96 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i386-- -mattr=sse2 | FileCheck %s --check-prefixes=X86,X86-SSE3; RUN: llc < %s -mtriple=i386-- -mattr=sse4.1 | FileCheck %s --check-prefixes=X86,X86-SSE4; RUN: llc < %s -mtriple=i386-- -mattr=avx2 | FileCheck %s --check-prefixes=X86,X86-AVX25; RUN: llc < %s -mtriple=i386-- -mattr=avx512f | FileCheck %s --check-prefixes=X86,X86-AVX512F6; RUN: llc < %s -mtriple=x86_64-- -mattr=sse2 | FileCheck %s --check-prefix=X64-SSE7; RUN: llc < %s -mtriple=x86_64-- -mattr=sse4.1 | FileCheck %s --check-prefix=X64-SSE8; RUN: llc < %s -mtriple=x86_64-- -mattr=avx2 | FileCheck %s --check-prefixes=X64-AVX,X64-AVX29; RUN: llc < %s -mtriple=x86_64-- -mattr=avx512f | FileCheck %s --check-prefixes=X64-AVX,X64-AVX512F10 11; This should do a single load into the fp stack for the return, not diddle with xmm registers.12 13define float @icmp_select_fp_constants(i32 %x) nounwind readnone {14; X86-LABEL: icmp_select_fp_constants:15; X86: # %bb.0:16; X86-NEXT: xorl %eax, %eax17; X86-NEXT: cmpl $0, {{[0-9]+}}(%esp)18; X86-NEXT: sete %al19; X86-NEXT: flds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)20; X86-NEXT: retl21;22; X64-SSE-LABEL: icmp_select_fp_constants:23; X64-SSE: # %bb.0:24; X64-SSE-NEXT: xorl %eax, %eax25; X64-SSE-NEXT: testl %edi, %edi26; X64-SSE-NEXT: sete %al27; X64-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero28; X64-SSE-NEXT: retq29;30; X64-AVX-LABEL: icmp_select_fp_constants:31; X64-AVX: # %bb.0:32; X64-AVX-NEXT: xorl %eax, %eax33; X64-AVX-NEXT: testl %edi, %edi34; X64-AVX-NEXT: sete %al35; X64-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero36; X64-AVX-NEXT: retq37 %c = icmp eq i32 %x, 038 %r = select i1 %c, float 42.0, float 23.039 ret float %r40}41 42define float @fcmp_select_fp_constants(float %x) nounwind readnone {43; X86-SSE-LABEL: fcmp_select_fp_constants:44; X86-SSE: # %bb.0:45; X86-SSE-NEXT: movss {{.*#+}} xmm0 = [-4.0E+0,0.0E+0,0.0E+0,0.0E+0]46; X86-SSE-NEXT: cmpneqss {{[0-9]+}}(%esp), %xmm047; X86-SSE-NEXT: movd %xmm0, %eax48; X86-SSE-NEXT: andl $1, %eax49; X86-SSE-NEXT: flds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)50; X86-SSE-NEXT: retl51;52; X86-AVX2-LABEL: fcmp_select_fp_constants:53; X86-AVX2: # %bb.0:54; X86-AVX2-NEXT: vmovss {{.*#+}} xmm0 = [-4.0E+0,0.0E+0,0.0E+0,0.0E+0]55; X86-AVX2-NEXT: vcmpneqss {{[0-9]+}}(%esp), %xmm0, %xmm056; X86-AVX2-NEXT: vmovd %xmm0, %eax57; X86-AVX2-NEXT: andl $1, %eax58; X86-AVX2-NEXT: flds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)59; X86-AVX2-NEXT: retl60;61; X86-AVX512F-LABEL: fcmp_select_fp_constants:62; X86-AVX512F: # %bb.0:63; X86-AVX512F-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero64; X86-AVX512F-NEXT: vcmpneqss {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %k065; X86-AVX512F-NEXT: kmovw %k0, %eax66; X86-AVX512F-NEXT: flds {{\.?LCPI[0-9]+_[0-9]+}}(,%eax,4)67; X86-AVX512F-NEXT: retl68;69; X64-SSE-LABEL: fcmp_select_fp_constants:70; X64-SSE: # %bb.0:71; X64-SSE-NEXT: cmpneqss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm072; X64-SSE-NEXT: movd %xmm0, %eax73; X64-SSE-NEXT: andl $1, %eax74; X64-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero75; X64-SSE-NEXT: retq76;77; X64-AVX2-LABEL: fcmp_select_fp_constants:78; X64-AVX2: # %bb.0:79; X64-AVX2-NEXT: vcmpneqss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm080; X64-AVX2-NEXT: vbroadcastss {{.*#+}} xmm1 = [4.2E+1,4.2E+1,4.2E+1,4.2E+1]81; X64-AVX2-NEXT: vbroadcastss {{.*#+}} xmm2 = [2.3E+1,2.3E+1,2.3E+1,2.3E+1]82; X64-AVX2-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm083; X64-AVX2-NEXT: retq84;85; X64-AVX512F-LABEL: fcmp_select_fp_constants:86; X64-AVX512F: # %bb.0:87; X64-AVX512F-NEXT: vcmpneqss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %k188; X64-AVX512F-NEXT: vmovss {{.*#+}} xmm0 = [2.3E+1,0.0E+0,0.0E+0,0.0E+0]89; X64-AVX512F-NEXT: vmovss {{.*#+}} xmm0 {%k1} = [4.2E+1,0.0E+0,0.0E+0,0.0E+0]90; X64-AVX512F-NEXT: retq91 %c = fcmp une float %x, -4.092 %r = select i1 %c, float 42.0, float 23.093 ret float %r94}95 96