54 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -asm-verbose=0 < %s | FileCheck -check-prefixes=ASM %s3; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -filetype=obj < %s | llvm-objdump --arch=amdgcn --mcpu=gfx942 --disassemble - | FileCheck -check-prefixes=OBJ %s4; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -amdgpu-kernarg-preload-count=1 -asm-verbose=0 < %s | llvm-mc -triple amdgcn-amd-amdhsa -mcpu=gfx942 -filetype=obj | llvm-objdump --arch=amdgcn --mcpu=gfx942 --disassemble - | FileCheck -check-prefixes=OBJ %s5 6; OBJ: preload_ptr_kernarg_header7; OBJ-COUNT-60: s_nop 08define amdgpu_kernel void @preload_ptr_kernarg_header(ptr inreg %arg) {9; ASM-LABEL: preload_ptr_kernarg_header:10; ASM: s_load_dwordx2 s[8:9], s[4:5], 0x011; ASM-NEXT: s_waitcnt lgkmcnt(0)12; ASM-NEXT: s_branch .LBB0_013; ASM-NEXT: .p2align 814; ASM-NEXT: .LBB0_0:15; ASM-NEXT: v_mov_b64_e32 v[0:1], s[8:9]16; ASM-NEXT: v_mov_b64_e32 v[2:3], s[8:9]17; ASM-NEXT: flat_store_dwordx2 v[0:1], v[2:3]18; ASM-NEXT: s_endpgm19 store ptr %arg, ptr %arg20 ret void21}22 23; OBJ: preload_i32_kernarg_header24; OBJ-COUNT-58: s_nop 025define amdgpu_kernel void @preload_i32_kernarg_header(ptr inreg %arg, i32 inreg %arg1) {26; ASM-LABEL: preload_i32_kernarg_header:27; ASM: s_load_dwordx2 s[8:9], s[4:5], 0x028; ASM-NEXT: s_load_dword s10, s[4:5], 0x829; ASM-NEXT: s_waitcnt lgkmcnt(0)30; ASM-NEXT: s_branch .LBB1_031; ASM-NEXT: .p2align 832; ASM-NEXT: .LBB1_0:33; ASM-NEXT: v_mov_b64_e32 v[0:1], s[8:9]34; ASM-NEXT: v_mov_b32_e32 v2, s1035; ASM-NEXT: flat_store_dword v[0:1], v236; ASM-NEXT: s_endpgm37 store i32 %arg1, ptr %arg38 ret void39}40 41; OBJ: non_kernel_function42; ASM: non_kernel_function43; OBJ-NOT: s_branch44; ASM-NOT: s_branch45define void @non_kernel_function(ptr %arg) {46; ASM-LABEL: non_kernel_function:47; ASM: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)48; ASM-NEXT: flat_store_dwordx2 v[0:1], v[0:1]49; ASM-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)50; ASM-NEXT: s_setpc_b64 s[30:31]51 store ptr %arg, ptr %arg52 ret void53}54