2 * Copyright (c) 2014 - 2015 Intel Corporation
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.
24 #include "util/ralloc.h"
25 #include "brw_context.h"
28 #include "intel_mipmap_tree.h"
29 #include "brw_state.h"
30 #include "intel_batchbuffer.h"
31 #include "compiler/brw_nir.h"
32 #include "brw_program.h"
33 #include "compiler/glsl/ir_uniform.h"
36 assign_cs_binding_table_offsets(const struct gen_device_info
*devinfo
,
37 const struct gl_program
*prog
,
38 struct brw_cs_prog_data
*prog_data
)
40 uint32_t next_binding_table_offset
= 0;
42 /* May not be used if the gl_NumWorkGroups variable is not accessed. */
43 prog_data
->binding_table
.work_groups_start
= next_binding_table_offset
;
44 next_binding_table_offset
++;
46 brw_assign_common_binding_table_offsets(devinfo
, prog
, &prog_data
->base
,
47 next_binding_table_offset
);
51 brw_codegen_cs_prog(struct brw_context
*brw
,
52 struct brw_program
*cp
,
53 struct brw_cs_prog_key
*key
)
55 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
56 struct gl_context
*ctx
= &brw
->ctx
;
57 const GLuint
*program
;
58 void *mem_ctx
= ralloc_context(NULL
);
60 struct brw_cs_prog_data prog_data
;
61 bool start_busy
= false;
62 double start_time
= 0;
64 memset(&prog_data
, 0, sizeof(prog_data
));
66 if (cp
->program
.info
.cs
.shared_size
> 64 * 1024) {
67 cp
->program
.sh
.data
->LinkStatus
= linking_failure
;
68 const char *error_str
=
69 "Compute shader used more than 64KB of shared variables";
70 ralloc_strcat(&cp
->program
.sh
.data
->InfoLog
, error_str
);
71 _mesa_problem(NULL
, "Failed to link compute shader: %s\n", error_str
);
76 prog_data
.base
.total_shared
= cp
->program
.info
.cs
.shared_size
;
79 assign_cs_binding_table_offsets(devinfo
, &cp
->program
, &prog_data
);
81 /* Allocate the references to the uniforms that will end up in the
82 * prog_data associated with the compiled program, and which will be freed
85 int param_count
= cp
->program
.nir
->num_uniforms
/ 4;
87 /* The backend also sometimes add a param for the thread local id. */
88 prog_data
.thread_local_id_index
= param_count
++;
90 /* The backend also sometimes adds params for texture size. */
91 param_count
+= 2 * ctx
->Const
.Program
[MESA_SHADER_COMPUTE
].MaxTextureImageUnits
;
92 prog_data
.base
.param
=
93 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
94 prog_data
.base
.pull_param
=
95 rzalloc_array(NULL
, const gl_constant_value
*, param_count
);
96 prog_data
.base
.image_param
=
97 rzalloc_array(NULL
, struct brw_image_param
,
98 cp
->program
.info
.num_images
);
99 prog_data
.base
.nr_params
= param_count
;
100 prog_data
.base
.nr_image_params
= cp
->program
.info
.num_images
;
102 brw_nir_setup_glsl_uniforms(cp
->program
.nir
, &cp
->program
,&prog_data
.base
,
105 if (unlikely(brw
->perf_debug
)) {
106 start_busy
= (brw
->batch
.last_bo
&&
107 brw_bo_busy(brw
->batch
.last_bo
));
108 start_time
= get_time();
112 if (INTEL_DEBUG
& DEBUG_SHADER_TIME
)
113 st_index
= brw_get_shader_time_index(brw
, &cp
->program
, ST_CS
, true);
116 program
= brw_compile_cs(brw
->screen
->compiler
, brw
, mem_ctx
, key
,
117 &prog_data
, cp
->program
.nir
, st_index
,
118 &program_size
, &error_str
);
119 if (program
== NULL
) {
120 cp
->program
.sh
.data
->LinkStatus
= linking_failure
;
121 ralloc_strcat(&cp
->program
.sh
.data
->InfoLog
, error_str
);
122 _mesa_problem(NULL
, "Failed to compile compute shader: %s\n", error_str
);
124 ralloc_free(mem_ctx
);
128 if (unlikely(brw
->perf_debug
)) {
129 if (cp
->compiled_once
) {
130 _mesa_problem(&brw
->ctx
, "CS programs shouldn't need recompiles");
132 cp
->compiled_once
= true;
134 if (start_busy
&& !brw_bo_busy(brw
->batch
.last_bo
)) {
135 perf_debug("CS compile took %.03f ms and stalled the GPU\n",
136 (get_time() - start_time
) * 1000);
140 const unsigned subslices
= MAX2(brw
->screen
->subslice_total
, 1);
142 /* WaCSScratchSize:hsw
144 * Haswell's scratch space address calculation appears to be sparse
145 * rather than tightly packed. The Thread ID has bits indicating
146 * which subslice, EU within a subslice, and thread within an EU
147 * it is. There's a maximum of two slices and two subslices, so these
148 * can be stored with a single bit. Even though there are only 10 EUs
149 * per subslice, this is stored in 4 bits, so there's an effective
150 * maximum value of 16 EUs. Similarly, although there are only 7
151 * threads per EU, this is stored in a 3 bit number, giving an effective
152 * maximum value of 8 threads per EU.
154 * This means that we need to use 16 * 8 instead of 10 * 7 for the
155 * number of threads per subslice.
157 const unsigned scratch_ids_per_subslice
=
158 devinfo
->is_haswell
? 16 * 8 : devinfo
->max_cs_threads
;
160 brw_alloc_stage_scratch(brw
, &brw
->cs
.base
,
161 prog_data
.base
.total_scratch
,
162 scratch_ids_per_subslice
* subslices
);
164 brw_upload_cache(&brw
->cache
, BRW_CACHE_CS_PROG
,
166 program
, program_size
,
167 &prog_data
, sizeof(prog_data
),
168 &brw
->cs
.base
.prog_offset
, &brw
->cs
.base
.prog_data
);
169 ralloc_free(mem_ctx
);
176 brw_cs_populate_key(struct brw_context
*brw
, struct brw_cs_prog_key
*key
)
178 struct gl_context
*ctx
= &brw
->ctx
;
179 /* BRW_NEW_COMPUTE_PROGRAM */
180 const struct brw_program
*cp
=
181 (struct brw_program
*) brw
->programs
[MESA_SHADER_COMPUTE
];
182 const struct gl_program
*prog
= (struct gl_program
*) cp
;
184 memset(key
, 0, sizeof(*key
));
187 brw_populate_sampler_prog_key_data(ctx
, prog
, &key
->tex
);
189 /* The unique compute program ID */
190 key
->program_string_id
= cp
->id
;
195 brw_upload_cs_prog(struct brw_context
*brw
)
197 struct gl_context
*ctx
= &brw
->ctx
;
198 struct brw_cs_prog_key key
;
199 struct brw_program
*cp
=
200 (struct brw_program
*) brw
->programs
[MESA_SHADER_COMPUTE
];
205 if (!brw_state_dirty(brw
, _NEW_TEXTURE
, BRW_NEW_COMPUTE_PROGRAM
))
208 brw
->cs
.base
.sampler_count
=
209 util_last_bit(ctx
->ComputeProgram
._Current
->SamplersUsed
);
211 brw_cs_populate_key(brw
, &key
);
213 if (!brw_search_cache(&brw
->cache
, BRW_CACHE_CS_PROG
,
215 &brw
->cs
.base
.prog_offset
,
216 &brw
->cs
.base
.prog_data
)) {
217 bool success
= brw_codegen_cs_prog(brw
, cp
, &key
);
225 brw_cs_precompile(struct gl_context
*ctx
, struct gl_program
*prog
)
227 struct brw_context
*brw
= brw_context(ctx
);
228 struct brw_cs_prog_key key
;
230 struct brw_program
*bcp
= brw_program(prog
);
232 memset(&key
, 0, sizeof(key
));
233 key
.program_string_id
= bcp
->id
;
235 brw_setup_tex_for_precompile(brw
, &key
.tex
, prog
);
237 uint32_t old_prog_offset
= brw
->cs
.base
.prog_offset
;
238 struct brw_stage_prog_data
*old_prog_data
= brw
->cs
.base
.prog_data
;
240 bool success
= brw_codegen_cs_prog(brw
, bcp
, &key
);
242 brw
->cs
.base
.prog_offset
= old_prog_offset
;
243 brw
->cs
.base
.prog_data
= old_prog_data
;