1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
|
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck -check-prefixes=GCN,MUBUF %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 -amdgpu-enable-flat-scratch < %s | FileCheck -check-prefixes=GCN,FLATSCR %s
; Test that the VGPR spiller correctly switches to SGPR offsets when the
; instruction offset field would overflow, and that it accounts for memory
; swizzling.
; GCN-LABEL: test_inst_offset_kernel
define amdgpu_kernel void @test_inst_offset_kernel() {
entry:
; Occupy 4092 bytes of scratch, so the offset of the spill of %a just fits in
; the instruction offset field.
%alloca = alloca i8, i32 4088, align 4, addrspace(5)
%buf = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], 0 offset:4092 ; 4-byte Folded Spill
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, s{{[0-9]+}} ; 4-byte Folded Spill
%a = load volatile i32, i32 addrspace(5)* %aptr
; Force %a to spill.
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
%outptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
store volatile i32 %a, i32 addrspace(5)* %outptr
ret void
}
; GCN-LABEL: test_sgpr_offset_kernel
define amdgpu_kernel void @test_sgpr_offset_kernel() {
entry:
; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
; fit in the instruction, and has to live in the SGPR offset.
%alloca = alloca i8, i32 4092, align 4, addrspace(5)
%buf = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
; 0x40000 / 64 = 4096 (for wave64)
; MUBUF: s_mov_b32 s4, 0x40000
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 ; 4-byte Folded Spill
; FLATSCR: s_movk_i32 s2, 0x1000
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, s2 ; 4-byte Folded Spill
%a = load volatile i32, i32 addrspace(5)* %aptr
; Force %a to spill
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
%outptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
store volatile i32 %a, i32 addrspace(5)* %outptr
ret void
}
; FIXME: If we fail to scavenge an SGPR in a kernel we don't have a stack
; pointer to temporarily update, so we just crash.
; GCN-LABEL: test_sgpr_offset_function_scavenge_fail_func
define void @test_sgpr_offset_function_scavenge_fail_func() #2 {
entry:
; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
; fit in the instruction, and has to live in the SGPR offset.
%alloca = alloca i8, i32 4096, align 4, addrspace(5)
%buf = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
%asm.0 = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
%asm0.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 0
%asm1.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 1
%asm2.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 2
%asm3.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 3
%asm4.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 4
%asm5.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 5
%asm6.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 6
%asm7.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 7
; 0x40000 / 64 = 4096 (for wave64)
%a = load volatile i32, i32 addrspace(5)* %aptr
; MUBUF: v_mov_b32_e32 [[OFFSET:v[0-9]+]], 0x1004
; MUBUF-NEXT: buffer_store_dword v{{[0-9]+}}, [[OFFSET]], s[{{[0-9]+:[0-9]+}}], s32 offen ; 4-byte Folded Spill
; FLATSCR: s_add_i32 [[SOFF:s[0-9]+]], s32, 0x1004
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, [[SOFF]] ; 4-byte Folded Spill
call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0.0, i32 %asm1.0, i32 %asm2.0, i32 %asm3.0, i32 %asm4.0, i32 %asm5.0, i32 %asm6.0, i32 %asm7.0, i32 %a)
%asm = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
%asm0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 0
%asm1 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 1
%asm2 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 2
%asm3 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 3
%asm4 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 4
%asm5 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 5
%asm6 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 6
%asm7 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 7
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}"() #0
; MUBUF: v_mov_b32_e32 [[OFFSET:v[0-9]+]], 0x1004
; MUBUF-NEXT: buffer_load_dword v{{[0-9]+}}, [[OFFSET]], s[{{[0-9]+:[0-9]+}}], s32 offen ; 4-byte Folded Reload
; FLATSCR: s_add_i32 [[SOFF:s[0-9]+]], s32, 0x1004
; FLATSCR: scratch_load_dword v{{[0-9]+}}, off, [[SOFF]] ; 4-byte Folded Reload
; Force %a to spill with no free SGPRs
call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0, i32 %asm1, i32 %asm2, i32 %asm3, i32 %asm4, i32 %asm5, i32 %asm6, i32 %asm7, i32 %a)
ret void
}
define amdgpu_kernel void @test_sgpr_offset_function_scavenge_fail_kernel() #3 {
entry:
; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
; fit in the instruction, and has to live in the SGPR offset.
%alloca = alloca i8, i32 4096, align 4, addrspace(5)
%buf = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
%asm.0 = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
%asm0.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 0
%asm1.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 1
%asm2.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 2
%asm3.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 3
%asm4.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 4
%asm5.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 5
%asm6.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 6
%asm7.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 7
; 0x40000 / 64 = 4096 (for wave64)
%a = load volatile i32, i32 addrspace(5)* %aptr
; MUBUF: v_mov_b32_e32 [[OFFSET:v[0-9]+]], 0x1004
; MUBUF: buffer_store_dword v{{[0-9]+}}, [[OFFSET]], s[{{[0-9]+:[0-9]+}}], 0 offen ; 4-byte Folded Spill
; FLATSCR: s_movk_i32 [[SOFF:s[0-9]+]], 0x1004
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, [[SOFF]] ; 4-byte Folded Spill
call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0.0, i32 %asm1.0, i32 %asm2.0, i32 %asm3.0, i32 %asm4.0, i32 %asm5.0, i32 %asm6.0, i32 %asm7.0, i32 %a)
%asm = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
%asm0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 0
%asm1 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 1
%asm2 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 2
%asm3 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 3
%asm4 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 4
%asm5 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 5
%asm6 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 6
%asm7 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 7
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}"() #0
; MUBUF: v_mov_b32_e32 [[OFFSET:v[0-9]+]], 0x1004
; MUBUF: buffer_load_dword v{{[0-9]+}}, [[OFFSET]], s[{{[0-9]+:[0-9]+}}], 0 offen ; 4-byte Folded Reload
; FLATSCR: s_movk_i32 [[SOFF:s[0-9]+]], 0x1004
; FLATSCR: scratch_load_dword v{{[0-9]+}}, off, [[SOFF]] ; 4-byte Folded Reload
; Force %a to spill with no free SGPRs
call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0, i32 %asm1, i32 %asm2, i32 %asm3, i32 %asm4, i32 %asm5, i32 %asm6, i32 %asm7, i32 %a)
ret void
}
; GCN-LABEL: test_sgpr_offset_subregs_kernel
define amdgpu_kernel void @test_sgpr_offset_subregs_kernel() {
entry:
; Occupy 4088 bytes of scratch, so that the spill of the last subreg of %a
; still fits below offset 4096 (4088 + 8 - 4 = 4092), and can be placed in
; the instruction offset field.
%alloca = alloca i8, i32 4084, align 4, addrspace(5)
%bufv1 = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], 0 offset:4088 ; 4-byte Folded Spill
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], 0 offset:4092 ; 4-byte Folded Spill
; FLATSCR: s_movk_i32 [[SOFF:s[0-9]+]], 0xff8
; FLATSCR: scratch_store_dwordx2 off, v[{{[0-9:]+}}], [[SOFF]] ; 8-byte Folded Spill
%aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1
%a = load volatile <2 x i32>, <2 x i32> addrspace(5)* %aptr
; Force %a to spill.
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
; Ensure the alloca sticks around.
%bptr = getelementptr i32, i32 addrspace(5)* %bufv1, i32 1
%b = load volatile i32, i32 addrspace(5)* %bptr
; Ensure the spill is of the full super-reg.
call void asm sideeffect "; $0", "r"(<2 x i32> %a)
ret void
}
; GCN-LABEL: test_inst_offset_subregs_kernel
define amdgpu_kernel void @test_inst_offset_subregs_kernel() {
entry:
; Occupy 4092 bytes of scratch, so that the spill of the last subreg of %a
; does not fit below offset 4096 (4092 + 8 - 4 = 4096), and has to live
; in the SGPR offset.
%alloca = alloca i8, i32 4088, align 4, addrspace(5)
%bufv1 = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
; 0x3ff00 / 64 = 4092 (for wave64)
; MUBUF: s_mov_b32 s4, 0x3ff00
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 ; 4-byte Folded Spill
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 offset:4 ; 4-byte Folded Spill
; FLATSCR: s_movk_i32 [[SOFF:s[0-9]+]], 0xffc
; FLATSCR: scratch_store_dwordx2 off, v[{{[0-9:]+}}], [[SOFF]] ; 8-byte Folded Spill
%aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1
%a = load volatile <2 x i32>, <2 x i32> addrspace(5)* %aptr
; Force %a to spill.
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
; Ensure the alloca sticks around.
%bptr = getelementptr i32, i32 addrspace(5)* %bufv1, i32 1
%b = load volatile i32, i32 addrspace(5)* %bptr
; Ensure the spill is of the full super-reg.
call void asm sideeffect "; $0", "r"(<2 x i32> %a)
ret void
}
; GCN-LABEL: test_inst_offset_function
define void @test_inst_offset_function() {
entry:
; Occupy enough bytes of scratch, so the offset of the spill of %a
; just fits in the instruction offset field when the emergency stack
; slot is added. It's hard to hit the actual limit since we're also
; going to insert the emergency stack slot for large frames.
%alloca = alloca i8, i32 4088, align 4, addrspace(5)
%buf = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4088 ; 4-byte Folded Spill
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, s{{[0-9]+}} offset:4088 ; 4-byte Folded Spill
%a = load volatile i32, i32 addrspace(5)* %aptr
; Force %a to spill.
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
%outptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
store volatile i32 %a, i32 addrspace(5)* %outptr
ret void
}
; GCN-LABEL: test_sgpr_offset_function
define void @test_sgpr_offset_function() {
entry:
; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
; fit in the instruction, and has to live in the SGPR offset.
%alloca = alloca i8, i32 4096, align 4, addrspace(5)
%buf = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
; 0x40000 / 64 = 4096 (for wave64)
; MUBUF: s_add_i32 s4, s32, 0x40100
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 ; 4-byte Folded Spill
; FLATSCR: s_add_i32 s0, s32, 0x1004
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, s0 ; 4-byte Folded Spill
%a = load volatile i32, i32 addrspace(5)* %aptr
; Force %a to spill
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
%outptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
store volatile i32 %a, i32 addrspace(5)* %outptr
ret void
}
; GCN-LABEL: test_sgpr_offset_subregs_function
define void @test_sgpr_offset_subregs_function() {
entry:
; We want to test the spill of the last subreg of %a is the highest
; valid value for the immediate offset. We enable the emergency
; stack slot for large frames, so it's hard to get the frame layout
; exactly as we want to test it.
;
; Occupy 4084 bytes of scratch, so that the spill of the last subreg of %a
; still fits below offset 4096 (4084 + 8 - 4 = 4092), and can be placed in
; the instruction offset field.
%alloca = alloca i8, i32 4084, align 4, addrspace(5)
%bufv1 = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4084 ; 4-byte Folded Spill
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4088 ; 4-byte Folded Spill
; FLATSCR: scratch_store_dwordx2 off, v[{{[0-9:]+}}], s32 offset:4084 ; 8-byte Folded Spill
%aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1
%a = load volatile <2 x i32>, <2 x i32> addrspace(5)* %aptr
; Force %a to spill.
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
; Ensure the alloca sticks around.
%bptr = getelementptr i32, i32 addrspace(5)* %bufv1, i32 1
%b = load volatile i32, i32 addrspace(5)* %bptr
; Ensure the spill is of the full super-reg.
call void asm sideeffect "; $0", "r"(<2 x i32> %a)
ret void
}
; GCN-LABEL: test_inst_offset_subregs_function
define void @test_inst_offset_subregs_function() {
entry:
; Occupy 4088 bytes of scratch, so that the spill of the last subreg of %a
; does not fit below offset 4096 (408 + 4 + 8 - 4 = 4096), and has to live
; in the SGPR offset.
%alloca = alloca i8, i32 4088, align 4, addrspace(5)
%bufv1 = bitcast i8 addrspace(5)* %alloca to i32 addrspace(5)*
%bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
; 0x3ff0000 / 64 = 4092 (for wave64)
; MUBUF: s_add_i32 s4, s32, 0x3ff00
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 ; 4-byte Folded Spill
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 offset:4 ; 4-byte Folded Spill
; FLATSCR: scratch_store_dwordx2 off, v[{{[0-9:]+}}], s32 offset:4092 ; 8-byte Folded Spill
%aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1
%a = load volatile <2 x i32>, <2 x i32> addrspace(5)* %aptr
; Force %a to spill.
call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()
; Ensure the alloca sticks around.
%bptr = getelementptr i32, i32 addrspace(5)* %bufv1, i32 1
%b = load volatile i32, i32 addrspace(5)* %bptr
; Ensure the spill is of the full super-reg.
call void asm sideeffect "; $0", "r"(<2 x i32> %a)
ret void
}
attributes #0 = { nounwind }
attributes #1 = { nounwind "amdgpu-num-sgpr"="17" "amdgpu-num-vgpr"="8" }
attributes #2 = { nounwind "amdgpu-num-sgpr"="14" "amdgpu-num-vgpr"="8" }
attributes #3 = { nounwind "amdgpu-num-sgpr"="16" "amdgpu-num-vgpr"="8" }
|
