306 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga < %s | FileCheck -check-prefixes=PREGFX10-UNPACKED %s3; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx810 < %s | FileCheck -check-prefixes=PREGFX10-PACKED %s4; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=PREGFX10-PACKED %s5; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-PACKED %s6; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX11-PACKED,GFX11-PACKED-TRUE16 %s7; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX11-PACKED,GFX11-PACKED-FAKE16 %s8; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX12-PACKED,GFX12-PACKED-SDAG,GFX12-PACKED-SDAG-TRUE16 %s9; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX12-PACKED,GFX12-PACKED-SDAG,GFX12-PACKED-SDAG-FAKE16 %s10; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX12-PACKED,GFX12-PACKED-SDAG,GFX12-PACKED-SDAG-TRUE16 %s11; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX12-PACKED,GFX12-PACKED-SDAG,GFX12-PACKED-SDAG-FAKE16 %s12; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=+real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX12-PACKED,GFX12-PACKED-GISEL,GFX12-PACKED-GISEL-TRUE16 %s13; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -mattr=-real-true16 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefixes=GFX12-PACKED,GFX12-PACKED-GISEL,GFX12-PACKED-GISEL-FAKE16 %s14 15define amdgpu_kernel void @tbuffer_store_d16_x(<4 x i32> %rsrc, half %data, i32 %vindex) {16; PREGFX10-UNPACKED-LABEL: tbuffer_store_d16_x:17; PREGFX10-UNPACKED: ; %bb.0: ; %main_body18; PREGFX10-UNPACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x1019; PREGFX10-UNPACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x020; PREGFX10-UNPACKED-NEXT: s_waitcnt lgkmcnt(0)21; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v0, s422; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v1, s523; PREGFX10-UNPACKED-NEXT: tbuffer_store_format_d16_x v0, v1, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen24; PREGFX10-UNPACKED-NEXT: s_endpgm25;26; PREGFX10-PACKED-LABEL: tbuffer_store_d16_x:27; PREGFX10-PACKED: ; %bb.0: ; %main_body28; PREGFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x1029; PREGFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x030; PREGFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)31; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v0, s432; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v1, s533; PREGFX10-PACKED-NEXT: tbuffer_store_format_d16_x v0, v1, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen34; PREGFX10-PACKED-NEXT: s_endpgm35;36; GFX10-PACKED-LABEL: tbuffer_store_d16_x:37; GFX10-PACKED: ; %bb.0: ; %main_body38; GFX10-PACKED-NEXT: s_clause 0x139; GFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x1040; GFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x041; GFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)42; GFX10-PACKED-NEXT: v_mov_b32_e32 v0, s443; GFX10-PACKED-NEXT: v_mov_b32_e32 v1, s544; GFX10-PACKED-NEXT: tbuffer_store_format_d16_x v0, v1, s[0:3], 0 format:[BUF_FMT_10_11_11_SSCALED] idxen45; GFX10-PACKED-NEXT: s_endpgm46;47; GFX11-PACKED-LABEL: tbuffer_store_d16_x:48; GFX11-PACKED: ; %bb.0: ; %main_body49; GFX11-PACKED-NEXT: s_clause 0x150; GFX11-PACKED-NEXT: s_load_b64 s[6:7], s[4:5], 0x1051; GFX11-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x052; GFX11-PACKED-NEXT: s_waitcnt lgkmcnt(0)53; GFX11-PACKED-NEXT: v_mov_b32_e32 v0, s654; GFX11-PACKED-NEXT: v_mov_b32_e32 v1, s755; GFX11-PACKED-NEXT: tbuffer_store_d16_format_x v0, v1, s[0:3], 0 format:[BUF_FMT_10_10_10_2_SNORM] idxen56; GFX11-PACKED-NEXT: s_endpgm57;58; GFX12-PACKED-LABEL: tbuffer_store_d16_x:59; GFX12-PACKED: ; %bb.0: ; %main_body60; GFX12-PACKED-NEXT: s_clause 0x161; GFX12-PACKED-NEXT: s_load_b64 s[6:7], s[4:5], 0x1062; GFX12-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x063; GFX12-PACKED-NEXT: s_wait_kmcnt 0x064; GFX12-PACKED-NEXT: v_mov_b32_e32 v0, s665; GFX12-PACKED-NEXT: v_mov_b32_e32 v1, s766; GFX12-PACKED-NEXT: tbuffer_store_d16_format_x v0, v1, s[0:3], null format:[BUF_FMT_10_10_10_2_SNORM] idxen67; GFX12-PACKED-NEXT: s_endpgm68main_body:69 call void @llvm.amdgcn.struct.tbuffer.store.f16(half %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 33, i32 0)70 ret void71}72 73define amdgpu_kernel void @tbuffer_store_d16_xy(<4 x i32> %rsrc, <2 x half> %data, i32 %vindex) {74; PREGFX10-UNPACKED-LABEL: tbuffer_store_d16_xy:75; PREGFX10-UNPACKED: ; %bb.0: ; %main_body76; PREGFX10-UNPACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x1077; PREGFX10-UNPACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x078; PREGFX10-UNPACKED-NEXT: s_waitcnt lgkmcnt(0)79; PREGFX10-UNPACKED-NEXT: s_lshr_b32 s6, s4, 1680; PREGFX10-UNPACKED-NEXT: s_and_b32 s4, s4, 0xffff81; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v0, s482; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v1, s683; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v2, s584; PREGFX10-UNPACKED-NEXT: tbuffer_store_format_d16_xy v[0:1], v2, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen85; PREGFX10-UNPACKED-NEXT: s_endpgm86;87; PREGFX10-PACKED-LABEL: tbuffer_store_d16_xy:88; PREGFX10-PACKED: ; %bb.0: ; %main_body89; PREGFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x1090; PREGFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x091; PREGFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)92; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v0, s493; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v1, s594; PREGFX10-PACKED-NEXT: tbuffer_store_format_d16_xy v0, v1, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen95; PREGFX10-PACKED-NEXT: s_endpgm96;97; GFX10-PACKED-LABEL: tbuffer_store_d16_xy:98; GFX10-PACKED: ; %bb.0: ; %main_body99; GFX10-PACKED-NEXT: s_clause 0x1100; GFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10101; GFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0102; GFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)103; GFX10-PACKED-NEXT: v_mov_b32_e32 v0, s4104; GFX10-PACKED-NEXT: v_mov_b32_e32 v1, s5105; GFX10-PACKED-NEXT: tbuffer_store_format_d16_xy v0, v1, s[0:3], 0 format:[BUF_FMT_10_11_11_SSCALED] idxen106; GFX10-PACKED-NEXT: s_endpgm107;108; GFX11-PACKED-LABEL: tbuffer_store_d16_xy:109; GFX11-PACKED: ; %bb.0: ; %main_body110; GFX11-PACKED-NEXT: s_clause 0x1111; GFX11-PACKED-NEXT: s_load_b64 s[6:7], s[4:5], 0x10112; GFX11-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x0113; GFX11-PACKED-NEXT: s_waitcnt lgkmcnt(0)114; GFX11-PACKED-NEXT: v_mov_b32_e32 v0, s6115; GFX11-PACKED-NEXT: v_mov_b32_e32 v1, s7116; GFX11-PACKED-NEXT: tbuffer_store_d16_format_xy v0, v1, s[0:3], 0 format:[BUF_FMT_10_10_10_2_SNORM] idxen117; GFX11-PACKED-NEXT: s_endpgm118;119; GFX12-PACKED-LABEL: tbuffer_store_d16_xy:120; GFX12-PACKED: ; %bb.0: ; %main_body121; GFX12-PACKED-NEXT: s_clause 0x1122; GFX12-PACKED-NEXT: s_load_b64 s[6:7], s[4:5], 0x10123; GFX12-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x0124; GFX12-PACKED-NEXT: s_wait_kmcnt 0x0125; GFX12-PACKED-NEXT: v_mov_b32_e32 v0, s6126; GFX12-PACKED-NEXT: v_mov_b32_e32 v1, s7127; GFX12-PACKED-NEXT: tbuffer_store_d16_format_xy v0, v1, s[0:3], null format:[BUF_FMT_10_10_10_2_SNORM] idxen128; GFX12-PACKED-NEXT: s_endpgm129main_body:130 call void @llvm.amdgcn.struct.tbuffer.store.v2f16(<2 x half> %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 33, i32 0)131 ret void132}133 134define amdgpu_kernel void @tbuffer_store_d16_xyz(<4 x i32> %rsrc, <4 x half> %data, i32 %vindex) {135; PREGFX10-UNPACKED-LABEL: tbuffer_store_d16_xyz:136; PREGFX10-UNPACKED: ; %bb.0: ; %main_body137; PREGFX10-UNPACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10138; PREGFX10-UNPACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0139; PREGFX10-UNPACKED-NEXT: s_load_dword s6, s[8:9], 0x18140; PREGFX10-UNPACKED-NEXT: s_waitcnt lgkmcnt(0)141; PREGFX10-UNPACKED-NEXT: s_and_b32 s5, s5, 0xffff142; PREGFX10-UNPACKED-NEXT: s_lshr_b32 s7, s4, 16143; PREGFX10-UNPACKED-NEXT: s_and_b32 s4, s4, 0xffff144; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v0, s4145; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v1, s7146; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v2, s5147; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v3, s6148; PREGFX10-UNPACKED-NEXT: tbuffer_store_format_d16_xyz v[0:2], v3, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen149; PREGFX10-UNPACKED-NEXT: s_endpgm150;151; PREGFX10-PACKED-LABEL: tbuffer_store_d16_xyz:152; PREGFX10-PACKED: ; %bb.0: ; %main_body153; PREGFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10154; PREGFX10-PACKED-NEXT: s_load_dword s6, s[8:9], 0x18155; PREGFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0156; PREGFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)157; PREGFX10-PACKED-NEXT: s_and_b32 s5, s5, 0xffff158; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v0, s4159; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v1, s5160; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v2, s6161; PREGFX10-PACKED-NEXT: tbuffer_store_format_d16_xyz v[0:1], v2, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen162; PREGFX10-PACKED-NEXT: s_endpgm163;164; GFX10-PACKED-LABEL: tbuffer_store_d16_xyz:165; GFX10-PACKED: ; %bb.0: ; %main_body166; GFX10-PACKED-NEXT: s_clause 0x2167; GFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10168; GFX10-PACKED-NEXT: s_load_dword s6, s[8:9], 0x18169; GFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0170; GFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)171; GFX10-PACKED-NEXT: s_and_b32 s5, s5, 0xffff172; GFX10-PACKED-NEXT: v_mov_b32_e32 v0, s4173; GFX10-PACKED-NEXT: v_mov_b32_e32 v1, s5174; GFX10-PACKED-NEXT: v_mov_b32_e32 v2, s6175; GFX10-PACKED-NEXT: tbuffer_store_format_d16_xyz v[0:1], v2, s[0:3], 0 format:[BUF_FMT_10_11_11_SSCALED] idxen176; GFX10-PACKED-NEXT: s_endpgm177;178; GFX11-PACKED-LABEL: tbuffer_store_d16_xyz:179; GFX11-PACKED: ; %bb.0: ; %main_body180; GFX11-PACKED-NEXT: s_clause 0x2181; GFX11-PACKED-NEXT: s_load_b64 s[6:7], s[4:5], 0x10182; GFX11-PACKED-NEXT: s_load_b32 s8, s[4:5], 0x18183; GFX11-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x0184; GFX11-PACKED-NEXT: s_waitcnt lgkmcnt(0)185; GFX11-PACKED-NEXT: s_and_b32 s4, s7, 0xffff186; GFX11-PACKED-NEXT: v_mov_b32_e32 v0, s6187; GFX11-PACKED-NEXT: v_mov_b32_e32 v1, s4188; GFX11-PACKED-NEXT: v_mov_b32_e32 v2, s8189; GFX11-PACKED-NEXT: tbuffer_store_d16_format_xyz v[0:1], v2, s[0:3], 0 format:[BUF_FMT_10_10_10_2_SNORM] idxen190; GFX11-PACKED-NEXT: s_endpgm191;192; GFX12-PACKED-SDAG-LABEL: tbuffer_store_d16_xyz:193; GFX12-PACKED-SDAG: ; %bb.0: ; %main_body194; GFX12-PACKED-SDAG-NEXT: s_clause 0x1195; GFX12-PACKED-SDAG-NEXT: s_load_b96 s[8:10], s[4:5], 0x10196; GFX12-PACKED-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x0197; GFX12-PACKED-SDAG-NEXT: s_wait_kmcnt 0x0198; GFX12-PACKED-SDAG-NEXT: s_and_b32 s4, s9, 0xffff199; GFX12-PACKED-SDAG-NEXT: v_mov_b32_e32 v0, s8200; GFX12-PACKED-SDAG-NEXT: v_mov_b32_e32 v1, s4201; GFX12-PACKED-SDAG-NEXT: v_mov_b32_e32 v2, s10202; GFX12-PACKED-SDAG-NEXT: tbuffer_store_d16_format_xyz v[0:1], v2, s[0:3], null format:[BUF_FMT_10_10_10_2_SNORM] idxen203; GFX12-PACKED-SDAG-NEXT: s_endpgm204;205; GFX12-PACKED-GISEL-LABEL: tbuffer_store_d16_xyz:206; GFX12-PACKED-GISEL: ; %bb.0: ; %main_body207; GFX12-PACKED-GISEL-NEXT: s_clause 0x1208; GFX12-PACKED-GISEL-NEXT: s_load_b96 s[8:10], s[4:5], 0x10209; GFX12-PACKED-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x0210; GFX12-PACKED-GISEL-NEXT: s_wait_kmcnt 0x0211; GFX12-PACKED-GISEL-NEXT: v_mov_b32_e32 v0, s8212; GFX12-PACKED-GISEL-NEXT: v_mov_b32_e32 v1, s9213; GFX12-PACKED-GISEL-NEXT: v_mov_b32_e32 v2, s10214; GFX12-PACKED-GISEL-NEXT: tbuffer_store_d16_format_xyz v[0:1], v2, s[0:3], null format:[BUF_FMT_10_10_10_2_SNORM] idxen215; GFX12-PACKED-GISEL-NEXT: s_endpgm216main_body:217 %data_subvec = shufflevector <4 x half> %data, <4 x half> poison, <3 x i32> <i32 0, i32 1, i32 2>218 call void @llvm.amdgcn.struct.tbuffer.store.v3f16(<3 x half> %data_subvec, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 33, i32 0)219 ret void220}221 222define amdgpu_kernel void @tbuffer_store_d16_xyzw(<4 x i32> %rsrc, <4 x half> %data, i32 %vindex) {223; PREGFX10-UNPACKED-LABEL: tbuffer_store_d16_xyzw:224; PREGFX10-UNPACKED: ; %bb.0: ; %main_body225; PREGFX10-UNPACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10226; PREGFX10-UNPACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0227; PREGFX10-UNPACKED-NEXT: s_load_dword s6, s[8:9], 0x18228; PREGFX10-UNPACKED-NEXT: s_waitcnt lgkmcnt(0)229; PREGFX10-UNPACKED-NEXT: s_lshr_b32 s7, s5, 16230; PREGFX10-UNPACKED-NEXT: s_and_b32 s5, s5, 0xffff231; PREGFX10-UNPACKED-NEXT: s_lshr_b32 s8, s4, 16232; PREGFX10-UNPACKED-NEXT: s_and_b32 s4, s4, 0xffff233; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v0, s4234; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v1, s8235; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v2, s5236; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v3, s7237; PREGFX10-UNPACKED-NEXT: v_mov_b32_e32 v4, s6238; PREGFX10-UNPACKED-NEXT: tbuffer_store_format_d16_xyzw v[0:3], v4, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen239; PREGFX10-UNPACKED-NEXT: s_endpgm240;241; PREGFX10-PACKED-LABEL: tbuffer_store_d16_xyzw:242; PREGFX10-PACKED: ; %bb.0: ; %main_body243; PREGFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10244; PREGFX10-PACKED-NEXT: s_load_dword s6, s[8:9], 0x18245; PREGFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0246; PREGFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)247; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v0, s4248; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v1, s5249; PREGFX10-PACKED-NEXT: v_mov_b32_e32 v2, s6250; PREGFX10-PACKED-NEXT: tbuffer_store_format_d16_xyzw v[0:1], v2, s[0:3], 0 format:[BUF_NUM_FORMAT_USCALED] idxen251; PREGFX10-PACKED-NEXT: s_endpgm252;253; GFX10-PACKED-LABEL: tbuffer_store_d16_xyzw:254; GFX10-PACKED: ; %bb.0: ; %main_body255; GFX10-PACKED-NEXT: s_clause 0x2256; GFX10-PACKED-NEXT: s_load_dwordx2 s[4:5], s[8:9], 0x10257; GFX10-PACKED-NEXT: s_load_dword s6, s[8:9], 0x18258; GFX10-PACKED-NEXT: s_load_dwordx4 s[0:3], s[8:9], 0x0259; GFX10-PACKED-NEXT: s_waitcnt lgkmcnt(0)260; GFX10-PACKED-NEXT: v_mov_b32_e32 v0, s4261; GFX10-PACKED-NEXT: v_mov_b32_e32 v1, s5262; GFX10-PACKED-NEXT: v_mov_b32_e32 v2, s6263; GFX10-PACKED-NEXT: tbuffer_store_format_d16_xyzw v[0:1], v2, s[0:3], 0 format:[BUF_FMT_10_11_11_SSCALED] idxen264; GFX10-PACKED-NEXT: s_endpgm265;266; GFX11-PACKED-LABEL: tbuffer_store_d16_xyzw:267; GFX11-PACKED: ; %bb.0: ; %main_body268; GFX11-PACKED-NEXT: s_clause 0x2269; GFX11-PACKED-NEXT: s_load_b64 s[6:7], s[4:5], 0x10270; GFX11-PACKED-NEXT: s_load_b32 s8, s[4:5], 0x18271; GFX11-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x0272; GFX11-PACKED-NEXT: s_waitcnt lgkmcnt(0)273; GFX11-PACKED-NEXT: v_mov_b32_e32 v0, s6274; GFX11-PACKED-NEXT: v_mov_b32_e32 v1, s7275; GFX11-PACKED-NEXT: v_mov_b32_e32 v2, s8276; GFX11-PACKED-NEXT: tbuffer_store_d16_format_xyzw v[0:1], v2, s[0:3], 0 format:[BUF_FMT_10_10_10_2_SNORM] idxen277; GFX11-PACKED-NEXT: s_endpgm278;279; GFX12-PACKED-LABEL: tbuffer_store_d16_xyzw:280; GFX12-PACKED: ; %bb.0: ; %main_body281; GFX12-PACKED-NEXT: s_clause 0x1282; GFX12-PACKED-NEXT: s_load_b96 s[8:10], s[4:5], 0x10283; GFX12-PACKED-NEXT: s_load_b128 s[0:3], s[4:5], 0x0284; GFX12-PACKED-NEXT: s_wait_kmcnt 0x0285; GFX12-PACKED-NEXT: v_mov_b32_e32 v0, s8286; GFX12-PACKED-NEXT: v_mov_b32_e32 v1, s9287; GFX12-PACKED-NEXT: v_mov_b32_e32 v2, s10288; GFX12-PACKED-NEXT: tbuffer_store_d16_format_xyzw v[0:1], v2, s[0:3], null format:[BUF_FMT_10_10_10_2_SNORM] idxen289; GFX12-PACKED-NEXT: s_endpgm290main_body:291 call void @llvm.amdgcn.struct.tbuffer.store.v4f16(<4 x half> %data, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 0, i32 33, i32 0)292 ret void293}294 295declare void @llvm.amdgcn.struct.tbuffer.store.f16(half, <4 x i32>, i32, i32, i32, i32, i32)296declare void @llvm.amdgcn.struct.tbuffer.store.v2f16(<2 x half>, <4 x i32>, i32, i32, i32, i32, i32)297declare void @llvm.amdgcn.struct.tbuffer.store.v3f16(<3 x half>, <4 x i32>, i32, i32, i32, i32, i32)298declare void @llvm.amdgcn.struct.tbuffer.store.v4f16(<4 x half>, <4 x i32>, i32, i32, i32, i32, i32)299;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:300; GFX11-PACKED-FAKE16: {{.*}}301; GFX11-PACKED-TRUE16: {{.*}}302; GFX12-PACKED-GISEL-FAKE16: {{.*}}303; GFX12-PACKED-GISEL-TRUE16: {{.*}}304; GFX12-PACKED-SDAG-FAKE16: {{.*}}305; GFX12-PACKED-SDAG-TRUE16: {{.*}}306