112 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s --check-prefixes=GCN3; RUN: llc -global-isel -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s --check-prefixes=GCN4 5declare void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) nocapture, i32 %size, i32 %vindex, i32 %voffset, i32 %soffset, i32 %offset, i32 %aux)6 7define amdgpu_ps float @buffer_load_lds_dword(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds) {8; GCN-LABEL: buffer_load_lds_dword:9; GCN: ; %bb.0: ; %main_body10; GCN-NEXT: s_mov_b32 m0, s411; GCN-NEXT: v_mov_b32_e32 v0, 812; GCN-NEXT: buffer_load_dword v0, s[0:3], 0 idxen lds13; GCN-NEXT: buffer_load_dword v0, s[0:3], 0 idxen offset:4 glc lds14; GCN-NEXT: buffer_load_dword v0, s[0:3], 0 idxen offset:8 slc lds15; GCN-NEXT: v_mov_b32_e32 v0, s416; GCN-NEXT: s_waitcnt vmcnt(0)17; GCN-NEXT: ds_read_b32 v0, v018; GCN-NEXT: s_waitcnt lgkmcnt(0)19; GCN-NEXT: ; return to shader part epilog20main_body:21 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 8, i32 0, i32 0, i32 0, i32 0)22 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 8, i32 0, i32 0, i32 4, i32 1)23 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 8, i32 0, i32 0, i32 8, i32 2)24 %res = load float, ptr addrspace(3) %lds25 ret float %res26}27 28define amdgpu_ps void @buffer_load_lds_dword_imm_offset(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex) {29; GCN-LABEL: buffer_load_lds_dword_imm_offset:30; GCN: ; %bb.0: ; %main_body31; GCN-NEXT: s_mov_b32 m0, s432; GCN-NEXT: s_nop 033; GCN-NEXT: buffer_load_dword v0, s[0:3], 0 idxen offset:2048 lds34; GCN-NEXT: s_endpgm35main_body:36 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 %vindex, i32 0, i32 0, i32 2048, i32 0)37 ret void38}39 40define amdgpu_ps void @buffer_load_lds_dword_v_offset(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex, i32 %voffset) {41; GCN-LABEL: buffer_load_lds_dword_v_offset:42; GCN: ; %bb.0: ; %main_body43; GCN-NEXT: s_mov_b32 m0, s444; GCN-NEXT: s_nop 045; GCN-NEXT: buffer_load_dword v[0:1], s[0:3], 0 idxen offen lds46; GCN-NEXT: s_endpgm47main_body:48 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 %vindex, i32 %voffset, i32 0, i32 0, i32 0)49 ret void50}51 52define amdgpu_ps void @buffer_load_lds_dword_s_offset(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex, i32 inreg %soffset) {53; GCN-LABEL: buffer_load_lds_dword_s_offset:54; GCN: ; %bb.0: ; %main_body55; GCN-NEXT: s_mov_b32 m0, s456; GCN-NEXT: s_nop 057; GCN-NEXT: buffer_load_dword v0, s[0:3], s5 idxen lds58; GCN-NEXT: s_endpgm59main_body:60 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 %vindex, i32 0, i32 %soffset, i32 0, i32 0)61 ret void62}63 64define amdgpu_ps void @buffer_load_lds_dword_vs_offset(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex, i32 %voffset, i32 inreg %soffset) {65; GCN-LABEL: buffer_load_lds_dword_vs_offset:66; GCN: ; %bb.0: ; %main_body67; GCN-NEXT: s_mov_b32 m0, s468; GCN-NEXT: s_nop 069; GCN-NEXT: buffer_load_dword v[0:1], s[0:3], s5 idxen offen lds70; GCN-NEXT: s_endpgm71main_body:72 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 %vindex, i32 %voffset, i32 %soffset, i32 0, i32 0)73 ret void74}75 76define amdgpu_ps void @buffer_load_lds_dword_vs_imm_offset(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex, i32 %voffset, i32 inreg %soffset) {77; GCN-LABEL: buffer_load_lds_dword_vs_imm_offset:78; GCN: ; %bb.0: ; %main_body79; GCN-NEXT: s_mov_b32 m0, s480; GCN-NEXT: s_nop 081; GCN-NEXT: buffer_load_dword v[0:1], s[0:3], s5 idxen offen offset:2048 lds82; GCN-NEXT: s_endpgm83main_body:84 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 4, i32 %vindex, i32 %voffset, i32 %soffset, i32 2048, i32 0)85 ret void86}87 88define amdgpu_ps void @buffer_load_lds_ushort(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex) {89; GCN-LABEL: buffer_load_lds_ushort:90; GCN: ; %bb.0: ; %main_body91; GCN-NEXT: v_mov_b32_e32 v1, 0x80092; GCN-NEXT: s_mov_b32 m0, s493; GCN-NEXT: s_nop 094; GCN-NEXT: buffer_load_ushort v[0:1], s[0:3], 0 idxen offen lds95; GCN-NEXT: s_endpgm96main_body:97 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 2, i32 %vindex, i32 2048, i32 0, i32 0, i32 0)98 ret void99}100 101define amdgpu_ps void @buffer_load_lds_ubyte(ptr addrspace(8) inreg %rsrc, ptr addrspace(3) inreg %lds, i32 %vindex) {102; GCN-LABEL: buffer_load_lds_ubyte:103; GCN: ; %bb.0: ; %main_body104; GCN-NEXT: s_mov_b32 m0, s4105; GCN-NEXT: s_nop 0106; GCN-NEXT: buffer_load_ubyte v0, s[0:3], 0 idxen offset:2048 lds107; GCN-NEXT: s_endpgm108main_body:109 call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) %rsrc, ptr addrspace(3) %lds, i32 1, i32 %vindex, i32 0, i32 0, i32 2048, i32 0)110 ret void111}112