i965: make vs and fs key generation helpers available to shader cache
[mesa.git] / src / mesa / drivers / dri / i965 / brw_vs.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keithw@vmware.com>
30 */
31
32
33 #include "main/compiler.h"
34 #include "main/context.h"
35 #include "brw_context.h"
36 #include "brw_vs.h"
37 #include "brw_util.h"
38 #include "brw_state.h"
39 #include "program/prog_print.h"
40 #include "program/prog_parameter.h"
41 #include "brw_nir.h"
42 #include "brw_program.h"
43
44 #include "util/ralloc.h"
45
46 GLbitfield64
47 brw_vs_outputs_written(struct brw_context *brw, struct brw_vs_prog_key *key,
48 GLbitfield64 user_varyings)
49 {
50 GLbitfield64 outputs_written = user_varyings;
51
52 if (key->copy_edgeflag) {
53 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
54 }
55
56 if (brw->gen < 6) {
57 /* Put dummy slots into the VUE for the SF to put the replaced
58 * point sprite coords in. We shouldn't need these dummy slots,
59 * which take up precious URB space, but it would mean that the SF
60 * doesn't get nice aligned pairs of input coords into output
61 * coords, which would be a pain to handle.
62 */
63 for (unsigned i = 0; i < 8; i++) {
64 if (key->point_coord_replace & (1 << i))
65 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
66 }
67
68 /* if back colors are written, allocate slots for front colors too */
69 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
70 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
71 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
72 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
73 }
74
75 /* In order for legacy clipping to work, we need to populate the clip
76 * distance varying slots whenever clipping is enabled, even if the vertex
77 * shader doesn't write to gl_ClipDistance.
78 */
79 if (key->nr_userclip_plane_consts > 0) {
80 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
81 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
82 }
83
84 return outputs_written;
85 }
86
87 bool
88 brw_codegen_vs_prog(struct brw_context *brw,
89 struct gl_shader_program *prog,
90 struct brw_vertex_program *vp,
91 struct brw_vs_prog_key *key)
92 {
93 const struct brw_compiler *compiler = brw->screen->compiler;
94 const struct gen_device_info *devinfo = &brw->screen->devinfo;
95 GLuint program_size;
96 const GLuint *program;
97 struct brw_vs_prog_data prog_data;
98 struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
99 void *mem_ctx;
100 struct brw_shader *vs = NULL;
101 bool start_busy = false;
102 double start_time = 0;
103
104 if (prog)
105 vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
106
107 memset(&prog_data, 0, sizeof(prog_data));
108
109 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
110 if (!prog)
111 stage_prog_data->use_alt_mode = true;
112
113 mem_ctx = ralloc_context(NULL);
114
115 brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX,
116 devinfo,
117 prog, &vp->program.Base,
118 &prog_data.base.base, 0);
119
120 /* Allocate the references to the uniforms that will end up in the
121 * prog_data associated with the compiled program, and which will be freed
122 * by the state cache.
123 */
124 int param_count = vp->program.Base.nir->num_uniforms / 4;
125
126 if (vs)
127 prog_data.base.base.nr_image_params = vs->base.NumImages;
128
129 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
130 * planes as uniforms.
131 */
132 param_count += key->nr_userclip_plane_consts * 4;
133
134 stage_prog_data->param =
135 rzalloc_array(NULL, const gl_constant_value *, param_count);
136 stage_prog_data->pull_param =
137 rzalloc_array(NULL, const gl_constant_value *, param_count);
138 stage_prog_data->image_param =
139 rzalloc_array(NULL, struct brw_image_param,
140 stage_prog_data->nr_image_params);
141 stage_prog_data->nr_params = param_count;
142
143 if (prog) {
144 brw_nir_setup_glsl_uniforms(vp->program.Base.nir, prog, &vp->program.Base,
145 &prog_data.base.base,
146 compiler->scalar_stage[MESA_SHADER_VERTEX]);
147 } else {
148 brw_nir_setup_arb_uniforms(vp->program.Base.nir, &vp->program.Base,
149 &prog_data.base.base);
150 }
151
152 GLbitfield64 outputs_written =
153 brw_vs_outputs_written(brw, key, vp->program.Base.OutputsWritten);
154 prog_data.inputs_read = vp->program.Base.InputsRead;
155
156 if (key->copy_edgeflag) {
157 prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
158 }
159
160 prog_data.base.cull_distance_mask =
161 ((1 << vp->program.Base.CullDistanceArraySize) - 1) <<
162 vp->program.Base.ClipDistanceArraySize;
163
164 brw_compute_vue_map(devinfo,
165 &prog_data.base.vue_map, outputs_written,
166 prog ? prog->SeparateShader ||
167 prog->_LinkedShaders[MESA_SHADER_TESS_EVAL]
168 : false);
169
170 if (0) {
171 _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
172 true);
173 }
174
175 if (unlikely(brw->perf_debug)) {
176 start_busy = (brw->batch.last_bo &&
177 drm_intel_bo_busy(brw->batch.last_bo));
178 start_time = get_time();
179 }
180
181 if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
182 brw_dump_ir("vertex", prog, vs ? &vs->base : NULL, &vp->program.Base);
183
184 fprintf(stderr, "VS Output ");
185 brw_print_vue_map(stderr, &prog_data.base.vue_map);
186 }
187
188 int st_index = -1;
189 if (INTEL_DEBUG & DEBUG_SHADER_TIME)
190 st_index = brw_get_shader_time_index(brw, prog, &vp->program.Base, ST_VS);
191
192 /* Emit GEN4 code.
193 */
194 char *error_str;
195 program = brw_compile_vs(compiler, brw, mem_ctx, key,
196 &prog_data, vp->program.Base.nir,
197 brw_select_clip_planes(&brw->ctx),
198 !_mesa_is_gles3(&brw->ctx),
199 st_index, &program_size, &error_str);
200 if (program == NULL) {
201 if (prog) {
202 prog->LinkStatus = false;
203 ralloc_strcat(&prog->InfoLog, error_str);
204 }
205
206 _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str);
207
208 ralloc_free(mem_ctx);
209 return false;
210 }
211
212 if (unlikely(brw->perf_debug) && vs) {
213 if (vs->compiled_once) {
214 brw_vs_debug_recompile(brw, prog, key);
215 }
216 if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
217 perf_debug("VS compile took %.03f ms and stalled the GPU\n",
218 (get_time() - start_time) * 1000);
219 }
220 vs->compiled_once = true;
221 }
222
223 /* Scratch space is used for register spilling */
224 brw_alloc_stage_scratch(brw, &brw->vs.base,
225 prog_data.base.base.total_scratch,
226 devinfo->max_vs_threads);
227
228 brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
229 key, sizeof(struct brw_vs_prog_key),
230 program, program_size,
231 &prog_data, sizeof(prog_data),
232 &brw->vs.base.prog_offset, &brw->vs.prog_data);
233 ralloc_free(mem_ctx);
234
235 return true;
236 }
237
238 void
239 brw_vs_debug_recompile(struct brw_context *brw,
240 struct gl_shader_program *prog,
241 const struct brw_vs_prog_key *key)
242 {
243 struct brw_cache_item *c = NULL;
244 const struct brw_vs_prog_key *old_key = NULL;
245 bool found = false;
246
247 perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
248
249 for (unsigned int i = 0; i < brw->cache.size; i++) {
250 for (c = brw->cache.items[i]; c; c = c->next) {
251 if (c->cache_id == BRW_CACHE_VS_PROG) {
252 old_key = c->key;
253
254 if (old_key->program_string_id == key->program_string_id)
255 break;
256 }
257 }
258 if (c)
259 break;
260 }
261
262 if (!c) {
263 perf_debug(" Didn't find previous compile in the shader cache for "
264 "debug\n");
265 return;
266 }
267
268 for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
269 found |= key_debug(brw, "Vertex attrib w/a flags",
270 old_key->gl_attrib_wa_flags[i],
271 key->gl_attrib_wa_flags[i]);
272 }
273
274 found |= key_debug(brw, "legacy user clipping",
275 old_key->nr_userclip_plane_consts,
276 key->nr_userclip_plane_consts);
277
278 found |= key_debug(brw, "copy edgeflag",
279 old_key->copy_edgeflag, key->copy_edgeflag);
280 found |= key_debug(brw, "PointCoord replace",
281 old_key->point_coord_replace, key->point_coord_replace);
282 found |= key_debug(brw, "vertex color clamping",
283 old_key->clamp_vertex_color, key->clamp_vertex_color);
284
285 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
286
287 if (!found) {
288 perf_debug(" Something else\n");
289 }
290 }
291
292 static bool
293 brw_vs_state_dirty(const struct brw_context *brw)
294 {
295 return brw_state_dirty(brw,
296 _NEW_BUFFERS |
297 _NEW_LIGHT |
298 _NEW_POINT |
299 _NEW_POLYGON |
300 _NEW_TEXTURE |
301 _NEW_TRANSFORM,
302 BRW_NEW_VERTEX_PROGRAM |
303 BRW_NEW_VS_ATTRIB_WORKAROUNDS);
304 }
305
306 void
307 brw_vs_populate_key(struct brw_context *brw,
308 struct brw_vs_prog_key *key)
309 {
310 struct gl_context *ctx = &brw->ctx;
311 /* BRW_NEW_VERTEX_PROGRAM */
312 struct brw_vertex_program *vp =
313 (struct brw_vertex_program *)brw->vertex_program;
314 struct gl_program *prog = (struct gl_program *) brw->vertex_program;
315
316 memset(key, 0, sizeof(*key));
317
318 /* Just upload the program verbatim for now. Always send it all
319 * the inputs it asks for, whether they are varying or not.
320 */
321 key->program_string_id = vp->id;
322
323 if (ctx->Transform.ClipPlanesEnabled != 0 &&
324 (ctx->API == API_OPENGL_COMPAT ||
325 ctx->API == API_OPENGLES) &&
326 vp->program.Base.ClipDistanceArraySize == 0) {
327 key->nr_userclip_plane_consts =
328 _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
329 }
330
331 if (brw->gen < 6) {
332 /* _NEW_POLYGON */
333 key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
334 ctx->Polygon.BackMode != GL_FILL);
335
336 /* _NEW_POINT */
337 if (ctx->Point.PointSprite) {
338 key->point_coord_replace = ctx->Point.CoordReplace & 0xff;
339 }
340 }
341
342 if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
343 VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
344 /* _NEW_LIGHT | _NEW_BUFFERS */
345 key->clamp_vertex_color = ctx->Light._ClampVertexColor;
346 }
347
348 /* _NEW_TEXTURE */
349 brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
350
351 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
352 if (brw->gen < 8 && !brw->is_haswell) {
353 memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
354 sizeof(brw->vb.attrib_wa_flags));
355 }
356 }
357
358 void
359 brw_upload_vs_prog(struct brw_context *brw)
360 {
361 struct gl_context *ctx = &brw->ctx;
362 struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
363 struct brw_vs_prog_key key;
364 /* BRW_NEW_VERTEX_PROGRAM */
365 struct brw_vertex_program *vp =
366 (struct brw_vertex_program *)brw->vertex_program;
367
368 if (!brw_vs_state_dirty(brw))
369 return;
370
371 brw_vs_populate_key(brw, &key);
372
373 if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
374 &key, sizeof(key),
375 &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
376 bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
377 vp, &key);
378 (void) success;
379 assert(success);
380 }
381 brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
382 }
383
384 bool
385 brw_vs_precompile(struct gl_context *ctx,
386 struct gl_shader_program *shader_prog,
387 struct gl_program *prog)
388 {
389 struct brw_context *brw = brw_context(ctx);
390 struct brw_vs_prog_key key;
391 uint32_t old_prog_offset = brw->vs.base.prog_offset;
392 struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
393 bool success;
394
395 struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
396 struct brw_vertex_program *bvp = brw_vertex_program(vp);
397
398 memset(&key, 0, sizeof(key));
399
400 brw_setup_tex_for_precompile(brw, &key.tex, prog);
401 key.program_string_id = bvp->id;
402 key.clamp_vertex_color =
403 (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
404 VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
405
406 success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
407
408 brw->vs.base.prog_offset = old_prog_offset;
409 brw->vs.prog_data = old_prog_data;
410
411 return success;
412 }