Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
PojavLauncherTeam
GitHub Repository: PojavLauncherTeam/mesa
Path: blob/21.2-virgl/src/freedreno/registers/adreno/adreno_pm4.xml
8362 views
1
<?xml version="1.0" encoding="UTF-8"?>
2
<database xmlns="http://nouveau.freedesktop.org/"
3
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
4
xsi:schemaLocation="http://nouveau.freedesktop.org/ rules-ng.xsd">
5
6
<enum name="vgt_event_type">
7
<value name="VS_DEALLOC" value="0"/>
8
<value name="PS_DEALLOC" value="1"/>
9
<value name="VS_DONE_TS" value="2"/>
10
<value name="PS_DONE_TS" value="3"/>
11
<value name="CACHE_FLUSH_TS" value="4"/>
12
<value name="CONTEXT_DONE" value="5"/>
13
<value name="CACHE_FLUSH" value="6"/>
14
<value name="VIZQUERY_START" value="7" varset="chip" variants="A2XX"/>
15
<value name="HLSQ_FLUSH" value="7" varset="chip" variants="A3XX-A4XX"/>
16
<value name="VIZQUERY_END" value="8" varset="chip" variants="A2XX"/>
17
<value name="SC_WAIT_WC" value="9" varset="chip" variants="A2XX"/>
18
<value name="WRITE_PRIMITIVE_COUNTS" value="9" varset="chip" variants="A6XX"/>
19
<value name="START_PRIMITIVE_CTRS" value="11" varset="chip" variants="A6XX"/>
20
<value name="STOP_PRIMITIVE_CTRS" value="12" varset="chip" variants="A6XX"/>
21
<value name="RST_PIX_CNT" value="13"/>
22
<value name="RST_VTX_CNT" value="14"/>
23
<value name="TILE_FLUSH" value="15"/>
24
<value name="STAT_EVENT" value="16"/>
25
<value name="CACHE_FLUSH_AND_INV_TS_EVENT" value="20" varset="chip" variants="A2XX-A4XX"/>
26
<value name="ZPASS_DONE" value="21"/>
27
<value name="CACHE_FLUSH_AND_INV_EVENT" value="22" varset="chip" variants="A2XX"/>
28
<value name="RB_DONE_TS" value="22" varset="chip" variants="A3XX-"/>
29
<value name="PERFCOUNTER_START" value="23" varset="chip" variants="A2XX-A4XX"/>
30
<value name="PERFCOUNTER_STOP" value="24" varset="chip" variants="A2XX-A4XX"/>
31
<value name="VS_FETCH_DONE" value="27"/>
32
<value name="FACENESS_FLUSH" value="28" varset="chip" variants="A2XX-A4XX"/>
33
34
<!-- a5xx events -->
35
<value name="WT_DONE_TS" value="8" varset="chip" variants="A5XX-"/>
36
<value name="FLUSH_SO_0" value="17" varset="chip" variants="A5XX-"/>
37
<value name="FLUSH_SO_1" value="18" varset="chip" variants="A5XX-"/>
38
<value name="FLUSH_SO_2" value="19" varset="chip" variants="A5XX-"/>
39
<value name="FLUSH_SO_3" value="20" varset="chip" variants="A5XX-"/>
40
<value name="PC_CCU_INVALIDATE_DEPTH" value="24" varset="chip" variants="A5XX-"/>
41
<value name="PC_CCU_INVALIDATE_COLOR" value="25" varset="chip" variants="A5XX-"/>
42
<value name="PC_CCU_RESOLVE_TS" value="26" varset="chip" variants="A6XX"/>
43
<value name="PC_CCU_FLUSH_DEPTH_TS" value="28" varset="chip" variants="A5XX-"/>
44
<value name="PC_CCU_FLUSH_COLOR_TS" value="29" varset="chip" variants="A5XX-"/>
45
<value name="BLIT" value="30" varset="chip" variants="A5XX-"/>
46
<value name="UNK_25" value="37" varset="chip" variants="A5XX"/>
47
<value name="LRZ_FLUSH" value="38" varset="chip" variants="A5XX-"/>
48
<value name="BLIT_OP_FILL_2D" value="39" varset="chip" variants="A5XX-"/>
49
<value name="BLIT_OP_COPY_2D" value="40" varset="chip" variants="A5XX-"/>
50
<value name="BLIT_OP_SCALE_2D" value="42" varset="chip" variants="A5XX-"/>
51
<value name="CONTEXT_DONE_2D" value="43" varset="chip" variants="A5XX-"/>
52
<value name="UNK_2C" value="44" varset="chip" variants="A5XX-"/>
53
<value name="UNK_2D" value="45" varset="chip" variants="A5XX-"/>
54
55
<!-- a6xx events -->
56
<value name="CACHE_INVALIDATE" value="49" varset="chip" variants="A6XX"/>
57
</enum>
58
59
<enum name="pc_di_primtype">
60
<value name="DI_PT_NONE" value="0"/>
61
<!-- POINTLIST_PSIZE is used on a3xx/a4xx when gl_PointSize is written: -->
62
<value name="DI_PT_POINTLIST_PSIZE" value="1"/>
63
<value name="DI_PT_LINELIST" value="2"/>
64
<value name="DI_PT_LINESTRIP" value="3"/>
65
<value name="DI_PT_TRILIST" value="4"/>
66
<value name="DI_PT_TRIFAN" value="5"/>
67
<value name="DI_PT_TRISTRIP" value="6"/>
68
<value name="DI_PT_LINELOOP" value="7"/> <!-- a22x, a3xx -->
69
<value name="DI_PT_RECTLIST" value="8"/>
70
<value name="DI_PT_POINTLIST" value="9"/>
71
<value name="DI_PT_LINE_ADJ" value="0xa"/>
72
<value name="DI_PT_LINESTRIP_ADJ" value="0xb"/>
73
<value name="DI_PT_TRI_ADJ" value="0xc"/>
74
<value name="DI_PT_TRISTRIP_ADJ" value="0xd"/>
75
76
<value name="DI_PT_PATCHES0" value="0x1f"/>
77
<value name="DI_PT_PATCHES1" value="0x20"/>
78
<value name="DI_PT_PATCHES2" value="0x21"/>
79
<value name="DI_PT_PATCHES3" value="0x22"/>
80
<value name="DI_PT_PATCHES4" value="0x23"/>
81
<value name="DI_PT_PATCHES5" value="0x24"/>
82
<value name="DI_PT_PATCHES6" value="0x25"/>
83
<value name="DI_PT_PATCHES7" value="0x26"/>
84
<value name="DI_PT_PATCHES8" value="0x27"/>
85
<value name="DI_PT_PATCHES9" value="0x28"/>
86
<value name="DI_PT_PATCHES10" value="0x29"/>
87
<value name="DI_PT_PATCHES11" value="0x2a"/>
88
<value name="DI_PT_PATCHES12" value="0x2b"/>
89
<value name="DI_PT_PATCHES13" value="0x2c"/>
90
<value name="DI_PT_PATCHES14" value="0x2d"/>
91
<value name="DI_PT_PATCHES15" value="0x2e"/>
92
<value name="DI_PT_PATCHES16" value="0x2f"/>
93
<value name="DI_PT_PATCHES17" value="0x30"/>
94
<value name="DI_PT_PATCHES18" value="0x31"/>
95
<value name="DI_PT_PATCHES19" value="0x32"/>
96
<value name="DI_PT_PATCHES20" value="0x33"/>
97
<value name="DI_PT_PATCHES21" value="0x34"/>
98
<value name="DI_PT_PATCHES22" value="0x35"/>
99
<value name="DI_PT_PATCHES23" value="0x36"/>
100
<value name="DI_PT_PATCHES24" value="0x37"/>
101
<value name="DI_PT_PATCHES25" value="0x38"/>
102
<value name="DI_PT_PATCHES26" value="0x39"/>
103
<value name="DI_PT_PATCHES27" value="0x3a"/>
104
<value name="DI_PT_PATCHES28" value="0x3b"/>
105
<value name="DI_PT_PATCHES29" value="0x3c"/>
106
<value name="DI_PT_PATCHES30" value="0x3d"/>
107
<value name="DI_PT_PATCHES31" value="0x3e"/>
108
</enum>
109
110
<enum name="pc_di_src_sel">
111
<value name="DI_SRC_SEL_DMA" value="0"/>
112
<value name="DI_SRC_SEL_IMMEDIATE" value="1"/>
113
<value name="DI_SRC_SEL_AUTO_INDEX" value="2"/>
114
<value name="DI_SRC_SEL_AUTO_XFB" value="3"/>
115
</enum>
116
117
<enum name="pc_di_face_cull_sel">
118
<value name="DI_FACE_CULL_NONE" value="0"/>
119
<value name="DI_FACE_CULL_FETCH" value="1"/>
120
<value name="DI_FACE_BACKFACE_CULL" value="2"/>
121
<value name="DI_FACE_FRONTFACE_CULL" value="3"/>
122
</enum>
123
124
<enum name="pc_di_index_size">
125
<value name="INDEX_SIZE_IGN" value="0"/>
126
<value name="INDEX_SIZE_16_BIT" value="0"/>
127
<value name="INDEX_SIZE_32_BIT" value="1"/>
128
<value name="INDEX_SIZE_8_BIT" value="2"/>
129
<value name="INDEX_SIZE_INVALID"/>
130
</enum>
131
132
<enum name="pc_di_vis_cull_mode">
133
<value name="IGNORE_VISIBILITY" value="0"/>
134
<value name="USE_VISIBILITY" value="1"/>
135
</enum>
136
137
<enum name="adreno_pm4_packet_type">
138
<value name="CP_TYPE0_PKT" value="0x00000000"/>
139
<value name="CP_TYPE1_PKT" value="0x40000000"/>
140
<value name="CP_TYPE2_PKT" value="0x80000000"/>
141
<value name="CP_TYPE3_PKT" value="0xc0000000"/>
142
<value name="CP_TYPE4_PKT" value="0x40000000"/>
143
<value name="CP_TYPE7_PKT" value="0x70000000"/>
144
</enum>
145
146
<!--
147
Note that in some cases, the same packet id is recycled on a later
148
generation, so variants attribute is used to distinguish. They
149
may not be completely accurate, we would probably have to analyze
150
the pfp and me/pm4 firmware to verify the packet is actually
151
handled on a particular generation. But it is at least enough to
152
disambiguate the packet-id's that were re-used for different
153
packets starting with a5xx.
154
-->
155
<enum name="adreno_pm4_type3_packets">
156
<doc>initialize CP's micro-engine</doc>
157
<value name="CP_ME_INIT" value="0x48"/>
158
<doc>skip N 32-bit words to get to the next packet</doc>
159
<value name="CP_NOP" value="0x10"/>
160
<doc>
161
indirect buffer dispatch. prefetch parser uses this packet
162
type to determine whether to pre-fetch the IB
163
</doc>
164
<value name="CP_PREEMPT_ENABLE" value="0x1c"/>
165
<value name="CP_PREEMPT_TOKEN" value="0x1e"/>
166
<value name="CP_INDIRECT_BUFFER" value="0x3f"/>
167
<doc>
168
Takes the same arguments as CP_INDIRECT_BUFFER, but jumps to
169
another buffer at the same level. Must be at the end of IB, and
170
doesn't work with draw state IB's.
171
</doc>
172
<value name="CP_INDIRECT_BUFFER_CHAIN" value="0x57" varset="chip" variants="A5XX-"/>
173
<doc>indirect buffer dispatch. same as IB, but init is pipelined</doc>
174
<value name="CP_INDIRECT_BUFFER_PFD" value="0x37"/>
175
<doc>wait for the IDLE state of the engine</doc>
176
<value name="CP_WAIT_FOR_IDLE" value="0x26"/>
177
<doc>wait until a register or memory location is a specific value</doc>
178
<value name="CP_WAIT_REG_MEM" value="0x3c"/>
179
<doc>wait until a register location is equal to a specific value</doc>
180
<value name="CP_WAIT_REG_EQ" value="0x52"/>
181
<doc>wait until a register location is >= a specific value</doc>
182
<value name="CP_WAIT_REG_GTE" value="0x53" varset="chip" variants="A2XX-A4XX"/>
183
<doc>wait until a read completes</doc>
184
<value name="CP_WAIT_UNTIL_READ" value="0x5c" varset="chip" variants="A2XX-A4XX"/>
185
<doc>wait until all base/size writes from an IB_PFD packet have completed</doc>
186
<value name="CP_WAIT_IB_PFD_COMPLETE" value="0x5d"/>
187
<doc>register read/modify/write</doc>
188
<value name="CP_REG_RMW" value="0x21"/>
189
<doc>Set binning configuration registers</doc>
190
<value name="CP_SET_BIN_DATA" value="0x2f" varset="chip" variants="A2XX-A4XX"/>
191
<value name="CP_SET_BIN_DATA5" value="0x2f" varset="chip" variants="A5XX-"/>
192
<doc>reads register in chip and writes to memory</doc>
193
<value name="CP_REG_TO_MEM" value="0x3e"/>
194
<doc>write N 32-bit words to memory</doc>
195
<value name="CP_MEM_WRITE" value="0x3d"/>
196
<doc>write CP_PROG_COUNTER value to memory</doc>
197
<value name="CP_MEM_WRITE_CNTR" value="0x4f"/>
198
<doc>conditional execution of a sequence of packets</doc>
199
<value name="CP_COND_EXEC" value="0x44"/>
200
<doc>conditional write to memory or register</doc>
201
<value name="CP_COND_WRITE" value="0x45" varset="chip" variants="A2XX-A4XX"/>
202
<value name="CP_COND_WRITE5" value="0x45" varset="chip" variants="A5XX-"/>
203
<doc>generate an event that creates a write to memory when completed</doc>
204
<value name="CP_EVENT_WRITE" value="0x46"/>
205
<doc>generate a VS|PS_done event</doc>
206
<value name="CP_EVENT_WRITE_SHD" value="0x58"/>
207
<doc>generate a cache flush done event</doc>
208
<value name="CP_EVENT_WRITE_CFL" value="0x59"/>
209
<doc>generate a z_pass done event</doc>
210
<value name="CP_EVENT_WRITE_ZPD" value="0x5b"/>
211
<doc>
212
not sure the real name, but this seems to be what is used for
213
opencl, instead of CP_DRAW_INDX..
214
</doc>
215
<value name="CP_RUN_OPENCL" value="0x31"/>
216
<doc>initiate fetch of index buffer and draw</doc>
217
<value name="CP_DRAW_INDX" value="0x22"/>
218
<doc>draw using supplied indices in packet</doc>
219
<value name="CP_DRAW_INDX_2" value="0x36" varset="chip" variants="A2XX-A4XX"/> <!-- this is something different on a6xx and unused on a5xx -->
220
<doc>initiate fetch of index buffer and binIDs and draw</doc>
221
<value name="CP_DRAW_INDX_BIN" value="0x34" varset="chip" variants="A2XX-A4XX"/>
222
<doc>initiate fetch of bin IDs and draw using supplied indices</doc>
223
<value name="CP_DRAW_INDX_2_BIN" value="0x35" varset="chip" variants="A2XX-A4XX"/>
224
<doc>begin/end initiator for viz query extent processing</doc>
225
<value name="CP_VIZ_QUERY" value="0x23" varset="chip" variants="A2XX-A4XX"/>
226
<doc>fetch state sub-blocks and initiate shader code DMAs</doc>
227
<value name="CP_SET_STATE" value="0x25"/>
228
<doc>load constant into chip and to memory</doc>
229
<value name="CP_SET_CONSTANT" value="0x2d"/>
230
<doc>load sequencer instruction memory (pointer-based)</doc>
231
<value name="CP_IM_LOAD" value="0x27"/>
232
<doc>load sequencer instruction memory (code embedded in packet)</doc>
233
<value name="CP_IM_LOAD_IMMEDIATE" value="0x2b"/>
234
<doc>load constants from a location in memory</doc>
235
<value name="CP_LOAD_CONSTANT_CONTEXT" value="0x2e" varset="chip" variants="A2XX"/>
236
<doc>selective invalidation of state pointers</doc>
237
<value name="CP_INVALIDATE_STATE" value="0x3b"/>
238
<doc>dynamically changes shader instruction memory partition</doc>
239
<value name="CP_SET_SHADER_BASES" value="0x4a" varset="chip" variants="A2XX-A4XX"/>
240
<doc>sets the 64-bit BIN_MASK register in the PFP</doc>
241
<value name="CP_SET_BIN_MASK" value="0x50" varset="chip" variants="A2XX-A4XX"/>
242
<doc>sets the 64-bit BIN_SELECT register in the PFP</doc>
243
<value name="CP_SET_BIN_SELECT" value="0x51"/>
244
<doc>updates the current context, if needed</doc>
245
<value name="CP_CONTEXT_UPDATE" value="0x5e"/>
246
<doc>generate interrupt from the command stream</doc>
247
<value name="CP_INTERRUPT" value="0x40"/>
248
<doc>copy sequencer instruction memory to system memory</doc>
249
<value name="CP_IM_STORE" value="0x2c" varset="chip" variants="A2XX"/>
250
251
<!-- For a20x -->
252
<!-- TODO handle variants..
253
<doc>
254
Program an offset that will added to the BIN_BASE value of
255
the 3D_DRAW_INDX_BIN packet
256
</doc>
257
<value name="CP_SET_BIN_BASE_OFFSET" value="0x4b"/>
258
-->
259
260
<!-- for a22x -->
261
<doc>
262
sets draw initiator flags register in PFP, gets bitwise-ORed into
263
every draw initiator
264
</doc>
265
<value name="CP_SET_DRAW_INIT_FLAGS" value="0x4b"/>
266
<doc>sets the register protection mode</doc>
267
<value name="CP_SET_PROTECTED_MODE" value="0x5f"/>
268
269
<value name="CP_BOOTSTRAP_UCODE" value="0x6f"/>
270
271
<!-- for a3xx -->
272
<doc>load high level sequencer command</doc>
273
<value name="CP_LOAD_STATE" value="0x30" varset="chip" variants="A3XX"/>
274
<value name="CP_LOAD_STATE4" value="0x30" varset="chip" variants="A4XX-A5XX"/>
275
<doc>Conditionally load a IB based on a flag, prefetch enabled</doc>
276
<value name="CP_COND_INDIRECT_BUFFER_PFE" value="0x3a"/>
277
<doc>Conditionally load a IB based on a flag, prefetch disabled</doc>
278
<value name="CP_COND_INDIRECT_BUFFER_PFD" value="0x32" varset="chip" variants="A3XX"/>
279
<doc>Load a buffer with pre-fetch enabled</doc>
280
<value name="CP_INDIRECT_BUFFER_PFE" value="0x3f" varset="chip" variants="A5XX"/>
281
<doc>Set bin (?)</doc>
282
<value name="CP_SET_BIN" value="0x4c" varset="chip" variants="A2XX"/>
283
284
<doc>test 2 memory locations to dword values specified</doc>
285
<value name="CP_TEST_TWO_MEMS" value="0x71"/>
286
287
<doc>Write register, ignoring context state for context sensitive registers</doc>
288
<value name="CP_REG_WR_NO_CTXT" value="0x78"/>
289
290
<doc>Record the real-time when this packet is processed by PFP</doc>
291
<value name="CP_RECORD_PFP_TIMESTAMP" value="0x11"/>
292
293
<!-- Used to switch GPU between secure and non-secure modes -->
294
<value name="CP_SET_SECURE_MODE" value="0x66"/>
295
296
<doc>PFP waits until the FIFO between the PFP and the ME is empty</doc>
297
<value name="CP_WAIT_FOR_ME" value="0x13"/>
298
299
<!-- for a4xx -->
300
<doc>
301
Used a bit like CP_SET_CONSTANT on a2xx, but can write multiple
302
groups of registers. Looks like it can be used to create state
303
objects in GPU memory, and on state change only emit pointer
304
(via CP_SET_DRAW_STATE), which should be nice for reducing CPU
305
overhead:
306
307
(A4x) save PM4 stream pointers to execute upon a visible draw
308
</doc>
309
<value name="CP_SET_DRAW_STATE" value="0x43" varset="chip" variants="A4XX-"/>
310
<value name="CP_DRAW_INDX_OFFSET" value="0x38"/>
311
<value name="CP_DRAW_INDIRECT" value="0x28" varset="chip" variants="A4XX-"/>
312
<value name="CP_DRAW_INDX_INDIRECT" value="0x29" varset="chip" variants="A4XX-"/>
313
<value name="CP_DRAW_INDIRECT_MULTI" value="0x2a" varset="chip" variants="A6XX"/>
314
<value name="CP_DRAW_AUTO" value="0x24"/>
315
316
<doc>
317
Enable or disable predication globally. Also resets the
318
predicate to "passing" and the local bit to enabled when
319
enabling global predication.
320
</doc>
321
<value name="CP_DRAW_PRED_ENABLE_GLOBAL" value="0x19"/>
322
323
<doc>
324
Enable or disable predication locally. Unlike globally enabling
325
predication, this packet doesn't touch any other state.
326
Predication only happens when enabled globally and locally and a
327
predicate has been set. This should be used for internal draws
328
which aren't supposed to use the predication state:
329
330
CP_DRAW_PRED_ENABLE_LOCAL(0)
331
... do draw...
332
CP_DRAW_PRED_ENABLE_LOCAL(1)
333
</doc>
334
<value name="CP_DRAW_PRED_ENABLE_LOCAL" value="0x1a"/>
335
336
<doc>
337
Latch a draw predicate into the internal register.
338
</doc>
339
<value name="CP_DRAW_PRED_SET" value="0x4e"/>
340
341
<doc>
342
for A4xx
343
Write to register with address that does not fit into type-0 pkt
344
</doc>
345
<value name="CP_WIDE_REG_WRITE" value="0x74" varset="chip" variants="A4XX"/>
346
347
<doc>copy from ME scratch RAM to a register</doc>
348
<value name="CP_SCRATCH_TO_REG" value="0x4d"/>
349
350
<doc>Copy from REG to ME scratch RAM</doc>
351
<value name="CP_REG_TO_SCRATCH" value="0x4a"/>
352
353
<doc>Wait for memory writes to complete</doc>
354
<value name="CP_WAIT_MEM_WRITES" value="0x12"/>
355
356
<doc>Conditional execution based on register comparison</doc>
357
<value name="CP_COND_REG_EXEC" value="0x47"/>
358
359
<doc>Memory to REG copy</doc>
360
<value name="CP_MEM_TO_REG" value="0x42"/>
361
362
<value name="CP_EXEC_CS_INDIRECT" value="0x41" varset="chip" variants="A4XX-"/>
363
<value name="CP_EXEC_CS" value="0x33"/>
364
365
<doc>
366
for a5xx
367
</doc>
368
<value name="CP_PERFCOUNTER_ACTION" value="0x50" varset="chip" variants="A5XX"/>
369
<!-- switches SMMU pagetable, used on a5xx+ only -->
370
<value name="CP_SMMU_TABLE_UPDATE" value="0x53" varset="chip" variants="A5XX-"/>
371
<!-- for a6xx -->
372
<doc>Tells CP the current mode of GPU operation</doc>
373
<value name="CP_SET_MARKER" value="0x65" varset="chip" variants="A6XX"/>
374
<doc>Instruct CP to set a few internal CP registers</doc>
375
<value name="CP_SET_PSEUDO_REG" value="0x56" varset="chip" variants="A6XX"/>
376
<!--
377
pairs of regid and value.. seems to be used to program some TF
378
related regs:
379
-->
380
<value name="CP_CONTEXT_REG_BUNCH" value="0x5c" varset="chip" variants="A5XX-"/>
381
<!-- A5XX Enable yield in RB only -->
382
<value name="CP_YIELD_ENABLE" value="0x1c" varset="chip" variants="A5XX"/>
383
<value name="CP_SKIP_IB2_ENABLE_GLOBAL" value="0x1d" varset="chip" variants="A5XX-"/>
384
<value name="CP_SKIP_IB2_ENABLE_LOCAL" value="0x23" varset="chip" variants="A5XX-"/>
385
<value name="CP_SET_SUBDRAW_SIZE" value="0x35" varset="chip" variants="A5XX-"/>
386
<value name="CP_WHERE_AM_I" value="0x62" varset="chip" variants="A5XX-"/>
387
<value name="CP_SET_VISIBILITY_OVERRIDE" value="0x64" varset="chip" variants="A5XX-"/>
388
<!-- Enable/Disable/Defer A5x global preemption model -->
389
<value name="CP_PREEMPT_ENABLE_GLOBAL" value="0x69" varset="chip" variants="A5XX"/>
390
<!-- Enable/Disable A5x local preemption model -->
391
<value name="CP_PREEMPT_ENABLE_LOCAL" value="0x6a" varset="chip" variants="A5XX"/>
392
<!-- Yield token on a5xx similar to CP_PREEMPT on a4xx -->
393
<value name="CP_CONTEXT_SWITCH_YIELD" value="0x6b" varset="chip" variants="A5XX"/>
394
<!-- Inform CP about current render mode (needed for a5xx preemption) -->
395
<value name="CP_SET_RENDER_MODE" value="0x6c" varset="chip" variants="A5XX"/>
396
<value name="CP_COMPUTE_CHECKPOINT" value="0x6e" varset="chip" variants="A5XX"/>
397
<!-- check if this works on earlier.. -->
398
<value name="CP_MEM_TO_MEM" value="0x73" varset="chip" variants="A5XX-"/>
399
<value name="CP_BLIT" value="0x2c" varset="chip" variants="A5XX-"/>
400
401
<!-- Test specified bit in specified register and set predicate -->
402
<value name="CP_REG_TEST" value="0x39" varset="chip" variants="A5XX-"/>
403
404
<!--
405
Seems to set the mode flags which control which CP_SET_DRAW_STATE
406
packets are executed, based on their ENABLE_MASK values
407
408
CP_SET_MODE w/ payload of 0x1 seems to cause CP_SET_DRAW_STATE
409
packets w/ ENABLE_MASK & 0x6 to execute immediately
410
-->
411
<value name="CP_SET_MODE" value="0x63" varset="chip" variants="A6XX"/>
412
413
<!--
414
Seems like there are now separate blocks of state for VS vs FS/CS
415
(probably these amounts to geometry vs fragments so that geometry
416
stage of the pipeline for next draw can start while fragment stage
417
of current draw is still running. The format of the payload of the
418
packets is the same, the only difference is the offsets of the regs
419
the firmware code that handles the packet writes.
420
421
Note that for CL, starting with a6xx, the preferred # of local
422
threads is no longer the same as the max, implying that the shader
423
core can now run warps from unrelated shaders (ie.
424
CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE vs
425
CL_KERNEL_WORK_GROUP_SIZE)
426
-->
427
<value name="CP_LOAD_STATE6_GEOM" value="0x32" varset="chip" variants="A6XX"/>
428
<value name="CP_LOAD_STATE6_FRAG" value="0x34" varset="chip" variants="A6XX"/>
429
<!--
430
Note: For IBO state (Image/SSBOs) which have shared state across
431
shader stages, for 3d pipeline CP_LOAD_STATE6 is used. But for
432
compute shaders, CP_LOAD_STATE6_FRAG is used. Possibly they are
433
interchangable.
434
-->
435
<value name="CP_LOAD_STATE6" value="0x36" varset="chip" variants="A6XX"/>
436
437
<!-- internal packets: -->
438
<value name="IN_IB_PREFETCH_END" value="0x17" varset="chip" variants="A2XX"/>
439
<value name="IN_SUBBLK_PREFETCH" value="0x1f" varset="chip" variants="A2XX"/>
440
<value name="IN_INSTR_PREFETCH" value="0x20" varset="chip" variants="A2XX"/>
441
<value name="IN_INSTR_MATCH" value="0x47" varset="chip" variants="A2XX"/>
442
<value name="IN_CONST_PREFETCH" value="0x49" varset="chip" variants="A2XX"/>
443
<value name="IN_INCR_UPDT_STATE" value="0x55" varset="chip" variants="A2XX"/>
444
<value name="IN_INCR_UPDT_CONST" value="0x56" varset="chip" variants="A2XX"/>
445
<value name="IN_INCR_UPDT_INSTR" value="0x57" varset="chip" variants="A2XX"/>
446
447
<!-- jmptable entry used to handle type4 packet on a5xx+: -->
448
<value name="PKT4" value="0x04" varset="chip" variants="A5XX-"/>
449
450
<!-- TODO do these exist on A5xx? -->
451
<value name="CP_SCRATCH_WRITE" value="0x4c" varset="chip" variants="A6XX"/>
452
<value name="CP_REG_TO_MEM_OFFSET_MEM" value="0x74" varset="chip" variants="A6XX"/>
453
<value name="CP_REG_TO_MEM_OFFSET_REG" value="0x72" varset="chip" variants="A6XX"/>
454
<value name="CP_WAIT_MEM_GTE" value="0x14" varset="chip" variants="A6XX"/>
455
<value name="CP_WAIT_TWO_REGS" value="0x70" varset="chip" variants="A6XX"/>
456
<value name="CP_MEMCPY" value="0x75" varset="chip" variants="A6XX"/>
457
<value name="CP_SET_BIN_DATA5_OFFSET" value="0x2e" varset="chip" variants="A6XX"/>
458
<value name="CP_SET_CTXSWITCH_IB" value="0x55" varset="chip" variants="A6XX"/>
459
460
<!--
461
Seems to always have the payload:
462
00000002 00008801 00004010
463
or:
464
00000002 00008801 00004090
465
or:
466
00000002 00008801 00000010
467
00000002 00008801 00010010
468
00000002 00008801 00d64010
469
...
470
Note set for compute shaders..
471
Is 0x8801 a register offset?
472
This appears to be a special sort of register write packet
473
more or less, but the firmware has some special handling..
474
Seems like it intercepts/modifies certain register offsets,
475
but others are treated like a normal PKT4 reg write. I
476
guess there are some registers that the fw controls certain
477
bits.
478
-->
479
<value name="CP_REG_WRITE" value="0x6d" varset="chip" variants="A6XX"/>
480
481
</enum>
482
483
484
<domain name="CP_LOAD_STATE" width="32">
485
<doc>Load state, a3xx (and later?)</doc>
486
<enum name="adreno_state_block">
487
<value name="SB_VERT_TEX" value="0"/>
488
<value name="SB_VERT_MIPADDR" value="1"/>
489
<value name="SB_FRAG_TEX" value="2"/>
490
<value name="SB_FRAG_MIPADDR" value="3"/>
491
<value name="SB_VERT_SHADER" value="4"/>
492
<value name="SB_GEOM_SHADER" value="5"/>
493
<value name="SB_FRAG_SHADER" value="6"/>
494
<value name="SB_COMPUTE_SHADER" value="7"/>
495
</enum>
496
<enum name="adreno_state_type">
497
<value name="ST_SHADER" value="0"/>
498
<value name="ST_CONSTANTS" value="1"/>
499
</enum>
500
<enum name="adreno_state_src">
501
<value name="SS_DIRECT" value="0">
502
<doc>inline with the CP_LOAD_STATE packet</doc>
503
</value>
504
<value name="SS_INVALID_ALL_IC" value="2"/>
505
<value name="SS_INVALID_PART_IC" value="3"/>
506
<value name="SS_INDIRECT" value="4">
507
<doc>in buffer pointed to by EXT_SRC_ADDR</doc>
508
</value>
509
<value name="SS_INDIRECT_TCM" value="5"/>
510
<value name="SS_INDIRECT_STM" value="6"/>
511
</enum>
512
<reg32 offset="0" name="0">
513
<bitfield name="DST_OFF" low="0" high="15" type="uint"/>
514
<bitfield name="STATE_SRC" low="16" high="18" type="adreno_state_src"/>
515
<bitfield name="STATE_BLOCK" low="19" high="21" type="adreno_state_block"/>
516
<bitfield name="NUM_UNIT" low="22" high="31" type="uint"/>
517
</reg32>
518
<reg32 offset="1" name="1">
519
<bitfield name="STATE_TYPE" low="0" high="1" type="adreno_state_type"/>
520
<bitfield name="EXT_SRC_ADDR" low="2" high="31" shr="2"/>
521
</reg32>
522
</domain>
523
524
<domain name="CP_LOAD_STATE4" width="32" varset="chip">
525
<doc>Load state, a4xx+</doc>
526
<enum name="a4xx_state_block">
527
<!--
528
unknown: 0x7 and 0xf <- seen in compute shader
529
530
STATE_BLOCK = 0x6, STATE_TYPE = 0x2 possibly used for preemption?
531
Seen in some GL shaders. Payload is NUM_UNIT dwords, and it contains
532
the gpuaddr of the following shader constants block. DST_OFF seems
533
to specify which shader stage:
534
535
16 -> vert
536
36 -> tcs
537
56 -> tes
538
76 -> geom
539
96 -> frag
540
541
Example:
542
543
opcode: CP_LOAD_STATE4 (30) (12 dwords)
544
{ DST_OFF = 16 | STATE_SRC = SS4_DIRECT | STATE_BLOCK = 0x6 | NUM_UNIT = 4 }
545
{ STATE_TYPE = 0x2 | EXT_SRC_ADDR = 0 }
546
{ EXT_SRC_ADDR_HI = 0 }
547
0000: c0264100 00000000 00000000 00000000
548
0000: 70b0000b 01180010 00000002 00000000 c0264100 00000000 00000000 00000000
549
550
opcode: CP_LOAD_STATE4 (30) (4 dwords)
551
{ DST_OFF = 16 | STATE_SRC = SS4_INDIRECT | STATE_BLOCK = SB4_VS_SHADER | NUM_UNIT = 4 }
552
{ STATE_TYPE = ST4_CONSTANTS | EXT_SRC_ADDR = 0xc0264100 }
553
{ EXT_SRC_ADDR_HI = 0 }
554
0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
555
0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
556
0000: 00000040 0000000c 00000000 00000000 00000000 00000000 00000000 00000000
557
558
STATE_BLOCK = 0x6, STATE_TYPE = 0x1, seen in compute shader. NUM_UNITS * 2 dwords.
559
560
-->
561
<value name="SB4_VS_TEX" value="0x0"/>
562
<value name="SB4_HS_TEX" value="0x1"/> <!-- aka. TCS -->
563
<value name="SB4_DS_TEX" value="0x2"/> <!-- aka. TES -->
564
<value name="SB4_GS_TEX" value="0x3"/>
565
<value name="SB4_FS_TEX" value="0x4"/>
566
<value name="SB4_CS_TEX" value="0x5"/>
567
<value name="SB4_VS_SHADER" value="0x8"/>
568
<value name="SB4_HS_SHADER" value="0x9"/>
569
<value name="SB4_DS_SHADER" value="0xa"/>
570
<value name="SB4_GS_SHADER" value="0xb"/>
571
<value name="SB4_FS_SHADER" value="0xc"/>
572
<value name="SB4_CS_SHADER" value="0xd"/>
573
<!--
574
for SSBO, STATE_TYPE=0 appears to be addresses (four dwords each),
575
STATE_TYPE=1 sizes, STATE_TYPE=2 addresses again (two dwords each)
576
577
Compute has it's own dedicated SSBO state, it seems, but the rest
578
of the stages share state
579
-->
580
<value name="SB4_SSBO" value="0xe"/>
581
<value name="SB4_CS_SSBO" value="0xf"/>
582
</enum>
583
<enum name="a4xx_state_type">
584
<value name="ST4_SHADER" value="0"/>
585
<value name="ST4_CONSTANTS" value="1"/>
586
<value name="ST4_UBO" value="2"/>
587
</enum>
588
<enum name="a4xx_state_src">
589
<value name="SS4_DIRECT" value="0"/>
590
<value name="SS4_INDIRECT" value="2"/>
591
</enum>
592
<reg32 offset="0" name="0">
593
<bitfield name="DST_OFF" low="0" high="13" type="uint"/>
594
<bitfield name="STATE_SRC" low="16" high="17" type="a4xx_state_src"/>
595
<bitfield name="STATE_BLOCK" low="18" high="21" type="a4xx_state_block"/>
596
<bitfield name="NUM_UNIT" low="22" high="31" type="uint"/>
597
</reg32>
598
<reg32 offset="1" name="1">
599
<bitfield name="STATE_TYPE" low="0" high="1" type="a4xx_state_type"/>
600
<bitfield name="EXT_SRC_ADDR" low="2" high="31" shr="2"/>
601
</reg32>
602
<reg32 offset="2" name="2" varset="chip" variants="A5XX-">
603
<bitfield name="EXT_SRC_ADDR_HI" low="0" high="31" shr="0"/>
604
</reg32>
605
</domain>
606
607
<!-- looks basically same CP_LOAD_STATE4 -->
608
<domain name="CP_LOAD_STATE6" width="32" varset="chip">
609
<doc>Load state, a6xx+</doc>
610
<enum name="a6xx_state_block">
611
<value name="SB6_VS_TEX" value="0x0"/>
612
<value name="SB6_HS_TEX" value="0x1"/> <!-- aka. TCS -->
613
<value name="SB6_DS_TEX" value="0x2"/> <!-- aka. TES -->
614
<value name="SB6_GS_TEX" value="0x3"/>
615
<value name="SB6_FS_TEX" value="0x4"/>
616
<value name="SB6_CS_TEX" value="0x5"/>
617
<value name="SB6_VS_SHADER" value="0x8"/>
618
<value name="SB6_HS_SHADER" value="0x9"/>
619
<value name="SB6_DS_SHADER" value="0xa"/>
620
<value name="SB6_GS_SHADER" value="0xb"/>
621
<value name="SB6_FS_SHADER" value="0xc"/>
622
<value name="SB6_CS_SHADER" value="0xd"/>
623
<value name="SB6_IBO" value="0xe"/>
624
<value name="SB6_CS_IBO" value="0xf"/>
625
</enum>
626
<enum name="a6xx_state_type">
627
<value name="ST6_SHADER" value="0"/>
628
<value name="ST6_CONSTANTS" value="1"/>
629
<value name="ST6_UBO" value="2"/>
630
<value name="ST6_IBO" value="3"/>
631
</enum>
632
<enum name="a6xx_state_src">
633
<value name="SS6_DIRECT" value="0"/>
634
<value name="SS6_BINDLESS" value="1"/> <!-- TODO does this exist on a4xx/a5xx? -->
635
<value name="SS6_INDIRECT" value="2"/>
636
<doc>
637
SS6_UBO used by the a6xx vulkan blob with tesselation constants
638
in this case, EXT_SRC_ADDR is (ubo_id shl 16 | offset)
639
to load constants from a UBO loaded with DST_OFF = 14 and offset 0,
640
EXT_SRC_ADDR = 0xe0000
641
(offset is a guess, should be in bytes given that maxUniformBufferRange=64k)
642
</doc>
643
<value name="SS6_UBO" value="3"/>
644
</enum>
645
<reg32 offset="0" name="0">
646
<bitfield name="DST_OFF" low="0" high="13" type="uint"/>
647
<bitfield name="STATE_TYPE" low="14" high="15" type="a6xx_state_type"/>
648
<bitfield name="STATE_SRC" low="16" high="17" type="a6xx_state_src"/>
649
<bitfield name="STATE_BLOCK" low="18" high="21" type="a6xx_state_block"/>
650
<bitfield name="NUM_UNIT" low="22" high="31" type="uint"/>
651
</reg32>
652
<reg32 offset="1" name="1">
653
<bitfield name="EXT_SRC_ADDR" low="2" high="31" shr="2"/>
654
</reg32>
655
<reg32 offset="2" name="2">
656
<bitfield name="EXT_SRC_ADDR_HI" low="0" high="31" shr="0"/>
657
</reg32>
658
<reg64 offset="1" name="EXT_SRC_ADDR" type="address"/>
659
</domain>
660
661
<bitset name="vgt_draw_initiator" inline="yes">
662
<bitfield name="PRIM_TYPE" low="0" high="5" type="pc_di_primtype"/>
663
<bitfield name="SOURCE_SELECT" low="6" high="7" type="pc_di_src_sel"/>
664
<bitfield name="VIS_CULL" low="9" high="10" type="pc_di_vis_cull_mode"/>
665
<bitfield name="INDEX_SIZE" pos="11" type="pc_di_index_size"/>
666
<bitfield name="NOT_EOP" pos="12" type="boolean"/>
667
<bitfield name="SMALL_INDEX" pos="13" type="boolean"/>
668
<bitfield name="PRE_DRAW_INITIATOR_ENABLE" pos="14" type="boolean"/>
669
<bitfield name="NUM_INSTANCES" low="24" high="31" type="uint"/>
670
</bitset>
671
672
<!-- changed on a4xx: -->
673
<enum name="a4xx_index_size">
674
<value name="INDEX4_SIZE_8_BIT" value="0"/>
675
<value name="INDEX4_SIZE_16_BIT" value="1"/>
676
<value name="INDEX4_SIZE_32_BIT" value="2"/>
677
</enum>
678
679
<enum name="a6xx_patch_type">
680
<value name="TESS_QUADS" value="0"/>
681
<value name="TESS_TRIANGLES" value="1"/>
682
<value name="TESS_ISOLINES" value="2"/>
683
</enum>
684
685
<bitset name="vgt_draw_initiator_a4xx" inline="yes">
686
<!-- When the 0x20 bit is set, it's the number of patch vertices - 1 -->
687
<bitfield name="PRIM_TYPE" low="0" high="5" type="pc_di_primtype"/>
688
<bitfield name="SOURCE_SELECT" low="6" high="7" type="pc_di_src_sel"/>
689
<bitfield name="VIS_CULL" low="8" high="9" type="pc_di_vis_cull_mode"/>
690
<bitfield name="INDEX_SIZE" low="10" high="11" type="a4xx_index_size"/>
691
<bitfield name="PATCH_TYPE" low="12" high="13" type="a6xx_patch_type"/>
692
<bitfield name="GS_ENABLE" pos="16" type="boolean"/>
693
<bitfield name="TESS_ENABLE" pos="17" type="boolean"/>
694
</bitset>
695
696
<domain name="CP_DRAW_INDX" width="32">
697
<reg32 offset="0" name="0">
698
<bitfield name="VIZ_QUERY" low="0" high="31"/>
699
</reg32>
700
<reg32 offset="1" name="1" type="vgt_draw_initiator"/>
701
<reg32 offset="2" name="2">
702
<bitfield name="NUM_INDICES" low="0" high="31" type="uint"/>
703
</reg32>
704
<reg32 offset="3" name="3">
705
<bitfield name="INDX_BASE" low="0" high="31"/>
706
</reg32>
707
<reg32 offset="4" name="4">
708
<bitfield name="INDX_SIZE" low="0" high="31"/>
709
</reg32>
710
</domain>
711
712
<domain name="CP_DRAW_INDX_2" width="32">
713
<reg32 offset="0" name="0">
714
<bitfield name="VIZ_QUERY" low="0" high="31"/>
715
</reg32>
716
<reg32 offset="1" name="1" type="vgt_draw_initiator"/>
717
<reg32 offset="2" name="2">
718
<bitfield name="NUM_INDICES" low="0" high="31" type="uint"/>
719
</reg32>
720
<!-- followed by NUM_INDICES indices.. -->
721
</domain>
722
723
<domain name="CP_DRAW_INDX_OFFSET" width="32">
724
<reg32 offset="0" name="0" type="vgt_draw_initiator_a4xx"/>
725
<reg32 offset="1" name="1">
726
<bitfield name="NUM_INSTANCES" low="0" high="31" type="uint"/>
727
</reg32>
728
<reg32 offset="2" name="2">
729
<bitfield name="NUM_INDICES" low="0" high="31" type="uint"/>
730
</reg32>
731
<reg32 offset="3" name="3">
732
<bitfield name="FIRST_INDX" low="0" high="31"/>
733
</reg32>
734
735
<stripe varset="chip" variants="A5XX-">
736
<reg32 offset="4" name="4">
737
<bitfield name="INDX_BASE_LO" low="0" high="31"/>
738
</reg32>
739
<reg32 offset="5" name="5">
740
<bitfield name="INDX_BASE_HI" low="0" high="31"/>
741
</reg32>
742
<reg64 offset="4" name="INDX_BASE" type="address"/>
743
<reg32 offset="6" name="6">
744
<!-- max # of elements in index buffer -->
745
<bitfield name="MAX_INDICES" low="0" high="31"/>
746
</reg32>
747
</stripe>
748
749
<reg32 offset="4" name="4">
750
<bitfield name="INDX_BASE" low="0" high="31" type="address"/>
751
</reg32>
752
753
<reg32 offset="5" name="5">
754
<bitfield name="INDX_SIZE" low="0" high="31" type="uint"/>
755
</reg32>
756
</domain>
757
758
<domain name="CP_DRAW_INDIRECT" width="32" varset="chip" prefix="chip" variants="A4XX-">
759
<reg32 offset="0" name="0" type="vgt_draw_initiator_a4xx"/>
760
<stripe varset="chip" variants="A4XX">
761
<reg32 offset="1" name="1">
762
<bitfield name="INDIRECT" low="0" high="31"/>
763
</reg32>
764
</stripe>
765
<stripe varset="chip" variants="A5XX-">
766
<reg32 offset="1" name="1">
767
<bitfield name="INDIRECT_LO" low="0" high="31"/>
768
</reg32>
769
<reg32 offset="2" name="2">
770
<bitfield name="INDIRECT_HI" low="0" high="31"/>
771
</reg32>
772
<reg64 offset="1" name="INDIRECT" type="address"/>
773
</stripe>
774
</domain>
775
776
<domain name="CP_DRAW_INDX_INDIRECT" width="32" varset="chip" prefix="chip" variants="A4XX-">
777
<reg32 offset="0" name="0" type="vgt_draw_initiator_a4xx"/>
778
<stripe varset="chip" variants="A4XX">
779
<reg32 offset="1" name="1">
780
<bitfield name="INDX_BASE" low="0" high="31"/>
781
</reg32>
782
<reg32 offset="2" name="2">
783
<!-- max # of bytes in index buffer -->
784
<bitfield name="INDX_SIZE" low="0" high="31" type="uint"/>
785
</reg32>
786
<reg32 offset="3" name="3">
787
<bitfield name="INDIRECT" low="0" high="31"/>
788
</reg32>
789
</stripe>
790
<stripe varset="chip" variants="A5XX-">
791
<reg32 offset="1" name="1">
792
<bitfield name="INDX_BASE_LO" low="0" high="31"/>
793
</reg32>
794
<reg32 offset="2" name="2">
795
<bitfield name="INDX_BASE_HI" low="0" high="31"/>
796
</reg32>
797
<reg64 offset="1" name="INDX_BASE" type="address"/>
798
<reg32 offset="3" name="3">
799
<!-- max # of elements in index buffer -->
800
<bitfield name="MAX_INDICES" low="0" high="31" type="uint"/>
801
</reg32>
802
<reg32 offset="4" name="4">
803
<bitfield name="INDIRECT_LO" low="0" high="31"/>
804
</reg32>
805
<reg32 offset="5" name="5">
806
<bitfield name="INDIRECT_HI" low="0" high="31"/>
807
</reg32>
808
<reg64 offset="4" name="INDIRECT" type="address"/>
809
</stripe>
810
</domain>
811
812
<domain name="CP_DRAW_INDIRECT_MULTI" width="32" varset="chip" prefix="chip" variants="A6XX-">
813
<enum name="a6xx_draw_indirect_opcode">
814
<value name="INDIRECT_OP_NORMAL" value="0x2"/>
815
<value name="INDIRECT_OP_INDEXED" value="0x4"/>
816
<value name="INDIRECT_OP_INDIRECT_COUNT" value="0x6"/>
817
<value name="INDIRECT_OP_INDIRECT_COUNT_INDEXED" value="0x7"/>
818
</enum>
819
<reg32 offset="0" name="0" type="vgt_draw_initiator_a4xx"/>
820
<reg32 offset="1" name="1">
821
<bitfield name="OPCODE" low="0" high="3" type="a6xx_draw_indirect_opcode" addvariant="yes"/>
822
<doc>
823
DST_OFF same as in CP_LOAD_STATE6 - vec4 VS const at this offset will
824
be updated for each draw to {draw_id, first_vertex, first_instance, 0}
825
value of 0 disables it
826
</doc>
827
<bitfield name="DST_OFF" low="8" high="21" type="hex"/>
828
</reg32>
829
<reg32 offset="2" name="DRAW_COUNT" type="uint"/>
830
<stripe varset="a6xx_draw_indirect_opcode" variants="INDIRECT_OP_NORMAL">
831
<reg64 offset="3" name="INDIRECT" type="address"/>
832
<reg32 offset="5" name="STRIDE" type="uint"/>
833
</stripe>
834
<stripe varset="a6xx_draw_indirect_opcode" variants="INDIRECT_OP_INDEXED" prefix="INDEXED">
835
<reg64 offset="3" name="INDEX" type="address"/>
836
<reg32 offset="5" name="MAX_INDICES" type="uint"/>
837
<reg64 offset="6" name="INDIRECT" type="address"/>
838
<reg32 offset="8" name="STRIDE" type="uint"/>
839
</stripe>
840
<stripe varset="a6xx_draw_indirect_opcode" variants="INDIRECT_OP_INDIRECT_COUNT" prefix="INDIRECT">
841
<reg64 offset="3" name="INDIRECT" type="address"/>
842
<reg64 offset="5" name="INDIRECT_COUNT" type="address"/>
843
<reg32 offset="7" name="STRIDE" type="uint"/>
844
</stripe>
845
<stripe varset="a6xx_draw_indirect_opcode" variants="INDIRECT_OP_INDIRECT_COUNT_INDEXED" prefix="INDIRECT_INDEXED">
846
<reg64 offset="3" name="INDEX" type="address"/>
847
<reg32 offset="5" name="MAX_INDICES" type="uint"/>
848
<reg64 offset="6" name="INDIRECT" type="address"/>
849
<reg64 offset="8" name="INDIRECT_COUNT" type="address"/>
850
<reg32 offset="10" name="STRIDE" type="uint"/>
851
</stripe>
852
</domain>
853
854
<domain name="CP_DRAW_PRED_ENABLE_GLOBAL" width="32" varset="chip">
855
<reg32 offset="0" name="0">
856
<bitfield name="ENABLE" pos="0" type="boolean"/>
857
</reg32>
858
</domain>
859
860
<domain name="CP_DRAW_PRED_ENABLE_LOCAL" width="32" varset="chip">
861
<reg32 offset="0" name="0">
862
<bitfield name="ENABLE" pos="0" type="boolean"/>
863
</reg32>
864
</domain>
865
866
<domain name="CP_DRAW_PRED_SET" width="32" varset="chip">
867
<enum name="cp_draw_pred_src">
868
<!--
869
Sources 1-4 seem to be about combining reading
870
SO/primitive queries and setting the predicate, which is
871
a DX11-specific optimization (since in DX11 you can only
872
predicate on the result of queries).
873
-->
874
<value name="PRED_SRC_MEM" value="5">
875
<doc>
876
Read a 64-bit value at the given address and
877
test if it equals/doesn't equal 0.
878
</doc>
879
</value>
880
</enum>
881
<enum name="cp_draw_pred_test">
882
<value name="NE_0_PASS" value="0"/>
883
<value name="EQ_0_PASS" value="1"/>
884
</enum>
885
<reg32 offset="0" name="0">
886
<bitfield name="SRC" low="4" high="7" type="cp_draw_pred_src"/>
887
<bitfield name="TEST" pos="8" type="cp_draw_pred_test"/>
888
</reg32>
889
<reg64 offset="1" name="MEM_ADDR" type="address"/>
890
</domain>
891
892
<domain name="CP_SET_DRAW_STATE" width="32" varset="chip" variants="A4XX-">
893
<array offset="0" stride="3" length="100">
894
<reg32 offset="0" name="0">
895
<bitfield name="COUNT" low="0" high="15" type="uint"/>
896
<bitfield name="DIRTY" pos="16" type="boolean"/>
897
<bitfield name="DISABLE" pos="17" type="boolean"/>
898
<bitfield name="DISABLE_ALL_GROUPS" pos="18" type="boolean"/>
899
<bitfield name="LOAD_IMMED" pos="19" type="boolean"/>
900
<bitfield name="BINNING" pos="20" varset="chip" variants="A6XX-" type="boolean"/>
901
<bitfield name="GMEM" pos="21" varset="chip" variants="A6XX-" type="boolean"/>
902
<bitfield name="SYSMEM" pos="22" varset="chip" variants="A6XX-" type="boolean"/>
903
<bitfield name="GROUP_ID" low="24" high="28" type="uint"/>
904
</reg32>
905
<reg32 offset="1" name="1">
906
<bitfield name="ADDR_LO" low="0" high="31" type="hex"/>
907
</reg32>
908
<reg32 offset="2" name="2" varset="chip" variants="A5XX-">
909
<bitfield name="ADDR_HI" low="0" high="31" type="hex"/>
910
</reg32>
911
</array>
912
</domain>
913
914
<domain name="CP_SET_BIN" width="32">
915
<doc>value at offset 0 always seems to be 0x00000000..</doc>
916
<reg32 offset="0" name="0"/>
917
<reg32 offset="1" name="1">
918
<bitfield name="X1" low="0" high="15" type="uint"/>
919
<bitfield name="Y1" low="16" high="31" type="uint"/>
920
</reg32>
921
<reg32 offset="2" name="2">
922
<bitfield name="X2" low="0" high="15" type="uint"/>
923
<bitfield name="Y2" low="16" high="31" type="uint"/>
924
</reg32>
925
</domain>
926
927
<domain name="CP_SET_BIN_DATA" width="32">
928
<reg32 offset="0" name="0">
929
<!-- corresponds to VSC_PIPE[n].DATA_ADDR -->
930
<bitfield name="BIN_DATA_ADDR" low="0" high="31" type="hex"/>
931
</reg32>
932
<reg32 offset="1" name="1">
933
<!-- seesm to correspond to VSC_SIZE_ADDRESS -->
934
<bitfield name="BIN_SIZE_ADDRESS" low="0" high="31"/>
935
</reg32>
936
</domain>
937
938
<domain name="CP_SET_BIN_DATA5" width="32">
939
<reg32 offset="0" name="0">
940
<!-- equiv to PC_VSTREAM_CONTROL.SIZE on a3xx/a4xx: -->
941
<bitfield name="VSC_SIZE" low="16" high="21" type="uint"/>
942
<!-- equiv to PC_VSTREAM_CONTROL.N on a3xx/a4xx: -->
943
<bitfield name="VSC_N" low="22" high="26" type="uint"/>
944
</reg32>
945
<!-- BIN_DATA_ADDR -> VSC_PIPE[p].DATA_ADDRESS -->
946
<reg32 offset="1" name="1">
947
<bitfield name="BIN_DATA_ADDR_LO" low="0" high="31" type="hex"/>
948
</reg32>
949
<reg32 offset="2" name="2">
950
<bitfield name="BIN_DATA_ADDR_HI" low="0" high="31" type="hex"/>
951
</reg32>
952
<!-- BIN_SIZE_ADDRESS -> VSC_SIZE_ADDRESS + (p * 4)-->
953
<reg32 offset="3" name="3">
954
<bitfield name="BIN_SIZE_ADDRESS_LO" low="0" high="31"/>
955
</reg32>
956
<reg32 offset="4" name="4">
957
<bitfield name="BIN_SIZE_ADDRESS_HI" low="0" high="31"/>
958
</reg32>
959
<!-- new on a6xx, where BIN_DATA_ADDR is the DRAW_STRM: -->
960
<reg32 offset="5" name="5">
961
<bitfield name="BIN_PRIM_STRM_LO" low="0" high="31"/>
962
</reg32>
963
<reg32 offset="6" name="6">
964
<bitfield name="BIN_PRIM_STRM_HI" low="0" high="31"/>
965
</reg32>
966
</domain>
967
968
<domain name="CP_SET_BIN_DATA5_OFFSET" width="32">
969
<doc>
970
Like CP_SET_BIN_DATA5, but set the pointers as offsets from the
971
pointers stored in VSC_PIPE_{DATA,DATA2,SIZE}_ADDRESS. Useful
972
for Vulkan where these values aren't known when the command
973
stream is recorded.
974
</doc>
975
<reg32 offset="0" name="0">
976
<!-- equiv to PC_VSTREAM_CONTROL.SIZE on a3xx/a4xx: -->
977
<bitfield name="VSC_SIZE" low="16" high="21" type="uint"/>
978
<!-- equiv to PC_VSTREAM_CONTROL.N on a3xx/a4xx: -->
979
<bitfield name="VSC_N" low="22" high="26" type="uint"/>
980
</reg32>
981
<!-- BIN_DATA_ADDR -> VSC_PIPE[p].DATA_ADDRESS -->
982
<reg32 offset="1" name="1">
983
<bitfield name="BIN_DATA_OFFSET" low="0" high="31" type="uint"/>
984
</reg32>
985
<!-- BIN_SIZE_ADDRESS -> VSC_SIZE_ADDRESS + (p * 4)-->
986
<reg32 offset="2" name="2">
987
<bitfield name="BIN_SIZE_OFFSET" low="0" high="31" type="uint"/>
988
</reg32>
989
<!-- BIN_DATA2_ADDR -> VSC_PIPE[p].DATA2_ADDRESS -->
990
<reg32 offset="3" name="3">
991
<bitfield name="BIN_DATA2_OFFSET" low="0" high="31" type="uint"/>
992
</reg32>
993
</domain>
994
995
<domain name="CP_REG_RMW" width="32">
996
<doc>
997
Modifies DST_REG using two sources that can either be registers
998
or immediates. If SRC1_ADD is set, then do the following:
999
1000
$dst = (($dst &amp; $src0) rot $rotate) + $src1
1001
1002
Otherwise:
1003
1004
$dst = (($dst &amp; $src0) rot $rotate) | $src1
1005
1006
Here "rot" means rotate left.
1007
</doc>
1008
<reg32 offset="0" name="0">
1009
<bitfield name="DST_REG" low="0" high="17" type="hex"/>
1010
<bitfield name="ROTATE" low="24" high="28" type="uint"/>
1011
<bitfield name="SRC1_ADD" pos="29" type="boolean"/>
1012
<bitfield name="SRC1_IS_REG" pos="30" type="boolean"/>
1013
<bitfield name="SRC0_IS_REG" pos="31" type="boolean"/>
1014
</reg32>
1015
<reg32 offset="1" name="1">
1016
<bitfield name="SRC0" low="0" high="31" type="uint"/>
1017
</reg32>
1018
<reg32 offset="2" name="2">
1019
<bitfield name="SRC1" low="0" high="31" type="uint"/>
1020
</reg32>
1021
</domain>
1022
1023
<domain name="CP_REG_TO_MEM" width="32">
1024
<reg32 offset="0" name="0">
1025
<bitfield name="REG" low="0" high="17" type="hex"/>
1026
<!-- number of registers/dwords copied is max(CNT, 1). -->
1027
<bitfield name="CNT" low="18" high="29" type="uint"/>
1028
<bitfield name="64B" pos="30" type="boolean"/>
1029
<bitfield name="ACCUMULATE" pos="31" type="boolean"/>
1030
</reg32>
1031
<reg32 offset="1" name="1">
1032
<bitfield name="DEST" low="0" high="31"/>
1033
</reg32>
1034
<reg32 offset="2" name="2" varset="chip" variants="A5XX-">
1035
<bitfield name="DEST_HI" low="0" high="31"/>
1036
</reg32>
1037
</domain>
1038
1039
<domain name="CP_REG_TO_MEM_OFFSET_REG" width="32">
1040
<doc>
1041
Like CP_REG_TO_MEM, but the memory address to write to can be
1042
offsetted using either one or two registers or scratch
1043
registers.
1044
</doc>
1045
<reg32 offset="0" name="0">
1046
<bitfield name="REG" low="0" high="17" type="hex"/>
1047
<!-- number of registers/dwords copied is max(CNT, 1). -->
1048
<bitfield name="CNT" low="18" high="29" type="uint"/>
1049
<bitfield name="64B" pos="30" type="boolean"/>
1050
<bitfield name="ACCUMULATE" pos="31" type="boolean"/>
1051
</reg32>
1052
<reg32 offset="1" name="1">
1053
<bitfield name="DEST" low="0" high="31"/>
1054
</reg32>
1055
<reg32 offset="2" name="2" varset="chip" variants="A5XX-">
1056
<bitfield name="DEST_HI" low="0" high="31"/>
1057
</reg32>
1058
<reg32 offset="3" name="3">
1059
<bitfield name="OFFSET0" low="0" high="17" type="hex"/>
1060
<bitfield name="OFFSET0_SCRATCH" pos="19" type="boolean"/>
1061
</reg32>
1062
<!-- followed by an optional identical OFFSET1 dword -->
1063
</domain>
1064
1065
<domain name="CP_REG_TO_MEM_OFFSET_MEM" width="32">
1066
<doc>
1067
Like CP_REG_TO_MEM, but the memory address to write to can be
1068
offsetted using a DWORD in memory.
1069
</doc>
1070
<reg32 offset="0" name="0">
1071
<bitfield name="REG" low="0" high="17" type="hex"/>
1072
<!-- number of registers/dwords copied is max(CNT, 1). -->
1073
<bitfield name="CNT" low="18" high="29" type="uint"/>
1074
<bitfield name="64B" pos="30" type="boolean"/>
1075
<bitfield name="ACCUMULATE" pos="31" type="boolean"/>
1076
</reg32>
1077
<reg32 offset="1" name="1">
1078
<bitfield name="DEST" low="0" high="31"/>
1079
</reg32>
1080
<reg32 offset="2" name="2" varset="chip" variants="A5XX-">
1081
<bitfield name="DEST_HI" low="0" high="31"/>
1082
</reg32>
1083
<reg32 offset="3" name="3">
1084
<bitfield name="OFFSET_LO" low="0" high="31" type="hex"/>
1085
</reg32>
1086
<reg32 offset="4" name="4">
1087
<bitfield name="OFFSET_HI" low="0" high="31" type="hex"/>
1088
</reg32>
1089
</domain>
1090
1091
<domain name="CP_MEM_TO_REG" width="32">
1092
<reg32 offset="0" name="0">
1093
<bitfield name="REG" low="0" high="17" type="hex"/>
1094
<!-- number of registers/dwords copied is max(CNT, 1). -->
1095
<bitfield name="CNT" low="19" high="29" type="uint"/>
1096
<!-- shift each DWORD left by 2 while copying -->
1097
<bitfield name="SHIFT_BY_2" pos="30" type="boolean"/>
1098
<!-- does the same thing as CP_MEM_TO_MEM::UNK31 -->
1099
<bitfield name="UNK31" pos="31" type="boolean"/>
1100
</reg32>
1101
<reg32 offset="1" name="1">
1102
<bitfield name="SRC" low="0" high="31"/>
1103
</reg32>
1104
<reg32 offset="2" name="2" varset="chip" variants="A5XX-">
1105
<bitfield name="SRC_HI" low="0" high="31"/>
1106
</reg32>
1107
</domain>
1108
1109
<domain name="CP_MEM_TO_MEM" width="32">
1110
<reg32 offset="0" name="0">
1111
<!--
1112
not sure how many src operands we have, but the low
1113
bits negate the n'th src argument.
1114
-->
1115
<bitfield name="NEG_A" pos="0" type="boolean"/>
1116
<bitfield name="NEG_B" pos="1" type="boolean"/>
1117
<bitfield name="NEG_C" pos="2" type="boolean"/>
1118
1119
<!-- if set treat src/dst as 64bit values -->
1120
<bitfield name="DOUBLE" pos="29" type="boolean"/>
1121
<!-- execute CP_WAIT_FOR_MEM_WRITES beforehand -->
1122
<bitfield name="WAIT_FOR_MEM_WRITES" pos="30" type="boolean"/>
1123
<!-- some other kind of wait -->
1124
<bitfield name="UNK31" pos="31" type="boolean"/>
1125
</reg32>
1126
<!--
1127
followed by sequence of addresses.. the first is the
1128
destination and the rest are N src addresses which are
1129
summed (after being negated if NEG_x bit set) allowing
1130
to do things like 'result += end - start' (which turns
1131
out to be useful for queries and accumulating results
1132
across multiple tiles)
1133
-->
1134
</domain>
1135
1136
<domain name="CP_MEMCPY" width="32">
1137
<reg32 offset="0" name="0">
1138
<bitfield name="DWORDS" low="0" high="31" type="uint"/>
1139
</reg32>
1140
<reg32 offset="1" name="1">
1141
<bitfield name="SRC_LO" low="0" high="31" type="hex"/>
1142
</reg32>
1143
<reg32 offset="2" name="2">
1144
<bitfield name="SRC_HI" low="0" high="31" type="hex"/>
1145
</reg32>
1146
<reg32 offset="3" name="3">
1147
<bitfield name="DST_LO" low="0" high="31" type="hex"/>
1148
</reg32>
1149
<reg32 offset="4" name="4">
1150
<bitfield name="DST_HI" low="0" high="31" type="hex"/>
1151
</reg32>
1152
</domain>
1153
1154
<domain name="CP_REG_TO_SCRATCH" width="32">
1155
<reg32 offset="0" name="0">
1156
<bitfield name="REG" low="0" high="17" type="hex"/>
1157
<bitfield name="SCRATCH" low="20" high="22" type="uint"/>
1158
<!-- number of registers/dwords copied is CNT + 1. -->
1159
<bitfield name="CNT" low="24" high="26" type="uint"/>
1160
</reg32>
1161
</domain>
1162
1163
<domain name="CP_SCRATCH_TO_REG" width="32">
1164
<reg32 offset="0" name="0">
1165
<bitfield name="REG" low="0" high="17" type="hex"/>
1166
<!-- note: CP_MEM_TO_REG always sets this when writing to the register -->
1167
<bitfield name="UNK18" pos="18" type="boolean"/>
1168
<bitfield name="SCRATCH" low="20" high="22" type="uint"/>
1169
<!-- number of registers/dwords copied is CNT + 1. -->
1170
<bitfield name="CNT" low="24" high="26" type="uint"/>
1171
</reg32>
1172
</domain>
1173
1174
<domain name="CP_SCRATCH_WRITE" width="32">
1175
<reg32 offset="0" name="0">
1176
<bitfield name="SCRATCH" low="20" high="22" type="uint"/>
1177
</reg32>
1178
<!-- followed by one or more DWORDs to write to scratch registers -->
1179
</domain>
1180
1181
<domain name="CP_MEM_WRITE" width="32">
1182
<reg32 offset="0" name="0">
1183
<bitfield name="ADDR_LO" low="0" high="31"/>
1184
</reg32>
1185
<reg32 offset="1" name="1">
1186
<bitfield name="ADDR_HI" low="0" high="31"/>
1187
</reg32>
1188
<!-- followed by the DWORDs to write -->
1189
</domain>
1190
1191
<enum name="cp_cond_function">
1192
<value value="0" name="WRITE_ALWAYS"/>
1193
<value value="1" name="WRITE_LT"/>
1194
<value value="2" name="WRITE_LE"/>
1195
<value value="3" name="WRITE_EQ"/>
1196
<value value="4" name="WRITE_NE"/>
1197
<value value="5" name="WRITE_GE"/>
1198
<value value="6" name="WRITE_GT"/>
1199
</enum>
1200
1201
<domain name="CP_COND_WRITE" width="32">
1202
<reg32 offset="0" name="0">
1203
<bitfield name="FUNCTION" low="0" high="2" type="cp_cond_function"/>
1204
<bitfield name="POLL_MEMORY" pos="4" type="boolean"/>
1205
<bitfield name="WRITE_MEMORY" pos="8" type="boolean"/>
1206
</reg32>
1207
<reg32 offset="1" name="1">
1208
<bitfield name="POLL_ADDR" low="0" high="31" type="hex"/>
1209
</reg32>
1210
<reg32 offset="2" name="2">
1211
<bitfield name="REF" low="0" high="31"/>
1212
</reg32>
1213
<reg32 offset="3" name="3">
1214
<bitfield name="MASK" low="0" high="31"/>
1215
</reg32>
1216
<reg32 offset="4" name="4">
1217
<bitfield name="WRITE_ADDR" low="0" high="31" type="hex"/>
1218
</reg32>
1219
<reg32 offset="5" name="5">
1220
<bitfield name="WRITE_DATA" low="0" high="31"/>
1221
</reg32>
1222
</domain>
1223
1224
<domain name="CP_COND_WRITE5" width="32">
1225
<reg32 offset="0" name="0">
1226
<bitfield name="FUNCTION" low="0" high="2" type="cp_cond_function"/>
1227
<bitfield name="SIGNED_COMPARE" pos="3" type="boolean"/>
1228
<!-- if both POLL_MEMORY and POLL_SCRATCH are false, it polls a register at POLL_ADDR_LO instead. -->
1229
<bitfield name="POLL_MEMORY" pos="4" type="boolean"/>
1230
<bitfield name="POLL_SCRATCH" pos="5" type="boolean"/>
1231
<bitfield name="WRITE_MEMORY" pos="8" type="boolean"/>
1232
</reg32>
1233
<reg32 offset="1" name="1">
1234
<bitfield name="POLL_ADDR_LO" low="0" high="31" type="hex"/>
1235
</reg32>
1236
<reg32 offset="2" name="2">
1237
<bitfield name="POLL_ADDR_HI" low="0" high="31" type="hex"/>
1238
</reg32>
1239
<reg32 offset="3" name="3">
1240
<bitfield name="REF" low="0" high="31"/>
1241
</reg32>
1242
<reg32 offset="4" name="4">
1243
<bitfield name="MASK" low="0" high="31"/>
1244
</reg32>
1245
<reg32 offset="5" name="5">
1246
<bitfield name="WRITE_ADDR_LO" low="0" high="31" type="hex"/>
1247
</reg32>
1248
<reg32 offset="6" name="6">
1249
<bitfield name="WRITE_ADDR_HI" low="0" high="31" type="hex"/>
1250
</reg32>
1251
<reg32 offset="7" name="7">
1252
<bitfield name="WRITE_DATA" low="0" high="31"/>
1253
</reg32>
1254
</domain>
1255
1256
<domain name="CP_WAIT_MEM_GTE" width="32">
1257
<doc>
1258
Wait until a memory value is greater than or equal to the
1259
reference, using signed comparison.
1260
</doc>
1261
<reg32 offset="0" name="0">
1262
<!-- Reserved for flags, presumably? Unused in FW -->
1263
<bitfield name="RESERVED" low="0" high="31" type="hex"/>
1264
</reg32>
1265
<reg32 offset="1" name="1">
1266
<bitfield name="POLL_ADDR_LO" low="0" high="31" type="hex"/>
1267
</reg32>
1268
<reg32 offset="2" name="2">
1269
<bitfield name="POLL_ADDR_HI" low="0" high="31" type="hex"/>
1270
</reg32>
1271
<reg32 offset="3" name="3">
1272
<bitfield name="REF" low="0" high="31"/>
1273
</reg32>
1274
</domain>
1275
1276
<domain name="CP_WAIT_REG_MEM" width="32">
1277
<doc>
1278
This uses the same internal comparison as CP_COND_WRITE,
1279
but waits until the comparison is true instead. It busy-loops in
1280
the CP for the given number of cycles before trying again.
1281
</doc>
1282
<reg32 offset="0" name="0">
1283
<bitfield name="FUNCTION" low="0" high="2" type="cp_cond_function"/>
1284
<bitfield name="SIGNED_COMPARE" pos="3" type="boolean"/>
1285
<bitfield name="POLL_MEMORY" pos="4" type="boolean"/>
1286
<bitfield name="POLL_SCRATCH" pos="5" type="boolean"/>
1287
<bitfield name="WRITE_MEMORY" pos="8" type="boolean"/>
1288
</reg32>
1289
<reg32 offset="1" name="1">
1290
<bitfield name="POLL_ADDR_LO" low="0" high="31" type="hex"/>
1291
</reg32>
1292
<reg32 offset="2" name="2">
1293
<bitfield name="POLL_ADDR_HI" low="0" high="31" type="hex"/>
1294
</reg32>
1295
<reg32 offset="3" name="3">
1296
<bitfield name="REF" low="0" high="31"/>
1297
</reg32>
1298
<reg32 offset="4" name="4">
1299
<bitfield name="MASK" low="0" high="31"/>
1300
</reg32>
1301
<reg32 offset="5" name="5">
1302
<bitfield name="DELAY_LOOP_CYCLES" low="0" high="31"/>
1303
</reg32>
1304
</domain>
1305
1306
<domain name="CP_WAIT_TWO_REGS" width="32">
1307
<doc>
1308
Waits for REG0 to not be 0 or REG1 to not equal REF
1309
</doc>
1310
<reg32 offset="0" name="0">
1311
<bitfield name="REG0" low="0" high="17" type="hex"/>
1312
</reg32>
1313
<reg32 offset="1" name="1">
1314
<bitfield name="REG1" low="0" high="17" type="hex"/>
1315
</reg32>
1316
<reg32 offset="2" name="2">
1317
<bitfield name="REF" low="0" high="31" type="uint"/>
1318
</reg32>
1319
</domain>
1320
1321
<domain name="CP_DISPATCH_COMPUTE" width="32">
1322
<reg32 offset="0" name="0"/>
1323
<reg32 offset="1" name="1">
1324
<bitfield name="X" low="0" high="31"/>
1325
</reg32>
1326
<reg32 offset="2" name="2">
1327
<bitfield name="Y" low="0" high="31"/>
1328
</reg32>
1329
<reg32 offset="3" name="3">
1330
<bitfield name="Z" low="0" high="31"/>
1331
</reg32>
1332
</domain>
1333
1334
<domain name="CP_SET_RENDER_MODE" width="32">
1335
<enum name="render_mode_cmd">
1336
<value value="1" name="BYPASS"/>
1337
<value value="2" name="BINNING"/>
1338
<value value="3" name="GMEM"/>
1339
<value value="5" name="BLIT2D"/>
1340
<!-- placeholder name.. used when CP_BLIT packets with BLIT_OP_SCALE?? -->
1341
<value value="7" name="BLIT2DSCALE"/>
1342
<!-- 8 set before going back to BYPASS exiting 2D -->
1343
<value value="8" name="END2D"/>
1344
</enum>
1345
<reg32 offset="0" name="0">
1346
<bitfield name="MODE" low="0" high="8" type="render_mode_cmd"/>
1347
<!--
1348
normally 0x1/0x3, sometimes see 0x5/0x8 with unknown registers in
1349
0x21xx range.. possibly (at least some) a5xx variants have a
1350
2d core?
1351
-->
1352
</reg32>
1353
<!-- I think first buffer is for GPU to save context in case of ctx switch? -->
1354
<reg32 offset="1" name="1">
1355
<bitfield name="ADDR_0_LO" low="0" high="31"/>
1356
</reg32>
1357
<reg32 offset="2" name="2">
1358
<bitfield name="ADDR_0_HI" low="0" high="31"/>
1359
</reg32>
1360
<reg32 offset="3" name="3">
1361
<!--
1362
set when in GMEM.. maybe indicates GMEM contents need to be
1363
preserved on ctx switch?
1364
-->
1365
<bitfield name="VSC_ENABLE" pos="3" type="boolean"/>
1366
<bitfield name="GMEM_ENABLE" pos="4" type="boolean"/>
1367
</reg32>
1368
<reg32 offset="4" name="4"/>
1369
<!-- second buffer looks like some cmdstream.. length in dwords: -->
1370
<reg32 offset="5" name="5">
1371
<bitfield name="ADDR_1_LEN" low="0" high="31" type="uint"/>
1372
</reg32>
1373
<reg32 offset="6" name="6">
1374
<bitfield name="ADDR_1_LO" low="0" high="31"/>
1375
</reg32>
1376
<reg32 offset="7" name="7">
1377
<bitfield name="ADDR_1_HI" low="0" high="31"/>
1378
</reg32>
1379
</domain>
1380
1381
<!-- this looks fairly similar to CP_SET_RENDER_MODE minus first dword -->
1382
<domain name="CP_COMPUTE_CHECKPOINT" width="32">
1383
<!-- I think first buffer is for GPU to save context in case of ctx switch? -->
1384
<reg32 offset="0" name="0">
1385
<bitfield name="ADDR_0_LO" low="0" high="31"/>
1386
</reg32>
1387
<reg32 offset="1" name="1">
1388
<bitfield name="ADDR_0_HI" low="0" high="31"/>
1389
</reg32>
1390
<reg32 offset="2" name="2">
1391
</reg32>
1392
<!-- second buffer looks like some cmdstream.. length in dwords: -->
1393
<reg32 offset="3" name="3">
1394
<bitfield name="ADDR_1_LEN" low="0" high="31" type="uint"/>
1395
</reg32>
1396
<reg32 offset="4" name="4"/>
1397
<reg32 offset="5" name="5">
1398
<bitfield name="ADDR_1_LO" low="0" high="31"/>
1399
</reg32>
1400
<reg32 offset="6" name="6">
1401
<bitfield name="ADDR_1_HI" low="0" high="31"/>
1402
</reg32>
1403
<reg32 offset="7" name="7"/>
1404
</domain>
1405
1406
<domain name="CP_PERFCOUNTER_ACTION" width="32">
1407
<reg32 offset="0" name="0">
1408
</reg32>
1409
<reg32 offset="1" name="1">
1410
<bitfield name="ADDR_0_LO" low="0" high="31"/>
1411
</reg32>
1412
<reg32 offset="2" name="2">
1413
<bitfield name="ADDR_0_HI" low="0" high="31"/>
1414
</reg32>
1415
</domain>
1416
1417
<domain name="CP_EVENT_WRITE" width="32">
1418
<reg32 offset="0" name="0">
1419
<bitfield name="EVENT" low="0" high="7" type="vgt_event_type"/>
1420
<!-- when set, write back timestamp instead of value from packet: -->
1421
<bitfield name="TIMESTAMP" pos="30" type="boolean"/>
1422
<bitfield name="IRQ" pos="31" type="boolean"/>
1423
</reg32>
1424
<!--
1425
TODO what is gpuaddr for, seems to be all 0's.. maybe needed for
1426
context switch?
1427
-->
1428
<reg32 offset="1" name="1">
1429
<bitfield name="ADDR_0_LO" low="0" high="31"/>
1430
</reg32>
1431
<reg32 offset="2" name="2">
1432
<bitfield name="ADDR_0_HI" low="0" high="31"/>
1433
</reg32>
1434
<reg32 offset="3" name="3">
1435
<!-- ??? -->
1436
</reg32>
1437
</domain>
1438
1439
<domain name="CP_BLIT" width="32">
1440
<enum name="cp_blit_cmd">
1441
<value value="0" name="BLIT_OP_FILL"/>
1442
<value value="1" name="BLIT_OP_COPY"/>
1443
<value value="3" name="BLIT_OP_SCALE"/> <!-- used for mipmap generation -->
1444
</enum>
1445
<reg32 offset="0" name="0">
1446
<bitfield name="OP" low="0" high="3" type="cp_blit_cmd"/>
1447
</reg32>
1448
<reg32 offset="1" name="1">
1449
<bitfield name="SRC_X1" low="0" high="13" type="uint"/>
1450
<bitfield name="SRC_Y1" low="16" high="29" type="uint"/>
1451
</reg32>
1452
<reg32 offset="2" name="2">
1453
<bitfield name="SRC_X2" low="0" high="13" type="uint"/>
1454
<bitfield name="SRC_Y2" low="16" high="29" type="uint"/>
1455
</reg32>
1456
<reg32 offset="3" name="3">
1457
<bitfield name="DST_X1" low="0" high="13" type="uint"/>
1458
<bitfield name="DST_Y1" low="16" high="29" type="uint"/>
1459
</reg32>
1460
<reg32 offset="4" name="4">
1461
<bitfield name="DST_X2" low="0" high="13" type="uint"/>
1462
<bitfield name="DST_Y2" low="16" high="29" type="uint"/>
1463
</reg32>
1464
</domain>
1465
1466
<domain name="CP_EXEC_CS" width="32">
1467
<reg32 offset="0" name="0">
1468
</reg32>
1469
<reg32 offset="1" name="1">
1470
<bitfield name="NGROUPS_X" low="0" high="31" type="uint"/>
1471
</reg32>
1472
<reg32 offset="2" name="2">
1473
<bitfield name="NGROUPS_Y" low="0" high="31" type="uint"/>
1474
</reg32>
1475
<reg32 offset="3" name="3">
1476
<bitfield name="NGROUPS_Z" low="0" high="31" type="uint"/>
1477
</reg32>
1478
</domain>
1479
1480
<domain name="CP_EXEC_CS_INDIRECT" width="32" varset="chip" prefix="chip" variants="A4XX-">
1481
<reg32 offset="0" name="0">
1482
</reg32>
1483
<stripe varset="chip" variants="A4XX">
1484
<reg32 offset="1" name="1">
1485
<bitfield name="ADDR" low="0" high="31"/>
1486
</reg32>
1487
<reg32 offset="2" name="2">
1488
<!-- localsize is value minus one: -->
1489
<bitfield name="LOCALSIZEX" low="2" high="11" type="uint"/>
1490
<bitfield name="LOCALSIZEY" low="12" high="21" type="uint"/>
1491
<bitfield name="LOCALSIZEZ" low="22" high="31" type="uint"/>
1492
</reg32>
1493
</stripe>
1494
<stripe varset="chip" variants="A5XX-">
1495
<reg32 offset="1" name="1">
1496
<bitfield name="ADDR_LO" low="0" high="31"/>
1497
</reg32>
1498
<reg32 offset="2" name="2">
1499
<bitfield name="ADDR_HI" low="0" high="31"/>
1500
</reg32>
1501
<reg32 offset="3" name="3">
1502
<!-- localsize is value minus one: -->
1503
<bitfield name="LOCALSIZEX" low="2" high="11" type="uint"/>
1504
<bitfield name="LOCALSIZEY" low="12" high="21" type="uint"/>
1505
<bitfield name="LOCALSIZEZ" low="22" high="31" type="uint"/>
1506
</reg32>
1507
</stripe>
1508
</domain>
1509
1510
<domain name="CP_SET_MARKER" width="32" varset="chip" prefix="chip" variants="A6XX-">
1511
<doc>Tell CP the current operation mode, indicates save and restore procedure</doc>
1512
<enum name="a6xx_render_mode">
1513
<value value="1" name="RM6_BYPASS"/>
1514
<value value="2" name="RM6_BINNING"/>
1515
<value value="4" name="RM6_GMEM"/>
1516
<value value="5" name="RM6_ENDVIS"/>
1517
<value value="6" name="RM6_RESOLVE"/>
1518
<value value="7" name="RM6_YIELD"/>
1519
<value value="8" name="RM6_COMPUTE"/>
1520
<value value="0xc" name="RM6_BLIT2DSCALE"/> <!-- no-op (at least on current sqe fw) -->
1521
1522
<!--
1523
These values come from a6xx_set_marker() in the
1524
downstream kernel, and they can only be set by the kernel
1525
-->
1526
<value value="0xd" name="RM6_IB1LIST_START"/>
1527
<value value="0xe" name="RM6_IB1LIST_END"/>
1528
<!-- IFPC - inter-frame power collapse -->
1529
<value value="0x100" name="RM6_IFPC_ENABLE"/>
1530
<value value="0x101" name="RM6_IFPC_DISABLE"/>
1531
</enum>
1532
<reg32 offset="0" name="0">
1533
<!--
1534
NOTE: blob driver and some versions of freedreno/turnip set
1535
b4, which is unused (at least by current sqe fw), but interferes
1536
with parsing if we extend the size of the bitfield to include
1537
b8 (only sent by kernel mode driver). Really, the way the
1538
parsing works in the firmware, only b0-b3 are considered, but
1539
if b8 is set, the low bits are interpreted differently. To
1540
model this, without getting confused by spurious b4, this is
1541
described as two overlapping bitfields:
1542
-->
1543
<bitfield name="MODE" low="0" high="8" type="a6xx_render_mode"/>
1544
<bitfield name="MARKER" low="0" high="3" type="a6xx_render_mode"/>
1545
</reg32>
1546
</domain>
1547
1548
<domain name="CP_SET_PSEUDO_REG" width="32" varset="chip" prefix="chip" variants="A6XX-">
1549
<doc>Set internal CP registers, used to indicate context save data addresses</doc>
1550
<enum name="pseudo_reg">
1551
<value value="0" name="SMMU_INFO"/>
1552
<value value="1" name="NON_SECURE_SAVE_ADDR"/>
1553
<value value="2" name="SECURE_SAVE_ADDR"/>
1554
<value value="3" name="NON_PRIV_SAVE_ADDR"/>
1555
<value value="4" name="COUNTER"/>
1556
</enum>
1557
<array offset="0" stride="3" length="100">
1558
<reg32 offset="0" name="0">
1559
<bitfield name="PSEUDO_REG" low="0" high="2" type="pseudo_reg"/>
1560
</reg32>
1561
<reg32 offset="1" name="1">
1562
<bitfield name="LO" low="0" high="31"/>
1563
</reg32>
1564
<reg32 offset="2" name="2">
1565
<bitfield name="HI" low="0" high="31"/>
1566
</reg32>
1567
</array>
1568
</domain>
1569
1570
<domain name="CP_REG_TEST" width="32" varset="chip" prefix="chip" variants="A6XX-">
1571
<doc>
1572
Tests bit in specified register and sets predicate for CP_COND_REG_EXEC.
1573
So:
1574
1575
opcode: CP_REG_TEST (39) (2 dwords)
1576
{ REG = 0xc10 | BIT = 0 }
1577
0000: 70b90001 00000c10
1578
opcode: CP_COND_REG_EXEC (47) (3 dwords)
1579
0000: 70c70002 10000000 00000004
1580
opcode: CP_INDIRECT_BUFFER (3f) (4 dwords)
1581
1582
Will execute the CP_INDIRECT_BUFFER only if b0 in the register at
1583
offset 0x0c10 is 1
1584
</doc>
1585
<reg32 offset="0" name="0">
1586
<!-- the register to test -->
1587
<bitfield name="REG" low="0" high="17"/>
1588
<!-- the bit to test -->
1589
<bitfield name="BIT" low="20" high="24" type="uint"/>
1590
<!-- execute CP_WAIT_FOR_ME beforehand -->
1591
<bitfield name="WAIT_FOR_ME" pos="25" type="boolean"/>
1592
</reg32>
1593
</domain>
1594
1595
<!-- I *think* this existed at least as far back as a4xx -->
1596
<domain name="CP_COND_REG_EXEC" width="32">
1597
<enum name="compare_mode">
1598
<!-- use the predicate bit set by CP_REG_TEST -->
1599
<value value="1" name="PRED_TEST"/>
1600
<!-- compare two registers directly for equality -->
1601
<value value="2" name="REG_COMPARE"/>
1602
<!-- test if certain render modes are set via CP_SET_MARKER -->
1603
<value value="3" name="RENDER_MODE" varset="chip" variants="A6XX-"/>
1604
</enum>
1605
<reg32 offset="0" name="0">
1606
<bitfield name="REG0" low="0" high="17" type="hex"/>
1607
1608
<!--
1609
Note: these bits have the same meaning, and use the same
1610
internal mechanism as the bits in CP_SET_DRAW_STATE.
1611
When RENDER_MODE is selected, they're used as
1612
a bitmask of which modes pass the test.
1613
-->
1614
1615
<!-- RM6_BINNING -->
1616
<bitfield name="BINNING" pos="25" varset="chip" variants="A6XX-" type="boolean"/>
1617
<!-- all others -->
1618
<bitfield name="GMEM" pos="26" varset="chip" variants="A6XX-" type="boolean"/>
1619
<!-- RM6_BYPASS -->
1620
<bitfield name="SYSMEM" pos="27" varset="chip" variants="A6XX-" type="boolean"/>
1621
1622
<bitfield name="MODE" low="28" high="31" type="compare_mode"/>
1623
</reg32>
1624
1625
<!-- in REG_COMPARE mode, there's an extra DWORD here with REG1 -->
1626
1627
<reg32 offset="1" name="1">
1628
<bitfield name="DWORDS" low="0" high="31" type="uint"/>
1629
</reg32>
1630
</domain>
1631
1632
<domain name="CP_COND_EXEC" width="32">
1633
<doc>
1634
Executes the following DWORDs of commands if the dword at ADDR0
1635
is not equal to 0 and the dword at ADDR1 is less than REF
1636
(signed comparison).
1637
</doc>
1638
<reg32 offset="0" name="0">
1639
<bitfield name="ADDR0_LO" low="0" high="31"/>
1640
</reg32>
1641
<reg32 offset="1" name="1">
1642
<bitfield name="ADDR0_HI" low="0" high="31"/>
1643
</reg32>
1644
<reg32 offset="2" name="2">
1645
<bitfield name="ADDR1_LO" low="0" high="31"/>
1646
</reg32>
1647
<reg32 offset="3" name="3">
1648
<bitfield name="ADDR1_HI" low="0" high="31"/>
1649
</reg32>
1650
<reg32 offset="4" name="4">
1651
<bitfield name="REF" low="0" high="31"/>
1652
</reg32>
1653
<reg32 offset="5" name="5">
1654
<bitfield name="DWORDS" low="0" high="31" type="uint"/>
1655
</reg32>
1656
</domain>
1657
1658
<domain name="CP_SET_CTXSWITCH_IB" width="32">
1659
<doc>
1660
Used by the userspace driver to set various IB's which are
1661
executed during context save/restore for handling
1662
state that isn't restored by the
1663
context switch routine itself.
1664
</doc>
1665
<enum name="ctxswitch_ib">
1666
<value name="RESTORE_IB" value="0">
1667
<doc>Executed unconditionally when switching back to the context.</doc>
1668
</value>
1669
<value name="YIELD_RESTORE_IB" value="1">
1670
<doc>
1671
Executed when switching back after switching
1672
away during execution of
1673
a CP_SET_MARKER packet with RM6_YIELD as the
1674
payload *and* the normal save routine was
1675
bypassed for a shorter one. I think this is
1676
connected to the "skipsaverestore" bit set by
1677
the kernel when preempting.
1678
</doc>
1679
</value>
1680
<value name="SAVE_IB" value="2">
1681
<doc>
1682
Executed when switching away from the context,
1683
except for context switches initiated via
1684
CP_YIELD.
1685
</doc>
1686
</value>
1687
<value name="RB_SAVE_IB" value="3">
1688
<doc>
1689
This can only be set by the RB (i.e. the kernel)
1690
and executes with protected mode off, but
1691
is otherwise similar to SAVE_IB.
1692
</doc>
1693
</value>
1694
</enum>
1695
<reg32 offset="0" name="0">
1696
<bitfield name="ADDR_LO" low="0" high="31"/>
1697
</reg32>
1698
<reg32 offset="1" name="1">
1699
<bitfield name="ADDR_HI" low="0" high="31"/>
1700
</reg32>
1701
<reg32 offset="2" name="2">
1702
<bitfield name="DWORDS" low="0" high="19" type="uint"/>
1703
<bitfield name="TYPE" low="20" high="21" type="ctxswitch_ib"/>
1704
</reg32>
1705
</domain>
1706
1707
<domain name="CP_REG_WRITE" width="32">
1708
<enum name="reg_tracker">
1709
<doc>
1710
Keep shadow copies of these registers and only set them
1711
when drawing, avoiding redundant writes:
1712
- VPC_CNTL_0
1713
- HLSQ_CONTROL_1_REG
1714
- HLSQ_UNKNOWN_B980
1715
</doc>
1716
<value name="TRACK_CNTL_REG" value="0x1"/>
1717
<doc>
1718
Track RB_RENDER_CNTL, and insert a WFI in the following
1719
situation:
1720
- There is a write that disables binning
1721
- There was a draw with binning left enabled, but in
1722
BYPASS mode
1723
Presumably this is a hang workaround?
1724
</doc>
1725
<value name="TRACK_RENDER_CNTL" value="0x2"/>
1726
<doc>
1727
Do a mysterious CP_EVENT_WRITE 0x3f when the low bit of
1728
the data to write is 0. Used by the Vulkan blob with
1729
PC_MULTIVIEW_CNTL, but this isn't predicated on particular
1730
register(s) like the others.
1731
</doc>
1732
<value name="UNK_EVENT_WRITE" value="0x4"/>
1733
</enum>
1734
<reg32 offset="0" name="0">
1735
<bitfield name="TRACKER" low="0" high="2" type="reg_tracker"/>
1736
</reg32>
1737
</domain>
1738
1739
<domain name="CP_SMMU_TABLE_UPDATE" width="32">
1740
<doc>
1741
Note that the SMMU's definition of TTBRn can take different forms
1742
depending on the pgtable format. But a5xx+ only uses aarch64
1743
format.
1744
</doc>
1745
<reg32 offset="0" name="0">
1746
<bitfield name="TTBR0_LO" low="0" high="31"/>
1747
</reg32>
1748
<reg32 offset="1" name="1">
1749
<bitfield name="TTBR0_HI" low="0" high="15"/>
1750
<bitfield name="ASID" low="16" high="31"/>
1751
</reg32>
1752
<reg32 offset="2" name="2">
1753
<doc>Unused, does not apply to aarch64 pgtable format</doc>
1754
<bitfield name="CONTEXTIDR" low="0" high="31"/>
1755
</reg32>
1756
<reg32 offset="3" name="3">
1757
<bitfield name="CONTEXTBANK" low="0" high="31"/>
1758
</reg32>
1759
</domain>
1760
1761
</database>
1762
1763
1764