73 lines · plain
1; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck --check-prefix=HSA %s2; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck --check-prefix=HSA-CI %s3; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo | FileCheck --check-prefix=HSA %s4; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo | FileCheck --check-prefix=HSA-VI %s5; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -filetype=obj | llvm-readobj --symbols -S --sd - | FileCheck --check-prefix=ELF %s6; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | llvm-mc -filetype=obj -triple amdgcn--amdhsa --amdhsa-code-object-version=4 -mcpu=kaveri | llvm-readobj --symbols -S --sd - | FileCheck %s --check-prefix=ELF7 8; The SHT_NOTE section contains the output from the .hsa_code_object_*9; directives.10 11; ELF: Section {12; ELF: Name: .text13; ELF: Type: SHT_PROGBITS (0x1)14; ELF: Flags [ (0x6)15; ELF: SHF_ALLOC (0x2)16; ELF: SHF_EXECINSTR (0x4)17; ELF: AddressAlignment: 418; ELF: }19 20; ELF: SHT_NOTE21; ELF: 0000: 07000000 4F000000 20000000 414D444722; ELF: 0010: 50550000 83AE616D 64687361 2E6B657223; ELF: 0020: 6E656C73 90AD616D 64687361 2E74617224; ELF: 0030: 676574BD 616D6467 636E2D75 6E6B6E6F25; ELF: 0040: 776E2D61 6D646873 612D2D67 6678373026; ELF: 0050: 30AE616D 64687361 2E766572 73696F6E27; ELF: 0060: 9201010028 29; ELF: Symbol {30; ELF: Name: simple31; ELF: Size: 3632; ELF: Type: Function (0x2)33; ELF: }34 35; HSA: .text36; HSA-CI: .amdgcn_target "amdgcn-unknown-amdhsa--gfx700" 37; HSA-VI: .amdgcn_target "amdgcn-unknown-amdhsa--gfx801" 38 39; HSA-NOT: .amdgpu_hsa_kernel simple40; HSA: .globl simple41; HSA: .p2align 242; HSA: {{^}}simple:43; HSA-NOT: amd_kernel_code_t44; HSA: flat_load_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v[0:1]45 46; Make sure we are setting the ATC bit:47; Make sure we generate flat store for HSA48; HSA: flat_store_dword v{{\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}}49 50; HSA: .Lfunc_end0:51; HSA: .size simple, .Lfunc_end0-simple52; HSA: ; Function info:53; HSA-NOT: COMPUTE_PGM_RSRC254define void @simple(ptr addrspace(4) %ptr.out) {55entry:56 %out = load ptr addrspace(1), ptr addrspace(4) %ptr.out57 store i32 0, ptr addrspace(1) %out58 ret void59}60 61; Ignore explicit alignment that is too low.62; HSA: .globl simple_align263; HSA: .p2align 264define void @simple_align2(ptr addrspace(4) %ptr.out) align 2 {65entry:66 %out = load ptr addrspace(1), ptr addrspace(4) %ptr.out67 store i32 0, ptr addrspace(1) %out68 ret void69}70 71!llvm.module.flags = !{!0}72!0 = !{i32 1, !"amdhsa_code_object_version", i32 400}73