xref: /llvm-project/llvm/test/CodeGen/AMDGPU/fold-imm-copy.mir (revision 9e9907f1cfa424366fba58d9520f9305b537cec9)
1# RUN: llc -mtriple=amdgcn -run-pass si-fold-operands -verify-machineinstrs %s -o - | FileCheck -check-prefix=GCN %s
2
3# GCN-LABEL:       name: fold-imm-copy
4# GCN:             V_AND_B32_e32 65535
5
6---
7name: fold-imm-copy
8tracksRegLiveness: true
9body:             |
10  bb.0:
11    liveins: $vgpr0, $sgpr0_sgpr1
12    %0:vgpr_32 = COPY $vgpr0
13    %1:sgpr_64 = COPY $sgpr0_sgpr1
14    %2:sgpr_128 = S_LOAD_DWORDX4_IMM %1, 9, 0
15    %3:sreg_32_xm0 = S_MOV_B32 2
16    %4:vgpr_32 = V_LSHLREV_B32_e64 killed %3, %0, implicit $exec
17    %5:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
18    %6:vreg_64 = REG_SEQUENCE killed %4, %subreg.sub0, killed %5, %subreg.sub1
19    %7:vgpr_32 = BUFFER_LOAD_DWORD_ADDR64 %6, %2, 0, 4, 0, 0, implicit $exec
20    %8:sreg_32_xm0 = S_MOV_B32 65535
21    %9:vgpr_32 = COPY %8
22    %10:vgpr_32 = V_AND_B32_e32 %7, %9, implicit $exec
23...
24
25---
26# GCN-LABEL:       name: no_extra_fold_on_same_opnd
27# The first XOR needs commuting to fold that immediate operand.
28# GCN:             V_XOR_B32_e32 0, %1
29# GCN:             V_XOR_B32_e32 %2, %4.sub0
30name: no_extra_fold_on_same_opnd
31tracksRegLiveness: true
32body:             |
33  bb.0:
34    %0:vgpr_32 = IMPLICIT_DEF
35    %1:vgpr_32 = IMPLICIT_DEF
36    %2:vgpr_32 = IMPLICIT_DEF
37    %3:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
38    %4:vreg_64 = REG_SEQUENCE killed %0, %subreg.sub0, killed %3, %subreg.sub1
39    %5:vgpr_32 = V_XOR_B32_e32 %1, %4.sub1, implicit $exec
40    %6:vgpr_32 = V_XOR_B32_e32 %2, %4.sub0, implicit $exec
41...
42
43---
44
45# Make sure the subreg index is not reinterpreted when folding
46# immediates
47#
48# GCN-LABEL: name: clear_subreg_imm_fold{{$}}
49# GCN: %1:sgpr_32 = S_MOV_B32 4294967288
50# GCN: %2:sgpr_32 = S_MOV_B32 4294967295
51name: clear_subreg_imm_fold
52tracksRegLiveness: true
53body:             |
54  bb.0:
55    %0:sreg_64 = S_MOV_B64 -8
56    %1:sgpr_32 = COPY %0.sub0
57    %2:sgpr_32 = COPY %0.sub1
58    S_ENDPGM 0, implicit %1, implicit %2
59
60...
61
62# GCN-LABEL: name: no_fold_imm_into_m0{{$}}
63# GCN: %0:sreg_32 = S_MOV_B32 -8
64# GCN-NEXT: $m0 = COPY %0
65
66---
67name: no_fold_imm_into_m0
68tracksRegLiveness: true
69body:             |
70  bb.0:
71    %0:sreg_32 = S_MOV_B32 -8
72    $m0 = COPY %0
73    S_ENDPGM 0, implicit $m0
74
75...
76
77# GCN-LABEL: name: fold_sgpr_imm_to_vgpr_copy{{$}}
78# GCN: $vgpr0 = V_MOV_B32_e32 -8, implicit $exec
79---
80name: fold_sgpr_imm_to_vgpr_copy
81tracksRegLiveness: true
82body:             |
83  bb.0:
84    %0:sreg_32 = S_MOV_B32 -8
85    $vgpr0 = COPY %0
86    S_ENDPGM 0, implicit $vgpr0
87
88...
89
90# The users of $vgpr1 should not be visited for further immediate
91# folding.
92
93# GCN-LABEL: name: no_fold_physreg_users_vgpr{{$}}
94# GCN: $vgpr1 = V_MOV_B32_e32 0, implicit $exec
95# GCN-NEXT: S_NOP 0, implicit-def $vgpr1
96# GCN-NEXT: %2:vgpr_32 = COPY $vgpr1
97# GCN-NEXT: $vgpr2 = COPY %2
98---
99name: no_fold_physreg_users_vgpr
100tracksRegLiveness: true
101body:             |
102  bb.0:
103    %0:sreg_32 = S_MOV_B32 0
104    %1:vgpr_32 = COPY %0
105    $vgpr1 = COPY %0
106    S_NOP 0, implicit-def $vgpr1
107    %2:vgpr_32 = COPY $vgpr1
108    $vgpr2 = COPY %2
109    S_ENDPGM 0
110
111...
112