glsl: move to compiler/
[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 #include "brw_program.h"
35 #include "compiler/glsl/ir_uniform.h"
36
37 static void
38 assign_cs_binding_table_offsets(const struct brw_device_info *devinfo,
39 const struct gl_shader_program *shader_prog,
40 const struct gl_program *prog,
41 struct brw_cs_prog_data *prog_data)
42 {
43 uint32_t next_binding_table_offset = 0;
44
45 /* May not be used if the gl_NumWorkGroups variable is not accessed. */
46 prog_data->binding_table.work_groups_start = next_binding_table_offset;
47 next_binding_table_offset++;
48
49 brw_assign_common_binding_table_offsets(MESA_SHADER_COMPUTE, devinfo,
50 shader_prog, prog, &prog_data->base,
51 next_binding_table_offset);
52 }
53
54 static bool
55 brw_codegen_cs_prog(struct brw_context *brw,
56 struct gl_shader_program *prog,
57 struct brw_compute_program *cp,
58 struct brw_cs_prog_key *key)
59 {
60 struct gl_context *ctx = &brw->ctx;
61 const GLuint *program;
62 void *mem_ctx = ralloc_context(NULL);
63 GLuint program_size;
64 struct brw_cs_prog_data prog_data;
65 bool start_busy = false;
66 double start_time = 0;
67
68 struct brw_shader *cs =
69 (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_COMPUTE];
70 assert (cs);
71
72 memset(&prog_data, 0, sizeof(prog_data));
73
74 if (prog->Comp.SharedSize > 64 * 1024) {
75 prog->LinkStatus = false;
76 const char *error_str =
77 "Compute shader used more than 64KB of shared variables";
78 ralloc_strcat(&prog->InfoLog, error_str);
79 _mesa_problem(NULL, "Failed to link compute shader: %s\n", error_str);
80
81 ralloc_free(mem_ctx);
82 return false;
83 } else {
84 prog_data.base.total_shared = prog->Comp.SharedSize;
85 }
86
87 assign_cs_binding_table_offsets(brw->intelScreen->devinfo, prog,
88 &cp->program.Base, &prog_data);
89
90 /* Allocate the references to the uniforms that will end up in the
91 * prog_data associated with the compiled program, and which will be freed
92 * by the state cache.
93 */
94 int param_count = cp->program.Base.nir->num_uniforms;
95
96 /* The backend also sometimes adds params for texture size. */
97 param_count += 2 * ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits;
98 prog_data.base.param =
99 rzalloc_array(NULL, const gl_constant_value *, param_count);
100 prog_data.base.pull_param =
101 rzalloc_array(NULL, const gl_constant_value *, param_count);
102 prog_data.base.image_param =
103 rzalloc_array(NULL, struct brw_image_param, cs->base.NumImages);
104 prog_data.base.nr_params = param_count;
105 prog_data.base.nr_image_params = cs->base.NumImages;
106
107 brw_nir_setup_glsl_uniforms(cp->program.Base.nir, prog, &cp->program.Base,
108 &prog_data.base, true);
109
110 if (unlikely(brw->perf_debug)) {
111 start_busy = (brw->batch.last_bo &&
112 drm_intel_bo_busy(brw->batch.last_bo));
113 start_time = get_time();
114 }
115
116 if (unlikely(INTEL_DEBUG & DEBUG_CS))
117 brw_dump_ir("compute", prog, &cs->base, &cp->program.Base);
118
119 int st_index = -1;
120 if (INTEL_DEBUG & DEBUG_SHADER_TIME)
121 st_index = brw_get_shader_time_index(brw, prog, &cp->program.Base, ST_CS);
122
123 char *error_str;
124 program = brw_compile_cs(brw->intelScreen->compiler, brw, mem_ctx,
125 key, &prog_data, cp->program.Base.nir,
126 st_index, &program_size, &error_str);
127 if (program == NULL) {
128 prog->LinkStatus = false;
129 ralloc_strcat(&prog->InfoLog, error_str);
130 _mesa_problem(NULL, "Failed to compile compute shader: %s\n", error_str);
131
132 ralloc_free(mem_ctx);
133 return false;
134 }
135
136 if (unlikely(brw->perf_debug) && cs) {
137 if (cs->compiled_once) {
138 _mesa_problem(&brw->ctx, "CS programs shouldn't need recompiles");
139 }
140 cs->compiled_once = true;
141
142 if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
143 perf_debug("CS compile took %.03f ms and stalled the GPU\n",
144 (get_time() - start_time) * 1000);
145 }
146 }
147
148 if (prog_data.base.total_scratch) {
149 brw_get_scratch_bo(brw, &brw->cs.base.scratch_bo,
150 prog_data.base.total_scratch * brw->max_cs_threads);
151 }
152
153 if (unlikely(INTEL_DEBUG & DEBUG_CS))
154 fprintf(stderr, "\n");
155
156 brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
157 key, sizeof(*key),
158 program, program_size,
159 &prog_data, sizeof(prog_data),
160 &brw->cs.base.prog_offset, &brw->cs.prog_data);
161 ralloc_free(mem_ctx);
162
163 return true;
164 }
165
166
167 static void
168 brw_cs_populate_key(struct brw_context *brw, struct brw_cs_prog_key *key)
169 {
170 struct gl_context *ctx = &brw->ctx;
171 /* BRW_NEW_COMPUTE_PROGRAM */
172 const struct brw_compute_program *cp =
173 (struct brw_compute_program *) brw->compute_program;
174 const struct gl_program *prog = (struct gl_program *) cp;
175
176 memset(key, 0, sizeof(*key));
177
178 /* _NEW_TEXTURE */
179 brw_populate_sampler_prog_key_data(ctx, prog, brw->cs.base.sampler_count,
180 &key->tex);
181
182 /* The unique compute program ID */
183 key->program_string_id = cp->id;
184 }
185
186
187 void
188 brw_upload_cs_prog(struct brw_context *brw)
189 {
190 struct gl_context *ctx = &brw->ctx;
191 struct brw_cs_prog_key key;
192 struct brw_compute_program *cp = (struct brw_compute_program *)
193 brw->compute_program;
194
195 if (!cp)
196 return;
197
198 if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_COMPUTE_PROGRAM))
199 return;
200
201 brw->cs.base.sampler_count =
202 _mesa_fls(ctx->ComputeProgram._Current->Base.SamplersUsed);
203
204 brw_cs_populate_key(brw, &key);
205
206 if (!brw_search_cache(&brw->cache, BRW_CACHE_CS_PROG,
207 &key, sizeof(key),
208 &brw->cs.base.prog_offset, &brw->cs.prog_data)) {
209 bool success =
210 brw_codegen_cs_prog(brw,
211 ctx->Shader.CurrentProgram[MESA_SHADER_COMPUTE],
212 cp, &key);
213 (void) success;
214 assert(success);
215 }
216 brw->cs.base.prog_data = &brw->cs.prog_data->base;
217 }
218
219
220 bool
221 brw_cs_precompile(struct gl_context *ctx,
222 struct gl_shader_program *shader_prog,
223 struct gl_program *prog)
224 {
225 struct brw_context *brw = brw_context(ctx);
226 struct brw_cs_prog_key key;
227
228 struct gl_compute_program *cp = (struct gl_compute_program *) prog;
229 struct brw_compute_program *bcp = brw_compute_program(cp);
230
231 memset(&key, 0, sizeof(key));
232 key.program_string_id = bcp->id;
233
234 brw_setup_tex_for_precompile(brw, &key.tex, prog);
235
236 uint32_t old_prog_offset = brw->cs.base.prog_offset;
237 struct brw_cs_prog_data *old_prog_data = brw->cs.prog_data;
238
239 bool success = brw_codegen_cs_prog(brw, shader_prog, bcp, &key);
240
241 brw->cs.base.prog_offset = old_prog_offset;
242 brw->cs.prog_data = old_prog_data;
243
244 return success;
245 }