1; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -march=amdgcn < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC %s 2; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-PROMOTE %s 3; RUN: llc -show-mc-encoding -mattr=-promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -march=amdgcn < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC 4; RUN: llc -show-mc-encoding -mattr=-promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-ALLOCA %s 5; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -march=amdgcn -mcpu=tonga -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC %s 6; RUN: llc -show-mc-encoding -mattr=-promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -march=amdgcn -mcpu=tonga -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC %s 7 8; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -enable-var-scope -check-prefix=HSAOPT -check-prefix=OPT %s 9; RUN: opt -S -mtriple=amdgcn-unknown-unknown -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -enable-var-scope -check-prefix=NOHSAOPT -check-prefix=OPT %s 10 11; RUN: llc -march=r600 -mcpu=cypress < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC 12 13 14; HSAOPT: @mova_same_clause.stack = internal unnamed_addr addrspace(3) global [256 x [5 x i32]] undef, align 4 15; HSAOPT: @high_alignment.stack = internal unnamed_addr addrspace(3) global [256 x [8 x i32]] undef, align 16 16 17 18; FUNC-LABEL: {{^}}mova_same_clause: 19; OPT-LABEL: @mova_same_clause( 20 21; R600: LDS_WRITE 22; R600: LDS_WRITE 23; R600: LDS_READ 24; R600: LDS_READ 25 26; HSA-PROMOTE: .amd_kernel_code_t 27; HSA-PROMOTE: workgroup_group_segment_byte_size = 5120 28; HSA-PROMOTE: .end_amd_kernel_code_t 29 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 !2 60; HSAOPT: [[WORKITEM_ID_Y:%[0-9]+]] = call i32 @llvm.amdgcn.workitem.id.y(), !range !2 61; HSAOPT: [[WORKITEM_ID_Z:%[0-9]+]] = call i32 @llvm.amdgcn.workitem.id.z(), !range !2 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 !1 79; NOHSAOPT: call i32 @llvm.amdgcn.workitem.id.y(), !range !1 80; NOHSAOPT: call i32 @llvm.amdgcn.workitem.id.z(), !range !1 81define amdgpu_kernel void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #0 { 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 amdgpu_kernel void @high_alignment(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #0 { 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 amdgpu_kernel void @no_replace_inbounds_gep(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #0 { 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 amdgpu_kernel void @multiple_structs(i32 addrspace(1)* %out) #0 { 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 amdgpu_kernel void @direct_loop(i32 addrspace(1)* %out, i32 addrspace(1)* %in) #0 { 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-ALLOCA-DAG: buffer_store_short v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:6 ; encoding: [0x06,0x00,0x68,0xe0 229; SI-ALLOCA-DAG: buffer_store_short v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding: [0x04,0x00,0x68,0xe0 230; Loaded value is 0 or 1, so sext will become zext, so we get buffer_load_ushort instead of buffer_load_sshort. 231; SI-ALLOCA: buffer_load_sshort v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} 232 233; SI-PROMOTE: s_load_dword [[IDX:s[0-9]+]] 234; SI-PROMOTE: s_lshl_b32 [[SCALED_IDX:s[0-9]+]], [[IDX]], 16 235; SI-PROMOTE: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[SCALED_IDX]], 16 236define amdgpu_kernel void @short_array(i32 addrspace(1)* %out, i32 %index) #0 { 237entry: 238 %0 = alloca [2 x i16] 239 %1 = getelementptr inbounds [2 x i16], [2 x i16]* %0, i32 0, i32 0 240 %2 = getelementptr inbounds [2 x i16], [2 x i16]* %0, i32 0, i32 1 241 store i16 0, i16* %1 242 store i16 1, i16* %2 243 %3 = getelementptr inbounds [2 x i16], [2 x i16]* %0, i32 0, i32 %index 244 %4 = load i16, i16* %3 245 %5 = sext i16 %4 to i32 246 store i32 %5, i32 addrspace(1)* %out 247 ret void 248} 249 250; FUNC-LABEL: {{^}}char_array: 251 252; R600: MOVA_INT 253 254; SI-PROMOTE-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding: 255; SI-PROMOTE-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:5 ; encoding: 256 257; SI-ALLOCA-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding: [0x04,0x00,0x60,0xe0 258; SI-ALLOCA-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:5 ; encoding: [0x05,0x00,0x60,0xe0 259define amdgpu_kernel void @char_array(i32 addrspace(1)* %out, i32 %index) #0 { 260entry: 261 %0 = alloca [2 x i8] 262 %1 = getelementptr inbounds [2 x i8], [2 x i8]* %0, i32 0, i32 0 263 %2 = getelementptr inbounds [2 x i8], [2 x i8]* %0, i32 0, i32 1 264 store i8 0, i8* %1 265 store i8 1, i8* %2 266 %3 = getelementptr inbounds [2 x i8], [2 x i8]* %0, i32 0, i32 %index 267 %4 = load i8, i8* %3 268 %5 = sext i8 %4 to i32 269 store i32 %5, i32 addrspace(1)* %out 270 ret void 271} 272 273; Test that two stack objects are not stored in the same register 274; The second stack object should be in T3.X 275; FUNC-LABEL: {{^}}no_overlap: 276; R600-CHECK: MOV 277; R600-CHECK: [[CHAN:[XYZW]]]+ 278; R600-NOT: [[CHAN]]+ 279; 280; A total of 5 bytes should be allocated and used. 281; SI: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; 282define amdgpu_kernel void @no_overlap(i32 addrspace(1)* %out, i32 %in) #0 { 283entry: 284 %0 = alloca [3 x i8], align 1 285 %1 = alloca [2 x i8], align 1 286 %2 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 0 287 %3 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 1 288 %4 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 2 289 %5 = getelementptr [2 x i8], [2 x i8]* %1, i32 0, i32 0 290 %6 = getelementptr [2 x i8], [2 x i8]* %1, i32 0, i32 1 291 store i8 0, i8* %2 292 store i8 1, i8* %3 293 store i8 2, i8* %4 294 store i8 1, i8* %5 295 store i8 0, i8* %6 296 %7 = getelementptr [3 x i8], [3 x i8]* %0, i32 0, i32 %in 297 %8 = getelementptr [2 x i8], [2 x i8]* %1, i32 0, i32 %in 298 %9 = load i8, i8* %7 299 %10 = load i8, i8* %8 300 %11 = add i8 %9, %10 301 %12 = sext i8 %11 to i32 302 store i32 %12, i32 addrspace(1)* %out 303 ret void 304} 305 306define amdgpu_kernel void @char_array_array(i32 addrspace(1)* %out, i32 %index) #0 { 307entry: 308 %alloca = alloca [2 x [2 x i8]] 309 %gep0 = getelementptr [2 x [2 x i8]], [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 0 310 %gep1 = getelementptr [2 x [2 x i8]], [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 1 311 store i8 0, i8* %gep0 312 store i8 1, i8* %gep1 313 %gep2 = getelementptr [2 x [2 x i8]], [2 x [2 x i8]]* %alloca, i32 0, i32 0, i32 %index 314 %load = load i8, i8* %gep2 315 %sext = sext i8 %load to i32 316 store i32 %sext, i32 addrspace(1)* %out 317 ret void 318} 319 320define amdgpu_kernel void @i32_array_array(i32 addrspace(1)* %out, i32 %index) #0 { 321entry: 322 %alloca = alloca [2 x [2 x i32]] 323 %gep0 = getelementptr [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 0 324 %gep1 = getelementptr [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 1 325 store i32 0, i32* %gep0 326 store i32 1, i32* %gep1 327 %gep2 = getelementptr [2 x [2 x i32]], [2 x [2 x i32]]* %alloca, i32 0, i32 0, i32 %index 328 %load = load i32, i32* %gep2 329 store i32 %load, i32 addrspace(1)* %out 330 ret void 331} 332 333define amdgpu_kernel void @i64_array_array(i64 addrspace(1)* %out, i32 %index) #0 { 334entry: 335 %alloca = alloca [2 x [2 x i64]] 336 %gep0 = getelementptr [2 x [2 x i64]], [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 0 337 %gep1 = getelementptr [2 x [2 x i64]], [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 1 338 store i64 0, i64* %gep0 339 store i64 1, i64* %gep1 340 %gep2 = getelementptr [2 x [2 x i64]], [2 x [2 x i64]]* %alloca, i32 0, i32 0, i32 %index 341 %load = load i64, i64* %gep2 342 store i64 %load, i64 addrspace(1)* %out 343 ret void 344} 345 346%struct.pair32 = type { i32, i32 } 347 348define amdgpu_kernel void @struct_array_array(i32 addrspace(1)* %out, i32 %index) #0 { 349entry: 350 %alloca = alloca [2 x [2 x %struct.pair32]] 351 %gep0 = getelementptr [2 x [2 x %struct.pair32]], [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 0, i32 1 352 %gep1 = getelementptr [2 x [2 x %struct.pair32]], [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 1, i32 1 353 store i32 0, i32* %gep0 354 store i32 1, i32* %gep1 355 %gep2 = getelementptr [2 x [2 x %struct.pair32]], [2 x [2 x %struct.pair32]]* %alloca, i32 0, i32 0, i32 %index, i32 0 356 %load = load i32, i32* %gep2 357 store i32 %load, i32 addrspace(1)* %out 358 ret void 359} 360 361define amdgpu_kernel void @struct_pair32_array(i32 addrspace(1)* %out, i32 %index) #0 { 362entry: 363 %alloca = alloca [2 x %struct.pair32] 364 %gep0 = getelementptr [2 x %struct.pair32], [2 x %struct.pair32]* %alloca, i32 0, i32 0, i32 1 365 %gep1 = getelementptr [2 x %struct.pair32], [2 x %struct.pair32]* %alloca, i32 0, i32 1, i32 0 366 store i32 0, i32* %gep0 367 store i32 1, i32* %gep1 368 %gep2 = getelementptr [2 x %struct.pair32], [2 x %struct.pair32]* %alloca, i32 0, i32 %index, i32 0 369 %load = load i32, i32* %gep2 370 store i32 %load, i32 addrspace(1)* %out 371 ret void 372} 373 374define amdgpu_kernel void @select_private(i32 addrspace(1)* %out, i32 %in) nounwind { 375entry: 376 %tmp = alloca [2 x i32] 377 %tmp1 = getelementptr [2 x i32], [2 x i32]* %tmp, i32 0, i32 0 378 %tmp2 = getelementptr [2 x i32], [2 x i32]* %tmp, i32 0, i32 1 379 store i32 0, i32* %tmp1 380 store i32 1, i32* %tmp2 381 %cmp = icmp eq i32 %in, 0 382 %sel = select i1 %cmp, i32* %tmp1, i32* %tmp2 383 %load = load i32, i32* %sel 384 store i32 %load, i32 addrspace(1)* %out 385 ret void 386} 387 388; AMDGPUPromoteAlloca does not know how to handle ptrtoint. When it 389; finds one, it should stop trying to promote. 390 391; FUNC-LABEL: ptrtoint: 392; SI-NOT: ds_write 393; SI: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen 394; SI: v_add_{{[iu]}}32_e32 [[ADD_OFFSET:v[0-9]+]], vcc, 5, 395; SI: buffer_load_dword v{{[0-9]+}}, [[ADD_OFFSET:v[0-9]+]], s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; 396define amdgpu_kernel void @ptrtoint(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 { 397 %alloca = alloca [16 x i32] 398 %tmp0 = getelementptr [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 399 store i32 5, i32* %tmp0 400 %tmp1 = ptrtoint [16 x i32]* %alloca to i32 401 %tmp2 = add i32 %tmp1, 5 402 %tmp3 = inttoptr i32 %tmp2 to i32* 403 %tmp4 = getelementptr i32, i32* %tmp3, i32 %b 404 %tmp5 = load i32, i32* %tmp4 405 store i32 %tmp5, i32 addrspace(1)* %out 406 ret void 407} 408 409; OPT-LABEL: @pointer_typed_alloca( 410; OPT: getelementptr inbounds [256 x i32 addrspace(1)*], [256 x i32 addrspace(1)*] addrspace(3)* @pointer_typed_alloca.A.addr, i32 0, i32 %{{[0-9]+}} 411; OPT: load i32 addrspace(1)*, i32 addrspace(1)* addrspace(3)* %{{[0-9]+}}, align 4 412define amdgpu_kernel void @pointer_typed_alloca(i32 addrspace(1)* %A) { 413entry: 414 %A.addr = alloca i32 addrspace(1)*, align 4 415 store i32 addrspace(1)* %A, i32 addrspace(1)** %A.addr, align 4 416 %ld0 = load i32 addrspace(1)*, i32 addrspace(1)** %A.addr, align 4 417 %arrayidx = getelementptr inbounds i32, i32 addrspace(1)* %ld0, i32 0 418 store i32 1, i32 addrspace(1)* %arrayidx, align 4 419 %ld1 = load i32 addrspace(1)*, i32 addrspace(1)** %A.addr, align 4 420 %arrayidx1 = getelementptr inbounds i32, i32 addrspace(1)* %ld1, i32 1 421 store i32 2, i32 addrspace(1)* %arrayidx1, align 4 422 %ld2 = load i32 addrspace(1)*, i32 addrspace(1)** %A.addr, align 4 423 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %ld2, i32 2 424 store i32 3, i32 addrspace(1)* %arrayidx2, align 4 425 ret void 426} 427 428; FUNC-LABEL: v16i32_stack: 429 430; R600: MOVA_INT 431; R600: MOVA_INT 432; R600: MOVA_INT 433; R600: MOVA_INT 434; R600: MOVA_INT 435; R600: MOVA_INT 436; R600: MOVA_INT 437; R600: MOVA_INT 438; R600: MOVA_INT 439; R600: MOVA_INT 440; R600: MOVA_INT 441; R600: MOVA_INT 442; R600: MOVA_INT 443; R600: MOVA_INT 444; R600: MOVA_INT 445; R600: MOVA_INT 446 447; SI: buffer_load_dword 448; SI: buffer_load_dword 449; SI: buffer_load_dword 450; SI: buffer_load_dword 451; SI: buffer_load_dword 452; SI: buffer_load_dword 453; SI: buffer_load_dword 454; SI: buffer_load_dword 455; SI: buffer_load_dword 456; SI: buffer_load_dword 457; SI: buffer_load_dword 458; SI: buffer_load_dword 459; SI: buffer_load_dword 460; SI: buffer_load_dword 461; SI: buffer_load_dword 462; SI: buffer_load_dword 463 464define amdgpu_kernel void @v16i32_stack(<16 x i32> addrspace(1)* %out, i32 %a) { 465 %alloca = alloca [2 x <16 x i32>] 466 %tmp0 = getelementptr [2 x <16 x i32>], [2 x <16 x i32>]* %alloca, i32 0, i32 %a 467 %tmp5 = load <16 x i32>, <16 x i32>* %tmp0 468 store <16 x i32> %tmp5, <16 x i32> addrspace(1)* %out 469 ret void 470} 471 472; FUNC-LABEL: v16float_stack: 473 474; R600: MOVA_INT 475; R600: MOVA_INT 476; R600: MOVA_INT 477; R600: MOVA_INT 478; R600: MOVA_INT 479; R600: MOVA_INT 480; R600: MOVA_INT 481; R600: MOVA_INT 482; R600: MOVA_INT 483; R600: MOVA_INT 484; R600: MOVA_INT 485; R600: MOVA_INT 486; R600: MOVA_INT 487; R600: MOVA_INT 488; R600: MOVA_INT 489; R600: MOVA_INT 490 491; SI: buffer_load_dword 492; SI: buffer_load_dword 493; SI: buffer_load_dword 494; SI: buffer_load_dword 495; SI: buffer_load_dword 496; SI: buffer_load_dword 497; SI: buffer_load_dword 498; SI: buffer_load_dword 499; SI: buffer_load_dword 500; SI: buffer_load_dword 501; SI: buffer_load_dword 502; SI: buffer_load_dword 503; SI: buffer_load_dword 504; SI: buffer_load_dword 505; SI: buffer_load_dword 506; SI: buffer_load_dword 507 508define amdgpu_kernel void @v16float_stack(<16 x float> addrspace(1)* %out, i32 %a) { 509 %alloca = alloca [2 x <16 x float>] 510 %tmp0 = getelementptr [2 x <16 x float>], [2 x <16 x float>]* %alloca, i32 0, i32 %a 511 %tmp5 = load <16 x float>, <16 x float>* %tmp0 512 store <16 x float> %tmp5, <16 x float> addrspace(1)* %out 513 ret void 514} 515 516; FUNC-LABEL: v2float_stack: 517 518; R600: MOVA_INT 519; R600: MOVA_INT 520 521; SI: buffer_load_dword 522; SI: buffer_load_dword 523 524define amdgpu_kernel void @v2float_stack(<2 x float> addrspace(1)* %out, i32 %a) { 525 %alloca = alloca [16 x <2 x float>] 526 %tmp0 = getelementptr [16 x <2 x float>], [16 x <2 x float>]* %alloca, i32 0, i32 %a 527 %tmp5 = load <2 x float>, <2 x float>* %tmp0 528 store <2 x float> %tmp5, <2 x float> addrspace(1)* %out 529 ret void 530} 531 532; OPT-LABEL: @direct_alloca_read_0xi32( 533; OPT: store [0 x i32] undef, [0 x i32] addrspace(3)* 534; OPT: load [0 x i32], [0 x i32] addrspace(3)* 535define amdgpu_kernel void @direct_alloca_read_0xi32([0 x i32] addrspace(1)* %out, i32 %index) { 536entry: 537 %tmp = alloca [0 x i32] 538 store [0 x i32] [], [0 x i32]* %tmp 539 %load = load [0 x i32], [0 x i32]* %tmp 540 store [0 x i32] %load, [0 x i32] addrspace(1)* %out 541 ret void 542} 543 544; OPT-LABEL: @direct_alloca_read_1xi32( 545; OPT: store [1 x i32] zeroinitializer, [1 x i32] addrspace(3)* 546; OPT: load [1 x i32], [1 x i32] addrspace(3)* 547define amdgpu_kernel void @direct_alloca_read_1xi32([1 x i32] addrspace(1)* %out, i32 %index) { 548entry: 549 %tmp = alloca [1 x i32] 550 store [1 x i32] [i32 0], [1 x i32]* %tmp 551 %load = load [1 x i32], [1 x i32]* %tmp 552 store [1 x i32] %load, [1 x i32] addrspace(1)* %out 553 ret void 554} 555 556attributes #0 = { nounwind "amdgpu-waves-per-eu"="1,2" } 557 558; HSAOPT: !0 = !{} 559; HSAOPT: !1 = !{i32 0, i32 257} 560; HSAOPT: !2 = !{i32 0, i32 256} 561 562; NOHSAOPT: !0 = !{i32 0, i32 257} 563; NOHSAOPT: !1 = !{i32 0, i32 256} 564