Skip to content

Commit 807ed69

Browse files
vpykhtinvpykhtin
authored
[AMDGPU] Use autogenerated test checks for sdwa-preserve.mir test. NFC. (llvm#82380)
1 parent ba3c1f9 commit 807ed69

File tree

1 file changed

+49
-31
lines changed

1 file changed

+49
-31
lines changed

‎llvm/test/CodeGen/AMDGPU/sdwa-preserve.mir

Lines changed: 49 additions & 31 deletions
Original file line numberDiff line numberDiff line change
@@ -1,16 +1,6 @@
1-
# RUN: llc -mtriple=amdgcn -mcpu=fiji -start-before=si-peephole-sdwa -verify-machineinstrs -o - %s | FileCheck -check-prefix=SDWA %s
2-
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -start-before=si-peephole-sdwa -verify-machineinstrs -o - %s | FileCheck -check-prefix=SDWA %s
3-
4-
# SDWA-LABEL: {{^}}add_f16_u32_preserve
5-
6-
# SDWA: flat_load_dword [[FIRST:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]
7-
# SDWA: flat_load_dword [[SECOND:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]
8-
9-
# SDWA: v_mul_f32_sdwa [[RES:v[0-9]+]], [[FIRST]], [[SECOND]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_3
10-
# SDWA: v_add_f16_sdwa [[RES:v[0-9]+]], [[FIRST]], [[SECOND]] dst_sel:BYTE_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_0 src1_sel:WORD_1
11-
12-
# SDWA: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], [[RES]]
13-
1+
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4
2+
# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass=si-peephole-sdwa -verify-machineinstrs -o - %s | FileCheck -check-prefix=SDWA %s
3+
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=si-peephole-sdwa -verify-machineinstrs -o - %s | FileCheck -check-prefix=SDWA %s
144
---
155
name: add_f16_u32_preserve
166
tracksRegLiveness: true
@@ -33,6 +23,23 @@ body: |
3323
bb.0:
3424
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
3525
26+
; SDWA-LABEL: name: add_f16_u32_preserve
27+
; SDWA: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
28+
; SDWA-NEXT: {{ $}}
29+
; SDWA-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr30_sgpr31
30+
; SDWA-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
31+
; SDWA-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
32+
; SDWA-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY2]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
33+
; SDWA-NEXT: [[FLAT_LOAD_DWORD1:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
34+
; SDWA-NEXT: [[V_AND_B32_e32_:%[0-9]+]]:vgpr_32 = V_AND_B32_e32 65535, [[FLAT_LOAD_DWORD]], implicit $exec
35+
; SDWA-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 16, [[FLAT_LOAD_DWORD1]], implicit $exec
36+
; SDWA-NEXT: [[V_BFE_U32_e64_:%[0-9]+]]:vgpr_32 = V_BFE_U32_e64 [[FLAT_LOAD_DWORD]], 8, 8, implicit $exec
37+
; SDWA-NEXT: [[V_LSHRREV_B32_e32_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e32 24, [[FLAT_LOAD_DWORD1]], implicit $exec
38+
; SDWA-NEXT: [[V_MUL_F32_sdwa:%[0-9]+]]:vgpr_32 = V_MUL_F32_sdwa 0, [[FLAT_LOAD_DWORD]], 0, [[FLAT_LOAD_DWORD1]], 0, 0, 5, 0, 1, 3, implicit $mode, implicit $exec
39+
; SDWA-NEXT: [[V_ADD_F16_sdwa:%[0-9]+]]:vgpr_32 = V_ADD_F16_sdwa 0, [[FLAT_LOAD_DWORD]], 0, [[FLAT_LOAD_DWORD1]], 0, 0, 1, 2, 4, 5, implicit $mode, implicit $exec, implicit killed [[V_MUL_F32_sdwa]](tied-def 0)
40+
; SDWA-NEXT: FLAT_STORE_DWORD [[COPY2]], [[V_ADD_F16_sdwa]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
41+
; SDWA-NEXT: $sgpr30_sgpr31 = COPY [[COPY]]
42+
; SDWA-NEXT: S_SETPC_B64_return $sgpr30_sgpr31
3643
%2 = COPY $sgpr30_sgpr31
3744
%1 = COPY $vgpr2_vgpr3
3845
%0 = COPY $vgpr0_vgpr1
@@ -54,17 +61,8 @@ body: |
5461
FLAT_STORE_DWORD %0, %13, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
5562
$sgpr30_sgpr31 = COPY %2
5663
S_SETPC_B64_return $sgpr30_sgpr31
57-
64+
...
5865
---
59-
# SDWA-LABEL: sdwa_preserve_keep
60-
# SDWA: flat_load_dword [[FIRST:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]
61-
# SDWA: flat_load_dword [[SECOND:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]
62-
63-
# SDWA: v_and_b32_e32 [[AND:v[0-9]+]], 0xff, [[FIRST]]
64-
# SDWA: v_mov_b32_sdwa [[AND]], [[SECOND]] dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_0
65-
66-
# SDWA: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], [[AND]]
67-
6866
name: sdwa_preserve_keep
6967
tracksRegLiveness: true
7068
registers:
@@ -85,6 +83,20 @@ body: |
8583
bb.0:
8684
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
8785
86+
; SDWA-LABEL: name: sdwa_preserve_keep
87+
; SDWA: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
88+
; SDWA-NEXT: {{ $}}
89+
; SDWA-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr30_sgpr31
90+
; SDWA-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
91+
; SDWA-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
92+
; SDWA-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY2]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
93+
; SDWA-NEXT: [[FLAT_LOAD_DWORD1:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
94+
; SDWA-NEXT: [[V_LSHRREV_B16_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B16_e64 8, [[FLAT_LOAD_DWORD]], implicit $exec
95+
; SDWA-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 255
96+
; SDWA-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[FLAT_LOAD_DWORD]], killed [[S_MOV_B32_]], implicit $exec
97+
; SDWA-NEXT: [[V_MOV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_MOV_B32_sdwa 0, [[FLAT_LOAD_DWORD1]], 0, 5, 2, 4, implicit $exec, implicit [[V_AND_B32_e64_]](tied-def 0)
98+
; SDWA-NEXT: FLAT_STORE_DWORD [[COPY2]], [[V_MOV_B32_sdwa]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
99+
; SDWA-NEXT: S_ENDPGM 0
88100
%2 = COPY $sgpr30_sgpr31
89101
%1 = COPY $vgpr2_vgpr3
90102
%0 = COPY $vgpr0_vgpr1
@@ -100,14 +112,6 @@ body: |
100112
101113
...
102114
---
103-
# SDWA-LABEL: sdwa_preserve_remove
104-
# SDWA: flat_load_dword [[FIRST:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]
105-
# SDWA: flat_load_dword [[SECOND:v[0-9]+]], v[{{[0-9]+}}:{{[0-9]+}}]
106-
107-
# SDWA: v_mov_b32_sdwa [[FIRST]], [[SECOND]] dst_sel:WORD_1 dst_unused:UNUSED_PRESERVE src0_sel:WORD_0
108-
109-
# SDWA: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], [[FIRST]]
110-
111115
name: sdwa_preserve_remove
112116
tracksRegLiveness: true
113117
registers:
@@ -128,6 +132,20 @@ body: |
128132
bb.0:
129133
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
130134
135+
; SDWA-LABEL: name: sdwa_preserve_remove
136+
; SDWA: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
137+
; SDWA-NEXT: {{ $}}
138+
; SDWA-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr30_sgpr31
139+
; SDWA-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
140+
; SDWA-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
141+
; SDWA-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY2]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
142+
; SDWA-NEXT: [[FLAT_LOAD_DWORD1:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
143+
; SDWA-NEXT: [[V_LSHRREV_B16_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B16_e64 8, [[FLAT_LOAD_DWORD]], implicit $exec
144+
; SDWA-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 65535
145+
; SDWA-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[FLAT_LOAD_DWORD]], killed [[S_MOV_B32_]], implicit $exec
146+
; SDWA-NEXT: [[V_MOV_B32_sdwa:%[0-9]+]]:vgpr_32 = V_MOV_B32_sdwa 0, [[FLAT_LOAD_DWORD1]], 0, 5, 2, 4, implicit $exec, implicit [[FLAT_LOAD_DWORD]](tied-def 0)
147+
; SDWA-NEXT: FLAT_STORE_DWORD [[COPY2]], [[V_MOV_B32_sdwa]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
148+
; SDWA-NEXT: S_ENDPGM 0
131149
%2 = COPY $sgpr30_sgpr31
132150
%1 = COPY $vgpr2_vgpr3
133151
%0 = COPY $vgpr0_vgpr1

0 commit comments

Comments
 (0)