2 * Copyright © 2017 Timothy Arceri
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 #include "st_program.h"
27 #include "st_shader_cache.h"
28 #include "compiler/glsl/program.h"
29 #include "compiler/nir/nir.h"
30 #include "compiler/nir/nir_serialize.h"
31 #include "pipe/p_shader_tokens.h"
32 #include "program/ir_to_mesa.h"
33 #include "util/u_memory.h"
36 st_get_program_binary_driver_sha1(struct gl_context
*ctx
, uint8_t *sha1
)
38 disk_cache_compute_key(ctx
->Cache
, NULL
, 0, sha1
);
42 write_stream_out_to_cache(struct blob
*blob
,
43 struct pipe_shader_state
*tgsi
)
45 blob_write_bytes(blob
, &tgsi
->stream_output
,
46 sizeof(tgsi
->stream_output
));
50 copy_blob_to_driver_cache_blob(struct blob
*blob
, struct gl_program
*prog
)
52 prog
->driver_cache_blob
= ralloc_size(NULL
, blob
->size
);
53 memcpy(prog
->driver_cache_blob
, blob
->data
, blob
->size
);
54 prog
->driver_cache_blob_size
= blob
->size
;
58 write_tgsi_to_cache(struct blob
*blob
, const struct tgsi_token
*tokens
,
59 struct gl_program
*prog
, unsigned num_tokens
)
61 blob_write_uint32(blob
, num_tokens
);
62 blob_write_bytes(blob
, tokens
, num_tokens
* sizeof(struct tgsi_token
));
63 copy_blob_to_driver_cache_blob(blob
, prog
);
67 write_nir_to_cache(struct blob
*blob
, struct gl_program
*prog
)
69 nir_serialize(blob
, prog
->nir
);
70 copy_blob_to_driver_cache_blob(blob
, prog
);
74 st_serialise_ir_program(struct gl_context
*ctx
, struct gl_program
*prog
,
80 switch (prog
->info
.stage
) {
81 case MESA_SHADER_VERTEX
: {
82 struct st_vertex_program
*stvp
= (struct st_vertex_program
*) prog
;
84 blob_write_uint32(&blob
, stvp
->num_inputs
);
85 blob_write_bytes(&blob
, stvp
->index_to_input
,
86 sizeof(stvp
->index_to_input
));
87 blob_write_bytes(&blob
, stvp
->result_to_output
,
88 sizeof(stvp
->result_to_output
));
90 write_stream_out_to_cache(&blob
, &stvp
->tgsi
);
93 write_nir_to_cache(&blob
, prog
);
95 write_tgsi_to_cache(&blob
, stvp
->tgsi
.tokens
, prog
,
96 stvp
->num_tgsi_tokens
);
99 case MESA_SHADER_TESS_CTRL
:
100 case MESA_SHADER_TESS_EVAL
:
101 case MESA_SHADER_GEOMETRY
: {
102 struct st_common_program
*stcp
= (struct st_common_program
*) prog
;
104 write_stream_out_to_cache(&blob
, &stcp
->tgsi
);
107 write_nir_to_cache(&blob
, prog
);
109 write_tgsi_to_cache(&blob
, stcp
->tgsi
.tokens
, prog
,
110 stcp
->num_tgsi_tokens
);
113 case MESA_SHADER_FRAGMENT
: {
114 struct st_fragment_program
*stfp
= (struct st_fragment_program
*) prog
;
117 write_nir_to_cache(&blob
, prog
);
119 write_tgsi_to_cache(&blob
, stfp
->tgsi
.tokens
, prog
,
120 stfp
->num_tgsi_tokens
);
123 case MESA_SHADER_COMPUTE
: {
124 struct st_compute_program
*stcp
= (struct st_compute_program
*) prog
;
127 write_nir_to_cache(&blob
, prog
);
129 write_tgsi_to_cache(&blob
, stcp
->tgsi
.prog
, prog
,
130 stcp
->num_tgsi_tokens
);
134 unreachable("Unsupported stage");
141 * Store tgsi and any other required state in on-disk shader cache.
144 st_store_ir_in_disk_cache(struct st_context
*st
, struct gl_program
*prog
,
150 /* Exit early when we are dealing with a ff shader with no source file to
151 * generate a source from.
153 static const char zero
[sizeof(prog
->sh
.data
->sha1
)] = {0};
154 if (memcmp(prog
->sh
.data
->sha1
, zero
, sizeof(prog
->sh
.data
->sha1
)) == 0)
157 st_serialise_ir_program(st
->ctx
, prog
, nir
);
159 if (st
->ctx
->_Shader
->Flags
& GLSL_CACHE_INFO
) {
160 fprintf(stderr
, "putting %s state tracker IR in cache\n",
161 _mesa_shader_stage_to_string(prog
->info
.stage
));
166 read_stream_out_from_cache(struct blob_reader
*blob_reader
,
167 struct pipe_shader_state
*tgsi
)
169 blob_copy_bytes(blob_reader
, (uint8_t *) &tgsi
->stream_output
,
170 sizeof(tgsi
->stream_output
));
174 read_tgsi_from_cache(struct blob_reader
*blob_reader
,
175 const struct tgsi_token
**tokens
,
176 unsigned *num_tokens
)
178 *num_tokens
= blob_read_uint32(blob_reader
);
179 unsigned tokens_size
= *num_tokens
* sizeof(struct tgsi_token
);
180 *tokens
= (const struct tgsi_token
*) MALLOC(tokens_size
);
181 blob_copy_bytes(blob_reader
, (uint8_t *) *tokens
, tokens_size
);
185 st_deserialise_ir_program(struct gl_context
*ctx
,
186 struct gl_shader_program
*shProg
,
187 struct gl_program
*prog
, bool nir
)
189 struct st_context
*st
= st_context(ctx
);
190 size_t size
= prog
->driver_cache_blob_size
;
191 uint8_t *buffer
= (uint8_t *) prog
->driver_cache_blob
;
192 const struct nir_shader_compiler_options
*options
=
193 ctx
->Const
.ShaderCompilerOptions
[prog
->info
.stage
].NirOptions
;
195 assert(prog
->driver_cache_blob
&& prog
->driver_cache_blob_size
> 0);
197 struct blob_reader blob_reader
;
198 blob_reader_init(&blob_reader
, buffer
, size
);
200 switch (prog
->info
.stage
) {
201 case MESA_SHADER_VERTEX
: {
202 struct st_vertex_program
*stvp
= (struct st_vertex_program
*) prog
;
204 st_release_vp_variants(st
, stvp
);
206 stvp
->num_inputs
= blob_read_uint32(&blob_reader
);
207 blob_copy_bytes(&blob_reader
, (uint8_t *) stvp
->index_to_input
,
208 sizeof(stvp
->index_to_input
));
209 blob_copy_bytes(&blob_reader
, (uint8_t *) stvp
->result_to_output
,
210 sizeof(stvp
->result_to_output
));
212 read_stream_out_from_cache(&blob_reader
, &stvp
->tgsi
);
215 stvp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
216 stvp
->shader_program
= shProg
;
217 stvp
->tgsi
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
219 read_tgsi_from_cache(&blob_reader
, &stvp
->tgsi
.tokens
,
220 &stvp
->num_tgsi_tokens
);
224 st
->dirty
|= ST_NEW_VERTEX_PROGRAM(st
, stvp
);
228 case MESA_SHADER_TESS_CTRL
: {
229 struct st_common_program
*sttcp
= st_common_program(prog
);
231 st_release_basic_variants(st
, sttcp
->Base
.Target
,
232 &sttcp
->variants
, &sttcp
->tgsi
);
234 read_stream_out_from_cache(&blob_reader
, &sttcp
->tgsi
);
237 sttcp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
238 sttcp
->shader_program
= shProg
;
239 sttcp
->tgsi
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
241 read_tgsi_from_cache(&blob_reader
, &sttcp
->tgsi
.tokens
,
242 &sttcp
->num_tgsi_tokens
);
245 if (st
->tcp
== sttcp
)
246 st
->dirty
|= sttcp
->affected_states
;
250 case MESA_SHADER_TESS_EVAL
: {
251 struct st_common_program
*sttep
= st_common_program(prog
);
253 st_release_basic_variants(st
, sttep
->Base
.Target
,
254 &sttep
->variants
, &sttep
->tgsi
);
256 read_stream_out_from_cache(&blob_reader
, &sttep
->tgsi
);
259 sttep
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
260 sttep
->shader_program
= shProg
;
261 sttep
->tgsi
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
263 read_tgsi_from_cache(&blob_reader
, &sttep
->tgsi
.tokens
,
264 &sttep
->num_tgsi_tokens
);
267 if (st
->tep
== sttep
)
268 st
->dirty
|= sttep
->affected_states
;
272 case MESA_SHADER_GEOMETRY
: {
273 struct st_common_program
*stgp
= st_common_program(prog
);
275 st_release_basic_variants(st
, stgp
->Base
.Target
, &stgp
->variants
,
278 read_stream_out_from_cache(&blob_reader
, &stgp
->tgsi
);
281 stgp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
282 stgp
->shader_program
= shProg
;
283 stgp
->tgsi
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
285 read_tgsi_from_cache(&blob_reader
, &stgp
->tgsi
.tokens
,
286 &stgp
->num_tgsi_tokens
);
290 st
->dirty
|= stgp
->affected_states
;
294 case MESA_SHADER_FRAGMENT
: {
295 struct st_fragment_program
*stfp
= (struct st_fragment_program
*) prog
;
297 st_release_fp_variants(st
, stfp
);
300 stfp
->tgsi
.type
= PIPE_SHADER_IR_NIR
;
301 stfp
->shader_program
= shProg
;
302 stfp
->tgsi
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
304 read_tgsi_from_cache(&blob_reader
, &stfp
->tgsi
.tokens
,
305 &stfp
->num_tgsi_tokens
);
309 st
->dirty
|= stfp
->affected_states
;
313 case MESA_SHADER_COMPUTE
: {
314 struct st_compute_program
*stcp
= (struct st_compute_program
*) prog
;
316 st_release_cp_variants(st
, stcp
);
319 stcp
->tgsi
.ir_type
= PIPE_SHADER_IR_NIR
;
320 stcp
->shader_program
= shProg
;
321 stcp
->tgsi
.prog
= nir_deserialize(NULL
, options
, &blob_reader
);
323 read_tgsi_from_cache(&blob_reader
,
324 (const struct tgsi_token
**) &stcp
->tgsi
.prog
,
325 &stcp
->num_tgsi_tokens
);
328 stcp
->tgsi
.req_local_mem
= stcp
->Base
.info
.cs
.shared_size
;
329 stcp
->tgsi
.req_private_mem
= 0;
330 stcp
->tgsi
.req_input_mem
= 0;
333 st
->dirty
|= stcp
->affected_states
;
338 unreachable("Unsupported stage");
341 /* Make sure we don't try to read more data than we wrote. This should
342 * never happen in release builds but its useful to have this check to
343 * catch development bugs.
345 if (blob_reader
.current
!= blob_reader
.end
|| blob_reader
.overrun
) {
346 assert(!"Invalid TGSI shader disk cache item!");
348 if (ctx
->_Shader
->Flags
& GLSL_CACHE_INFO
) {
349 fprintf(stderr
, "Error reading program from cache (invalid "
350 "TGSI cache item)\n");
354 st_set_prog_affected_state_flags(prog
);
355 _mesa_associate_uniform_storage(ctx
, shProg
, prog
, false);
357 /* Create Gallium shaders now instead of on demand. */
358 if (ST_DEBUG
& DEBUG_PRECOMPILE
||
359 st
->shader_has_one_variant
[prog
->info
.stage
])
360 st_precompile_shader_variant(st
, prog
);
364 st_load_ir_from_disk_cache(struct gl_context
*ctx
,
365 struct gl_shader_program
*prog
,
371 /* If we didn't load the GLSL metadata from cache then we could not have
372 * loaded the tgsi either.
374 if (prog
->data
->LinkStatus
!= LINKING_SKIPPED
)
377 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
378 if (prog
->_LinkedShaders
[i
] == NULL
)
381 struct gl_program
*glprog
= prog
->_LinkedShaders
[i
]->Program
;
382 st_deserialise_ir_program(ctx
, prog
, glprog
, nir
);
384 /* We don't need the cached blob anymore so free it */
385 ralloc_free(glprog
->driver_cache_blob
);
386 glprog
->driver_cache_blob
= NULL
;
387 glprog
->driver_cache_blob_size
= 0;
389 if (ctx
->_Shader
->Flags
& GLSL_CACHE_INFO
) {
390 fprintf(stderr
, "%s state tracker IR retrieved from cache\n",
391 _mesa_shader_stage_to_string(i
));
399 st_serialise_tgsi_program(struct gl_context
*ctx
, struct gl_program
*prog
)
401 st_serialise_ir_program(ctx
, prog
, false);
405 st_deserialise_tgsi_program(struct gl_context
*ctx
,
406 struct gl_shader_program
*shProg
,
407 struct gl_program
*prog
)
409 st_deserialise_ir_program(ctx
, shProg
, prog
, false);
413 st_serialise_nir_program(struct gl_context
*ctx
, struct gl_program
*prog
)
415 st_serialise_ir_program(ctx
, prog
, true);
419 st_deserialise_nir_program(struct gl_context
*ctx
,
420 struct gl_shader_program
*shProg
,
421 struct gl_program
*prog
)
423 st_deserialise_ir_program(ctx
, shProg
, prog
, true);