279 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefix=GFX6 %s3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s4; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s5; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s6 7define amdgpu_ps <4 x float> @load_3d_v4f32_xyzw(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %r) {8; GFX6-LABEL: load_3d_v4f32_xyzw:9; GFX6: ; %bb.0:10; GFX6-NEXT: s_mov_b32 s0, s211; GFX6-NEXT: s_mov_b32 s1, s312; GFX6-NEXT: s_mov_b32 s2, s413; GFX6-NEXT: s_mov_b32 s3, s514; GFX6-NEXT: s_mov_b32 s4, s615; GFX6-NEXT: s_mov_b32 s5, s716; GFX6-NEXT: s_mov_b32 s6, s817; GFX6-NEXT: s_mov_b32 s7, s918; GFX6-NEXT: image_load v[0:3], v[0:2], s[0:7] dmask:0xf unorm19; GFX6-NEXT: s_waitcnt vmcnt(0)20; GFX6-NEXT: ; return to shader part epilog21;22; GFX10PLUS-LABEL: load_3d_v4f32_xyzw:23; GFX10PLUS: ; %bb.0:24; GFX10PLUS-NEXT: s_mov_b32 s0, s225; GFX10PLUS-NEXT: s_mov_b32 s1, s326; GFX10PLUS-NEXT: s_mov_b32 s2, s427; GFX10PLUS-NEXT: s_mov_b32 s3, s528; GFX10PLUS-NEXT: s_mov_b32 s4, s629; GFX10PLUS-NEXT: s_mov_b32 s5, s730; GFX10PLUS-NEXT: s_mov_b32 s6, s831; GFX10PLUS-NEXT: s_mov_b32 s7, s932; GFX10PLUS-NEXT: image_load v[0:3], v[0:2], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm33; GFX10PLUS-NEXT: s_waitcnt vmcnt(0)34; GFX10PLUS-NEXT: ; return to shader part epilog35;36; GFX12-LABEL: load_3d_v4f32_xyzw:37; GFX12: ; %bb.0:38; GFX12-NEXT: s_mov_b32 s0, s239; GFX12-NEXT: s_mov_b32 s1, s340; GFX12-NEXT: s_mov_b32 s2, s441; GFX12-NEXT: s_mov_b32 s3, s542; GFX12-NEXT: s_mov_b32 s4, s643; GFX12-NEXT: s_mov_b32 s5, s744; GFX12-NEXT: s_mov_b32 s6, s845; GFX12-NEXT: s_mov_b32 s7, s946; GFX12-NEXT: image_load v[0:3], [v0, v1, v2], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D47; GFX12-NEXT: s_wait_loadcnt 0x048; GFX12-NEXT: ; return to shader part epilog49 %v = call <4 x float> @llvm.amdgcn.image.load.3d.v4f32.i32(i32 15, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 0, i32 0)50 ret <4 x float> %v51}52 53define amdgpu_ps <4 x float> @load_3d_v4f32_xyzw_tfe(<8 x i32> inreg %rsrc, ptr addrspace(1) inreg %out, i32 %s, i32 %t, i32 %r) {54; GFX6-LABEL: load_3d_v4f32_xyzw_tfe:55; GFX6: ; %bb.0:56; GFX6-NEXT: v_mov_b32_e32 v5, v057; GFX6-NEXT: v_mov_b32_e32 v0, 058; GFX6-NEXT: s_mov_b32 s0, s259; GFX6-NEXT: s_mov_b32 s1, s360; GFX6-NEXT: s_mov_b32 s2, s461; GFX6-NEXT: s_mov_b32 s3, s562; GFX6-NEXT: s_mov_b32 s4, s663; GFX6-NEXT: s_mov_b32 s5, s764; GFX6-NEXT: s_mov_b32 s6, s865; GFX6-NEXT: s_mov_b32 s7, s966; GFX6-NEXT: v_mov_b32_e32 v6, v167; GFX6-NEXT: v_mov_b32_e32 v7, v268; GFX6-NEXT: v_mov_b32_e32 v1, v069; GFX6-NEXT: v_mov_b32_e32 v2, v070; GFX6-NEXT: v_mov_b32_e32 v3, v071; GFX6-NEXT: v_mov_b32_e32 v4, v072; GFX6-NEXT: image_load v[0:4], v[5:7], s[0:7] dmask:0xf unorm tfe73; GFX6-NEXT: s_mov_b32 s8, s1074; GFX6-NEXT: s_mov_b32 s9, s1175; GFX6-NEXT: s_mov_b32 s10, -176; GFX6-NEXT: s_mov_b32 s11, 0xf00077; GFX6-NEXT: s_waitcnt vmcnt(0)78; GFX6-NEXT: buffer_store_dword v4, off, s[8:11], 079; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)80; GFX6-NEXT: ; return to shader part epilog81;82; GFX10-LABEL: load_3d_v4f32_xyzw_tfe:83; GFX10: ; %bb.0:84; GFX10-NEXT: v_mov_b32_e32 v8, 085; GFX10-NEXT: v_mov_b32_e32 v5, v086; GFX10-NEXT: v_mov_b32_e32 v6, v187; GFX10-NEXT: v_mov_b32_e32 v7, v288; GFX10-NEXT: s_mov_b32 s0, s289; GFX10-NEXT: v_mov_b32_e32 v9, v890; GFX10-NEXT: v_mov_b32_e32 v10, v891; GFX10-NEXT: v_mov_b32_e32 v11, v892; GFX10-NEXT: v_mov_b32_e32 v12, v893; GFX10-NEXT: s_mov_b32 s1, s394; GFX10-NEXT: s_mov_b32 s2, s495; GFX10-NEXT: s_mov_b32 s3, s596; GFX10-NEXT: s_mov_b32 s4, s697; GFX10-NEXT: s_mov_b32 s5, s798; GFX10-NEXT: s_mov_b32 s6, s899; GFX10-NEXT: s_mov_b32 s7, s9100; GFX10-NEXT: v_mov_b32_e32 v0, v8101; GFX10-NEXT: v_mov_b32_e32 v1, v9102; GFX10-NEXT: v_mov_b32_e32 v2, v10103; GFX10-NEXT: v_mov_b32_e32 v3, v11104; GFX10-NEXT: v_mov_b32_e32 v4, v12105; GFX10-NEXT: image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe106; GFX10-NEXT: s_waitcnt vmcnt(0)107; GFX10-NEXT: global_store_dword v8, v4, s[10:11]108; GFX10-NEXT: ; return to shader part epilog109;110; GFX11-LABEL: load_3d_v4f32_xyzw_tfe:111; GFX11: ; %bb.0:112; GFX11-NEXT: v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0113; GFX11-NEXT: v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1114; GFX11-NEXT: s_mov_b32 s0, s2115; GFX11-NEXT: s_mov_b32 s1, s3116; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)117; GFX11-NEXT: v_mov_b32_e32 v9, v8118; GFX11-NEXT: v_mov_b32_e32 v10, v8119; GFX11-NEXT: v_mov_b32_e32 v11, v8120; GFX11-NEXT: v_mov_b32_e32 v12, v8121; GFX11-NEXT: s_mov_b32 s2, s4122; GFX11-NEXT: s_mov_b32 s3, s5123; GFX11-NEXT: s_mov_b32 s4, s6124; GFX11-NEXT: s_mov_b32 s5, s7125; GFX11-NEXT: s_mov_b32 s6, s8126; GFX11-NEXT: s_mov_b32 s7, s9127; GFX11-NEXT: v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9128; GFX11-NEXT: v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11129; GFX11-NEXT: v_mov_b32_e32 v4, v12130; GFX11-NEXT: image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe131; GFX11-NEXT: s_waitcnt vmcnt(0)132; GFX11-NEXT: global_store_b32 v8, v4, s[10:11]133; GFX11-NEXT: ; return to shader part epilog134;135; GFX12-LABEL: load_3d_v4f32_xyzw_tfe:136; GFX12: ; %bb.0:137; GFX12-NEXT: v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0138; GFX12-NEXT: v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1139; GFX12-NEXT: s_mov_b32 s0, s2140; GFX12-NEXT: s_mov_b32 s1, s3141; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)142; GFX12-NEXT: v_dual_mov_b32 v9, v8 :: v_dual_mov_b32 v10, v8143; GFX12-NEXT: v_dual_mov_b32 v11, v8 :: v_dual_mov_b32 v12, v8144; GFX12-NEXT: s_mov_b32 s2, s4145; GFX12-NEXT: s_mov_b32 s3, s5146; GFX12-NEXT: s_mov_b32 s4, s6147; GFX12-NEXT: s_mov_b32 s5, s7148; GFX12-NEXT: s_mov_b32 s6, s8149; GFX12-NEXT: s_mov_b32 s7, s9150; GFX12-NEXT: v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9151; GFX12-NEXT: v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11152; GFX12-NEXT: v_mov_b32_e32 v4, v12153; GFX12-NEXT: image_load v[0:4], [v5, v6, v7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D tfe154; GFX12-NEXT: s_wait_loadcnt 0x0155; GFX12-NEXT: global_store_b32 v8, v4, s[10:11]156; GFX12-NEXT: ; return to shader part epilog157 %v = call { <4 x float>, i32 } @llvm.amdgcn.image.load.3d.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 1, i32 0)158 %v.vec = extractvalue { <4 x float>, i32 } %v, 0159 %v.err = extractvalue { <4 x float>, i32 } %v, 1160 store i32 %v.err, ptr addrspace(1) %out, align 4161 ret <4 x float> %v.vec162}163 164define amdgpu_ps <4 x float> @load_3d_v4f32_xyzw_tfe_lwe(<8 x i32> inreg %rsrc, ptr addrspace(1) inreg %out, i32 %s, i32 %t, i32 %r) {165; GFX6-LABEL: load_3d_v4f32_xyzw_tfe_lwe:166; GFX6: ; %bb.0:167; GFX6-NEXT: v_mov_b32_e32 v5, v0168; GFX6-NEXT: v_mov_b32_e32 v0, 0169; GFX6-NEXT: s_mov_b32 s0, s2170; GFX6-NEXT: s_mov_b32 s1, s3171; GFX6-NEXT: s_mov_b32 s2, s4172; GFX6-NEXT: s_mov_b32 s3, s5173; GFX6-NEXT: s_mov_b32 s4, s6174; GFX6-NEXT: s_mov_b32 s5, s7175; GFX6-NEXT: s_mov_b32 s6, s8176; GFX6-NEXT: s_mov_b32 s7, s9177; GFX6-NEXT: v_mov_b32_e32 v6, v1178; GFX6-NEXT: v_mov_b32_e32 v7, v2179; GFX6-NEXT: v_mov_b32_e32 v1, v0180; GFX6-NEXT: v_mov_b32_e32 v2, v0181; GFX6-NEXT: v_mov_b32_e32 v3, v0182; GFX6-NEXT: v_mov_b32_e32 v4, v0183; GFX6-NEXT: image_load v[0:4], v[5:7], s[0:7] dmask:0xf unorm tfe lwe184; GFX6-NEXT: s_mov_b32 s8, s10185; GFX6-NEXT: s_mov_b32 s9, s11186; GFX6-NEXT: s_mov_b32 s10, -1187; GFX6-NEXT: s_mov_b32 s11, 0xf000188; GFX6-NEXT: s_waitcnt vmcnt(0)189; GFX6-NEXT: buffer_store_dword v4, off, s[8:11], 0190; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)191; GFX6-NEXT: ; return to shader part epilog192;193; GFX10-LABEL: load_3d_v4f32_xyzw_tfe_lwe:194; GFX10: ; %bb.0:195; GFX10-NEXT: v_mov_b32_e32 v8, 0196; GFX10-NEXT: v_mov_b32_e32 v5, v0197; GFX10-NEXT: v_mov_b32_e32 v6, v1198; GFX10-NEXT: v_mov_b32_e32 v7, v2199; GFX10-NEXT: s_mov_b32 s0, s2200; GFX10-NEXT: v_mov_b32_e32 v9, v8201; GFX10-NEXT: v_mov_b32_e32 v10, v8202; GFX10-NEXT: v_mov_b32_e32 v11, v8203; GFX10-NEXT: v_mov_b32_e32 v12, v8204; GFX10-NEXT: s_mov_b32 s1, s3205; GFX10-NEXT: s_mov_b32 s2, s4206; GFX10-NEXT: s_mov_b32 s3, s5207; GFX10-NEXT: s_mov_b32 s4, s6208; GFX10-NEXT: s_mov_b32 s5, s7209; GFX10-NEXT: s_mov_b32 s6, s8210; GFX10-NEXT: s_mov_b32 s7, s9211; GFX10-NEXT: v_mov_b32_e32 v0, v8212; GFX10-NEXT: v_mov_b32_e32 v1, v9213; GFX10-NEXT: v_mov_b32_e32 v2, v10214; GFX10-NEXT: v_mov_b32_e32 v3, v11215; GFX10-NEXT: v_mov_b32_e32 v4, v12216; GFX10-NEXT: image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe lwe217; GFX10-NEXT: s_waitcnt vmcnt(0)218; GFX10-NEXT: global_store_dword v8, v4, s[10:11]219; GFX10-NEXT: ; return to shader part epilog220;221; GFX11-LABEL: load_3d_v4f32_xyzw_tfe_lwe:222; GFX11: ; %bb.0:223; GFX11-NEXT: v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0224; GFX11-NEXT: v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1225; GFX11-NEXT: s_mov_b32 s0, s2226; GFX11-NEXT: s_mov_b32 s1, s3227; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2)228; GFX11-NEXT: v_mov_b32_e32 v9, v8229; GFX11-NEXT: v_mov_b32_e32 v10, v8230; GFX11-NEXT: v_mov_b32_e32 v11, v8231; GFX11-NEXT: v_mov_b32_e32 v12, v8232; GFX11-NEXT: s_mov_b32 s2, s4233; GFX11-NEXT: s_mov_b32 s3, s5234; GFX11-NEXT: s_mov_b32 s4, s6235; GFX11-NEXT: s_mov_b32 s5, s7236; GFX11-NEXT: s_mov_b32 s6, s8237; GFX11-NEXT: s_mov_b32 s7, s9238; GFX11-NEXT: v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9239; GFX11-NEXT: v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11240; GFX11-NEXT: v_mov_b32_e32 v4, v12241; GFX11-NEXT: image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe lwe242; GFX11-NEXT: s_waitcnt vmcnt(0)243; GFX11-NEXT: global_store_b32 v8, v4, s[10:11]244; GFX11-NEXT: ; return to shader part epilog245;246; GFX12-LABEL: load_3d_v4f32_xyzw_tfe_lwe:247; GFX12: ; %bb.0:248; GFX12-NEXT: v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0249; GFX12-NEXT: v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1250; GFX12-NEXT: s_mov_b32 s0, s2251; GFX12-NEXT: s_mov_b32 s1, s3252; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)253; GFX12-NEXT: v_dual_mov_b32 v9, v8 :: v_dual_mov_b32 v10, v8254; GFX12-NEXT: v_dual_mov_b32 v11, v8 :: v_dual_mov_b32 v12, v8255; GFX12-NEXT: s_mov_b32 s2, s4256; GFX12-NEXT: s_mov_b32 s3, s5257; GFX12-NEXT: s_mov_b32 s4, s6258; GFX12-NEXT: s_mov_b32 s5, s7259; GFX12-NEXT: s_mov_b32 s6, s8260; GFX12-NEXT: s_mov_b32 s7, s9261; GFX12-NEXT: v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9262; GFX12-NEXT: v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11263; GFX12-NEXT: v_mov_b32_e32 v4, v12264; GFX12-NEXT: image_load v[0:4], [v5, v6, v7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D tfe265; GFX12-NEXT: s_wait_loadcnt 0x0266; GFX12-NEXT: global_store_b32 v8, v4, s[10:11]267; GFX12-NEXT: ; return to shader part epilog268 %v = call { <4 x float>, i32 } @llvm.amdgcn.image.load.3d.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 3, i32 0)269 %v.vec = extractvalue { <4 x float>, i32 } %v, 0270 %v.err = extractvalue { <4 x float>, i32 } %v, 1271 store i32 %v.err, ptr addrspace(1) %out, align 4272 ret <4 x float> %v.vec273}274 275declare <4 x float> @llvm.amdgcn.image.load.3d.v4f32.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0276declare { <4 x float>, i32 } @llvm.amdgcn.image.load.3d.sl_v4f32i32s.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0277 278attributes #0 = { nounwind readonly }279