155 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s3; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s4 5declare i32 @llvm.amdgcn.ds.bpermute.fi.b32(i32, i32) #06 7define amdgpu_kernel void @ds_bpermute_fi_b32(ptr addrspace(1) %out, i32 %index, i32 %src) nounwind {8; GFX12-LABEL: ds_bpermute_fi_b32:9; GFX12: ; %bb.0:10; GFX12-NEXT: s_load_b128 s[0:3], s[4:5], 0x2411; GFX12-NEXT: s_wait_kmcnt 0x012; GFX12-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s313; GFX12-NEXT: ds_bpermute_fi_b32 v0, v0, v114; GFX12-NEXT: v_mov_b32_e32 v1, 015; GFX12-NEXT: s_wait_dscnt 0x016; GFX12-NEXT: global_store_b32 v1, v0, s[0:1]17; GFX12-NEXT: s_endpgm18 %bpermute = call i32 @llvm.amdgcn.ds.bpermute.fi.b32(i32 %index, i32 %src) #019 store i32 %bpermute, ptr addrspace(1) %out, align 420 ret void21}22 23define amdgpu_kernel void @ds_bpermute_fi_b32_imm_offset(ptr addrspace(1) %out, i32 %base_index, i32 %src) nounwind {24; GFX12-SDAG-LABEL: ds_bpermute_fi_b32_imm_offset:25; GFX12-SDAG: ; %bb.0:26; GFX12-SDAG-NEXT: s_load_b128 s[0:3], s[4:5], 0x2427; GFX12-SDAG-NEXT: s_wait_kmcnt 0x028; GFX12-SDAG-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s329; GFX12-SDAG-NEXT: ds_bpermute_fi_b32 v0, v0, v1 offset:430; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, 031; GFX12-SDAG-NEXT: s_wait_dscnt 0x032; GFX12-SDAG-NEXT: global_store_b32 v1, v0, s[0:1]33; GFX12-SDAG-NEXT: s_endpgm34;35; GFX12-GISEL-LABEL: ds_bpermute_fi_b32_imm_offset:36; GFX12-GISEL: ; %bb.0:37; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[4:5], 0x2438; GFX12-GISEL-NEXT: s_wait_kmcnt 0x039; GFX12-GISEL-NEXT: s_add_co_i32 s2, s2, 440; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_1)41; GFX12-GISEL-NEXT: v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s242; GFX12-GISEL-NEXT: ds_bpermute_fi_b32 v0, v0, v143; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 044; GFX12-GISEL-NEXT: s_wait_dscnt 0x045; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]46; GFX12-GISEL-NEXT: s_endpgm47 %index = add i32 %base_index, 448 %bpermute = call i32 @llvm.amdgcn.ds.bpermute.fi.b32(i32 %index, i32 %src) #049 store i32 %bpermute, ptr addrspace(1) %out, align 450 ret void51}52 53define amdgpu_kernel void @ds_bpermute_fi_b32_imm_index(ptr addrspace(1) %out, i32 %base_index, i32 %src) nounwind {54; GFX12-SDAG-LABEL: ds_bpermute_fi_b32_imm_index:55; GFX12-SDAG: ; %bb.0:56; GFX12-SDAG-NEXT: s_load_b32 s0, s[4:5], 0x3057; GFX12-SDAG-NEXT: s_wait_kmcnt 0x058; GFX12-SDAG-NEXT: v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s059; GFX12-SDAG-NEXT: s_load_b64 s[0:1], s[4:5], 0x2460; GFX12-SDAG-NEXT: ds_bpermute_fi_b32 v1, v0, v1 offset:6461; GFX12-SDAG-NEXT: s_wait_dscnt 0x062; GFX12-SDAG-NEXT: s_wait_kmcnt 0x063; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]64; GFX12-SDAG-NEXT: s_endpgm65;66; GFX12-GISEL-LABEL: ds_bpermute_fi_b32_imm_index:67; GFX12-GISEL: ; %bb.0:68; GFX12-GISEL-NEXT: s_load_b32 s0, s[4:5], 0x3069; GFX12-GISEL-NEXT: s_wait_kmcnt 0x070; GFX12-GISEL-NEXT: v_dual_mov_b32 v0, 64 :: v_dual_mov_b32 v1, s071; GFX12-GISEL-NEXT: s_load_b64 s[0:1], s[4:5], 0x2472; GFX12-GISEL-NEXT: ds_bpermute_fi_b32 v0, v0, v173; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 074; GFX12-GISEL-NEXT: s_wait_dscnt 0x075; GFX12-GISEL-NEXT: s_wait_kmcnt 0x076; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]77; GFX12-GISEL-NEXT: s_endpgm78 %bpermute = call i32 @llvm.amdgcn.ds.bpermute.fi.b32(i32 64, i32 %src) #079 store i32 %bpermute, ptr addrspace(1) %out, align 480 ret void81}82 83define void @ds_bpermute_fi_b32_add_shl(ptr addrspace(1) %out, i32 %base_index, i32 %src) nounwind {84; GFX12-SDAG-LABEL: ds_bpermute_fi_b32_add_shl:85; GFX12-SDAG: ; %bb.0:86; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x087; GFX12-SDAG-NEXT: s_wait_expcnt 0x088; GFX12-SDAG-NEXT: s_wait_samplecnt 0x089; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x090; GFX12-SDAG-NEXT: s_wait_kmcnt 0x091; GFX12-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v292; GFX12-SDAG-NEXT: ds_bpermute_fi_b32 v2, v2, v3 offset:493; GFX12-SDAG-NEXT: s_wait_dscnt 0x094; GFX12-SDAG-NEXT: global_store_b32 v[0:1], v2, off95; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]96;97; GFX12-GISEL-LABEL: ds_bpermute_fi_b32_add_shl:98; GFX12-GISEL: ; %bb.0:99; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0100; GFX12-GISEL-NEXT: s_wait_expcnt 0x0101; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0102; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0103; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0104; GFX12-GISEL-NEXT: v_lshl_add_u32 v2, v2, 2, 4105; GFX12-GISEL-NEXT: ds_bpermute_fi_b32 v2, v2, v3106; GFX12-GISEL-NEXT: s_wait_dscnt 0x0107; GFX12-GISEL-NEXT: global_store_b32 v[0:1], v2, off108; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]109 %index = add i32 %base_index, 1110 %byte_index = shl i32 %index, 2111 %bpermute = call i32 @llvm.amdgcn.ds.bpermute.fi.b32(i32 %byte_index, i32 %src) #0112 store i32 %bpermute, ptr addrspace(1) %out, align 4113 ret void114}115 116define void @ds_bpermute_fi_b32_or_shl(ptr addrspace(1) %out, i32 %base_index, i32 %src) nounwind {117; GFX12-SDAG-LABEL: ds_bpermute_fi_b32_or_shl:118; GFX12-SDAG: ; %bb.0:119; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0120; GFX12-SDAG-NEXT: s_wait_expcnt 0x0121; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0122; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0123; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0124; GFX12-SDAG-NEXT: v_and_b32_e32 v2, 62, v2125; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_1)126; GFX12-SDAG-NEXT: v_lshlrev_b32_e32 v2, 2, v2127; GFX12-SDAG-NEXT: ds_bpermute_fi_b32 v2, v2, v3 offset:4128; GFX12-SDAG-NEXT: s_wait_dscnt 0x0129; GFX12-SDAG-NEXT: global_store_b32 v[0:1], v2, off130; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]131;132; GFX12-GISEL-LABEL: ds_bpermute_fi_b32_or_shl:133; GFX12-GISEL: ; %bb.0:134; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0135; GFX12-GISEL-NEXT: s_wait_expcnt 0x0136; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0137; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0138; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0139; GFX12-GISEL-NEXT: v_and_b32_e32 v2, 62, v2140; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)141; GFX12-GISEL-NEXT: v_lshl_or_b32 v2, v2, 2, 4142; GFX12-GISEL-NEXT: ds_bpermute_fi_b32 v2, v2, v3143; GFX12-GISEL-NEXT: s_wait_dscnt 0x0144; GFX12-GISEL-NEXT: global_store_b32 v[0:1], v2, off145; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]146 %masked = and i32 %base_index, 62147 %index = or i32 %masked, 1148 %byte_index = shl i32 %index, 2149 %bpermute = call i32 @llvm.amdgcn.ds.bpermute.fi.b32(i32 %byte_index, i32 %src) #0150 store i32 %bpermute, ptr addrspace(1) %out, align 4151 ret void152}153 154attributes #0 = { nounwind readnone convergent }155