1; RUN: llc -mtriple=amdgcn-amd-amdhsa -verify-machineinstrs -amdgpu-codegenprepare-break-large-phis=0 < %s | FileCheck %s 2 3%struct.wombat = type { [4 x i32], [4 x i32], [4 x i32] } 4 5define amdgpu_kernel void @wobble(ptr addrspace(1) nocapture readonly %arg) #0 !dbg !4 { 6bb: 7 %tmp = load i32, ptr addrspace(1) undef, align 4 8 %tmp1 = load <4 x float>, ptr addrspace(1) undef, align 16 9 %tmp2 = sext i32 %tmp to i64 10 %tmp3 = shufflevector <4 x float> undef, <4 x float> %tmp1, <2 x i32> <i32 3, i32 7> 11 %tmp4 = call float @barney() #2 12 %tmp9 = getelementptr inbounds %struct.wombat, ptr addrspace(1) %arg, i64 %tmp2, i32 2, i64 0 13 %tmp10 = load i32, ptr addrspace(1) %tmp9, align 4 14 %tmp11 = sext i32 %tmp10 to i64 15 %tmp12 = getelementptr inbounds <2 x float>, ptr addrspace(1) %arg, i64 %tmp11 16 %tmp14 = getelementptr inbounds i8, ptr addrspace(1) %arg, i64 undef 17 %tmp16 = getelementptr inbounds <4 x float>, ptr addrspace(1) %tmp14, i64 undef 18 %tmp17 = load <4 x float>, ptr addrspace(1) %tmp16, align 16 19 %tmp18 = fsub <4 x float> %tmp17, %tmp17 20 %ext = extractelement <4 x float> %tmp18, i32 1 21 %tmp19 = fadd float %ext, 0.000000e+00 22 %tmp20 = fcmp oeq float %tmp19, 0.000000e+00 23 br i1 %tmp20, label %bb21, label %bb25 24 25bb21: ; preds = %bb 26 %tmp22 = fmul <4 x float> %tmp18, %tmp18 27 %tmp23 = fadd <4 x float> %tmp22, %tmp22 28 %tmp24 = fmul <4 x float> %tmp23, %tmp23 29 br label %bb28 30 31bb25: ; preds = %bb 32 %tmp26 = insertelement <4 x float> undef, float 0.000000e+00, i32 1 33 %tmp27 = insertelement <4 x float> %tmp26, float undef, i32 2 34 br label %bb28 35 36bb28: ; preds = %bb25, %bb21 37 %tmp29 = phi <4 x float> [ %tmp27, %bb25 ], [ %tmp24, %bb21 ] 38 store <4 x float> %tmp29, ptr addrspace(5) undef, align 16 39 %tmp30 = getelementptr inbounds %struct.wombat, ptr addrspace(1) %arg, i64 %tmp2, i32 2, i64 2 40 %tmp31 = load i32, ptr addrspace(1) %tmp30, align 4 41 %tmp32 = sext i32 %tmp31 to i64 42 %tmp33 = getelementptr inbounds <2 x float>, ptr addrspace(1) %arg, i64 %tmp32 43 %tmp35 = load i64, ptr addrspace(1) %tmp33, align 8 44 %tmp36 = load i32, ptr addrspace(1) undef, align 4 45 %tmp37 = sext i32 %tmp36 to i64 46 %tmp38 = getelementptr inbounds <4 x float>, ptr addrspace(1) null, i64 %tmp37 47 %tmp39 = load <4 x float>, ptr addrspace(1) %tmp38, align 16 48 %tmp40 = load <4 x float>, ptr addrspace(1) undef, align 16 49 %tmp41 = fsub <4 x float> zeroinitializer, %tmp40 50 %tmp42 = fsub <4 x float> %tmp39, %tmp40 51 %tmp43 = extractelement <4 x float> %tmp40, i32 1 52 %tmp44 = fsub float %tmp43, undef 53 %tmp45 = fadd float undef, undef 54 %tmp46 = fdiv float %tmp44, %tmp45 55 %tmp47 = insertelement <4 x float> undef, float %tmp46, i32 0 56 %tmp48 = shufflevector <4 x float> %tmp47, <4 x float> undef, <4 x i32> zeroinitializer 57 %tmp49 = fsub <4 x float> %tmp48, %tmp40 58 %tmp50 = extractelement <4 x float> %tmp41, i32 1 59 %tmp51 = extractelement <4 x float> %tmp42, i32 2 60 %tmp52 = fmul float undef, undef 61 %tmp53 = fadd float %tmp52, undef 62 %tmp54 = fadd float %tmp51, %tmp53 63 %tmp55 = extractelement <4 x float> %tmp49, i32 1 64 %tmp56 = fmul float %tmp55, %tmp50 65 %tmp57 = fmul float %tmp54, %tmp56 66 %tmp58 = fdiv float %tmp57, 0.000000e+00 67 ; Make sure this isn't double emitted 68 ; CHECK-NOT: ;DEBUG_VALUE: 69 ; CHECK: ;DEBUG_VALUE: foo:var <- [DW_OP_constu 1, DW_OP_swap, DW_OP_xderef] 70 ; CHECK-NOT: ;DEBUG_VALUE: 71 call void @llvm.dbg.value(metadata <4 x float> %tmp29, metadata !3, metadata !DIExpression(DW_OP_constu, 1, DW_OP_swap, DW_OP_xderef)) #2, !dbg !5 72 %tmp59 = bitcast i64 %tmp35 to <2 x float> 73 %tmp60 = insertelement <2 x float> undef, float %tmp58, i32 0 74 %tmp61 = shufflevector <2 x float> %tmp60, <2 x float> undef, <2 x i32> zeroinitializer 75 %tmp62 = fmul <2 x float> %tmp61, undef 76 %tmp63 = fsub <2 x float> %tmp62, %tmp59 77 %tmp64 = extractelement <2 x float> %tmp63, i64 0 78 call void @eggs(float %tmp64) #2 79 store <2 x float> %tmp3, ptr addrspace(1) undef, align 8 80 store float 0.000000e+00, ptr addrspace(1) undef, align 4 81 ret void 82} 83 84declare float @barney() #2 85declare void @eggs(float) #2 86declare void @llvm.dbg.value(metadata, metadata, metadata) #1 87 88attributes #0 = { convergent nounwind "target-cpu"="gfx900" } 89attributes #1 = { nounwind readnone speculatable } 90attributes #2 = { nounwind } 91 92!llvm.dbg.cu = !{!0} 93!llvm.module.flags = !{!2} 94 95!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug) 96!1 = !DIFile(filename: "foo.cl", directory: "/tmp") 97!2 = !{i32 2, !"Debug Info Version", i32 3} 98!3 = !DILocalVariable(name: "var", arg: 8, scope: !4) 99!4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, type: !12, isLocal: false, isDefinition: true, flags: DIFlagPrototyped, isOptimized: true, unit: !0) 100!5 = !DILocation(line: 69, scope: !4) 101!12 = !DISubroutineType(types: !13) 102!13 = !{null, !14} 103!14 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed) 104