i965: Move brw_get_shader_time_index() call out of emit functions
[mesa.git] / src / mesa / drivers / dri / i965 / brw_cs.c
1 /*
2 * Copyright (c) 2014 - 2015 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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.
22 */
23
24 #include "util/ralloc.h"
25 #include "brw_context.h"
26 #include "brw_cs.h"
27 #include "brw_eu.h"
28 #include "brw_wm.h"
29 #include "brw_shader.h"
30 #include "intel_mipmap_tree.h"
31 #include "brw_state.h"
32 #include "intel_batchbuffer.h"
33 #include "brw_nir.h"
34
35 static void
36 assign_cs_binding_table_offsets(const struct brw_device_info *devinfo,
37 const struct gl_shader_program *shader_prog,
38 const struct gl_program *prog,
39 struct brw_cs_prog_data *prog_data)
40 {
41 uint32_t next_binding_table_offset = 0;
42
43 /* May not be used if the gl_NumWorkGroups variable is not accessed. */
44 prog_data->binding_table.work_groups_start = next_binding_table_offset;
45 next_binding_table_offset++;
46
47 brw_assign_common_binding_table_offsets(MESA_SHADER_COMPUTE, devinfo,
48 shader_prog, prog, &prog_data->base,
49 next_binding_table_offset);
50 }
51
52 static bool
53 brw_codegen_cs_prog(struct brw_context *brw,
54 struct gl_shader_program *prog,
55 struct brw_compute_program *cp,
56 struct brw_cs_prog_key *key)
57 {
58 struct gl_context *ctx = &brw->ctx;
59 const GLuint *program;
60 void *mem_ctx = ralloc_context(NULL);
61 GLuint program_size;
62 struct brw_cs_prog_data prog_data;
63 bool start_busy = false;
64 double start_time = 0;
65
66 struct brw_shader *cs =
67 (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_COMPUTE];
68 assert (cs);
69
70 memset(&prog_data, 0, sizeof(prog_data));
71
72 assign_cs_binding_table_offsets(brw->intelScreen->devinfo, prog,
73 &cp->program.Base, &prog_data);
74
75 /* Allocate the references to the uniforms that will end up in the
76 * prog_data associated with the compiled program, and which will be freed
77 * by the state cache.
78 */
79 int param_count = cp->program.Base.nir->num_uniforms;
80
81 /* The backend also sometimes adds params for texture size. */
82 param_count += 2 * ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits;
83 prog_data.base.param =
84 rzalloc_array(NULL, const gl_constant_value *, param_count);
85 prog_data.base.pull_param =
86 rzalloc_array(NULL, const gl_constant_value *, param_count);
87 prog_data.base.image_param =
88 rzalloc_array(NULL, struct brw_image_param, cs->base.NumImages);
89 prog_data.base.nr_params = param_count;
90 prog_data.base.nr_image_params = cs->base.NumImages;
91
92 brw_nir_setup_glsl_uniforms(cp->program.Base.nir, prog, &cp->program.Base,
93 &prog_data.base, true);
94
95 if (unlikely(brw->perf_debug)) {
96 start_busy = (brw->batch.last_bo &&
97 drm_intel_bo_busy(brw->batch.last_bo));
98 start_time = get_time();
99 }
100
101 if (unlikely(INTEL_DEBUG & DEBUG_CS))
102 brw_dump_ir("compute", prog, &cs->base, &cp->program.Base);
103
104 int st_index = -1;
105 if (INTEL_DEBUG & DEBUG_SHADER_TIME)
106 st_index = brw_get_shader_time_index(brw, prog, &cp->program.Base, ST_CS);
107
108 program = brw_cs_emit(brw, mem_ctx, key, &prog_data,
109 &cp->program, prog, st_index, &program_size);
110 if (program == NULL) {
111 ralloc_free(mem_ctx);
112 return false;
113 }
114
115 if (unlikely(brw->perf_debug) && cs) {
116 if (cs->compiled_once) {
117 _mesa_problem(&brw->ctx, "CS programs shouldn't need recompiles");
118 }
119 cs->compiled_once = true;
120
121 if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
122 perf_debug("CS compile took %.03f ms and stalled the GPU\n",
123 (get_time() - start_time) * 1000);
124 }
125 }
126
127 if (prog_data.base.total_scratch) {
128 brw_get_scratch_bo(brw, &brw->cs.base.scratch_bo,
129 prog_data.base.total_scratch * brw->max_cs_threads);
130 }
131
132 if (unlikely(INTEL_DEBUG & DEBUG_CS))
133 fprintf(stderr, "\n");
134
135 brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
136 key, sizeof(*key),
137 program, program_size,
138 &prog_data, sizeof(prog_data),
139 &brw->cs.base.prog_offset, &brw->cs.prog_data);
140 ralloc_free(mem_ctx);
141
142 return true;
143 }
144
145
146 static void
147 brw_cs_populate_key(struct brw_context *brw, struct brw_cs_prog_key *key)
148 {
149 struct gl_context *ctx = &brw->ctx;
150 /* BRW_NEW_COMPUTE_PROGRAM */
151 const struct brw_compute_program *cp =
152 (struct brw_compute_program *) brw->compute_program;
153 const struct gl_program *prog = (struct gl_program *) cp;
154
155 memset(key, 0, sizeof(*key));
156
157 /* _NEW_TEXTURE */
158 brw_populate_sampler_prog_key_data(ctx, prog, brw->cs.base.sampler_count,
159 &key->tex);
160
161 /* The unique compute program ID */
162 key->program_string_id = cp->id;
163 }
164
165
166 void
167 brw_upload_cs_prog(struct brw_context *brw)
168 {
169 struct gl_context *ctx = &brw->ctx;
170 struct brw_cs_prog_key key;
171 struct brw_compute_program *cp = (struct brw_compute_program *)
172 brw->compute_program;
173
174 if (!cp)
175 return;
176
177 if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_COMPUTE_PROGRAM))
178 return;
179
180 brw->cs.base.sampler_count =
181 _mesa_fls(ctx->ComputeProgram._Current->Base.SamplersUsed);
182
183 brw_cs_populate_key(brw, &key);
184
185 if (!brw_search_cache(&brw->cache, BRW_CACHE_CS_PROG,
186 &key, sizeof(key),
187 &brw->cs.base.prog_offset, &brw->cs.prog_data)) {
188 bool success =
189 brw_codegen_cs_prog(brw,
190 ctx->Shader.CurrentProgram[MESA_SHADER_COMPUTE],
191 cp, &key);
192 (void) success;
193 assert(success);
194 }
195 brw->cs.base.prog_data = &brw->cs.prog_data->base;
196 }
197
198
199 bool
200 brw_cs_precompile(struct gl_context *ctx,
201 struct gl_shader_program *shader_prog,
202 struct gl_program *prog)
203 {
204 struct brw_context *brw = brw_context(ctx);
205 struct brw_cs_prog_key key;
206
207 struct gl_compute_program *cp = (struct gl_compute_program *) prog;
208 struct brw_compute_program *bcp = brw_compute_program(cp);
209
210 memset(&key, 0, sizeof(key));
211 key.program_string_id = bcp->id;
212
213 brw_setup_tex_for_precompile(brw, &key.tex, prog);
214
215 uint32_t old_prog_offset = brw->cs.base.prog_offset;
216 struct brw_cs_prog_data *old_prog_data = brw->cs.prog_data;
217
218 bool success = brw_codegen_cs_prog(brw, shader_prog, bcp, &key);
219
220 brw->cs.base.prog_offset = old_prog_offset;
221 brw->cs.prog_data = old_prog_data;
222
223 return success;
224 }