1; RUN: llc -march=r600 -mcpu=cypress < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC 2; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC 3; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-PROMOTE 4; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC 5; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-ALLOCA 6; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC 7; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC 8 9; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -check-prefix=HSAOPT -check-prefix=OPT %s 10; RUN: opt -S -mtriple=amdgcn-unknown-unknown -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -check-prefix=NOHSAOPT -check-prefix=OPT %s 11 12; HSAOPT: @mova_same_clause.stack = internal unnamed_addr addrspace(3) global [256 x [5 x i32]] undef, align 4 13; HSAOPT: @high_alignment.stack = internal unnamed_addr addrspace(3) global [256 x [8 x i32]] undef, align 16 14 15 16; FUNC-LABEL: {{^}}mova_same_clause: 17; OPT-LABEL: @mova_same_clause( 18 19; R600: LDS_WRITE 20; R600: LDS_WRITE 21; R600: LDS_READ 22; R600: LDS_READ 23 24; HSA-PROMOTE: .amd_kernel_code_t 25; HSA-PROMOTE: workgroup_group_segment_byte_size = 5120 26; HSA-PROMOTE: .end_amd_kernel_code_t 27 28; FIXME: These should be merged 29; HSA-PROMOTE: s_load_dword s{{[0-9]+}}, s[4:5], 0x1 30; HSA-PROMOTE: s_load_dword s{{[0-9]+}}, s[4:5], 0x2 31 32; SI-PROMOTE: ds_write_b32 33; SI-PROMOTE: ds_write_b32 34; SI-PROMOTE: ds_read_b32 35; SI-PROMOTE: ds_read_b32 36 37; HSA-ALLOCA: .amd_kernel_code_t 38; FIXME: Creating the emergency stack slots causes us to over-estimate scratch 39; by 4 bytes. 40; HSA-ALLOCA: workitem_private_segment_byte_size = 24 41; HSA-ALLOCA: .end_amd_kernel_code_t 42 43; HSA-ALLOCA: s_mov_b32 flat_scratch_lo, s7 44; HSA-ALLOCA: s_add_u32 s6, s6, s9 45; HSA-ALLOCA: s_lshr_b32 flat_scratch_hi, s6, 8 46 47; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0 48; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0 49 50 51; HSAOPT: [[DISPATCH_PTR:%[0-9]+]] = call noalias nonnull dereferenceable(64) i8 addrspace(2)* @llvm.amdgcn.dispatch.ptr() 52; HSAOPT: [[CAST_DISPATCH_PTR:%[0-9]+]] = bitcast i8 addrspace(2)* [[DISPATCH_PTR]] to i32 addrspace(2)* 53; HSAOPT: [[GEP0:%[0-9]+]] = getelementptr inbounds i32, i32 addrspace(2)* [[CAST_DISPATCH_PTR]], i64 1 54; HSAOPT: [[LDXY:%[0-9]+]] = load i32, i32 addrspace(2)* [[GEP0]], align 4, !invariant.load !0 55; HSAOPT: [[GEP1:%[0-9]+]] = getelementptr inbounds i32, i32 addrspace(2)* [[CAST_DISPATCH_PTR]], i64 2 56; HSAOPT: [[LDZU:%[0-9]+]] = load i32, i32 addrspace(2)* [[GEP1]], align 4, !range !1, !invariant.load !0 57; HSAOPT: [[EXTRACTY:%[0-9]+]] = lshr i32 [[LDXY]], 16 58 59; HSAOPT: [[WORKITEM_ID_X:%[0-9]+]] = call i32 @llvm.amdgcn.workitem.id.x(), !range !1 60; HSAOPT: [[WORKITEM_ID_Y:%[0-9]+]] = call i32 @llvm.amdgcn.workitem.id.y(), !range !1 61; HSAOPT: [[WORKITEM_ID_Z:%[0-9]+]] = call i32 @llvm.amdgcn.workitem.id.z(), !range !1 62 63; HSAOPT: [[Y_SIZE_X_Z_SIZE:%[0-9]+]] = mul nuw nsw i32 [[EXTRACTY]], [[LDZU]] 64; HSAOPT: [[YZ_X_XID:%[0-9]+]] = mul i32 [[Y_SIZE_X_Z_SIZE]], [[WORKITEM_ID_X]] 65; HSAOPT: [[Y_X_Z_SIZE:%[0-9]+]] = mul nuw nsw i32 [[WORKITEM_ID_Y]], [[LDZU]] 66; HSAOPT: [[ADD_YZ_X_X_YZ_SIZE:%[0-9]+]] = add i32 [[YZ_X_XID]], [[Y_X_Z_SIZE]] 67; HSAOPT: [[ADD_ZID:%[0-9]+]] = add i32 [[ADD_YZ_X_X_YZ_SIZE]], [[WORKITEM_ID_Z]] 68 69; HSAOPT: [[LOCAL_GEP:%[0-9]+]] = getelementptr inbounds [256 x [5 x i32]], [256 x [5 x i32]] addrspace(3)* @mova_same_clause.stack, i32 0, i32 [[ADD_ZID]] 70; HSAOPT: %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 {{%[0-9]+}} 71; HSAOPT: %arrayidx3 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 {{%[0-9]+}} 72; HSAOPT: %arrayidx10 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 0 73; HSAOPT: %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32] addrspace(3)* [[LOCAL_GEP]], i32 0, i32 1 74 75 76; NOHSAOPT: call i32 @llvm.r600.read.local.size.y(), !range !0 77; NOHSAOPT: call i32 @llvm.r600.read.local.size.z(), !range !0 78; NOHSAOPT: call i32 @llvm.amdgcn.workitem.id.x(), !range !0 79; NOHSAOPT: call i32 @llvm.amdgcn.workitem.id.y(), !range !0 80; NOHSAOPT: call i32 @llvm.amdgcn.workitem.id.z(), !range !0 81define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) { 82entry: 83 %stack = alloca [5 x i32], align 4 84 %0 = load i32, i32 addrspace(1)* %in, align 4 85 %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %0 86 store i32 4, i32* %arrayidx1, align 4 87 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1 88 %1 = load i32, i32 addrspace(1)* %arrayidx2, align 4 89 %arrayidx3 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %1 90 store i32 5, i32* %arrayidx3, align 4 91 %arrayidx10 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 0 92 %2 = load i32, i32* %arrayidx10, align 4 93 store i32 %2, i32 addrspace(1)* %out, align 4 94 %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 1 95 %3 = load i32, i32* %arrayidx12 96 %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1 97 store i32 %3, i32 addrspace(1)* %arrayidx13 98 ret void 99} 100 101; OPT-LABEL: @high_alignment( 102; OPT: getelementptr inbounds [256 x [8 x i32]], [256 x [8 x i32]] addrspace(3)* @high_alignment.stack, i32 0, i32 %{{[0-9]+}} 103define void @high_alignment(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) { 104entry: 105 %stack = alloca [8 x i32], align 16 106 %0 = load i32, i32 addrspace(1)* %in, align 4 107 %arrayidx1 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 %0 108 store i32 4, i32* %arrayidx1, align 4 109 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1 110 %1 = load i32, i32 addrspace(1)* %arrayidx2, align 4 111 %arrayidx3 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 %1 112 store i32 5, i32* %arrayidx3, align 4 113 %arrayidx10 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 0 114 %2 = load i32, i32* %arrayidx10, align 4 115 store i32 %2, i32 addrspace(1)* %out, align 4 116 %arrayidx12 = getelementptr inbounds [8 x i32], [8 x i32]* %stack, i32 0, i32 1 117 %3 = load i32, i32* %arrayidx12 118 %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1 119 store i32 %3, i32 addrspace(1)* %arrayidx13 120 ret void 121} 122 123; FUNC-LABEL: {{^}}no_replace_inbounds_gep: 124; OPT-LABEL: @no_replace_inbounds_gep( 125; OPT: alloca [5 x i32] 126 127; SI-NOT: ds_write 128define void @no_replace_inbounds_gep(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) { 129entry: 130 %stack = alloca [5 x i32], align 4 131 %0 = load i32, i32 addrspace(1)* %in, align 4 132 %arrayidx1 = getelementptr [5 x i32], [5 x i32]* %stack, i32 0, i32 %0 133 store i32 4, i32* %arrayidx1, align 4 134 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1 135 %1 = load i32, i32 addrspace(1)* %arrayidx2, align 4 136 %arrayidx3 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %1 137 store i32 5, i32* %arrayidx3, align 4 138 %arrayidx10 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 0 139 %2 = load i32, i32* %arrayidx10, align 4 140 store i32 %2, i32 addrspace(1)* %out, align 4 141 %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 1 142 %3 = load i32, i32* %arrayidx12 143 %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1 144 store i32 %3, i32 addrspace(1)* %arrayidx13 145 ret void 146} 147 148; This test checks that the stack offset is calculated correctly for structs. 149; All register loads/stores should be optimized away, so there shouldn't be 150; any MOVA instructions. 151; 152; XXX: This generated code has unnecessary MOVs, we should be able to optimize 153; this. 154 155; FUNC-LABEL: {{^}}multiple_structs: 156; OPT-LABEL: @multiple_structs( 157 158; R600-NOT: MOVA_INT 159; SI-NOT: v_movrel 160; SI-NOT: v_movrel 161%struct.point = type { i32, i32 } 162 163define void @multiple_structs(i32 addrspace(1)* %out) { 164entry: 165 %a = alloca %struct.point 166 %b = alloca %struct.point 167 %a.x.ptr = getelementptr %struct.point, %struct.point* %a, i32 0, i32 0 168 %a.y.ptr = getelementptr %struct.point, %struct.point* %a, i32 0, i32 1 169 %b.x.ptr = getelementptr %struct.point, %struct.point* %b, i32 0, i32 0 170 %b.y.ptr = getelementptr %struct.point, %struct.point* %b, i32 0, i32 1 171 store i32 0, i32* %a.x.ptr 172 store i32 1, i32* %a.y.ptr 173 store i32 2, i32* %b.x.ptr 174 store i32 3, i32* %b.y.ptr 175 %a.indirect.ptr = getelementptr %struct.point, %struct.point* %a, i32 0, i32 0 176 %b.indirect.ptr = getelementptr %struct.point, %struct.point* %b, i32 0, i32 0 177 %a.indirect = load i32, i32* %a.indirect.ptr 178 %b.indirect = load i32, i32* %b.indirect.ptr 179 %0 = add i32 %a.indirect, %b.indirect 180 store i32 %0, i32 addrspace(1)* %out 181 ret void 182} 183 184; Test direct access of a private array inside a loop. The private array 185; loads and stores should be lowered to copies, so there shouldn't be any 186; MOVA instructions. 187 188; FUNC-LABEL: {{^}}direct_loop: 189; R600-NOT: MOVA_INT 190; SI-NOT: v_movrel 191 192define void @direct_loop(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 193entry: 194 %prv_array_const = alloca [2 x i32] 195 %prv_array = alloca [2 x i32] 196 %a = load i32, i32 addrspace(1)* %in 197 %b_src_ptr = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1 198 %b = load i32, i32 addrspace(1)* %b_src_ptr 199 %a_dst_ptr = getelementptr inbounds [2 x i32], [2 x i32]* %prv_array_const, i32 0, i32 0 200 store i32 %a, i32* %a_dst_ptr 201 %b_dst_ptr = getelementptr inbounds [2 x i32], [2 x i32]* %prv_array_const, i32 0, i32 1 202 store i32 %b, i32* %b_dst_ptr 203 br label %for.body 204 205for.body: 206 %inc = phi i32 [0, %entry], [%count, %for.body] 207 %x_ptr = getelementptr inbounds [2 x i32], [2 x i32]* %prv_array_const, i32 0, i32 0 208 %x = load i32, i32* %x_ptr 209 %y_ptr = getelementptr inbounds [2 x i32], [2 x i32]* %prv_array, i32 0, i32 0 210 %y = load i32, i32* %y_ptr 211 %xy = add i32 %x, %y 212 store i32 %xy, i32* %y_ptr 213 %count = add i32 %inc, 1 214 %done = icmp eq i32 %count, 4095 215 br i1 %done, label %for.end, label %for.body 216 217for.end: 218 %value_ptr = getelementptr inbounds [2 x i32], [2 x i32]* %prv_array, i32 0, i32 0 219 %value = load i32, i32* %value_ptr 220 store i32 %value, i32 addrspace(1)* %out 221 ret void 222} 223 224; FUNC-LABEL: {{^}}short_array: 225 226; R600: MOVA_INT 227 228; SI-PROMOTE-DAG: buffer_store_short v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x68,0xe0 229; SI-PROMOTE-DAG: buffer_store_short v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen offset:2 ; encoding: [0x02,0x10,0x68,0xe0 230; SI-PROMOTE: buffer_load_sshort v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} 231define void @short_array(i32 addrspace(1)* %out, i32 %index) { 232entry: 233 %0 = alloca [2 x i16] 234 %1 = getelementptr inbounds [2 x i16], [2 x i16]* %0, i32 0, i32 0 235 %2 = getelementptr inbounds [2 x i16], [2 x i16]* %0, i32 0, i32 1 236 store i16 0, i16* %1 237 store i16 1, i16* %2 238 %3 = getelementptr inbounds [2 x i16], [2 x i16]* %0, i32 0, i32 %index 239 %4 = load i16, i16* %3 240 %5 = sext i16 %4 to i32 241 store i32 %5, i32 addrspace(1)* %out 242 ret void 243} 244 245; FUNC-LABEL: {{^}}char_array: 246 247; R600: MOVA_INT 248 249; SI-DAG: buffer_store_byte v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x60,0xe0 250; SI-DAG: buffer_store_byte v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen offset:1 ; encoding: [0x01,0x10,0x60,0xe0 251define void @char_array(i32 addrspace(1)* %out, i32 %index) { 252entry: 253 %0 = alloca [2 x i8] 254 %1 = getelementptr inbounds [2 x i8], [2 x i8]* %0, i32 0, i32 0 255 %2 = getelementptr inbounds [2 x i8], [2 x i8]* %0, i32 0, i32 1 256 store i8 0, i8* %1 257 store i8 1, i8* %2 258 %3 = getelementptr inbounds [2 x i8], [2 x i8]* %0, i32 0, i32 %index 259 %4 = load i8, i8* %3 260 %5 = sext i8 %4 to i32 261 store i32 %5, i32 addrspace(1)* %out 262 ret void 263 264} 265 266 267; Test that two stack objects are not stored in the same register 268; The second stack object should be in T3.X 269; FUNC-LABEL: {{^}}no_overlap: 270; R600_CHECK: MOV 271; R600_CHECK: [[CHAN:[XYZW]]]+ 272; R600-NOT: [[CHAN]]+ 273; SI: v_mov_b32_e32 v3 274define void @no_overlap(i32 addrspace(1)* %out, i32 %in) { 275entry: 276 %0 = alloca [3 x i8], align 1 277 %1 = alloca [2 x i8], align 1 278 %2 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 0 279 %3 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 1 280 %4 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 2 281 %5 = getelementptr [2 x i8], [2 x i8]* %1, i32 0, i32 0 282 %6 = getelementptr [2 x i8], [2 x i8]* %1, i32 0, i32 1 283 store i8 0, i8* %2 284 store i8 1, i8* %3 285 store i8 2, i8* %4 286 store i8 1, i8* %5 287 store i8 0, i8* %6 288 %7 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 %in 289 %8 = getelementptr [2 x i8], [2 x i8]* %1, i32 0, i32 %in 290 %9 = load i8, i8* %7 291 %10 = load i8, i8* %8 292 %11 = add i8 %9, %10 293 %12 = sext i8 %11 to i32 294 store i32 %12, i32 addrspace(1)* %out 295 ret void 296} 297 298define void @char_array_array(i32 addrspace(1)* %out, i32 %index) { 299entry: 300 %alloca = alloca [2 x [2 x i8]] 301 %gep0 = getelementptr [2 x [2 x i8]], [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 0 302 %gep1 = getelementptr [2 x [2 x i8]], [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 1 303 store i8 0, i8* %gep0 304 store i8 1, i8* %gep1 305 %gep2 = getelementptr [2 x [2 x i8]], [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 %index 306 %load = load i8, i8* %gep2 307 %sext = sext i8 %load to i32 308 store i32 %sext, i32 addrspace(1)* %out 309 ret void 310} 311 312define void @i32_array_array(i32 addrspace(1)* %out, i32 %index) { 313entry: 314 %alloca = alloca [2 x [2 x i32]] 315 %gep0 = getelementptr [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 0 316 %gep1 = getelementptr [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 1 317 store i32 0, i32* %gep0 318 store i32 1, i32* %gep1 319 %gep2 = getelementptr [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 %index 320 %load = load i32, i32* %gep2 321 store i32 %load, i32 addrspace(1)* %out 322 ret void 323} 324 325define void @i64_array_array(i64 addrspace(1)* %out, i32 %index) { 326entry: 327 %alloca = alloca [2 x [2 x i64]] 328 %gep0 = getelementptr [2 x [2 x i64]], [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 0 329 %gep1 = getelementptr [2 x [2 x i64]], [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 1 330 store i64 0, i64* %gep0 331 store i64 1, i64* %gep1 332 %gep2 = getelementptr [2 x [2 x i64]], [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 %index 333 %load = load i64, i64* %gep2 334 store i64 %load, i64 addrspace(1)* %out 335 ret void 336} 337 338%struct.pair32 = type { i32, i32 } 339 340define void @struct_array_array(i32 addrspace(1)* %out, i32 %index) { 341entry: 342 %alloca = alloca [2 x [2 x %struct.pair32]] 343 %gep0 = getelementptr [2 x [2 x %struct.pair32]], [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 0, i32 1 344 %gep1 = getelementptr [2 x [2 x %struct.pair32]], [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 1, i32 1 345 store i32 0, i32* %gep0 346 store i32 1, i32* %gep1 347 %gep2 = getelementptr [2 x [2 x %struct.pair32]], [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 %index, i32 0 348 %load = load i32, i32* %gep2 349 store i32 %load, i32 addrspace(1)* %out 350 ret void 351} 352 353define void @struct_pair32_array(i32 addrspace(1)* %out, i32 %index) { 354entry: 355 %alloca = alloca [2 x %struct.pair32] 356 %gep0 = getelementptr [2 x %struct.pair32], [2 x %struct.pair32]* %alloca, i32 0, i32 0, i32 1 357 %gep1 = getelementptr [2 x %struct.pair32], [2 x %struct.pair32]* %alloca, i32 0, i32 1, i32 0 358 store i32 0, i32* %gep0 359 store i32 1, i32* %gep1 360 %gep2 = getelementptr [2 x %struct.pair32], [2 x %struct.pair32]* %alloca, i32 0, i32 %index, i32 0 361 %load = load i32, i32* %gep2 362 store i32 %load, i32 addrspace(1)* %out 363 ret void 364} 365 366define void @select_private(i32 addrspace(1)* %out, i32 %in) nounwind { 367entry: 368 %tmp = alloca [2 x i32] 369 %tmp1 = getelementptr [2 x i32], [2 x i32]* %tmp, i32 0, i32 0 370 %tmp2 = getelementptr [2 x i32], [2 x i32]* %tmp, i32 0, i32 1 371 store i32 0, i32* %tmp1 372 store i32 1, i32* %tmp2 373 %cmp = icmp eq i32 %in, 0 374 %sel = select i1 %cmp, i32* %tmp1, i32* %tmp2 375 %load = load i32, i32* %sel 376 store i32 %load, i32 addrspace(1)* %out 377 ret void 378} 379 380; AMDGPUPromoteAlloca does not know how to handle ptrtoint. When it 381; finds one, it should stop trying to promote. 382 383; FUNC-LABEL: ptrtoint: 384; SI-NOT: ds_write 385; SI: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen 386; SI: buffer_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen offset:5 ; 387define void @ptrtoint(i32 addrspace(1)* %out, i32 %a, i32 %b) { 388 %alloca = alloca [16 x i32] 389 %tmp0 = getelementptr [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 390 store i32 5, i32* %tmp0 391 %tmp1 = ptrtoint [16 x i32]* %alloca to i32 392 %tmp2 = add i32 %tmp1, 5 393 %tmp3 = inttoptr i32 %tmp2 to i32* 394 %tmp4 = getelementptr i32, i32* %tmp3, i32 %b 395 %tmp5 = load i32, i32* %tmp4 396 store i32 %tmp5, i32 addrspace(1)* %out 397 ret void 398} 399 400; HSAOPT: !0 = !{} 401; HSAOPT: !1 = !{i32 0, i32 2048} 402 403; NOHSAOPT: !0 = !{i32 0, i32 2048} 404