1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,MOVREL,PREGFX9 %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,MOVREL,PREGFX9 %s
3 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -amdgpu-vgpr-index-mode -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,IDXMODE,PREGFX9 %s
5 ; Tests for indirect addressing on SI, which is implemented using dynamic
8 ; Subtest below moved from file test/CodeGen/AMDGPU/indirect-addressing-si.ll
9 ; to avoid gfx9 scheduling induced issues.
12 ; GCN-LABEL: {{^}}insert_vgpr_offset_multiple_in_block:
13 ; GCN-DAG: s_load_dwordx16 s{{\[}}[[S_ELT0:[0-9]+]]:[[S_ELT15:[0-9]+]]{{\]}}
14 ; GCN-DAG: {{buffer|flat|global}}_load_dword [[IDX0:v[0-9]+]]
15 ; GCN-DAG: v_mov_b32 [[INS0:v[0-9]+]], 62
17 ; GCN-DAG: v_mov_b32_e32 v[[VEC_ELT15:[0-9]+]], s[[S_ELT15]]
18 ; GCN-DAG: v_mov_b32_e32 v[[VEC_ELT0:[0-9]+]], s[[S_ELT0]]
20 ; GCN-DAG: v_add_{{i32|u32}}_e32 [[IDX1:v[0-9]+]], vcc, 1, [[IDX0]]
22 ; GCN: [[LOOP0:BB[0-9]+_[0-9]+]]:
23 ; GCN-NEXT: v_readfirstlane_b32 [[READLANE:s[0-9]+]], [[IDX0]]
24 ; GCN: v_cmp_eq_u32_e32 vcc, [[READLANE]], [[IDX0]]
25 ; GCN: s_and_saveexec_b64 vcc, vcc
27 ; MOVREL: s_mov_b32 m0, [[READLANE]]
28 ; MOVREL-NEXT: v_movreld_b32_e32 v[[VEC_ELT0]], [[INS0]]
30 ; IDXMODE: s_set_gpr_idx_on [[READLANE]], gpr_idx(DST)
31 ; IDXMODE-NEXT: v_mov_b32_e32 v[[VEC_ELT0]], [[INS0]]
32 ; IDXMODE: s_set_gpr_idx_off
34 ; GCN-NEXT: s_xor_b64 exec, exec, vcc
35 ; GCN: s_cbranch_execnz [[LOOP0]]
37 ; FIXME: Redundant copy
38 ; GCN: s_mov_b64 exec, [[MASK:s\[[0-9]+:[0-9]+\]]]
40 ; GCN: s_mov_b64 [[MASK]], exec
42 ; GCN: [[LOOP1:BB[0-9]+_[0-9]+]]:
43 ; GCN-NEXT: v_readfirstlane_b32 [[READLANE:s[0-9]+]], [[IDX1]]
44 ; GCN: v_cmp_eq_u32_e32 vcc, [[READLANE]], [[IDX1]]
45 ; GCN: s_and_saveexec_b64 vcc, vcc
47 ; MOVREL: s_mov_b32 m0, [[READLANE]]
48 ; MOVREL-NEXT: v_movreld_b32_e32 v{{[0-9]+}}, 63
50 ; IDXMODE: s_set_gpr_idx_on [[READLANE]], gpr_idx(DST)
51 ; IDXMODE-NEXT: v_mov_b32_e32 v{{[0-9]+}}, 63
52 ; IDXMODE: s_set_gpr_idx_off
54 ; GCN-NEXT: s_xor_b64 exec, exec, vcc
55 ; GCN: s_cbranch_execnz [[LOOP1]]
57 ; GCN: buffer_store_dwordx4 v{{\[}}[[VEC_ELT0]]:
59 ; GCN: buffer_store_dword [[INS0]]
60 define amdgpu_kernel void @insert_vgpr_offset_multiple_in_block(<16 x i32> addrspace(1)* %out0, <16 x i32> addrspace(1)* %out1, i32 addrspace(1)* %in, <16 x i32> %vec0) #0 {
62 %id = call i32 @llvm.amdgcn.workitem.id.x() #1
63 %id.ext = zext i32 %id to i64
64 %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i64 %id.ext
65 %idx0 = load volatile i32, i32 addrspace(1)* %gep
66 %idx1 = add i32 %idx0, 1
67 %live.out.val = call i32 asm sideeffect "v_mov_b32 $0, 62", "=v"()
68 %vec1 = insertelement <16 x i32> %vec0, i32 %live.out.val, i32 %idx0
69 %vec2 = insertelement <16 x i32> %vec1, i32 63, i32 %idx1
70 store volatile <16 x i32> %vec2, <16 x i32> addrspace(1)* %out0
71 %cmp = icmp eq i32 %id, 0
72 br i1 %cmp, label %bb1, label %bb2
75 store volatile i32 %live.out.val, i32 addrspace(1)* undef
82 declare i32 @llvm.amdgcn.workitem.id.x() #1
83 declare void @llvm.amdgcn.s.barrier() #2
85 attributes #0 = { nounwind }
86 attributes #1 = { nounwind readnone }
87 attributes #2 = { nounwind convergent }