117 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s3 4define <4 x i32> @foo(<4 x i32> %t, <4 x i32> %u) nounwind {5; CHECK-LABEL: foo:6; CHECK: # %bb.0:7; CHECK-NEXT: pshufd {{.*#+}} xmm2 = xmm0[3,3,3,3]8; CHECK-NEXT: movd %xmm2, %eax9; CHECK-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,3,3,3]10; CHECK-NEXT: movd %xmm2, %ecx11; CHECK-NEXT: cltd12; CHECK-NEXT: idivl %ecx13; CHECK-NEXT: movd %edx, %xmm214; CHECK-NEXT: pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]15; CHECK-NEXT: movd %xmm3, %eax16; CHECK-NEXT: pshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]17; CHECK-NEXT: movd %xmm3, %ecx18; CHECK-NEXT: cltd19; CHECK-NEXT: idivl %ecx20; CHECK-NEXT: movd %edx, %xmm321; CHECK-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]22; CHECK-NEXT: movd %xmm0, %eax23; CHECK-NEXT: movd %xmm1, %ecx24; CHECK-NEXT: cltd25; CHECK-NEXT: idivl %ecx26; CHECK-NEXT: movd %edx, %xmm227; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]28; CHECK-NEXT: movd %xmm0, %eax29; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]30; CHECK-NEXT: movd %xmm0, %ecx31; CHECK-NEXT: cltd32; CHECK-NEXT: idivl %ecx33; CHECK-NEXT: movd %edx, %xmm034; CHECK-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]35; CHECK-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]36; CHECK-NEXT: movdqa %xmm2, %xmm037; CHECK-NEXT: retq38 %m = srem <4 x i32> %t, %u39 ret <4 x i32> %m40}41 42define <4 x i32> @bar(<4 x i32> %t, <4 x i32> %u) nounwind {43; CHECK-LABEL: bar:44; CHECK: # %bb.0:45; CHECK-NEXT: pshufd {{.*#+}} xmm2 = xmm0[3,3,3,3]46; CHECK-NEXT: movd %xmm2, %eax47; CHECK-NEXT: pshufd {{.*#+}} xmm2 = xmm1[3,3,3,3]48; CHECK-NEXT: movd %xmm2, %ecx49; CHECK-NEXT: xorl %edx, %edx50; CHECK-NEXT: divl %ecx51; CHECK-NEXT: movd %edx, %xmm252; CHECK-NEXT: pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]53; CHECK-NEXT: movd %xmm3, %eax54; CHECK-NEXT: pshufd {{.*#+}} xmm3 = xmm1[2,3,2,3]55; CHECK-NEXT: movd %xmm3, %ecx56; CHECK-NEXT: xorl %edx, %edx57; CHECK-NEXT: divl %ecx58; CHECK-NEXT: movd %edx, %xmm359; CHECK-NEXT: punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]60; CHECK-NEXT: movd %xmm0, %eax61; CHECK-NEXT: movd %xmm1, %ecx62; CHECK-NEXT: xorl %edx, %edx63; CHECK-NEXT: divl %ecx64; CHECK-NEXT: movd %edx, %xmm265; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]66; CHECK-NEXT: movd %xmm0, %eax67; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]68; CHECK-NEXT: movd %xmm0, %ecx69; CHECK-NEXT: xorl %edx, %edx70; CHECK-NEXT: divl %ecx71; CHECK-NEXT: movd %edx, %xmm072; CHECK-NEXT: punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]73; CHECK-NEXT: punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]74; CHECK-NEXT: movdqa %xmm2, %xmm075; CHECK-NEXT: retq76 %m = urem <4 x i32> %t, %u77 ret <4 x i32> %m78}79 80define <4 x float> @qux(<4 x float> %t, <4 x float> %u) nounwind {81; CHECK-LABEL: qux:82; CHECK: # %bb.0:83; CHECK-NEXT: subq $72, %rsp84; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill85; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill86; CHECK-NEXT: shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]87; CHECK-NEXT: shufps {{.*#+}} xmm1 = xmm1[3,3,3,3]88; CHECK-NEXT: callq fmodf@PLT89; CHECK-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill90; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload91; CHECK-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]92; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload93; CHECK-NEXT: movhlps {{.*#+}} xmm1 = xmm1[1,1]94; CHECK-NEXT: callq fmodf@PLT95; CHECK-NEXT: unpcklps (%rsp), %xmm0 # 16-byte Folded Reload96; CHECK-NEXT: # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]97; CHECK-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill98; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload99; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload100; CHECK-NEXT: callq fmodf@PLT101; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill102; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload103; CHECK-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]104; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload105; CHECK-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,1,1]106; CHECK-NEXT: callq fmodf@PLT107; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload108; CHECK-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]109; CHECK-NEXT: unpcklpd (%rsp), %xmm1 # 16-byte Folded Reload110; CHECK-NEXT: # xmm1 = xmm1[0],mem[0]111; CHECK-NEXT: movaps %xmm1, %xmm0112; CHECK-NEXT: addq $72, %rsp113; CHECK-NEXT: retq114 %m = frem <4 x float> %t, %u115 ret <4 x float> %m116}117