82 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE23; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE414; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX15; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX26 7define <2 x i16> @test_urem_unary_v2i16() nounwind {8; SSE-LABEL: test_urem_unary_v2i16:9; SSE: # %bb.0:10; SSE-NEXT: xorps %xmm0, %xmm011; SSE-NEXT: retq12;13; AVX-LABEL: test_urem_unary_v2i16:14; AVX: # %bb.0:15; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm016; AVX-NEXT: retq17 %I8 = insertelement <2 x i16> zeroinitializer, i16 -1, i32 018 %I9 = insertelement <2 x i16> %I8, i16 -1, i32 119 %B9 = urem <2 x i16> %I9, %I920 ret <2 x i16> %B921}22 23define <4 x i32> @PR20355(<4 x i32> %a) nounwind {24; SSE2-LABEL: PR20355:25; SSE2: # %bb.0: # %entry26; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [1431655766,1431655766,1431655766,1431655766]27; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]28; SSE2-NEXT: pxor %xmm3, %xmm329; SSE2-NEXT: pcmpgtd %xmm0, %xmm330; SSE2-NEXT: pmuludq %xmm1, %xmm031; SSE2-NEXT: pshufd {{.*#+}} xmm4 = xmm0[1,3,2,3]32; SSE2-NEXT: pmuludq %xmm1, %xmm233; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[1,3,2,3]34; SSE2-NEXT: punpckldq {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1]35; SSE2-NEXT: pand %xmm1, %xmm336; SSE2-NEXT: psubd %xmm3, %xmm437; SSE2-NEXT: movdqa %xmm4, %xmm038; SSE2-NEXT: psrld $31, %xmm039; SSE2-NEXT: paddd %xmm4, %xmm040; SSE2-NEXT: retq41;42; SSE41-LABEL: PR20355:43; SSE41: # %bb.0: # %entry44; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]45; SSE41-NEXT: movdqa {{.*#+}} xmm2 = [1431655766,1431655766,1431655766,1431655766]46; SSE41-NEXT: pmuldq %xmm2, %xmm147; SSE41-NEXT: pmuldq %xmm2, %xmm048; SSE41-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]49; SSE41-NEXT: pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7]50; SSE41-NEXT: movdqa %xmm2, %xmm051; SSE41-NEXT: psrld $31, %xmm052; SSE41-NEXT: paddd %xmm2, %xmm053; SSE41-NEXT: retq54;55; AVX1-LABEL: PR20355:56; AVX1: # %bb.0: # %entry57; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]58; AVX1-NEXT: vbroadcastss {{.*#+}} xmm2 = [1431655766,1431655766,1431655766,1431655766]59; AVX1-NEXT: vpmuldq %xmm2, %xmm1, %xmm160; AVX1-NEXT: vpmuldq %xmm2, %xmm0, %xmm061; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]62; AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]63; AVX1-NEXT: vpsrld $31, %xmm0, %xmm164; AVX1-NEXT: vpaddd %xmm1, %xmm0, %xmm065; AVX1-NEXT: retq66;67; AVX2-LABEL: PR20355:68; AVX2: # %bb.0: # %entry69; AVX2-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]70; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm2 = [1431655766,1431655766,1431655766,1431655766]71; AVX2-NEXT: vpmuldq %xmm2, %xmm1, %xmm172; AVX2-NEXT: vpmuldq %xmm2, %xmm0, %xmm073; AVX2-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]74; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]75; AVX2-NEXT: vpsrld $31, %xmm0, %xmm176; AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm077; AVX2-NEXT: retq78entry:79 %sdiv = sdiv <4 x i32> %a, <i32 3, i32 3, i32 3, i32 3>80 ret <4 x i32> %sdiv81}82