-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
; CHECK-SSE2-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
; CHECK-SSE2-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_even_poweroftwo_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,268435456,0,1374389535]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm3
; CHECK-SSE41-NEXT: movdqa %xmm3, %xmm2
; CHECK-SSE41-NEXT: psrld $5, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-SSE41-NEXT: pmuludq %xmm0, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: psrld $2, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm1
; CHECK-SSE41-NEXT: psubd %xmm1, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_even_poweroftwo_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,0,1374389535]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX1-NEXT: vpsrld $5, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm3[4,5,6,7]
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_even_poweroftwo_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,268435456,0,1374389535]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_even_poweroftwo_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 16, i32 1, i32 100>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
;------------------------------------------------------------------------------;
define <4 x i32> @test_urem_odd_allones_and_poweroftwo_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_odd_allones_and_poweroftwo_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,0]
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm3[2,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [5,4294967295,16,1]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %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 {{.*#+}} xmm4 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm4
; CHECK-SSE2-NEXT: psrld $31, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_odd_allones_and_poweroftwo_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,0]
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa %xmm2, %xmm3
; CHECK-SSE41-NEXT: psrld $2, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $31, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5],xmm0[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_odd_allones_and_poweroftwo_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,0]
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm0, %xmm2
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm2, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm1, %xmm3, %xmm1
; CHECK-AVX1-NEXT: vpsrld $31, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5],xmm0[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_odd_allones_and_poweroftwo_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [3435973837,2147483649,268435456,0]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1,2],xmm0[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_odd_allones_and_poweroftwo_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 5, i32 4294967295, i32 16, i32 1>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
define <4 x i32> @test_urem_even_allones_and_poweroftwo_and_one(<4 x i32> %X) nounwind {
; CHECK-SSE2-LABEL: test_urem_even_allones_and_poweroftwo_and_one:
; CHECK-SSE2: # %bb.0:
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE2-NEXT: psrld $1, %xmm1
; CHECK-SSE2-NEXT: movdqa %xmm0, %xmm2
; CHECK-SSE2-NEXT: movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [2454267027,2147483649,268435456,0]
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,3,2,3]
; CHECK-SSE2-NEXT: psrld $2, %xmm2
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,3],xmm3[2,3]
; CHECK-SSE2-NEXT: movdqa {{.*#+}} xmm3 = [14,4294967295,16,1]
; CHECK-SSE2-NEXT: pmuludq %xmm3, %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 {{.*#+}} xmm4 = xmm0[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm1, %xmm4
; CHECK-SSE2-NEXT: psrld $31, %xmm4
; CHECK-SSE2-NEXT: shufps {{.*#+}} xmm4 = xmm4[1,1],xmm0[3,3]
; CHECK-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm3[1,1,3,3]
; CHECK-SSE2-NEXT: pmuludq %xmm4, %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: psubd %xmm2, %xmm0
; CHECK-SSE2-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE2-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE2-NEXT: psrld $31, %xmm0
; CHECK-SSE2-NEXT: retq
;
; CHECK-SSE41-LABEL: test_urem_even_allones_and_poweroftwo_and_one:
; CHECK-SSE41: # %bb.0:
; CHECK-SSE41-NEXT: movdqa %xmm0, %xmm1
; CHECK-SSE41-NEXT: psrld $1, %xmm1
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5,6,7]
; CHECK-SSE41-NEXT: movdqa {{.*#+}} xmm2 = [2454267027,2147483649,268435456,0]
; CHECK-SSE41-NEXT: pmuludq %xmm2, %xmm1
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-SSE41-NEXT: movdqa %xmm1, %xmm3
; CHECK-SSE41-NEXT: psrld $2, %xmm3
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm2[1,1,3,3]
; CHECK-SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
; CHECK-SSE41-NEXT: pmuludq %xmm1, %xmm2
; CHECK-SSE41-NEXT: psrld $31, %xmm2
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
; CHECK-SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5],xmm0[6,7]
; CHECK-SSE41-NEXT: pmulld {{.*}}(%rip), %xmm2
; CHECK-SSE41-NEXT: psubd %xmm2, %xmm0
; CHECK-SSE41-NEXT: pxor %xmm1, %xmm1
; CHECK-SSE41-NEXT: pcmpeqd %xmm1, %xmm0
; CHECK-SSE41-NEXT: psrld $31, %xmm0
; CHECK-SSE41-NEXT: retq
;
; CHECK-AVX1-LABEL: test_urem_even_allones_and_poweroftwo_and_one:
; CHECK-AVX1: # %bb.0:
; CHECK-AVX1-NEXT: vpsrld $1, %xmm0, %xmm1
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3,4,5,6,7]
; CHECK-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [2454267027,2147483649,268435456,0]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
; CHECK-AVX1-NEXT: vpsrld $2, %xmm1, %xmm3
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
; CHECK-AVX1-NEXT: vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
; CHECK-AVX1-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
; CHECK-AVX1-NEXT: vpsrld $31, %xmm2, %xmm2
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
; CHECK-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5],xmm0[6,7]
; CHECK-AVX1-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX1-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX1-NEXT: retq
;
; CHECK-AVX2-LABEL: test_urem_even_allones_and_poweroftwo_and_one:
; CHECK-AVX2: # %bb.0:
; CHECK-AVX2-NEXT: vmovdqa {{.*#+}} xmm1 = [2454267027,2147483649,268435456,0]
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
; CHECK-AVX2-NEXT: vpsrlvd {{.*}}(%rip), %xmm0, %xmm3
; CHECK-AVX2-NEXT: vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
; CHECK-AVX2-NEXT: vpmuludq %xmm2, %xmm4, %xmm2
; CHECK-AVX2-NEXT: vpmuludq %xmm1, %xmm3, %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: vpsrlvd {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1,2],xmm0[3]
; CHECK-AVX2-NEXT: vpmulld {{.*}}(%rip), %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpsubd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; CHECK-AVX2-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
; CHECK-AVX2-NEXT: vpsrld $31, %xmm0, %xmm0
; CHECK-AVX2-NEXT: retq
;
; CHECK-AVX512VL-LABEL: test_urem_even_allones_and_poweroftwo_and_one:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vpmulld {{.*}}(%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
%urem = urem <4 x i32> %X, <i32 14, i32 4294967295, i32 16, i32 1>
%cmp = icmp eq <4 x i32> %urem, <i32 0, i32 0, i32 0, i32 0>
%ret = zext <4 x i1> %cmp to <4 x i32>
ret <4 x i32> %ret
}
Reference in New Issue View Git Blame Copy Permalink