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"
29 #include "compiler/glsl/program.h"
30 #include "compiler/nir/nir.h"
31 #include "compiler/nir/nir_serialize.h"
32 #include "pipe/p_shader_tokens.h"
33 #include "program/ir_to_mesa.h"
34 #include "tgsi/tgsi_parse.h"
35 #include "util/u_memory.h"
38 st_get_program_binary_driver_sha1(struct gl_context
*ctx
, uint8_t *sha1
)
40 disk_cache_compute_key(ctx
->Cache
, NULL
, 0, sha1
);
44 write_stream_out_to_cache(struct blob
*blob
,
45 struct pipe_shader_state
*state
)
47 blob_write_bytes(blob
, &state
->stream_output
,
48 sizeof(state
->stream_output
));
52 copy_blob_to_driver_cache_blob(struct blob
*blob
, struct gl_program
*prog
)
54 prog
->driver_cache_blob
= ralloc_size(NULL
, blob
->size
);
55 memcpy(prog
->driver_cache_blob
, blob
->data
, blob
->size
);
56 prog
->driver_cache_blob_size
= blob
->size
;
60 write_tgsi_to_cache(struct blob
*blob
, const struct tgsi_token
*tokens
,
61 struct gl_program
*prog
)
63 unsigned num_tokens
= tgsi_num_tokens(tokens
);
65 blob_write_uint32(blob
, num_tokens
);
66 blob_write_bytes(blob
, tokens
, num_tokens
* sizeof(struct tgsi_token
));
67 copy_blob_to_driver_cache_blob(blob
, prog
);
71 write_nir_to_cache(struct blob
*blob
, struct gl_program
*prog
)
73 nir_serialize(blob
, prog
->nir
, false);
74 copy_blob_to_driver_cache_blob(blob
, prog
);
78 st_serialise_ir_program(struct gl_context
*ctx
, struct gl_program
*prog
,
81 if (prog
->driver_cache_blob
)
87 switch (prog
->info
.stage
) {
88 case MESA_SHADER_VERTEX
: {
89 struct st_common_program
*stvp
= (struct st_common_program
*) prog
;
91 blob_write_uint32(&blob
, stvp
->num_inputs
);
92 blob_write_bytes(&blob
, stvp
->index_to_input
,
93 sizeof(stvp
->index_to_input
));
94 blob_write_bytes(&blob
, stvp
->input_to_index
,
95 sizeof(stvp
->input_to_index
));
96 blob_write_bytes(&blob
, stvp
->result_to_output
,
97 sizeof(stvp
->result_to_output
));
99 write_stream_out_to_cache(&blob
, &stvp
->state
);
102 write_nir_to_cache(&blob
, prog
);
104 write_tgsi_to_cache(&blob
, stvp
->state
.tokens
, prog
);
107 case MESA_SHADER_TESS_CTRL
:
108 case MESA_SHADER_TESS_EVAL
:
109 case MESA_SHADER_GEOMETRY
:
110 case MESA_SHADER_FRAGMENT
:
111 case MESA_SHADER_COMPUTE
: {
112 struct st_common_program
*stcp
= (struct st_common_program
*) prog
;
114 if (prog
->info
.stage
== MESA_SHADER_TESS_EVAL
||
115 prog
->info
.stage
== MESA_SHADER_GEOMETRY
)
116 write_stream_out_to_cache(&blob
, &stcp
->state
);
119 write_nir_to_cache(&blob
, prog
);
121 write_tgsi_to_cache(&blob
, stcp
->state
.tokens
, prog
);
125 unreachable("Unsupported stage");
132 * Store TGSI or NIR and any other required state in on-disk shader cache.
135 st_store_ir_in_disk_cache(struct st_context
*st
, struct gl_program
*prog
,
141 /* Exit early when we are dealing with a ff shader with no source file to
142 * generate a source from.
144 static const char zero
[sizeof(prog
->sh
.data
->sha1
)] = {0};
145 if (memcmp(prog
->sh
.data
->sha1
, zero
, sizeof(prog
->sh
.data
->sha1
)) == 0)
148 st_serialise_ir_program(st
->ctx
, prog
, nir
);
150 if (st
->ctx
->_Shader
->Flags
& GLSL_CACHE_INFO
) {
151 fprintf(stderr
, "putting %s state tracker IR in cache\n",
152 _mesa_shader_stage_to_string(prog
->info
.stage
));
157 read_stream_out_from_cache(struct blob_reader
*blob_reader
,
158 struct pipe_shader_state
*state
)
160 blob_copy_bytes(blob_reader
, (uint8_t *) &state
->stream_output
,
161 sizeof(state
->stream_output
));
165 read_tgsi_from_cache(struct blob_reader
*blob_reader
,
166 const struct tgsi_token
**tokens
)
168 unsigned num_tokens
= blob_read_uint32(blob_reader
);
169 unsigned tokens_size
= num_tokens
* sizeof(struct tgsi_token
);
170 *tokens
= (const struct tgsi_token
*) MALLOC(tokens_size
);
171 blob_copy_bytes(blob_reader
, (uint8_t *) *tokens
, tokens_size
);
175 st_deserialise_ir_program(struct gl_context
*ctx
,
176 struct gl_shader_program
*shProg
,
177 struct gl_program
*prog
, bool nir
)
179 struct st_context
*st
= st_context(ctx
);
180 size_t size
= prog
->driver_cache_blob_size
;
181 uint8_t *buffer
= (uint8_t *) prog
->driver_cache_blob
;
182 const struct nir_shader_compiler_options
*options
=
183 ctx
->Const
.ShaderCompilerOptions
[prog
->info
.stage
].NirOptions
;
185 st_set_prog_affected_state_flags(prog
);
186 _mesa_associate_uniform_storage(ctx
, shProg
, prog
);
188 assert(prog
->driver_cache_blob
&& prog
->driver_cache_blob_size
> 0);
190 struct blob_reader blob_reader
;
191 blob_reader_init(&blob_reader
, buffer
, size
);
193 switch (prog
->info
.stage
) {
194 case MESA_SHADER_VERTEX
: {
195 struct st_common_program
*stvp
= (struct st_common_program
*) prog
;
197 st_release_vp_variants(st
, stvp
);
199 stvp
->num_inputs
= blob_read_uint32(&blob_reader
);
200 blob_copy_bytes(&blob_reader
, (uint8_t *) stvp
->index_to_input
,
201 sizeof(stvp
->index_to_input
));
202 blob_copy_bytes(&blob_reader
, (uint8_t *) stvp
->input_to_index
,
203 sizeof(stvp
->input_to_index
));
204 blob_copy_bytes(&blob_reader
, (uint8_t *) stvp
->result_to_output
,
205 sizeof(stvp
->result_to_output
));
207 read_stream_out_from_cache(&blob_reader
, &stvp
->state
);
210 stvp
->state
.type
= PIPE_SHADER_IR_NIR
;
211 stvp
->shader_program
= shProg
;
212 stvp
->state
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
213 prog
->nir
= stvp
->state
.ir
.nir
;
215 read_tgsi_from_cache(&blob_reader
, &stvp
->state
.tokens
);
219 case MESA_SHADER_TESS_CTRL
:
220 case MESA_SHADER_TESS_EVAL
:
221 case MESA_SHADER_GEOMETRY
:
222 case MESA_SHADER_FRAGMENT
:
223 case MESA_SHADER_COMPUTE
: {
224 struct st_common_program
*stcp
= st_common_program(prog
);
226 if (prog
->info
.stage
== MESA_SHADER_FRAGMENT
)
227 st_release_fp_variants(st
, stcp
);
229 st_release_common_variants(st
, stcp
);
231 if (prog
->info
.stage
== MESA_SHADER_TESS_EVAL
||
232 prog
->info
.stage
== MESA_SHADER_GEOMETRY
)
233 read_stream_out_from_cache(&blob_reader
, &stcp
->state
);
236 stcp
->state
.type
= PIPE_SHADER_IR_NIR
;
237 stcp
->state
.ir
.nir
= nir_deserialize(NULL
, options
, &blob_reader
);
238 stcp
->shader_program
= shProg
;
239 prog
->nir
= stcp
->state
.ir
.nir
;
241 read_tgsi_from_cache(&blob_reader
, &stcp
->state
.tokens
);
246 unreachable("Unsupported stage");
249 /* Make sure we don't try to read more data than we wrote. This should
250 * never happen in release builds but its useful to have this check to
251 * catch development bugs.
253 if (blob_reader
.current
!= blob_reader
.end
|| blob_reader
.overrun
) {
254 assert(!"Invalid TGSI shader disk cache item!");
256 if (ctx
->_Shader
->Flags
& GLSL_CACHE_INFO
) {
257 fprintf(stderr
, "Error reading program from cache (invalid "
258 "TGSI cache item)\n");
262 st_finalize_program(st
, prog
);
266 st_load_ir_from_disk_cache(struct gl_context
*ctx
,
267 struct gl_shader_program
*prog
,
273 /* If we didn't load the GLSL metadata from cache then we could not have
274 * loaded TGSI or NIR either.
276 if (prog
->data
->LinkStatus
!= LINKING_SKIPPED
)
279 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
280 if (prog
->_LinkedShaders
[i
] == NULL
)
283 struct gl_program
*glprog
= prog
->_LinkedShaders
[i
]->Program
;
284 st_deserialise_ir_program(ctx
, prog
, glprog
, nir
);
286 /* We don't need the cached blob anymore so free it */
287 ralloc_free(glprog
->driver_cache_blob
);
288 glprog
->driver_cache_blob
= NULL
;
289 glprog
->driver_cache_blob_size
= 0;
291 if (ctx
->_Shader
->Flags
& GLSL_CACHE_INFO
) {
292 fprintf(stderr
, "%s state tracker IR retrieved from cache\n",
293 _mesa_shader_stage_to_string(i
));
301 st_serialise_tgsi_program(struct gl_context
*ctx
, struct gl_program
*prog
)
303 st_serialise_ir_program(ctx
, prog
, false);
307 st_serialise_tgsi_program_binary(struct gl_context
*ctx
,
308 struct gl_shader_program
*shProg
,
309 struct gl_program
*prog
)
311 st_serialise_ir_program(ctx
, prog
, false);
315 st_deserialise_tgsi_program(struct gl_context
*ctx
,
316 struct gl_shader_program
*shProg
,
317 struct gl_program
*prog
)
319 st_deserialise_ir_program(ctx
, shProg
, prog
, false);
323 st_serialise_nir_program(struct gl_context
*ctx
, struct gl_program
*prog
)
325 st_serialise_ir_program(ctx
, prog
, true);
329 st_serialise_nir_program_binary(struct gl_context
*ctx
,
330 struct gl_shader_program
*shProg
,
331 struct gl_program
*prog
)
333 st_serialise_ir_program(ctx
, prog
, true);
337 st_deserialise_nir_program(struct gl_context
*ctx
,
338 struct gl_shader_program
*shProg
,
339 struct gl_program
*prog
)
341 st_deserialise_ir_program(ctx
, shProg
, prog
, true);