brintos

brintos / llvm-project-archived public Read only

0
0
Text · 23.3 KiB · f21c3f7 Raw
693 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-prefixes=SSE2-SSSE3,SSE23; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE34; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX12,AVX15; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX12,AVX26; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=AVX512F7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefix=AVX512BW8 9define i8 @v8i16(<8 x i16> %a, <8 x i16> %b) {10; SSE2-SSSE3-LABEL: v8i16:11; SSE2-SSSE3:       # %bb.0:12; SSE2-SSSE3-NEXT:    pcmpgtw %xmm1, %xmm013; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm014; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %eax15; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax16; SSE2-SSSE3-NEXT:    retq17;18; AVX12-LABEL: v8i16:19; AVX12:       # %bb.0:20; AVX12-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm021; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm022; AVX12-NEXT:    vpmovmskb %xmm0, %eax23; AVX12-NEXT:    # kill: def $al killed $al killed $eax24; AVX12-NEXT:    retq25;26; AVX512F-LABEL: v8i16:27; AVX512F:       # %bb.0:28; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm029; AVX512F-NEXT:    vpmovsxwd %xmm0, %ymm030; AVX512F-NEXT:    vptestmd %ymm0, %ymm0, %k031; AVX512F-NEXT:    kmovw %k0, %eax32; AVX512F-NEXT:    # kill: def $al killed $al killed $eax33; AVX512F-NEXT:    vzeroupper34; AVX512F-NEXT:    retq35;36; AVX512BW-LABEL: v8i16:37; AVX512BW:       # %bb.0:38; AVX512BW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k039; AVX512BW-NEXT:    kmovd %k0, %eax40; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax41; AVX512BW-NEXT:    retq42  %x = icmp sgt <8 x i16> %a, %b43  %res = bitcast <8 x i1> %x to i844  ret i8 %res45}46 47define i4 @v4i32(<4 x i32> %a, <4 x i32> %b) {48; SSE2-SSSE3-LABEL: v4i32:49; SSE2-SSSE3:       # %bb.0:50; SSE2-SSSE3-NEXT:    pcmpgtd %xmm1, %xmm051; SSE2-SSSE3-NEXT:    movmskps %xmm0, %eax52; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax53; SSE2-SSSE3-NEXT:    retq54;55; AVX12-LABEL: v4i32:56; AVX12:       # %bb.0:57; AVX12-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm058; AVX12-NEXT:    vmovmskps %xmm0, %eax59; AVX12-NEXT:    # kill: def $al killed $al killed $eax60; AVX12-NEXT:    retq61;62; AVX512F-LABEL: v4i32:63; AVX512F:       # %bb.0:64; AVX512F-NEXT:    vpcmpgtd %xmm1, %xmm0, %k065; AVX512F-NEXT:    kmovw %k0, %eax66; AVX512F-NEXT:    # kill: def $al killed $al killed $eax67; AVX512F-NEXT:    retq68;69; AVX512BW-LABEL: v4i32:70; AVX512BW:       # %bb.0:71; AVX512BW-NEXT:    vpcmpgtd %xmm1, %xmm0, %k072; AVX512BW-NEXT:    kmovd %k0, %eax73; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax74; AVX512BW-NEXT:    retq75  %x = icmp sgt <4 x i32> %a, %b76  %res = bitcast <4 x i1> %x to i477  ret i4 %res78}79 80define i4 @v4f32(<4 x float> %a, <4 x float> %b) {81; SSE2-SSSE3-LABEL: v4f32:82; SSE2-SSSE3:       # %bb.0:83; SSE2-SSSE3-NEXT:    cmpltps %xmm0, %xmm184; SSE2-SSSE3-NEXT:    movmskps %xmm1, %eax85; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax86; SSE2-SSSE3-NEXT:    retq87;88; AVX12-LABEL: v4f32:89; AVX12:       # %bb.0:90; AVX12-NEXT:    vcmpltps %xmm0, %xmm1, %xmm091; AVX12-NEXT:    vmovmskps %xmm0, %eax92; AVX12-NEXT:    # kill: def $al killed $al killed $eax93; AVX12-NEXT:    retq94;95; AVX512F-LABEL: v4f32:96; AVX512F:       # %bb.0:97; AVX512F-NEXT:    vcmpltps %xmm0, %xmm1, %k098; AVX512F-NEXT:    kmovw %k0, %eax99; AVX512F-NEXT:    # kill: def $al killed $al killed $eax100; AVX512F-NEXT:    retq101;102; AVX512BW-LABEL: v4f32:103; AVX512BW:       # %bb.0:104; AVX512BW-NEXT:    vcmpltps %xmm0, %xmm1, %k0105; AVX512BW-NEXT:    kmovd %k0, %eax106; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax107; AVX512BW-NEXT:    retq108  %x = fcmp ogt <4 x float> %a, %b109  %res = bitcast <4 x i1> %x to i4110  ret i4 %res111}112 113define i16 @v16i8(<16 x i8> %a, <16 x i8> %b) {114; SSE2-SSSE3-LABEL: v16i8:115; SSE2-SSSE3:       # %bb.0:116; SSE2-SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0117; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %eax118; SSE2-SSSE3-NEXT:    # kill: def $ax killed $ax killed $eax119; SSE2-SSSE3-NEXT:    retq120;121; AVX12-LABEL: v16i8:122; AVX12:       # %bb.0:123; AVX12-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0124; AVX12-NEXT:    vpmovmskb %xmm0, %eax125; AVX12-NEXT:    # kill: def $ax killed $ax killed $eax126; AVX12-NEXT:    retq127;128; AVX512F-LABEL: v16i8:129; AVX512F:       # %bb.0:130; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0131; AVX512F-NEXT:    vpmovmskb %xmm0, %eax132; AVX512F-NEXT:    # kill: def $ax killed $ax killed $eax133; AVX512F-NEXT:    retq134;135; AVX512BW-LABEL: v16i8:136; AVX512BW:       # %bb.0:137; AVX512BW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0138; AVX512BW-NEXT:    kmovd %k0, %eax139; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax140; AVX512BW-NEXT:    retq141  %x = icmp sgt <16 x i8> %a, %b142  %res = bitcast <16 x i1> %x to i16143  ret i16 %res144}145 146define i2 @v2i8(<2 x i8> %a, <2 x i8> %b) {147; SSE2-LABEL: v2i8:148; SSE2:       # %bb.0:149; SSE2-NEXT:    pcmpgtb %xmm1, %xmm0150; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]151; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,1,4,5,6,7]152; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]153; SSE2-NEXT:    movmskpd %xmm0, %eax154; SSE2-NEXT:    # kill: def $al killed $al killed $eax155; SSE2-NEXT:    retq156;157; SSSE3-LABEL: v2i8:158; SSSE3:       # %bb.0:159; SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0160; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[u,u,u,u,u,u,u,0,u,u,u,u,u,u,u,1]161; SSSE3-NEXT:    movmskpd %xmm0, %eax162; SSSE3-NEXT:    # kill: def $al killed $al killed $eax163; SSSE3-NEXT:    retq164;165; AVX12-LABEL: v2i8:166; AVX12:       # %bb.0:167; AVX12-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0168; AVX12-NEXT:    vpmovsxbq %xmm0, %xmm0169; AVX12-NEXT:    vmovmskpd %xmm0, %eax170; AVX12-NEXT:    # kill: def $al killed $al killed $eax171; AVX12-NEXT:    retq172;173; AVX512F-LABEL: v2i8:174; AVX512F:       # %bb.0:175; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0176; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0177; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0178; AVX512F-NEXT:    kmovw %k0, %eax179; AVX512F-NEXT:    # kill: def $al killed $al killed $eax180; AVX512F-NEXT:    vzeroupper181; AVX512F-NEXT:    retq182;183; AVX512BW-LABEL: v2i8:184; AVX512BW:       # %bb.0:185; AVX512BW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0186; AVX512BW-NEXT:    kmovd %k0, %eax187; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax188; AVX512BW-NEXT:    retq189  %x = icmp sgt <2 x i8> %a, %b190  %res = bitcast <2 x i1> %x to i2191  ret i2 %res192}193 194define i2 @v2i16(<2 x i16> %a, <2 x i16> %b) {195; SSE2-SSSE3-LABEL: v2i16:196; SSE2-SSSE3:       # %bb.0:197; SSE2-SSSE3-NEXT:    pcmpgtw %xmm1, %xmm0198; SSE2-SSSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,1,1,4,5,6,7]199; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]200; SSE2-SSSE3-NEXT:    movmskpd %xmm0, %eax201; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax202; SSE2-SSSE3-NEXT:    retq203;204; AVX12-LABEL: v2i16:205; AVX12:       # %bb.0:206; AVX12-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0207; AVX12-NEXT:    vpmovsxwq %xmm0, %xmm0208; AVX12-NEXT:    vmovmskpd %xmm0, %eax209; AVX12-NEXT:    # kill: def $al killed $al killed $eax210; AVX12-NEXT:    retq211;212; AVX512F-LABEL: v2i16:213; AVX512F:       # %bb.0:214; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0215; AVX512F-NEXT:    vpmovsxwd %xmm0, %ymm0216; AVX512F-NEXT:    vptestmd %ymm0, %ymm0, %k0217; AVX512F-NEXT:    kmovw %k0, %eax218; AVX512F-NEXT:    # kill: def $al killed $al killed $eax219; AVX512F-NEXT:    vzeroupper220; AVX512F-NEXT:    retq221;222; AVX512BW-LABEL: v2i16:223; AVX512BW:       # %bb.0:224; AVX512BW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0225; AVX512BW-NEXT:    kmovd %k0, %eax226; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax227; AVX512BW-NEXT:    retq228  %x = icmp sgt <2 x i16> %a, %b229  %res = bitcast <2 x i1> %x to i2230  ret i2 %res231}232 233define i2 @v2i32(<2 x i32> %a, <2 x i32> %b) {234; SSE2-SSSE3-LABEL: v2i32:235; SSE2-SSSE3:       # %bb.0:236; SSE2-SSSE3-NEXT:    pcmpgtd %xmm1, %xmm0237; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]238; SSE2-SSSE3-NEXT:    movmskpd %xmm0, %eax239; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax240; SSE2-SSSE3-NEXT:    retq241;242; AVX12-LABEL: v2i32:243; AVX12:       # %bb.0:244; AVX12-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0245; AVX12-NEXT:    vpmovsxdq %xmm0, %xmm0246; AVX12-NEXT:    vmovmskpd %xmm0, %eax247; AVX12-NEXT:    # kill: def $al killed $al killed $eax248; AVX12-NEXT:    retq249;250; AVX512F-LABEL: v2i32:251; AVX512F:       # %bb.0:252; AVX512F-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0253; AVX512F-NEXT:    kmovw %k0, %eax254; AVX512F-NEXT:    # kill: def $al killed $al killed $eax255; AVX512F-NEXT:    retq256;257; AVX512BW-LABEL: v2i32:258; AVX512BW:       # %bb.0:259; AVX512BW-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0260; AVX512BW-NEXT:    kmovd %k0, %eax261; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax262; AVX512BW-NEXT:    retq263  %x = icmp sgt <2 x i32> %a, %b264  %res = bitcast <2 x i1> %x to i2265  ret i2 %res266}267 268define i2 @v2i64(<2 x i64> %a, <2 x i64> %b) {269; SSE2-SSSE3-LABEL: v2i64:270; SSE2-SSSE3:       # %bb.0:271; SSE2-SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]272; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm1273; SSE2-SSSE3-NEXT:    pxor %xmm2, %xmm0274; SSE2-SSSE3-NEXT:    movdqa %xmm0, %xmm2275; SSE2-SSSE3-NEXT:    pcmpeqd %xmm1, %xmm2276; SSE2-SSSE3-NEXT:    pcmpgtd %xmm1, %xmm0277; SSE2-SSSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,0,2,2]278; SSE2-SSSE3-NEXT:    pand %xmm2, %xmm1279; SSE2-SSSE3-NEXT:    por %xmm0, %xmm1280; SSE2-SSSE3-NEXT:    movmskpd %xmm1, %eax281; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax282; SSE2-SSSE3-NEXT:    retq283;284; AVX12-LABEL: v2i64:285; AVX12:       # %bb.0:286; AVX12-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0287; AVX12-NEXT:    vmovmskpd %xmm0, %eax288; AVX12-NEXT:    # kill: def $al killed $al killed $eax289; AVX12-NEXT:    retq290;291; AVX512F-LABEL: v2i64:292; AVX512F:       # %bb.0:293; AVX512F-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0294; AVX512F-NEXT:    kmovw %k0, %eax295; AVX512F-NEXT:    # kill: def $al killed $al killed $eax296; AVX512F-NEXT:    retq297;298; AVX512BW-LABEL: v2i64:299; AVX512BW:       # %bb.0:300; AVX512BW-NEXT:    vpcmpgtq %xmm1, %xmm0, %k0301; AVX512BW-NEXT:    kmovd %k0, %eax302; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax303; AVX512BW-NEXT:    retq304  %x = icmp sgt <2 x i64> %a, %b305  %res = bitcast <2 x i1> %x to i2306  ret i2 %res307}308 309define i2 @v2f64(<2 x double> %a, <2 x double> %b) {310; SSE2-SSSE3-LABEL: v2f64:311; SSE2-SSSE3:       # %bb.0:312; SSE2-SSSE3-NEXT:    cmpltpd %xmm0, %xmm1313; SSE2-SSSE3-NEXT:    movmskpd %xmm1, %eax314; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax315; SSE2-SSSE3-NEXT:    retq316;317; AVX12-LABEL: v2f64:318; AVX12:       # %bb.0:319; AVX12-NEXT:    vcmpltpd %xmm0, %xmm1, %xmm0320; AVX12-NEXT:    vmovmskpd %xmm0, %eax321; AVX12-NEXT:    # kill: def $al killed $al killed $eax322; AVX12-NEXT:    retq323;324; AVX512F-LABEL: v2f64:325; AVX512F:       # %bb.0:326; AVX512F-NEXT:    vcmpltpd %xmm0, %xmm1, %k0327; AVX512F-NEXT:    kmovw %k0, %eax328; AVX512F-NEXT:    # kill: def $al killed $al killed $eax329; AVX512F-NEXT:    retq330;331; AVX512BW-LABEL: v2f64:332; AVX512BW:       # %bb.0:333; AVX512BW-NEXT:    vcmpltpd %xmm0, %xmm1, %k0334; AVX512BW-NEXT:    kmovd %k0, %eax335; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax336; AVX512BW-NEXT:    retq337  %x = fcmp ogt <2 x double> %a, %b338  %res = bitcast <2 x i1> %x to i2339  ret i2 %res340}341 342define i4 @v4i8(<4 x i8> %a, <4 x i8> %b) {343; SSE2-SSSE3-LABEL: v4i8:344; SSE2-SSSE3:       # %bb.0:345; SSE2-SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0346; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]347; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]348; SSE2-SSSE3-NEXT:    movmskps %xmm0, %eax349; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax350; SSE2-SSSE3-NEXT:    retq351;352; AVX12-LABEL: v4i8:353; AVX12:       # %bb.0:354; AVX12-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0355; AVX12-NEXT:    vpmovsxbd %xmm0, %xmm0356; AVX12-NEXT:    vmovmskps %xmm0, %eax357; AVX12-NEXT:    # kill: def $al killed $al killed $eax358; AVX12-NEXT:    retq359;360; AVX512F-LABEL: v4i8:361; AVX512F:       # %bb.0:362; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0363; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0364; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0365; AVX512F-NEXT:    kmovw %k0, %eax366; AVX512F-NEXT:    # kill: def $al killed $al killed $eax367; AVX512F-NEXT:    vzeroupper368; AVX512F-NEXT:    retq369;370; AVX512BW-LABEL: v4i8:371; AVX512BW:       # %bb.0:372; AVX512BW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0373; AVX512BW-NEXT:    kmovd %k0, %eax374; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax375; AVX512BW-NEXT:    retq376  %x = icmp sgt <4 x i8> %a, %b377  %res = bitcast <4 x i1> %x to i4378  ret i4 %res379}380 381define i4 @v4i16(<4 x i16> %a, <4 x i16> %b) {382; SSE2-SSSE3-LABEL: v4i16:383; SSE2-SSSE3:       # %bb.0:384; SSE2-SSSE3-NEXT:    pcmpgtw %xmm1, %xmm0385; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]386; SSE2-SSSE3-NEXT:    movmskps %xmm0, %eax387; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax388; SSE2-SSSE3-NEXT:    retq389;390; AVX12-LABEL: v4i16:391; AVX12:       # %bb.0:392; AVX12-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0393; AVX12-NEXT:    vpmovsxwd %xmm0, %xmm0394; AVX12-NEXT:    vmovmskps %xmm0, %eax395; AVX12-NEXT:    # kill: def $al killed $al killed $eax396; AVX12-NEXT:    retq397;398; AVX512F-LABEL: v4i16:399; AVX512F:       # %bb.0:400; AVX512F-NEXT:    vpcmpgtw %xmm1, %xmm0, %xmm0401; AVX512F-NEXT:    vpmovsxwd %xmm0, %ymm0402; AVX512F-NEXT:    vptestmd %ymm0, %ymm0, %k0403; AVX512F-NEXT:    kmovw %k0, %eax404; AVX512F-NEXT:    # kill: def $al killed $al killed $eax405; AVX512F-NEXT:    vzeroupper406; AVX512F-NEXT:    retq407;408; AVX512BW-LABEL: v4i16:409; AVX512BW:       # %bb.0:410; AVX512BW-NEXT:    vpcmpgtw %xmm1, %xmm0, %k0411; AVX512BW-NEXT:    kmovd %k0, %eax412; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax413; AVX512BW-NEXT:    retq414  %x = icmp sgt <4 x i16> %a, %b415  %res = bitcast <4 x i1> %x to i4416  ret i4 %res417}418 419define i8 @v8i8(<8 x i8> %a, <8 x i8> %b) {420; SSE2-SSSE3-LABEL: v8i8:421; SSE2-SSSE3:       # %bb.0:422; SSE2-SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0423; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %eax424; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax425; SSE2-SSSE3-NEXT:    retq426;427; AVX12-LABEL: v8i8:428; AVX12:       # %bb.0:429; AVX12-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0430; AVX12-NEXT:    vpmovmskb %xmm0, %eax431; AVX12-NEXT:    # kill: def $al killed $al killed $eax432; AVX12-NEXT:    retq433;434; AVX512F-LABEL: v8i8:435; AVX512F:       # %bb.0:436; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0437; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0438; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0439; AVX512F-NEXT:    kmovw %k0, %eax440; AVX512F-NEXT:    # kill: def $al killed $al killed $eax441; AVX512F-NEXT:    vzeroupper442; AVX512F-NEXT:    retq443;444; AVX512BW-LABEL: v8i8:445; AVX512BW:       # %bb.0:446; AVX512BW-NEXT:    vpcmpgtb %xmm1, %xmm0, %k0447; AVX512BW-NEXT:    kmovd %k0, %eax448; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax449; AVX512BW-NEXT:    retq450  %x = icmp sgt <8 x i8> %a, %b451  %res = bitcast <8 x i1> %x to i8452  ret i8 %res453}454 455define i64 @v16i8_widened_with_zeroes(<16 x i8> %a, <16 x i8> %b) {456; SSE2-SSSE3-LABEL: v16i8_widened_with_zeroes:457; SSE2-SSSE3:       # %bb.0: # %entry458; SSE2-SSSE3-NEXT:    pcmpeqb %xmm1, %xmm0459; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %eax460; SSE2-SSSE3-NEXT:    retq461;462; AVX1-LABEL: v16i8_widened_with_zeroes:463; AVX1:       # %bb.0: # %entry464; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0465; AVX1-NEXT:    vpmovmskb %xmm0, %eax466; AVX1-NEXT:    retq467;468; AVX2-LABEL: v16i8_widened_with_zeroes:469; AVX2:       # %bb.0: # %entry470; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0471; AVX2-NEXT:    vpmovmskb %ymm0, %eax472; AVX2-NEXT:    vzeroupper473; AVX2-NEXT:    retq474;475; AVX512F-LABEL: v16i8_widened_with_zeroes:476; AVX512F:       # %bb.0: # %entry477; AVX512F-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0478; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0479; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0480; AVX512F-NEXT:    kmovw %k0, %eax481; AVX512F-NEXT:    vzeroupper482; AVX512F-NEXT:    retq483;484; AVX512BW-LABEL: v16i8_widened_with_zeroes:485; AVX512BW:       # %bb.0: # %entry486; AVX512BW-NEXT:    vpcmpeqb %xmm1, %xmm0, %k0487; AVX512BW-NEXT:    kmovq %k0, %rax488; AVX512BW-NEXT:    retq489entry:490  %c = icmp eq <16 x i8> %a, %b491  %d = shufflevector <16 x i1> %c, <16 x i1> zeroinitializer, <64 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, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>492  %e = bitcast <64 x i1> %d to i64493  ret i64 %e494}495 496define i64 @v16i8_widened_with_ones(<16 x i8> %a, <16 x i8> %b) {497; SSE2-SSSE3-LABEL: v16i8_widened_with_ones:498; SSE2-SSSE3:       # %bb.0: # %entry499; SSE2-SSSE3-NEXT:    pcmpeqb %xmm1, %xmm0500; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %ecx501; SSE2-SSSE3-NEXT:    orl $-65536, %ecx # imm = 0xFFFF0000502; SSE2-SSSE3-NEXT:    movabsq $-4294967296, %rax # imm = 0xFFFFFFFF00000000503; SSE2-SSSE3-NEXT:    orq %rcx, %rax504; SSE2-SSSE3-NEXT:    retq505;506; AVX1-LABEL: v16i8_widened_with_ones:507; AVX1:       # %bb.0: # %entry508; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0509; AVX1-NEXT:    vpmovmskb %xmm0, %ecx510; AVX1-NEXT:    orl $-65536, %ecx # imm = 0xFFFF0000511; AVX1-NEXT:    movabsq $-4294967296, %rax # imm = 0xFFFFFFFF00000000512; AVX1-NEXT:    orq %rcx, %rax513; AVX1-NEXT:    retq514;515; AVX2-LABEL: v16i8_widened_with_ones:516; AVX2:       # %bb.0: # %entry517; AVX2-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0518; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1519; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0520; AVX2-NEXT:    vpmovmskb %ymm0, %ecx521; AVX2-NEXT:    movabsq $-4294967296, %rax # imm = 0xFFFFFFFF00000000522; AVX2-NEXT:    orq %rcx, %rax523; AVX2-NEXT:    vzeroupper524; AVX2-NEXT:    retq525;526; AVX512F-LABEL: v16i8_widened_with_ones:527; AVX512F:       # %bb.0: # %entry528; AVX512F-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0529; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0530; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0531; AVX512F-NEXT:    kmovw %k0, %ecx532; AVX512F-NEXT:    orl $-65536, %ecx # imm = 0xFFFF0000533; AVX512F-NEXT:    movabsq $-4294967296, %rax # imm = 0xFFFFFFFF00000000534; AVX512F-NEXT:    orq %rcx, %rax535; AVX512F-NEXT:    vzeroupper536; AVX512F-NEXT:    retq537;538; AVX512BW-LABEL: v16i8_widened_with_ones:539; AVX512BW:       # %bb.0: # %entry540; AVX512BW-NEXT:    vpcmpeqb %xmm1, %xmm0, %k0541; AVX512BW-NEXT:    kxnorw %k0, %k0, %k1542; AVX512BW-NEXT:    kunpckwd %k0, %k1, %k0543; AVX512BW-NEXT:    kxnord %k0, %k0, %k1544; AVX512BW-NEXT:    kunpckdq %k0, %k1, %k0545; AVX512BW-NEXT:    kmovq %k0, %rax546; AVX512BW-NEXT:    retq547entry:548  %c = icmp eq <16 x i8> %a, %b549  %d = shufflevector <16 x i1> %c, <16 x i1> <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1>, <64 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, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>550  %e = bitcast <64 x i1> %d to i64551  ret i64 %e552}553 554define void @bitcast_16i8_store(ptr %p, <16 x i8> %a0) {555; SSE2-SSSE3-LABEL: bitcast_16i8_store:556; SSE2-SSSE3:       # %bb.0:557; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %eax558; SSE2-SSSE3-NEXT:    movw %ax, (%rdi)559; SSE2-SSSE3-NEXT:    retq560;561; AVX12-LABEL: bitcast_16i8_store:562; AVX12:       # %bb.0:563; AVX12-NEXT:    vpmovmskb %xmm0, %eax564; AVX12-NEXT:    movw %ax, (%rdi)565; AVX12-NEXT:    retq566;567; AVX512F-LABEL: bitcast_16i8_store:568; AVX512F:       # %bb.0:569; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1570; AVX512F-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0571; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0572; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0573; AVX512F-NEXT:    kmovw %k0, (%rdi)574; AVX512F-NEXT:    vzeroupper575; AVX512F-NEXT:    retq576;577; AVX512BW-LABEL: bitcast_16i8_store:578; AVX512BW:       # %bb.0:579; AVX512BW-NEXT:    vpmovb2m %xmm0, %k0580; AVX512BW-NEXT:    kmovw %k0, (%rdi)581; AVX512BW-NEXT:    retq582  %a1 = icmp slt <16 x i8> %a0, zeroinitializer583  %a2 = bitcast <16 x i1> %a1 to i16584  store i16 %a2, ptr %p585  ret void586}587 588define void @bitcast_8i16_store(ptr %p, <8 x i16> %a0) {589; SSE2-SSSE3-LABEL: bitcast_8i16_store:590; SSE2-SSSE3:       # %bb.0:591; SSE2-SSSE3-NEXT:    packsswb %xmm0, %xmm0592; SSE2-SSSE3-NEXT:    pmovmskb %xmm0, %eax593; SSE2-SSSE3-NEXT:    movb %al, (%rdi)594; SSE2-SSSE3-NEXT:    retq595;596; AVX12-LABEL: bitcast_8i16_store:597; AVX12:       # %bb.0:598; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0599; AVX12-NEXT:    vpmovmskb %xmm0, %eax600; AVX12-NEXT:    movb %al, (%rdi)601; AVX12-NEXT:    retq602;603; AVX512F-LABEL: bitcast_8i16_store:604; AVX512F:       # %bb.0:605; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1606; AVX512F-NEXT:    vpcmpgtw %xmm0, %xmm1, %xmm0607; AVX512F-NEXT:    vpmovsxwd %xmm0, %ymm0608; AVX512F-NEXT:    vptestmd %ymm0, %ymm0, %k0609; AVX512F-NEXT:    kmovw %k0, %eax610; AVX512F-NEXT:    movb %al, (%rdi)611; AVX512F-NEXT:    vzeroupper612; AVX512F-NEXT:    retq613;614; AVX512BW-LABEL: bitcast_8i16_store:615; AVX512BW:       # %bb.0:616; AVX512BW-NEXT:    vpmovw2m %xmm0, %k0617; AVX512BW-NEXT:    kmovd %k0, %eax618; AVX512BW-NEXT:    movb %al, (%rdi)619; AVX512BW-NEXT:    retq620  %a1 = icmp slt <8 x i16> %a0, zeroinitializer621  %a2 = bitcast <8 x i1> %a1 to i8622  store i8 %a2, ptr %p623  ret void624}625 626define void @bitcast_4i32_store(ptr %p, <4 x i32> %a0) {627; SSE2-SSSE3-LABEL: bitcast_4i32_store:628; SSE2-SSSE3:       # %bb.0:629; SSE2-SSSE3-NEXT:    movmskps %xmm0, %eax630; SSE2-SSSE3-NEXT:    movb %al, (%rdi)631; SSE2-SSSE3-NEXT:    retq632;633; AVX12-LABEL: bitcast_4i32_store:634; AVX12:       # %bb.0:635; AVX12-NEXT:    vmovmskps %xmm0, %eax636; AVX12-NEXT:    movb %al, (%rdi)637; AVX12-NEXT:    retq638;639; AVX512F-LABEL: bitcast_4i32_store:640; AVX512F:       # %bb.0:641; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1642; AVX512F-NEXT:    vpcmpgtd %xmm0, %xmm1, %k0643; AVX512F-NEXT:    kmovw %k0, %eax644; AVX512F-NEXT:    movb %al, (%rdi)645; AVX512F-NEXT:    retq646;647; AVX512BW-LABEL: bitcast_4i32_store:648; AVX512BW:       # %bb.0:649; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1650; AVX512BW-NEXT:    vpcmpgtd %xmm0, %xmm1, %k0651; AVX512BW-NEXT:    kmovd %k0, %eax652; AVX512BW-NEXT:    movb %al, (%rdi)653; AVX512BW-NEXT:    retq654  %a1 = icmp slt <4 x i32> %a0, zeroinitializer655  %a2 = bitcast <4 x i1> %a1 to i4656  store i4 %a2, ptr %p657  ret void658}659 660define void @bitcast_2i64_store(ptr %p, <2 x i64> %a0) {661; SSE2-SSSE3-LABEL: bitcast_2i64_store:662; SSE2-SSSE3:       # %bb.0:663; SSE2-SSSE3-NEXT:    movmskpd %xmm0, %eax664; SSE2-SSSE3-NEXT:    movb %al, (%rdi)665; SSE2-SSSE3-NEXT:    retq666;667; AVX12-LABEL: bitcast_2i64_store:668; AVX12:       # %bb.0:669; AVX12-NEXT:    vmovmskpd %xmm0, %eax670; AVX12-NEXT:    movb %al, (%rdi)671; AVX12-NEXT:    retq672;673; AVX512F-LABEL: bitcast_2i64_store:674; AVX512F:       # %bb.0:675; AVX512F-NEXT:    vpxor %xmm1, %xmm1, %xmm1676; AVX512F-NEXT:    vpcmpgtq %xmm0, %xmm1, %k0677; AVX512F-NEXT:    kmovw %k0, %eax678; AVX512F-NEXT:    movb %al, (%rdi)679; AVX512F-NEXT:    retq680;681; AVX512BW-LABEL: bitcast_2i64_store:682; AVX512BW:       # %bb.0:683; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1684; AVX512BW-NEXT:    vpcmpgtq %xmm0, %xmm1, %k0685; AVX512BW-NEXT:    kmovd %k0, %eax686; AVX512BW-NEXT:    movb %al, (%rdi)687; AVX512BW-NEXT:    retq688  %a1 = icmp slt <2 x i64> %a0, zeroinitializer689  %a2 = bitcast <2 x i1> %a1 to i2690  store i2 %a2, ptr %p691  ret void692}693