brintos

brintos / llvm-project-archived public Read only

0
0
Text · 8.4 KiB · 15c2aab Raw
230 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_mem_shuffle2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,X863; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,X644 5define <32 x i8> @funcA(<32 x i8> %a) nounwind uwtable readnone ssp {6; CHECK-LABEL: funcA:7; CHECK:       # %bb.0: # %entry8; CHECK-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5]9; CHECK-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm010; CHECK-NEXT:    ret{{[l|q]}}11entry:12  %shuffle = shufflevector <32 x i8> %a, <32 x i8> undef, <32 x i32> <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>13  ret <32 x i8> %shuffle14}15 16define <16 x i16> @funcB(<16 x i16> %a) nounwind uwtable readnone ssp {17; CHECK-LABEL: funcB:18; CHECK:       # %bb.0: # %entry19; CHECK-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,5,5]20; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]21; CHECK-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm022; CHECK-NEXT:    ret{{[l|q]}}23entry:24  %shuffle = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>25  ret <16 x i16> %shuffle26}27 28define <4 x i64> @funcC(i64 %q) nounwind uwtable readnone ssp {29; X86-LABEL: funcC:30; X86:       # %bb.0: # %entry31; X86-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %ymm032; X86-NEXT:    retl33;34; X64-LABEL: funcC:35; X64:       # %bb.0: # %entry36; X64-NEXT:    vmovq %rdi, %xmm037; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]38; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm039; X64-NEXT:    retq40entry:41  %vecinit.i = insertelement <4 x i64> undef, i64 %q, i32 042  %vecinit2.i = insertelement <4 x i64> %vecinit.i, i64 %q, i32 143  %vecinit4.i = insertelement <4 x i64> %vecinit2.i, i64 %q, i32 244  %vecinit6.i = insertelement <4 x i64> %vecinit4.i, i64 %q, i32 345  ret <4 x i64> %vecinit6.i46}47 48define <4 x double> @funcD(double %q) nounwind uwtable readnone ssp {49; X86-LABEL: funcD:50; X86:       # %bb.0: # %entry51; X86-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %ymm052; X86-NEXT:    retl53;54; X64-LABEL: funcD:55; X64:       # %bb.0: # %entry56; X64-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]57; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm058; X64-NEXT:    retq59entry:60  %vecinit.i = insertelement <4 x double> undef, double %q, i32 061  %vecinit2.i = insertelement <4 x double> %vecinit.i, double %q, i32 162  %vecinit4.i = insertelement <4 x double> %vecinit2.i, double %q, i32 263  %vecinit6.i = insertelement <4 x double> %vecinit4.i, double %q, i32 364  ret <4 x double> %vecinit6.i65}66 67; Test this turns into a broadcast:68;   shuffle (scalar_to_vector (load (ptr + 4))), undef, <0, 0, 0, 0>69;70define <8 x float> @funcE() nounwind {71; X86-LABEL: funcE:72; X86:       # %bb.0: # %allocas73; X86-NEXT:    xorl %eax, %eax74; X86-NEXT:    testb %al, %al75; X86-NEXT:    # implicit-def: $ymm076; X86-NEXT:    jne .LBB4_277; X86-NEXT:  # %bb.1: # %load.i124778; X86-NEXT:    pushl %ebp79; X86-NEXT:    movl %esp, %ebp80; X86-NEXT:    andl $-32, %esp81; X86-NEXT:    subl $1312, %esp # imm = 0x52082; X86-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %ymm083; X86-NEXT:    movl %ebp, %esp84; X86-NEXT:    popl %ebp85; X86-NEXT:  .LBB4_2: # %__load_and_broadcast_32.exit124986; X86-NEXT:    retl87;88; X64-LABEL: funcE:89; X64:       # %bb.0: # %allocas90; X64-NEXT:    xorl %eax, %eax91; X64-NEXT:    testb %al, %al92; X64-NEXT:    # implicit-def: $ymm093; X64-NEXT:    jne .LBB4_294; X64-NEXT:  # %bb.1: # %load.i124795; X64-NEXT:    pushq %rbp96; X64-NEXT:    movq %rsp, %rbp97; X64-NEXT:    andq $-32, %rsp98; X64-NEXT:    subq $1312, %rsp # imm = 0x52099; X64-NEXT:    vbroadcastss {{[0-9]+}}(%rsp), %ymm0100; X64-NEXT:    movq %rbp, %rsp101; X64-NEXT:    popq %rbp102; X64-NEXT:  .LBB4_2: # %__load_and_broadcast_32.exit1249103; X64-NEXT:    retq104allocas:105  %udx495 = alloca [18 x [18 x float]], align 32106  br label %for_test505.preheader107 108for_test505.preheader:                            ; preds = %for_test505.preheader, %allocas109  br i1 undef, label %for_exit499, label %for_test505.preheader110 111for_exit499:                                      ; preds = %for_test505.preheader112  br i1 undef, label %__load_and_broadcast_32.exit1249, label %load.i1247113 114load.i1247:                                       ; preds = %for_exit499115  %ptr1227 = getelementptr [18 x [18 x float]], ptr %udx495, i64 0, i64 1, i64 1116  %val.i1238 = load i32, ptr %ptr1227, align 4117  %ret6.i1245 = insertelement <8 x i32> undef, i32 %val.i1238, i32 6118  %ret7.i1246 = insertelement <8 x i32> %ret6.i1245, i32 %val.i1238, i32 7119  %phitmp = bitcast <8 x i32> %ret7.i1246 to <8 x float>120  br label %__load_and_broadcast_32.exit1249121 122__load_and_broadcast_32.exit1249:                 ; preds = %load.i1247, %for_exit499123  %load_broadcast12281250 = phi <8 x float> [ %phitmp, %load.i1247 ], [ undef, %for_exit499 ]124  ret <8 x float> %load_broadcast12281250125}126 127define <8 x float> @funcF(i32 %val) nounwind {128; X86-LABEL: funcF:129; X86:       # %bb.0:130; X86-NEXT:    vbroadcastss {{[0-9]+}}(%esp), %ymm0131; X86-NEXT:    retl132;133; X64-LABEL: funcF:134; X64:       # %bb.0:135; X64-NEXT:    vmovd %edi, %xmm0136; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]137; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0138; X64-NEXT:    retq139  %ret6 = insertelement <8 x i32> undef, i32 %val, i32 6140  %ret7 = insertelement <8 x i32> %ret6, i32 %val, i32 7141  %tmp = bitcast <8 x i32> %ret7 to <8 x float>142  ret <8 x float> %tmp143}144 145define <8 x float> @funcG(<8 x float> %a) nounwind uwtable readnone ssp {146; CHECK-LABEL: funcG:147; CHECK:       # %bb.0: # %entry148; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,0,0,0]149; CHECK-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0150; CHECK-NEXT:    ret{{[l|q]}}151entry:152  %shuffle = shufflevector <8 x float> %a, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>153  ret <8 x float> %shuffle154}155 156define <8 x float> @funcH(<8 x float> %a) nounwind uwtable readnone ssp {157; CHECK-LABEL: funcH:158; CHECK:       # %bb.0: # %entry159; CHECK-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,2,3]160; CHECK-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,1,1,1,5,5,5,5]161; CHECK-NEXT:    ret{{[l|q]}}162entry:163  %shuffle = shufflevector <8 x float> %a, <8 x float> undef, <8 x i32> <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>164  ret <8 x float> %shuffle165}166 167define <2 x double> @splat_load_2f64_11(ptr %ptr) {168; X86-LABEL: splat_load_2f64_11:169; X86:       # %bb.0:170; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax171; X86-NEXT:    vmovddup 8(%eax), %xmm0 # xmm0 = mem[0,0]172; X86-NEXT:    retl173;174; X64-LABEL: splat_load_2f64_11:175; X64:       # %bb.0:176; X64-NEXT:    vmovddup 8(%rdi), %xmm0 # xmm0 = mem[0,0]177; X64-NEXT:    retq178  %x = load <2 x double>, ptr %ptr179  %x1 = shufflevector <2 x double> %x, <2 x double> undef, <2 x i32> <i32 1, i32 1>180  ret <2 x double> %x1181}182 183define <4 x double> @splat_load_4f64_2222(ptr %ptr) {184; X86-LABEL: splat_load_4f64_2222:185; X86:       # %bb.0:186; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax187; X86-NEXT:    vbroadcastsd 16(%eax), %ymm0188; X86-NEXT:    retl189;190; X64-LABEL: splat_load_4f64_2222:191; X64:       # %bb.0:192; X64-NEXT:    vbroadcastsd 16(%rdi), %ymm0193; X64-NEXT:    retq194  %x = load <4 x double>, ptr %ptr195  %x1 = shufflevector <4 x double> %x, <4 x double> undef, <4 x i32> <i32 2, i32 2, i32 2, i32 2>196  ret <4 x double> %x1197}198 199define <4 x float> @splat_load_4f32_0000(ptr %ptr) {200; X86-LABEL: splat_load_4f32_0000:201; X86:       # %bb.0:202; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax203; X86-NEXT:    vbroadcastss (%eax), %xmm0204; X86-NEXT:    retl205;206; X64-LABEL: splat_load_4f32_0000:207; X64:       # %bb.0:208; X64-NEXT:    vbroadcastss (%rdi), %xmm0209; X64-NEXT:    retq210  %x = load <4 x float>, ptr %ptr211  %x1 = shufflevector <4 x float> %x, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>212  ret <4 x float> %x1213}214 215define <8 x float> @splat_load_8f32_77777777(ptr %ptr) {216; X86-LABEL: splat_load_8f32_77777777:217; X86:       # %bb.0:218; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax219; X86-NEXT:    vbroadcastss 28(%eax), %ymm0220; X86-NEXT:    retl221;222; X64-LABEL: splat_load_8f32_77777777:223; X64:       # %bb.0:224; X64-NEXT:    vbroadcastss 28(%rdi), %ymm0225; X64-NEXT:    retq226  %x = load <8 x float>, ptr %ptr227  %x1 = shufflevector <8 x float> %x, <8 x float> undef, <8 x i32> <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>228  ret <8 x float> %x1229}230