c34620934415dbe16201c39024cd238525898478
[mesa.git] / src / mesa / drivers / dri / i965 / brw_gs.c
1 /*
2 * Copyright © 2013 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 /**
25 * \file brw_vec4_gs.c
26 *
27 * State atom for client-programmable geometry shaders, and support code.
28 */
29
30 #include "brw_gs.h"
31 #include "brw_context.h"
32 #include "brw_vec4_gs_visitor.h"
33 #include "brw_state.h"
34 #include "brw_ff_gs.h"
35 #include "brw_nir.h"
36 #include "brw_program.h"
37 #include "compiler/glsl/ir_uniform.h"
38
39 static void
40 brw_gs_debug_recompile(struct brw_context *brw,
41 struct gl_shader_program *shader_prog,
42 const struct brw_gs_prog_key *key)
43 {
44 struct brw_cache_item *c = NULL;
45 const struct brw_gs_prog_key *old_key = NULL;
46 bool found = false;
47
48 perf_debug("Recompiling geometry shader for program %d\n",
49 shader_prog->Name);
50
51 for (unsigned int i = 0; i < brw->cache.size; i++) {
52 for (c = brw->cache.items[i]; c; c = c->next) {
53 if (c->cache_id == BRW_CACHE_GS_PROG) {
54 old_key = c->key;
55
56 if (old_key->program_string_id == key->program_string_id)
57 break;
58 }
59 }
60 if (c)
61 break;
62 }
63
64 if (!c) {
65 perf_debug(" Didn't find previous compile in the shader cache for "
66 "debug\n");
67 return;
68 }
69
70 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
71
72 if (!found) {
73 perf_debug(" Something else\n");
74 }
75 }
76
77 static void
78 assign_gs_binding_table_offsets(const struct gen_device_info *devinfo,
79 const struct gl_shader_program *shader_prog,
80 const struct gl_program *prog,
81 struct brw_gs_prog_data *prog_data)
82 {
83 /* In gen6 we reserve the first BRW_MAX_SOL_BINDINGS entries for transform
84 * feedback surfaces.
85 */
86 uint32_t reserved = devinfo->gen == 6 ? BRW_MAX_SOL_BINDINGS : 0;
87
88 brw_assign_common_binding_table_offsets(MESA_SHADER_GEOMETRY, devinfo,
89 shader_prog, prog,
90 &prog_data->base.base,
91 reserved);
92 }
93
94 bool
95 brw_codegen_gs_prog(struct brw_context *brw,
96 struct gl_shader_program *prog,
97 struct brw_program *gp,
98 struct brw_gs_prog_key *key)
99 {
100 struct brw_compiler *compiler = brw->screen->compiler;
101 const struct gen_device_info *devinfo = &brw->screen->devinfo;
102 struct brw_stage_state *stage_state = &brw->gs.base;
103 struct brw_gs_prog_data prog_data;
104 bool start_busy = false;
105 double start_time = 0;
106
107 memset(&prog_data, 0, sizeof(prog_data));
108
109 assign_gs_binding_table_offsets(devinfo, prog, &gp->program, &prog_data);
110
111 /* Allocate the references to the uniforms that will end up in the
112 * prog_data associated with the compiled program, and which will be freed
113 * by the state cache.
114 *
115 * Note: param_count needs to be num_uniform_components * 4, since we add
116 * padding around uniform values below vec4 size, so the worst case is that
117 * every uniform is a float which gets padded to the size of a vec4.
118 */
119 int param_count = gp->program.nir->num_uniforms / 4;
120
121 prog_data.base.base.param =
122 rzalloc_array(NULL, const gl_constant_value *, param_count);
123 prog_data.base.base.pull_param =
124 rzalloc_array(NULL, const gl_constant_value *, param_count);
125 prog_data.base.base.image_param =
126 rzalloc_array(NULL, struct brw_image_param,
127 gp->program.info.num_images);
128 prog_data.base.base.nr_params = param_count;
129 prog_data.base.base.nr_image_params = gp->program.info.num_images;
130
131 brw_nir_setup_glsl_uniforms(gp->program.nir, prog, &gp->program,
132 &prog_data.base.base,
133 compiler->scalar_stage[MESA_SHADER_GEOMETRY]);
134
135 uint64_t outputs_written = gp->program.info.outputs_written;
136
137 brw_compute_vue_map(devinfo,
138 &prog_data.base.vue_map, outputs_written,
139 prog->SeparateShader);
140
141 int st_index = -1;
142 if (INTEL_DEBUG & DEBUG_SHADER_TIME)
143 st_index = brw_get_shader_time_index(brw, prog, NULL, ST_GS);
144
145 if (unlikely(brw->perf_debug)) {
146 start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
147 start_time = get_time();
148 }
149
150 void *mem_ctx = ralloc_context(NULL);
151 unsigned program_size;
152 char *error_str;
153 const unsigned *program =
154 brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
155 &prog_data, gp->program.nir, prog,
156 st_index, &program_size, &error_str);
157 if (program == NULL) {
158 ralloc_strcat(&prog->data->InfoLog, error_str);
159 _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
160
161 ralloc_free(mem_ctx);
162 return false;
163 }
164
165 if (unlikely(brw->perf_debug)) {
166 if (gp->compiled_once) {
167 brw_gs_debug_recompile(brw, prog, key);
168 }
169 if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
170 perf_debug("GS compile took %.03f ms and stalled the GPU\n",
171 (get_time() - start_time) * 1000);
172 }
173 gp->compiled_once = true;
174 }
175
176 /* Scratch space is used for register spilling */
177 brw_alloc_stage_scratch(brw, stage_state,
178 prog_data.base.base.total_scratch,
179 devinfo->max_gs_threads);
180
181 brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
182 key, sizeof(*key),
183 program, program_size,
184 &prog_data, sizeof(prog_data),
185 &stage_state->prog_offset, &brw->gs.base.prog_data);
186 ralloc_free(mem_ctx);
187
188 return true;
189 }
190
191 static bool
192 brw_gs_state_dirty(const struct brw_context *brw)
193 {
194 return brw_state_dirty(brw,
195 _NEW_TEXTURE,
196 BRW_NEW_GEOMETRY_PROGRAM |
197 BRW_NEW_TRANSFORM_FEEDBACK);
198 }
199
200 void
201 brw_gs_populate_key(struct brw_context *brw,
202 struct brw_gs_prog_key *key)
203 {
204 struct gl_context *ctx = &brw->ctx;
205 struct brw_program *gp = (struct brw_program *) brw->geometry_program;
206
207 memset(key, 0, sizeof(*key));
208
209 key->program_string_id = gp->id;
210
211 /* _NEW_TEXTURE */
212 brw_populate_sampler_prog_key_data(ctx, &gp->program, &key->tex);
213 }
214
215 void
216 brw_upload_gs_prog(struct brw_context *brw)
217 {
218 struct gl_context *ctx = &brw->ctx;
219 struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
220 struct brw_stage_state *stage_state = &brw->gs.base;
221 struct brw_gs_prog_key key;
222 /* BRW_NEW_GEOMETRY_PROGRAM */
223 struct brw_program *gp = (struct brw_program *) brw->geometry_program;
224
225 if (!brw_gs_state_dirty(brw))
226 return;
227
228 if (gp == NULL) {
229 /* No geometry shader. Vertex data just passes straight through. */
230 if (brw->gen == 6 &&
231 (brw->ctx.NewDriverState & BRW_NEW_TRANSFORM_FEEDBACK)) {
232 gen6_brw_upload_ff_gs_prog(brw);
233 return;
234 }
235
236 /* Other state atoms had better not try to access prog_data, since
237 * there's no GS program.
238 */
239 brw->gs.base.prog_data = NULL;
240
241 return;
242 }
243
244 brw_gs_populate_key(brw, &key);
245
246 if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
247 &key, sizeof(key),
248 &stage_state->prog_offset,
249 &brw->gs.base.prog_data)) {
250 bool success = brw_codegen_gs_prog(brw, current[MESA_SHADER_GEOMETRY],
251 gp, &key);
252 assert(success);
253 (void)success;
254 }
255 }
256
257 bool
258 brw_gs_precompile(struct gl_context *ctx,
259 struct gl_shader_program *shader_prog,
260 struct gl_program *prog)
261 {
262 struct brw_context *brw = brw_context(ctx);
263 struct brw_gs_prog_key key;
264 uint32_t old_prog_offset = brw->gs.base.prog_offset;
265 struct brw_stage_prog_data *old_prog_data = brw->gs.base.prog_data;
266 bool success;
267
268 struct brw_program *bgp = brw_program(prog);
269
270 memset(&key, 0, sizeof(key));
271
272 brw_setup_tex_for_precompile(brw, &key.tex, prog);
273 key.program_string_id = bgp->id;
274
275 success = brw_codegen_gs_prog(brw, shader_prog, bgp, &key);
276
277 brw->gs.base.prog_offset = old_prog_offset;
278 brw->gs.base.prog_data = old_prog_data;
279
280 return success;
281 }