67 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefix=GFX942 %s3; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -O0 < %s | FileCheck -check-prefix=GFX90A %s4 5@G = addrspace(1) global <2 x i32> splat (i32 5)6 7define amdgpu_ps void @global_load_lds_dword_saddr(ptr addrspace(1) inreg nocapture %gptr, ptr addrspace(3) nocapture %lptr) {8; GFX942-LABEL: global_load_lds_dword_saddr:9; GFX942: ; %bb.0: ; %main_body10; GFX942-NEXT: v_readfirstlane_b32 s2, v011; GFX942-NEXT: v_mov_b32_e32 v1, 012; GFX942-NEXT: s_mov_b32 m0, s213; GFX942-NEXT: s_nop 014; GFX942-NEXT: global_load_lds_dword v1, s[0:1] offset:32 nt15; GFX942-NEXT: s_getpc_b64 s[0:1]16; GFX942-NEXT: s_add_u32 s0, s0, G@gotpcrel32@lo+417; GFX942-NEXT: s_addc_u32 s1, s1, G@gotpcrel32@hi+1218; GFX942-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x019; GFX942-NEXT: s_waitcnt lgkmcnt(0)20; GFX942-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x021; GFX942-NEXT: s_waitcnt lgkmcnt(0)22; GFX942-NEXT: s_mul_i32 s3, s3, 1023; GFX942-NEXT: s_mul_i32 s2, s2, 1024; GFX942-NEXT: v_mov_b32_e32 v2, s225; GFX942-NEXT: v_mov_b32_e32 v3, s326; GFX942-NEXT: global_store_dwordx2 v1, v[2:3], s[0:1]27; GFX942-NEXT: s_endpgm28;29; GFX90A-LABEL: global_load_lds_dword_saddr:30; GFX90A: ; %bb.0: ; %main_body31; GFX90A-NEXT: v_mov_b32_e32 v1, v032; GFX90A-NEXT: s_mov_b32 s2, s033; GFX90A-NEXT: ; kill: def $sgpr2 killed $sgpr2 def $sgpr2_sgpr334; GFX90A-NEXT: s_mov_b32 s3, s135; GFX90A-NEXT: ; kill: def $sgpr0_sgpr1 killed $sgpr2_sgpr336; GFX90A-NEXT: s_getpc_b64 s[0:1]37; GFX90A-NEXT: s_add_u32 s0, s0, G@gotpcrel32@lo+438; GFX90A-NEXT: s_addc_u32 s1, s1, G@gotpcrel32@hi+1239; GFX90A-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x040; GFX90A-NEXT: s_waitcnt lgkmcnt(0)41; GFX90A-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x042; GFX90A-NEXT: v_mov_b32_e32 v0, 043; GFX90A-NEXT: s_waitcnt lgkmcnt(0)44; GFX90A-NEXT: s_mov_b32 s4, s945; GFX90A-NEXT: s_mov_b32 s6, 1046; GFX90A-NEXT: s_mul_i32 s4, s4, s647; GFX90A-NEXT: s_mov_b32 s5, s848; GFX90A-NEXT: s_mul_i32 s5, s5, s649; GFX90A-NEXT: v_mov_b32_e32 v2, s550; GFX90A-NEXT: v_mov_b32_e32 v4, s451; GFX90A-NEXT: ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec52; GFX90A-NEXT: v_mov_b32_e32 v3, v453; GFX90A-NEXT: v_readfirstlane_b32 s4, v154; GFX90A-NEXT: s_mov_b32 m0, s455; GFX90A-NEXT: s_nop 056; GFX90A-NEXT: global_load_dword v0, s[2:3] offset:32 slc lds57; GFX90A-NEXT: s_waitcnt lgkmcnt(0)58; GFX90A-NEXT: global_store_dwordx2 v0, v[2:3], s[0:1]59; GFX90A-NEXT: s_endpgm60main_body:61 %LGV = load <2 x i32>, ptr addrspace(1) @G, align 862 %B = mul <2 x i32> %LGV, splat (i32 10)63 call void @llvm.amdgcn.global.load.lds(ptr addrspace(1) %gptr, ptr addrspace(3) %lptr, i32 4, i32 32, i32 2)64 store <2 x i32> %B, ptr addrspace(1) @G, align 865 ret void66}67