1; RUN: opt < %s -instcombine -S | FileCheck %s 2target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64" 3@C.0.1248 = internal constant [128 x float] [ float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float -1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float -1.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float -1.000000e+00, float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00 ], align 32 ; <[128 x float]*> [#uses=1] 4 5define float @test1(i32 %hash, float %x, float %y, float %z, float %w) { 6entry: 7 %lookupTable = alloca [128 x float], align 16 ; <[128 x float]*> [#uses=5] 8 %lookupTable1 = bitcast [128 x float]* %lookupTable to i8* ; <i8*> [#uses=1] 9 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 %lookupTable1, i8* align 16 bitcast ([128 x float]* @C.0.1248 to i8*), i64 512, i1 false) 10 11; CHECK-LABEL: @test1( 12; CHECK-NOT: alloca 13; CHECK-NOT: call{{.*}}@llvm.memcpy 14 15 %tmp3 = shl i32 %hash, 2 ; <i32> [#uses=1] 16 %tmp5 = and i32 %tmp3, 124 ; <i32> [#uses=4] 17 %tmp753 = getelementptr [128 x float], [128 x float]* %lookupTable, i32 0, i32 %tmp5 ; <float*> [#uses=1] 18 %tmp9 = load float, float* %tmp753 ; <float> [#uses=1] 19 %tmp11 = fmul float %tmp9, %x ; <float> [#uses=1] 20 %tmp13 = fadd float %tmp11, 0.000000e+00 ; <float> [#uses=1] 21 %tmp17.sum52 = or i32 %tmp5, 1 ; <i32> [#uses=1] 22 %tmp1851 = getelementptr [128 x float], [128 x float]* %lookupTable, i32 0, i32 %tmp17.sum52 ; <float*> [#uses=1] 23 %tmp19 = load float, float* %tmp1851 ; <float> [#uses=1] 24 %tmp21 = fmul float %tmp19, %y ; <float> [#uses=1] 25 %tmp23 = fadd float %tmp21, %tmp13 ; <float> [#uses=1] 26 %tmp27.sum50 = or i32 %tmp5, 2 ; <i32> [#uses=1] 27 %tmp2849 = getelementptr [128 x float], [128 x float]* %lookupTable, i32 0, i32 %tmp27.sum50 ; <float*> [#uses=1] 28 %tmp29 = load float, float* %tmp2849 ; <float> [#uses=1] 29 %tmp31 = fmul float %tmp29, %z ; <float> [#uses=1] 30 %tmp33 = fadd float %tmp31, %tmp23 ; <float> [#uses=1] 31 %tmp37.sum48 = or i32 %tmp5, 3 ; <i32> [#uses=1] 32 %tmp3847 = getelementptr [128 x float], [128 x float]* %lookupTable, i32 0, i32 %tmp37.sum48 ; <float*> [#uses=1] 33 %tmp39 = load float, float* %tmp3847 ; <float> [#uses=1] 34 %tmp41 = fmul float %tmp39, %w ; <float> [#uses=1] 35 %tmp43 = fadd float %tmp41, %tmp33 ; <float> [#uses=1] 36 ret float %tmp43 37} 38 39declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1) nounwind 40declare void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* nocapture, i8* nocapture, i64, i1) nounwind 41declare void @llvm.memcpy.p0i8.p1i8.i64(i8* nocapture, i8 addrspace(1)* nocapture, i64, i1) nounwind 42declare void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* nocapture, i8 addrspace(1)* nocapture, i64, i1) nounwind 43 44%T = type { i8, [123 x i8] } 45%U = type { i32, i32, i32, i32, i32 } 46 47@G = constant %T {i8 1, [123 x i8] zeroinitializer } 48@H = constant [2 x %U] zeroinitializer, align 16 49 50define void @test2() { 51 %A = alloca %T 52 %B = alloca %T 53 %a = bitcast %T* %A to i8* 54 %b = bitcast %T* %B to i8* 55 56; CHECK-LABEL: @test2( 57 58; %A alloca is deleted 59; CHECK-NEXT: alloca [124 x i8] 60; CHECK-NEXT: getelementptr inbounds [124 x i8], [124 x i8]* 61 62; use @G instead of %A 63; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 8 %{{.*}}, i8* align 16 getelementptr inbounds (%T, %T* @G, i64 0, i32 0) 64 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%T* @G to i8*), i64 124, i1 false) 65 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %b, i8* align 4 %a, i64 124, i1 false) 66 call void @bar(i8* %b) 67 ret void 68} 69 70define void @test2_no_null_opt() #0 { 71 %A = alloca %T 72 %B = alloca %T 73 %a = bitcast %T* %A to i8* 74 %b = bitcast %T* %B to i8* 75 76; CHECK-LABEL: @test2_no_null_opt( 77 78; %A alloca is deleted 79; CHECK-NEXT: alloca [124 x i8] 80; CHECK-NEXT: getelementptr inbounds [124 x i8], [124 x i8]* 81 82; use @G instead of %A 83; CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %{{.*}}, i8* align 16 getelementptr inbounds (%T, %T* @G, i64 0, i32 0) 84 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%T* @G to i8*), i64 124, i1 false) 85 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %b, i8* align 4 %a, i64 124, i1 false) 86 call void @bar(i8* %b) 87 ret void 88} 89 90define void @test2_addrspacecast() { 91 %A = alloca %T 92 %B = alloca %T 93 %a = addrspacecast %T* %A to i8 addrspace(1)* 94 %b = addrspacecast %T* %B to i8 addrspace(1)* 95 96; CHECK-LABEL: @test2_addrspacecast( 97 98; %A alloca is deleted 99; This doesn't exactly match what test2 does, because folding the type 100; cast into the alloca doesn't work for the addrspacecast yet. 101; CHECK-NEXT: alloca [124 x i8] 102; CHECK-NEXT: getelementptr 103; CHECK-NEXT: addrspacecast 104 105; use @G instead of %A 106; CHECK-NEXT: call void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* align 4 %{{.*}}, 107 call void @llvm.memcpy.p1i8.p0i8.i64(i8 addrspace(1)* align 4 %a, i8* align 4 bitcast (%T* @G to i8*), i64 124, i1 false) 108 call void @llvm.memcpy.p1i8.p1i8.i64(i8 addrspace(1)* align 4 %b, i8 addrspace(1)* align 4 %a, i64 124, i1 false) 109 call void @bar_as1(i8 addrspace(1)* %b) 110 ret void 111} 112 113declare void @bar(i8*) 114declare void @bar_as1(i8 addrspace(1)*) 115 116 117;; Should be able to eliminate the alloca. 118define void @test3() { 119 %A = alloca %T 120 %a = bitcast %T* %A to i8* 121 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%T* @G to i8*), i64 124, i1 false) 122 call void @bar(i8* %a) readonly 123; CHECK-LABEL: @test3( 124; CHECK-NEXT: call void @bar(i8* getelementptr inbounds (%T, %T* @G, i64 0, i32 0)) 125 ret void 126} 127 128define void @test3_addrspacecast() { 129 %A = alloca %T 130 %a = bitcast %T* %A to i8* 131 call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %a, i8 addrspace(1)* align 4 addrspacecast (%T* @G to i8 addrspace(1)*), i64 124, i1 false) 132 call void @bar(i8* %a) readonly 133; CHECK-LABEL: @test3_addrspacecast( 134; CHECK-NEXT: call void @bar(i8* getelementptr inbounds (%T, %T* @G, i64 0, i32 0)) 135 ret void 136} 137 138 139define void @test4() { 140 %A = alloca %T 141 %a = bitcast %T* %A to i8* 142 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%T* @G to i8*), i64 124, i1 false) 143 call void @baz(i8* byval %a) 144; CHECK-LABEL: @test4( 145; CHECK-NEXT: call void @baz(i8* byval getelementptr inbounds (%T, %T* @G, i64 0, i32 0)) 146 ret void 147} 148 149declare void @llvm.lifetime.start.p0i8(i64, i8*) 150define void @test5() { 151 %A = alloca %T 152 %a = bitcast %T* %A to i8* 153 call void @llvm.lifetime.start.p0i8(i64 -1, i8* %a) 154 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%T* @G to i8*), i64 124, i1 false) 155 call void @baz(i8* byval %a) 156; CHECK-LABEL: @test5( 157; CHECK-NEXT: call void @baz(i8* byval getelementptr inbounds (%T, %T* @G, i64 0, i32 0)) 158 ret void 159} 160 161 162declare void @baz(i8* byval) 163 164 165define void @test6() { 166 %A = alloca %U, align 16 167 %a = bitcast %U* %A to i8* 168 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 %a, i8* align 16 bitcast ([2 x %U]* @H to i8*), i64 20, i1 false) 169 call void @bar(i8* %a) readonly 170; CHECK-LABEL: @test6( 171; CHECK-NEXT: call void @bar(i8* bitcast ([2 x %U]* @H to i8*)) 172 ret void 173} 174 175define void @test7() { 176 %A = alloca %U, align 16 177 %a = bitcast %U* %A to i8* 178 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%U* getelementptr ([2 x %U], [2 x %U]* @H, i64 0, i32 0) to i8*), i64 20, i1 false) 179 call void @bar(i8* %a) readonly 180; CHECK-LABEL: @test7( 181; CHECK-NEXT: call void @bar(i8* bitcast ([2 x %U]* @H to i8*)) 182 ret void 183} 184 185define void @test8() { 186 %A = alloca %U, align 16 187 %a = bitcast %U* %A to i8* 188 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%U* getelementptr ([2 x %U], [2 x %U]* @H, i64 0, i32 1) to i8*), i64 20, i1 false) 189 call void @bar(i8* %a) readonly 190; CHECK-LABEL: @test8( 191; CHECK: llvm.memcpy 192; CHECK: bar 193 ret void 194} 195 196 197define void @test8_addrspacecast() { 198 %A = alloca %U, align 16 199 %a = bitcast %U* %A to i8* 200 call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %a, i8 addrspace(1)* align 4 addrspacecast (%U* getelementptr ([2 x %U], [2 x %U]* @H, i64 0, i32 1) to i8 addrspace(1)*), i64 20, i1 false) 201 call void @bar(i8* %a) readonly 202; CHECK-LABEL: @test8_addrspacecast( 203; CHECK: llvm.memcpy 204; CHECK: bar 205 ret void 206} 207 208define void @test9() { 209 %A = alloca %U, align 4 210 %a = bitcast %U* %A to i8* 211 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a, i8* align 4 bitcast (%U* getelementptr ([2 x %U], [2 x %U]* @H, i64 0, i32 1) to i8*), i64 20, i1 false) 212 call void @bar(i8* %a) readonly 213; CHECK-LABEL: @test9( 214; CHECK-NEXT: call void @bar(i8* bitcast (%U* getelementptr inbounds ([2 x %U], [2 x %U]* @H, i64 0, i64 1) to i8*)) 215 ret void 216} 217 218define void @test9_addrspacecast() { 219 %A = alloca %U, align 4 220 %a = bitcast %U* %A to i8* 221 call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %a, i8 addrspace(1)* align 4 addrspacecast (%U* getelementptr ([2 x %U], [2 x %U]* @H, i64 0, i32 1) to i8 addrspace(1)*), i64 20, i1 false) 222 call void @bar(i8* %a) readonly 223; CHECK-LABEL: @test9_addrspacecast( 224; CHECK-NEXT: call void @bar(i8* bitcast (%U* getelementptr inbounds ([2 x %U], [2 x %U]* @H, i64 0, i64 1) to i8*)) 225 ret void 226} 227 228@bbb = local_unnamed_addr global [1000000 x i8] zeroinitializer, align 16 229@_ZL3KKK = internal unnamed_addr constant [3 x i8] c"\01\01\02", align 1 230 231; Should not replace alloca with global because of size mismatch. 232define void @test9_small_global() { 233; CHECK-LABEL: @test9_small_global( 234; CHECK-NOT: call void @llvm.memcpy.p0i8.p0i8.i64({{.*}}@bbb,{{.*}}@_ZL3KKK, 235; CHECK: alloca [1000000 x i8] 236entry: 237 %cc = alloca [1000000 x i8], align 16 238 %cc.0..sroa_idx = getelementptr inbounds [1000000 x i8], [1000000 x i8]* %cc, i64 0, i64 0 239 %arraydecay = getelementptr inbounds [1000000 x i8], [1000000 x i8]* %cc, i32 0, i32 0 240 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %arraydecay, i8* getelementptr inbounds ([3 x i8], [3 x i8]* @_ZL3KKK, i32 0, i32 0), i64 3, i1 false) 241 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 16 getelementptr inbounds ([1000000 x i8], [1000000 x i8]* @bbb, i32 0, i32 0), i8* align 16 %arraydecay, i64 1000000, i1 false) 242 ret void 243} 244 245; Should replace alloca with global as they have exactly the same size. 246define void @test10_same_global() { 247; CHECK-LABEL: @test10_same_global( 248; CHECK-NOT: alloca 249; CHECK: call void @llvm.memcpy.p0i8.p0i8.i64({{.*}}@bbb,{{.*}}@_ZL3KKK,{{.*}}, i64 3, 250entry: 251 %cc = alloca [3 x i8], align 1 252 %cc.0..sroa_idx = getelementptr inbounds [3 x i8], [3 x i8]* %cc, i64 0, i64 0 253 %arraydecay = getelementptr inbounds [3 x i8], [3 x i8]* %cc, i32 0, i32 0 254 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %arraydecay, i8* getelementptr inbounds ([3 x i8], [3 x i8]* @_ZL3KKK, i32 0, i32 0), i64 3, i1 false) 255 call void @llvm.memcpy.p0i8.p0i8.i64(i8* getelementptr inbounds ([1000000 x i8], [1000000 x i8]* @bbb, i32 0, i32 0), i8* %arraydecay, i64 3, i1 false) 256 ret void 257} 258 259attributes #0 = { "null-pointer-is-valid"="true" } 260