163 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX13; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX24; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX,AVX5125 6define <4 x double> @concat_roundpd_v4f64_v2f64(<2 x double> %a0, <2 x double> %a1) {7; AVX-LABEL: concat_roundpd_v4f64_v2f64:8; AVX: # %bb.0:9; AVX-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm010; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm011; AVX-NEXT: vroundpd $4, %ymm0, %ymm012; AVX-NEXT: retq13 %v0 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a0, i32 4)14 %v1 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a1, i32 4)15 %res = shufflevector <2 x double> %v0, <2 x double> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>16 ret <4 x double> %res17}18 19define <8 x float> @concat_roundps_v8f32_v4f32(<4 x float> %a0, <4 x float> %a1) {20; AVX-LABEL: concat_roundps_v8f32_v4f32:21; AVX: # %bb.0:22; AVX-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm023; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm024; AVX-NEXT: vroundps $4, %ymm0, %ymm025; AVX-NEXT: retq26 %v0 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a0, i32 4)27 %v1 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a1, i32 4)28 %res = shufflevector <4 x float> %v0, <4 x float> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>29 ret <8 x float> %res30}31 32define <8 x double> @concat_roundpd_v8f64_v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, <2 x double> %a3) {33; AVX1-LABEL: concat_roundpd_v8f64_v2f64:34; AVX1: # %bb.0:35; AVX1-NEXT: # kill: def $xmm2 killed $xmm2 def $ymm236; AVX1-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm037; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm038; AVX1-NEXT: vroundpd $4, %ymm0, %ymm039; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm140; AVX1-NEXT: vroundpd $4, %ymm1, %ymm141; AVX1-NEXT: retq42;43; AVX2-LABEL: concat_roundpd_v8f64_v2f64:44; AVX2: # %bb.0:45; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm046; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm047; AVX2-NEXT: vroundpd $4, %ymm0, %ymm048; AVX2-NEXT: # kill: def $xmm2 killed $xmm2 def $ymm249; AVX2-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm150; AVX2-NEXT: vroundpd $4, %ymm1, %ymm151; AVX2-NEXT: retq52;53; AVX512-LABEL: concat_roundpd_v8f64_v2f64:54; AVX512: # %bb.0:55; AVX512-NEXT: # kill: def $xmm2 killed $xmm2 def $ymm256; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm057; AVX512-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm258; AVX512-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm059; AVX512-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm060; AVX512-NEXT: vrndscalepd $4, %zmm0, %zmm061; AVX512-NEXT: retq62 %v0 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a0, i32 4)63 %v1 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a1, i32 4)64 %v2 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a2, i32 4)65 %v3 = call <2 x double> @llvm.x86.sse41.round.pd(<2 x double> %a3, i32 4)66 %r01 = shufflevector <2 x double> %v0, <2 x double> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>67 %r23 = shufflevector <2 x double> %v2, <2 x double> %v3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>68 %res = shufflevector <4 x double> %r01, <4 x double> %r23, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>69 ret <8 x double> %res70}71 72define <16 x float> @concat_roundps_v16f32_v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, <4 x float> %a3) {73; AVX1-LABEL: concat_roundps_v16f32_v4f32:74; AVX1: # %bb.0:75; AVX1-NEXT: # kill: def $xmm2 killed $xmm2 def $ymm276; AVX1-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm077; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm078; AVX1-NEXT: vroundps $4, %ymm0, %ymm079; AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm180; AVX1-NEXT: vroundps $4, %ymm1, %ymm181; AVX1-NEXT: retq82;83; AVX2-LABEL: concat_roundps_v16f32_v4f32:84; AVX2: # %bb.0:85; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm086; AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm087; AVX2-NEXT: vroundps $4, %ymm0, %ymm088; AVX2-NEXT: # kill: def $xmm2 killed $xmm2 def $ymm289; AVX2-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm190; AVX2-NEXT: vroundps $4, %ymm1, %ymm191; AVX2-NEXT: retq92;93; AVX512-LABEL: concat_roundps_v16f32_v4f32:94; AVX512: # %bb.0:95; AVX512-NEXT: # kill: def $xmm2 killed $xmm2 def $ymm296; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm097; AVX512-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm298; AVX512-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm099; AVX512-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0100; AVX512-NEXT: vrndscaleps $4, %zmm0, %zmm0101; AVX512-NEXT: retq102 %v0 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a0, i32 4)103 %v1 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a1, i32 4)104 %v2 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a2, i32 4)105 %v3 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a3, i32 4)106 %r01 = shufflevector <4 x float> %v0, <4 x float> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>107 %r23 = shufflevector <4 x float> %v2, <4 x float> %v3, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>108 %res = shufflevector <8 x float> %r01, <8 x float> %r23, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>109 ret <16 x float> %res110}111 112define <8 x double> @concat_roundpd_v8f64_v4f64(<4 x double> %a0, <4 x double> %a1) {113; AVX1OR2-LABEL: concat_roundpd_v8f64_v4f64:114; AVX1OR2: # %bb.0:115; AVX1OR2-NEXT: vroundpd $4, %ymm0, %ymm0116; AVX1OR2-NEXT: vroundpd $4, %ymm1, %ymm1117; AVX1OR2-NEXT: retq118;119; AVX512-LABEL: concat_roundpd_v8f64_v4f64:120; AVX512: # %bb.0:121; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0122; AVX512-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0123; AVX512-NEXT: vrndscalepd $4, %zmm0, %zmm0124; AVX512-NEXT: retq125 %v0 = call <4 x double> @llvm.x86.avx.round.pd.256(<4 x double> %a0, i32 4)126 %v1 = call <4 x double> @llvm.x86.avx.round.pd.256(<4 x double> %a1, i32 4)127 %res = shufflevector <4 x double> %v0, <4 x double> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>128 ret <8 x double> %res129}130 131define <16 x float> @concat_roundps_v16f32_v8f32(<8 x float> %a0, <8 x float> %a1) {132; AVX1OR2-LABEL: concat_roundps_v16f32_v8f32:133; AVX1OR2: # %bb.0:134; AVX1OR2-NEXT: vroundps $4, %ymm0, %ymm0135; AVX1OR2-NEXT: vroundps $4, %ymm1, %ymm1136; AVX1OR2-NEXT: retq137;138; AVX512-LABEL: concat_roundps_v16f32_v8f32:139; AVX512: # %bb.0:140; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0141; AVX512-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0142; AVX512-NEXT: vrndscaleps $4, %zmm0, %zmm0143; AVX512-NEXT: retq144 %v0 = call <8 x float> @llvm.x86.avx.round.ps.256(<8 x float> %a0, i32 4)145 %v1 = call <8 x float> @llvm.x86.avx.round.ps.256(<8 x float> %a1, i32 4)146 %res = shufflevector <8 x float> %v0, <8 x float> %v1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>147 ret <16 x float> %res148}149 150; negative test - rounding mode mismatch151define <8 x float> @concat_roundps_v8f32_v4f32_mismatch(<4 x float> %a0, <4 x float> %a1) {152; AVX-LABEL: concat_roundps_v8f32_v4f32_mismatch:153; AVX: # %bb.0:154; AVX-NEXT: vroundps $0, %xmm0, %xmm0155; AVX-NEXT: vroundps $4, %xmm1, %xmm1156; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0157; AVX-NEXT: retq158 %v0 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a0, i32 0)159 %v1 = call <4 x float> @llvm.x86.sse41.round.ps(<4 x float> %a1, i32 4)160 %res = shufflevector <4 x float> %v0, <4 x float> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>161 ret <8 x float> %res162}163