221 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16,+avx512vl | FileCheck %s3 4declare half @llvm.minimum.f16(half, half)5declare half @llvm.maximum.f16(half, half)6declare <8 x half> @llvm.minimum.v8f16(<8 x half>, <8 x half>)7declare <8 x half> @llvm.maximum.v8f16(<8 x half>, <8 x half>)8declare <16 x half> @llvm.minimum.v16f16(<16 x half>, <16 x half>)9declare <16 x half> @llvm.maximum.v16f16(<16 x half>, <16 x half>)10declare <32 x half> @llvm.minimum.v32f16(<32 x half>, <32 x half>)11declare <32 x half> @llvm.maximum.v32f16(<32 x half>, <32 x half>)12 13define half @test_fminimum(half %x, half %y) {14; CHECK-LABEL: test_fminimum:15; CHECK: # %bb.0:16; CHECK-NEXT: vmovw %xmm0, %eax17; CHECK-NEXT: testw %ax, %ax18; CHECK-NEXT: sets %al19; CHECK-NEXT: kmovd %eax, %k120; CHECK-NEXT: vmovaps %xmm1, %xmm221; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm2 {%k1}22; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0 {%k1}23; CHECK-NEXT: vminsh %xmm2, %xmm0, %xmm124; CHECK-NEXT: vcmpunordsh %xmm0, %xmm0, %k125; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm1 {%k1}26; CHECK-NEXT: vmovaps %xmm1, %xmm027; CHECK-NEXT: retq28 %z = call half @llvm.minimum.f16(half %x, half %y)29 ret half %z30}31 32define <8 x half> @test_fminimum_v8f16(<8 x half> %x, <8 x half> %y) "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" {33; CHECK-LABEL: test_fminimum_v8f16:34; CHECK: # %bb.0:35; CHECK-NEXT: vminph %xmm1, %xmm0, %xmm036; CHECK-NEXT: retq37 %r = call <8 x half> @llvm.minimum.v8f16(<8 x half> %x, <8 x half> %y)38 ret <8 x half> %r39}40 41define half @test_fminimum_nnan(half %x, half %y) "no-nans-fp-math"="true" {42; CHECK-LABEL: test_fminimum_nnan:43; CHECK: # %bb.0:44; CHECK-NEXT: vfpclasssh $5, %xmm1, %k1 # k1 = isQuietNaN(xmm1) | isNegativeZero(xmm1)45; CHECK-NEXT: vmovaps %xmm0, %xmm246; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm2 {%k1}47; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm1 {%k1}48; CHECK-NEXT: vminsh %xmm2, %xmm1, %xmm049; CHECK-NEXT: retq50 %1 = tail call half @llvm.minimum.f16(half %x, half %y)51 ret half %152}53 54define half @test_fminimum_zero(half %x, half %y) {55; CHECK-LABEL: test_fminimum_zero:56; CHECK: # %bb.0:57; CHECK-NEXT: vcmpunordsh %xmm1, %xmm1, %k158; CHECK-NEXT: vminsh {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm059; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0 {%k1}60; CHECK-NEXT: retq61 %1 = tail call half @llvm.minimum.f16(half -0.0, half %y)62 ret half %163}64 65define half @test_fminimum_nsz(half %x, half %y) {66; CHECK-LABEL: test_fminimum_nsz:67; CHECK: # %bb.0:68; CHECK-NEXT: vminsh %xmm1, %xmm0, %xmm169; CHECK-NEXT: vcmpunordsh %xmm0, %xmm0, %k170; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm1 {%k1}71; CHECK-NEXT: vmovaps %xmm1, %xmm072; CHECK-NEXT: retq73 %1 = tail call nsz half @llvm.minimum.f16(half %x, half %y)74 ret half %175}76 77define half @test_fminimum_combine_cmps(half %x, half %y) {78; CHECK-LABEL: test_fminimum_combine_cmps:79; CHECK: # %bb.0:80; CHECK-NEXT: vdivsh %xmm0, %xmm1, %xmm181; CHECK-NEXT: vfpclasssh $5, %xmm0, %k1 # k1 = isQuietNaN(xmm0) | isNegativeZero(xmm0)82; CHECK-NEXT: vmovaps %xmm1, %xmm283; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm2 {%k1}84; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0 {%k1}85; CHECK-NEXT: vminsh %xmm2, %xmm0, %xmm086; CHECK-NEXT: retq87 %1 = fdiv nnan half %y, %x88 %2 = tail call half @llvm.minimum.f16(half %x, half %1)89 ret half %290}91 92define half @test_fmaximum(half %x, half %y) {93; CHECK-LABEL: test_fmaximum:94; CHECK: # %bb.0:95; CHECK-NEXT: vmovw %xmm0, %eax96; CHECK-NEXT: testw %ax, %ax97; CHECK-NEXT: sets %al98; CHECK-NEXT: kmovd %eax, %k199; CHECK-NEXT: vmovaps %xmm0, %xmm2100; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm2 {%k1}101; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm1 {%k1}102; CHECK-NEXT: vmaxsh %xmm2, %xmm1, %xmm0103; CHECK-NEXT: vcmpunordsh %xmm1, %xmm1, %k1104; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0 {%k1}105; CHECK-NEXT: retq106 %r = call half @llvm.maximum.f16(half %x, half %y)107 ret half %r108}109 110define <8 x half> @test_fmaximum_v8f16(<8 x half> %x, <8 x half> %y) "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" {111; CHECK-LABEL: test_fmaximum_v8f16:112; CHECK: # %bb.0:113; CHECK-NEXT: vmaxph %xmm1, %xmm0, %xmm0114; CHECK-NEXT: retq115 %r = call <8 x half> @llvm.maximum.v8f16(<8 x half> %x, <8 x half> %y)116 ret <8 x half> %r117}118 119define half @test_fmaximum_nnan(half %x, half %y) {120; CHECK-LABEL: test_fmaximum_nnan:121; CHECK: # %bb.0:122; CHECK-NEXT: vaddsh %xmm1, %xmm0, %xmm2123; CHECK-NEXT: vsubsh %xmm1, %xmm0, %xmm0124; CHECK-NEXT: vfpclasssh $3, %xmm0, %k1 # k1 = isQuietNaN(xmm0) | isPositiveZero(xmm0)125; CHECK-NEXT: vmovaps %xmm2, %xmm1126; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm1 {%k1}127; CHECK-NEXT: vmovsh %xmm2, %xmm0, %xmm0 {%k1}128; CHECK-NEXT: vmaxsh %xmm1, %xmm0, %xmm0129; CHECK-NEXT: retq130 %1 = fadd nnan half %x, %y131 %2 = fsub nnan half %x, %y132 %3 = tail call half @llvm.maximum.f16(half %1, half %2)133 ret half %3134}135 136define half @test_fmaximum_zero(half %x, half %y) {137; CHECK-LABEL: test_fmaximum_zero:138; CHECK: # %bb.0:139; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0140; CHECK-NEXT: vmaxsh %xmm0, %xmm1, %xmm0141; CHECK-NEXT: vcmpunordsh %xmm1, %xmm1, %k1142; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0 {%k1}143; CHECK-NEXT: retq144 %1 = tail call half @llvm.maximum.f16(half 0.0, half %y)145 ret half %1146}147 148define half @test_fmaximum_nsz(half %x, half %y) "no-signed-zeros-fp-math"="true" {149; CHECK-LABEL: test_fmaximum_nsz:150; CHECK: # %bb.0:151; CHECK-NEXT: vmaxsh %xmm1, %xmm0, %xmm1152; CHECK-NEXT: vcmpunordsh %xmm0, %xmm0, %k1153; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm1 {%k1}154; CHECK-NEXT: vmovaps %xmm1, %xmm0155; CHECK-NEXT: retq156 %1 = tail call half @llvm.maximum.f16(half %x, half %y)157 ret half %1158}159 160define half @test_fmaximum_combine_cmps(half %x, half %y) {161; CHECK-LABEL: test_fmaximum_combine_cmps:162; CHECK: # %bb.0:163; CHECK-NEXT: vdivsh %xmm0, %xmm1, %xmm1164; CHECK-NEXT: vfpclasssh $3, %xmm0, %k1 # k1 = isQuietNaN(xmm0) | isPositiveZero(xmm0)165; CHECK-NEXT: vmovaps %xmm1, %xmm2166; CHECK-NEXT: vmovsh %xmm0, %xmm0, %xmm2 {%k1}167; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0 {%k1}168; CHECK-NEXT: vmaxsh %xmm2, %xmm0, %xmm0169; CHECK-NEXT: retq170 %1 = fdiv nnan half %y, %x171 %2 = tail call half @llvm.maximum.f16(half %x, half %1)172 ret half %2173}174 175define <16 x half> @test_fminimum_v16f16(<16 x half> %x, <16 x half> %y) "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" {176; CHECK-LABEL: test_fminimum_v16f16:177; CHECK: # %bb.0:178; CHECK-NEXT: vminph %ymm1, %ymm0, %ymm0179; CHECK-NEXT: retq180 %r = call <16 x half> @llvm.minimum.v16f16(<16 x half> %x, <16 x half> %y)181 ret <16 x half> %r182}183 184define <16 x half> @test_fmaximum_v16f16_nans(<16 x half> %x, <16 x half> %y) "no-signed-zeros-fp-math"="true" {185; CHECK-LABEL: test_fmaximum_v16f16_nans:186; CHECK: # %bb.0:187; CHECK-NEXT: vmaxph %ymm1, %ymm0, %ymm1188; CHECK-NEXT: vcmpunordph %ymm0, %ymm0, %k1189; CHECK-NEXT: vmovdqu16 %ymm0, %ymm1 {%k1}190; CHECK-NEXT: vmovdqa %ymm1, %ymm0191; CHECK-NEXT: retq192 %r = call <16 x half> @llvm.maximum.v16f16(<16 x half> %x, <16 x half> %y)193 ret <16 x half> %r194}195 196define <32 x half> @test_fminimum_v32f16_szero(<32 x half> %x, <32 x half> %y) "no-nans-fp-math"="true" {197; CHECK-LABEL: test_fminimum_v32f16_szero:198; CHECK: # %bb.0:199; CHECK-NEXT: vpmovw2m %zmm0, %k1200; CHECK-NEXT: vpblendmw %zmm0, %zmm1, %zmm2 {%k1}201; CHECK-NEXT: vmovdqu16 %zmm1, %zmm0 {%k1}202; CHECK-NEXT: vminph %zmm2, %zmm0, %zmm0203; CHECK-NEXT: retq204 %r = call <32 x half> @llvm.minimum.v32f16(<32 x half> %x, <32 x half> %y)205 ret <32 x half> %r206}207 208define <32 x half> @test_fmaximum_v32f16_nans_szero(<32 x half> %x, <32 x half> %y) {209; CHECK-LABEL: test_fmaximum_v32f16_nans_szero:210; CHECK: # %bb.0:211; CHECK-NEXT: vpmovw2m %zmm0, %k1212; CHECK-NEXT: vpblendmw %zmm1, %zmm0, %zmm2 {%k1}213; CHECK-NEXT: vmovdqu16 %zmm0, %zmm1 {%k1}214; CHECK-NEXT: vmaxph %zmm2, %zmm1, %zmm0215; CHECK-NEXT: vcmpunordph %zmm1, %zmm1, %k1216; CHECK-NEXT: vmovdqu16 %zmm1, %zmm0 {%k1}217; CHECK-NEXT: retq218 %r = call <32 x half> @llvm.maximum.v32f16(<32 x half> %x, <32 x half> %y)219 ret <32 x half> %r220}221