Page MenuHomePhabricator
Differential D61494 Diff 206420 llvm/trunk/test/CodeGen/AMDGPU/local-memory.ll

Changeset View

Standalone View

View Options

llvm/trunk/test/CodeGen/AMDGPU/local-memory.ll

1; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s 1​; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s 2​; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=FUNC %s
3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 3​; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
4 4​
5@local_memory.local_mem = internal unnamed_addr addrspace(3) global [128 x i32] undef, align 4 5​@local_memory.local_mem = internal unnamed_addr addrspace(3) global [128 x i32] undef, align 4
6 6​
7@lds = addrspace(3) global [512 x i32] undef, align 4 7​@lds = addrspace(3) global [512 x i32] undef, align 4
8 8​
9; On SI we need to make sure that the base offset is a register and 9​; On SI we need to make sure that the base offset is a register and
10; not an immediate. 10​; not an immediate.
11 11​
12; FUNC-LABEL: {{^}}load_i32_local_const_ptr: 12​; FUNC-LABEL: {{^}}load_i32_local_const_ptr:
13; GCN: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0 13​; GCN: v_mov_b32_e32 v[[PTR:[0-9]+]], lds@abs32@lo
14; GCN: ds_read_b32 v{{[0-9]+}}, v[[ZERO]] offset:4 14​; GCN: ds_read_b32 v{{[0-9]+}}, v[[PTR]] offset:4
15 15​
16; R600: LDS_READ_RET 16​; R600: LDS_READ_RET
17define amdgpu_kernel void @load_i32_local_const_ptr(i32 addrspace(1)* %out, i32 addrspace(3)* %in) #0 { 17​define amdgpu_kernel void @load_i32_local_const_ptr(i32 addrspace(1)* %out, i32 addrspace(3)* %in) #0 {
18entry: 18​entry:
19 %tmp0 = getelementptr [512 x i32], [512 x i32] addrspace(3)* @lds, i32 0, i32 1 19​ %tmp0 = getelementptr [512 x i32], [512 x i32] addrspace(3)* @lds, i32 0, i32 1
20 %tmp1 = load i32, i32 addrspace(3)* %tmp0 20​ %tmp1 = load i32, i32 addrspace(3)* %tmp0
21 %tmp2 = getelementptr i32, i32 addrspace(1)* %out, i32 1 21​ %tmp2 = getelementptr i32, i32 addrspace(1)* %out, i32 1
22 store i32 %tmp1, i32 addrspace(1)* %tmp2 22​ store i32 %tmp1, i32 addrspace(1)* %tmp2
Show All 22 Lines