328 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2;RUN: llc < %s -mtriple=amdgcn -mcpu=verde | FileCheck -check-prefix=PREGFX10 %s3;RUN: llc < %s -mtriple=amdgcn -mcpu=tonga | FileCheck -check-prefix=PREGFX10 %s4;RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1010 | FileCheck -check-prefix=GFX10 %s5;RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1100 | FileCheck -check-prefix=GFX11 %s6;RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1200 | FileCheck -check-prefix=GFX12 %s7;RUN: llc < %s -global-isel -mtriple=amdgcn -mcpu=gfx1200 | FileCheck -check-prefix=GFX12 %s8 9define amdgpu_ps void @tbuffer_store(<4 x i32> inreg, <4 x float>, <4 x float>, <4 x float>) {10; PREGFX10-LABEL: tbuffer_store:11; PREGFX10: ; %bb.0: ; %main_body12; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_DATA_FORMAT_16_16_16_16,BUF_NUM_FORMAT_USCALED]13; PREGFX10-NEXT: tbuffer_store_format_xyzw v[4:7], off, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32,BUF_NUM_FORMAT_SSCALED] glc14; PREGFX10-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] slc15; PREGFX10-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] glc16; PREGFX10-NEXT: s_endpgm17;18; GFX10-LABEL: tbuffer_store:19; GFX10: ; %bb.0: ; %main_body20; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_FMT_10_10_10_2_UNORM]21; GFX10-NEXT: tbuffer_store_format_xyzw v[4:7], off, s[0:3], 0 format:[BUF_FMT_8_8_8_8_SINT] glc22; GFX10-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], 0 format:78 slc23; GFX10-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], 0 format:78 glc dlc24; GFX10-NEXT: s_endpgm25;26; GFX11-LABEL: tbuffer_store:27; GFX11: ; %bb.0: ; %main_body28; GFX11-NEXT: s_clause 0x329; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_FMT_8_8_8_8_USCALED]30; GFX11-NEXT: tbuffer_store_format_xyzw v[4:7], off, s[0:3], 0 format:[BUF_FMT_32_32_32_32_UINT] glc31; GFX11-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], 0 format:78 slc32; GFX11-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], 0 format:78 glc dlc33; GFX11-NEXT: s_endpgm34;35; GFX12-LABEL: tbuffer_store:36; GFX12: ; %bb.0: ; %main_body37; GFX12-NEXT: s_clause 0x338; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], null format:[BUF_FMT_8_8_8_8_USCALED]39; GFX12-NEXT: tbuffer_store_format_xyzw v[4:7], off, s[0:3], null format:[BUF_FMT_32_32_32_32_UINT] th:TH_STORE_NT40; GFX12-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], null format:78 th:TH_STORE_HT41; GFX12-NEXT: tbuffer_store_format_xyzw v[8:11], off, s[0:3], null format:78 th:TH_STORE_RT_NT42; GFX12-NEXT: s_endpgm43main_body:44 %in1 = bitcast <4 x float> %1 to <4 x i32>45 %in2 = bitcast <4 x float> %2 to <4 x i32>46 %in3 = bitcast <4 x float> %3 to <4 x i32>47 call void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32> %in1, <4 x i32> %0, i32 0, i32 0, i32 44, i32 0)48 call void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32> %in2, <4 x i32> %0, i32 0, i32 0, i32 61, i32 1)49 call void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32> %in3, <4 x i32> %0, i32 0, i32 0, i32 78, i32 2)50 call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %3, <4 x i32> %0, i32 0, i32 0, i32 78, i32 5)51 ret void52}53 54define amdgpu_ps void @tbuffer_store_immoffs(<4 x i32> inreg, <4 x float>) {55; PREGFX10-LABEL: tbuffer_store_immoffs:56; PREGFX10: ; %bb.0: ; %main_body57; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_DATA_FORMAT_16_16,BUF_NUM_FORMAT_FLOAT] offset:4258; PREGFX10-NEXT: s_endpgm59;60; GFX10-LABEL: tbuffer_store_immoffs:61; GFX10: ; %bb.0: ; %main_body62; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:117 offset:4263; GFX10-NEXT: s_endpgm64;65; GFX11-LABEL: tbuffer_store_immoffs:66; GFX11: ; %bb.0: ; %main_body67; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:117 offset:4268; GFX11-NEXT: s_endpgm69;70; GFX12-LABEL: tbuffer_store_immoffs:71; GFX12: ; %bb.0: ; %main_body72; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], null format:117 offset:4273; GFX12-NEXT: s_endpgm74main_body:75 %in1 = bitcast <4 x float> %1 to <4 x i32>76 call void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32> %in1, <4 x i32> %0, i32 42, i32 0, i32 117, i32 0)77 ret void78}79 80define amdgpu_ps void @tbuffer_store_scalar_and_imm_offs(<4 x i32> inreg, <4 x float> %vdata, i32 inreg %soffset) {81; PREGFX10-LABEL: tbuffer_store_scalar_and_imm_offs:82; PREGFX10: ; %bb.0: ; %main_body83; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], s4 format:[BUF_DATA_FORMAT_16_16,BUF_NUM_FORMAT_FLOAT] offset:4284; PREGFX10-NEXT: s_endpgm85;86; GFX10-LABEL: tbuffer_store_scalar_and_imm_offs:87; GFX10: ; %bb.0: ; %main_body88; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], s4 format:117 offset:4289; GFX10-NEXT: s_endpgm90;91; GFX11-LABEL: tbuffer_store_scalar_and_imm_offs:92; GFX11: ; %bb.0: ; %main_body93; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], s4 format:117 offset:4294; GFX11-NEXT: s_endpgm95;96; GFX12-LABEL: tbuffer_store_scalar_and_imm_offs:97; GFX12: ; %bb.0: ; %main_body98; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], s4 format:117 offset:4299; GFX12-NEXT: s_endpgm100main_body:101 %in1 = bitcast <4 x float> %vdata to <4 x i32>102 call void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32> %in1, <4 x i32> %0, i32 42, i32 %soffset, i32 117, i32 0)103 ret void104}105 106define amdgpu_ps void @buffer_store_ofs(<4 x i32> inreg, <4 x float> %vdata, i32 %voffset) {107; PREGFX10-LABEL: buffer_store_ofs:108; PREGFX10: ; %bb.0: ; %main_body109; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_DATA_FORMAT_8_8,BUF_NUM_FORMAT_FLOAT] offen110; PREGFX10-NEXT: s_endpgm111;112; GFX10-LABEL: buffer_store_ofs:113; GFX10: ; %bb.0: ; %main_body114; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:115 offen115; GFX10-NEXT: s_endpgm116;117; GFX11-LABEL: buffer_store_ofs:118; GFX11: ; %bb.0: ; %main_body119; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:115 offen120; GFX11-NEXT: s_endpgm121;122; GFX12-LABEL: buffer_store_ofs:123; GFX12: ; %bb.0: ; %main_body124; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], null format:115 offen125; GFX12-NEXT: s_endpgm126main_body:127 %in1 = bitcast <4 x float> %vdata to <4 x i32>128 call void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32> %in1, <4 x i32> %0, i32 %voffset, i32 0, i32 115, i32 0)129 ret void130}131 132define amdgpu_ps void @buffer_store_x1(<4 x i32> inreg %rsrc, float %data) {133; PREGFX10-LABEL: buffer_store_x1:134; PREGFX10: ; %bb.0: ; %main_body135; PREGFX10-NEXT: tbuffer_store_format_x v0, off, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32,BUF_NUM_FORMAT_FLOAT]136; PREGFX10-NEXT: s_endpgm137;138; GFX10-LABEL: buffer_store_x1:139; GFX10: ; %bb.0: ; %main_body140; GFX10-NEXT: tbuffer_store_format_x v0, off, s[0:3], 0 format:125141; GFX10-NEXT: s_endpgm142;143; GFX11-LABEL: buffer_store_x1:144; GFX11: ; %bb.0: ; %main_body145; GFX11-NEXT: tbuffer_store_format_x v0, off, s[0:3], 0 format:125146; GFX11-NEXT: s_endpgm147;148; GFX12-LABEL: buffer_store_x1:149; GFX12: ; %bb.0: ; %main_body150; GFX12-NEXT: tbuffer_store_format_x v0, off, s[0:3], null format:125151; GFX12-NEXT: s_endpgm152main_body:153 %data.i = bitcast float %data to i32154 call void @llvm.amdgcn.raw.tbuffer.store.i32(i32 %data.i, <4 x i32> %rsrc, i32 0, i32 0, i32 125, i32 0)155 ret void156}157 158define amdgpu_ps void @buffer_store_x2(<4 x i32> inreg %rsrc, <2 x float> %data) {159; PREGFX10-LABEL: buffer_store_x2:160; PREGFX10: ; %bb.0: ; %main_body161; PREGFX10-NEXT: tbuffer_store_format_xy v[0:1], off, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED]162; PREGFX10-NEXT: s_endpgm163;164; GFX10-LABEL: buffer_store_x2:165; GFX10: ; %bb.0: ; %main_body166; GFX10-NEXT: tbuffer_store_format_xy v[0:1], off, s[0:3], 0 format:[BUF_FMT_10_11_11_SSCALED]167; GFX10-NEXT: s_endpgm168;169; GFX11-LABEL: buffer_store_x2:170; GFX11: ; %bb.0: ; %main_body171; GFX11-NEXT: tbuffer_store_format_xy v[0:1], off, s[0:3], 0 format:[BUF_FMT_10_10_10_2_SNORM]172; GFX11-NEXT: s_endpgm173;174; GFX12-LABEL: buffer_store_x2:175; GFX12: ; %bb.0: ; %main_body176; GFX12-NEXT: tbuffer_store_format_xy v[0:1], off, s[0:3], null format:[BUF_FMT_10_10_10_2_SNORM]177; GFX12-NEXT: s_endpgm178main_body:179 %data.i = bitcast <2 x float> %data to <2 x i32>180 call void @llvm.amdgcn.raw.tbuffer.store.v2i32(<2 x i32> %data.i, <4 x i32> %rsrc, i32 0, i32 0, i32 33, i32 0)181 ret void182}183 184define amdgpu_ps void @buffer_store_voffset_large_12bit(<4 x i32> inreg %rsrc, <4 x float> %data) {185; PREGFX10-LABEL: buffer_store_voffset_large_12bit:186; PREGFX10: ; %bb.0: ; %main_body187; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] offset:4092188; PREGFX10-NEXT: s_endpgm189;190; GFX10-LABEL: buffer_store_voffset_large_12bit:191; GFX10: ; %bb.0: ; %main_body192; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_FMT_32_32_SINT] offset:4092193; GFX10-NEXT: s_endpgm194;195; GFX11-LABEL: buffer_store_voffset_large_12bit:196; GFX11: ; %bb.0: ; %main_body197; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] offset:4092198; GFX11-NEXT: s_endpgm199;200; GFX12-LABEL: buffer_store_voffset_large_12bit:201; GFX12: ; %bb.0: ; %main_body202; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], null format:[BUF_FMT_32_32_32_32_FLOAT] offset:4092203; GFX12-NEXT: s_endpgm204main_body:205 call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %data, <4 x i32> %rsrc, i32 4092, i32 0, i32 63, i32 0)206 ret void207}208 209define amdgpu_ps void @buffer_store_voffset_large_13bit(<4 x i32> inreg %rsrc, <4 x float> %data) {210; PREGFX10-LABEL: buffer_store_voffset_large_13bit:211; PREGFX10: ; %bb.0: ; %main_body212; PREGFX10-NEXT: v_mov_b32_e32 v4, 0x1000213; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] offen offset:4092214; PREGFX10-NEXT: s_endpgm215;216; GFX10-LABEL: buffer_store_voffset_large_13bit:217; GFX10: ; %bb.0: ; %main_body218; GFX10-NEXT: v_mov_b32_e32 v4, 0x1000219; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_SINT] offen offset:4092220; GFX10-NEXT: s_endpgm221;222; GFX11-LABEL: buffer_store_voffset_large_13bit:223; GFX11: ; %bb.0: ; %main_body224; GFX11-NEXT: v_mov_b32_e32 v4, 0x1000225; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] offen offset:4092226; GFX11-NEXT: s_endpgm227;228; GFX12-LABEL: buffer_store_voffset_large_13bit:229; GFX12: ; %bb.0: ; %main_body230; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], null format:[BUF_FMT_32_32_32_32_FLOAT] offset:8188231; GFX12-NEXT: s_endpgm232main_body:233 call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %data, <4 x i32> %rsrc, i32 8188, i32 0, i32 63, i32 0)234 ret void235}236 237define amdgpu_ps void @buffer_store_voffset_large_16bit(<4 x i32> inreg %rsrc, <4 x float> %data) {238; PREGFX10-LABEL: buffer_store_voffset_large_16bit:239; PREGFX10: ; %bb.0: ; %main_body240; PREGFX10-NEXT: v_mov_b32_e32 v4, 0xf000241; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] offen offset:4092242; PREGFX10-NEXT: s_endpgm243;244; GFX10-LABEL: buffer_store_voffset_large_16bit:245; GFX10: ; %bb.0: ; %main_body246; GFX10-NEXT: v_mov_b32_e32 v4, 0xf000247; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_SINT] offen offset:4092248; GFX10-NEXT: s_endpgm249;250; GFX11-LABEL: buffer_store_voffset_large_16bit:251; GFX11: ; %bb.0: ; %main_body252; GFX11-NEXT: v_mov_b32_e32 v4, 0xf000253; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] offen offset:4092254; GFX11-NEXT: s_endpgm255;256; GFX12-LABEL: buffer_store_voffset_large_16bit:257; GFX12: ; %bb.0: ; %main_body258; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], null format:[BUF_FMT_32_32_32_32_FLOAT] offset:65532259; GFX12-NEXT: s_endpgm260main_body:261 call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %data, <4 x i32> %rsrc, i32 65532, i32 0, i32 63, i32 0)262 ret void263}264 265define amdgpu_ps void @buffer_store_voffset_large_23bit(<4 x i32> inreg %rsrc, <4 x float> %data) {266; PREGFX10-LABEL: buffer_store_voffset_large_23bit:267; PREGFX10: ; %bb.0: ; %main_body268; PREGFX10-NEXT: v_mov_b32_e32 v4, 0x7ff000269; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] offen offset:4092270; PREGFX10-NEXT: s_endpgm271;272; GFX10-LABEL: buffer_store_voffset_large_23bit:273; GFX10: ; %bb.0: ; %main_body274; GFX10-NEXT: v_mov_b32_e32 v4, 0x7ff000275; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_SINT] offen offset:4092276; GFX10-NEXT: s_endpgm277;278; GFX11-LABEL: buffer_store_voffset_large_23bit:279; GFX11: ; %bb.0: ; %main_body280; GFX11-NEXT: v_mov_b32_e32 v4, 0x7ff000281; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] offen offset:4092282; GFX11-NEXT: s_endpgm283;284; GFX12-LABEL: buffer_store_voffset_large_23bit:285; GFX12: ; %bb.0: ; %main_body286; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], off, s[0:3], null format:[BUF_FMT_32_32_32_32_FLOAT] offset:8388604287; GFX12-NEXT: s_endpgm288main_body:289 call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %data, <4 x i32> %rsrc, i32 8388604, i32 0, i32 63, i32 0)290 ret void291}292 293define amdgpu_ps void @buffer_store_voffset_large_24bit(<4 x i32> inreg %rsrc, <4 x float> %data) {294; PREGFX10-LABEL: buffer_store_voffset_large_24bit:295; PREGFX10: ; %bb.0: ; %main_body296; PREGFX10-NEXT: v_mov_b32_e32 v4, 0xfff000297; PREGFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] offen offset:4092298; PREGFX10-NEXT: s_endpgm299;300; GFX10-LABEL: buffer_store_voffset_large_24bit:301; GFX10: ; %bb.0: ; %main_body302; GFX10-NEXT: v_mov_b32_e32 v4, 0xfff000303; GFX10-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_SINT] offen offset:4092304; GFX10-NEXT: s_endpgm305;306; GFX11-LABEL: buffer_store_voffset_large_24bit:307; GFX11: ; %bb.0: ; %main_body308; GFX11-NEXT: v_mov_b32_e32 v4, 0xfff000309; GFX11-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] offen offset:4092310; GFX11-NEXT: s_endpgm311;312; GFX12-LABEL: buffer_store_voffset_large_24bit:313; GFX12: ; %bb.0: ; %main_body314; GFX12-NEXT: v_mov_b32_e32 v4, 0x800000315; GFX12-NEXT: tbuffer_store_format_xyzw v[0:3], v4, s[0:3], null format:[BUF_FMT_32_32_32_32_FLOAT] offen offset:8388604316; GFX12-NEXT: s_endpgm317main_body:318 call void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float> %data, <4 x i32> %rsrc, i32 16777212, i32 0, i32 63, i32 0)319 ret void320}321 322declare void @llvm.amdgcn.raw.tbuffer.store.i32(i32, <4 x i32>, i32, i32, i32, i32) #0323declare void @llvm.amdgcn.raw.tbuffer.store.v2i32(<2 x i32>, <4 x i32>, i32, i32, i32, i32) #0324declare void @llvm.amdgcn.raw.tbuffer.store.v4i32(<4 x i32>, <4 x i32>, i32, i32, i32, i32) #0325declare void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32, i32) #0326attributes #0 = { nounwind }327attributes #1 = { nounwind readonly }328