i965: Move brw_get_shader_time_index() call out of emit functions
[mesa.git] / src / mesa / drivers / dri / i965 / brw_wm.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 #include "brw_context.h"
27 #include "brw_wm.h"
28 #include "brw_state.h"
29 #include "brw_shader.h"
30 #include "main/enums.h"
31 #include "main/formats.h"
32 #include "main/fbobject.h"
33 #include "main/samplerobj.h"
34 #include "main/framebuffer.h"
35 #include "program/prog_parameter.h"
36 #include "program/program.h"
37 #include "intel_mipmap_tree.h"
38 #include "brw_nir.h"
39
40 #include "util/ralloc.h"
41
42 /**
43 * Return a bitfield where bit n is set if barycentric interpolation mode n
44 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
45 */
46 static unsigned
47 brw_compute_barycentric_interp_modes(const struct brw_device_info *devinfo,
48 bool shade_model_flat,
49 bool persample_shading,
50 nir_shader *shader)
51 {
52 unsigned barycentric_interp_modes = 0;
53
54 nir_foreach_variable(var, &shader->inputs) {
55 enum glsl_interp_qualifier interp_qualifier = var->data.interpolation;
56 bool is_centroid = var->data.centroid && !persample_shading;
57 bool is_sample = var->data.sample || persample_shading;
58 bool is_gl_Color = (var->data.location == VARYING_SLOT_COL0) ||
59 (var->data.location == VARYING_SLOT_COL1);
60
61 /* Ignore WPOS and FACE, because they don't require interpolation. */
62 if (var->data.location == VARYING_SLOT_POS ||
63 var->data.location == VARYING_SLOT_FACE)
64 continue;
65
66 /* Determine the set (or sets) of barycentric coordinates needed to
67 * interpolate this variable. Note that when
68 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
69 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
70 * for lit pixels, so we need both sets of barycentric coordinates.
71 */
72 if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
73 if (is_centroid) {
74 barycentric_interp_modes |=
75 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
76 } else if (is_sample) {
77 barycentric_interp_modes |=
78 1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
79 }
80 if ((!is_centroid && !is_sample) ||
81 devinfo->needs_unlit_centroid_workaround) {
82 barycentric_interp_modes |=
83 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
84 }
85 } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
86 (!(shade_model_flat && is_gl_Color) &&
87 interp_qualifier == INTERP_QUALIFIER_NONE)) {
88 if (is_centroid) {
89 barycentric_interp_modes |=
90 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
91 } else if (is_sample) {
92 barycentric_interp_modes |=
93 1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
94 }
95 if ((!is_centroid && !is_sample) ||
96 devinfo->needs_unlit_centroid_workaround) {
97 barycentric_interp_modes |=
98 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
99 }
100 }
101 }
102
103 return barycentric_interp_modes;
104 }
105
106 static uint8_t
107 computed_depth_mode(struct gl_fragment_program *fp)
108 {
109 if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
110 switch (fp->FragDepthLayout) {
111 case FRAG_DEPTH_LAYOUT_NONE:
112 case FRAG_DEPTH_LAYOUT_ANY:
113 return BRW_PSCDEPTH_ON;
114 case FRAG_DEPTH_LAYOUT_GREATER:
115 return BRW_PSCDEPTH_ON_GE;
116 case FRAG_DEPTH_LAYOUT_LESS:
117 return BRW_PSCDEPTH_ON_LE;
118 case FRAG_DEPTH_LAYOUT_UNCHANGED:
119 return BRW_PSCDEPTH_OFF;
120 }
121 }
122 return BRW_PSCDEPTH_OFF;
123 }
124
125 static void
126 assign_fs_binding_table_offsets(const struct brw_device_info *devinfo,
127 const struct gl_shader_program *shader_prog,
128 const struct gl_program *prog,
129 const struct brw_wm_prog_key *key,
130 struct brw_wm_prog_data *prog_data)
131 {
132 uint32_t next_binding_table_offset = 0;
133
134 /* If there are no color regions, we still perform an FB write to a null
135 * renderbuffer, which we place at surface index 0.
136 */
137 prog_data->binding_table.render_target_start = next_binding_table_offset;
138 next_binding_table_offset += MAX2(key->nr_color_regions, 1);
139
140 brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
141 shader_prog, prog, &prog_data->base,
142 next_binding_table_offset);
143 }
144
145 /**
146 * All Mesa program -> GPU code generation goes through this function.
147 * Depending on the instructions used (i.e. flow control instructions)
148 * we'll use one of two code generators.
149 */
150 bool
151 brw_codegen_wm_prog(struct brw_context *brw,
152 struct gl_shader_program *prog,
153 struct brw_fragment_program *fp,
154 struct brw_wm_prog_key *key)
155 {
156 struct gl_context *ctx = &brw->ctx;
157 void *mem_ctx = ralloc_context(NULL);
158 struct brw_wm_prog_data prog_data;
159 const GLuint *program;
160 struct brw_shader *fs = NULL;
161 GLuint program_size;
162 bool start_busy = false;
163 double start_time = 0;
164
165 if (prog)
166 fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
167
168 memset(&prog_data, 0, sizeof(prog_data));
169 /* key->alpha_test_func means simulating alpha testing via discards,
170 * so the shader definitely kills pixels.
171 */
172 prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
173 prog_data.uses_omask =
174 fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
175 prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
176
177 prog_data.early_fragment_tests = fs && fs->base.EarlyFragmentTests;
178
179 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
180 if (!prog)
181 prog_data.base.use_alt_mode = true;
182
183 assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
184 &fp->program.Base, key, &prog_data);
185
186 /* Allocate the references to the uniforms that will end up in the
187 * prog_data associated with the compiled program, and which will be freed
188 * by the state cache.
189 */
190 int param_count = fp->program.Base.nir->num_uniforms;
191 if (fs)
192 prog_data.base.nr_image_params = fs->base.NumImages;
193 /* The backend also sometimes adds params for texture size. */
194 param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
195 prog_data.base.param =
196 rzalloc_array(NULL, const gl_constant_value *, param_count);
197 prog_data.base.pull_param =
198 rzalloc_array(NULL, const gl_constant_value *, param_count);
199 prog_data.base.image_param =
200 rzalloc_array(NULL, struct brw_image_param,
201 prog_data.base.nr_image_params);
202 prog_data.base.nr_params = param_count;
203
204 if (prog) {
205 brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base,
206 &prog_data.base, true);
207 } else {
208 brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
209 &prog_data.base);
210 }
211
212 prog_data.barycentric_interp_modes =
213 brw_compute_barycentric_interp_modes(brw->intelScreen->devinfo,
214 key->flat_shade,
215 key->persample_shading,
216 fp->program.Base.nir);
217
218 if (unlikely(brw->perf_debug)) {
219 start_busy = (brw->batch.last_bo &&
220 drm_intel_bo_busy(brw->batch.last_bo));
221 start_time = get_time();
222 }
223
224 if (unlikely(INTEL_DEBUG & DEBUG_WM))
225 brw_dump_ir("fragment", prog, &fs->base, &fp->program.Base);
226
227 int st_index8 = -1, st_index16 = -1;
228 if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
229 st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8);
230 st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16);
231 }
232
233 program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
234 &fp->program, prog, st_index8, st_index16, &program_size);
235 if (program == NULL) {
236 ralloc_free(mem_ctx);
237 return false;
238 }
239
240 if (unlikely(brw->perf_debug) && fs) {
241 if (fs->compiled_once)
242 brw_wm_debug_recompile(brw, prog, key);
243 fs->compiled_once = true;
244
245 if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
246 perf_debug("FS compile took %.03f ms and stalled the GPU\n",
247 (get_time() - start_time) * 1000);
248 }
249 }
250
251 if (prog_data.base.total_scratch) {
252 brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
253 prog_data.base.total_scratch * brw->max_wm_threads);
254 }
255
256 if (unlikely(INTEL_DEBUG & DEBUG_WM))
257 fprintf(stderr, "\n");
258
259 brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
260 key, sizeof(struct brw_wm_prog_key),
261 program, program_size,
262 &prog_data, sizeof(prog_data),
263 &brw->wm.base.prog_offset, &brw->wm.prog_data);
264
265 ralloc_free(mem_ctx);
266
267 return true;
268 }
269
270 static bool
271 key_debug(struct brw_context *brw, const char *name, int a, int b)
272 {
273 if (a != b) {
274 perf_debug(" %s %d->%d\n", name, a, b);
275 return true;
276 } else {
277 return false;
278 }
279 }
280
281 bool
282 brw_debug_recompile_sampler_key(struct brw_context *brw,
283 const struct brw_sampler_prog_key_data *old_key,
284 const struct brw_sampler_prog_key_data *key)
285 {
286 bool found = false;
287
288 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
289 found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
290 old_key->swizzles[i], key->swizzles[i]);
291 }
292 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
293 old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
294 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
295 old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
296 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
297 old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
298 found |= key_debug(brw, "gather channel quirk on any texture unit",
299 old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
300 found |= key_debug(brw, "compressed multisample layout",
301 old_key->compressed_multisample_layout_mask,
302 key->compressed_multisample_layout_mask);
303
304 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
305 found |= key_debug(brw, "textureGather workarounds",
306 old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
307 }
308
309 return found;
310 }
311
312 void
313 brw_wm_debug_recompile(struct brw_context *brw,
314 struct gl_shader_program *prog,
315 const struct brw_wm_prog_key *key)
316 {
317 struct brw_cache_item *c = NULL;
318 const struct brw_wm_prog_key *old_key = NULL;
319 bool found = false;
320
321 perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
322
323 for (unsigned int i = 0; i < brw->cache.size; i++) {
324 for (c = brw->cache.items[i]; c; c = c->next) {
325 if (c->cache_id == BRW_CACHE_FS_PROG) {
326 old_key = c->key;
327
328 if (old_key->program_string_id == key->program_string_id)
329 break;
330 }
331 }
332 if (c)
333 break;
334 }
335
336 if (!c) {
337 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
338 return;
339 }
340
341 found |= key_debug(brw, "alphatest, computed depth, depth test, or "
342 "depth write",
343 old_key->iz_lookup, key->iz_lookup);
344 found |= key_debug(brw, "depth statistics",
345 old_key->stats_wm, key->stats_wm);
346 found |= key_debug(brw, "flat shading",
347 old_key->flat_shade, key->flat_shade);
348 found |= key_debug(brw, "per-sample shading",
349 old_key->persample_shading, key->persample_shading);
350 found |= key_debug(brw, "per-sample shading and 2x MSAA",
351 old_key->persample_2x, key->persample_2x);
352 found |= key_debug(brw, "number of color buffers",
353 old_key->nr_color_regions, key->nr_color_regions);
354 found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
355 old_key->replicate_alpha, key->replicate_alpha);
356 found |= key_debug(brw, "rendering to FBO",
357 old_key->render_to_fbo, key->render_to_fbo);
358 found |= key_debug(brw, "fragment color clamping",
359 old_key->clamp_fragment_color, key->clamp_fragment_color);
360 found |= key_debug(brw, "line smoothing",
361 old_key->line_aa, key->line_aa);
362 found |= key_debug(brw, "renderbuffer height",
363 old_key->drawable_height, key->drawable_height);
364 found |= key_debug(brw, "input slots valid",
365 old_key->input_slots_valid, key->input_slots_valid);
366 found |= key_debug(brw, "mrt alpha test function",
367 old_key->alpha_test_func, key->alpha_test_func);
368 found |= key_debug(brw, "mrt alpha test reference value",
369 old_key->alpha_test_ref, key->alpha_test_ref);
370
371 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
372
373 if (!found) {
374 perf_debug(" Something else\n");
375 }
376 }
377
378 static uint8_t
379 gen6_gather_workaround(GLenum internalformat)
380 {
381 switch (internalformat) {
382 case GL_R8I: return WA_SIGN | WA_8BIT;
383 case GL_R8UI: return WA_8BIT;
384 case GL_R16I: return WA_SIGN | WA_16BIT;
385 case GL_R16UI: return WA_16BIT;
386 default:
387 /* Note that even though GL_R32I and GL_R32UI have format overrides in
388 * the surface state, there is no shader w/a required.
389 */
390 return 0;
391 }
392 }
393
394 void
395 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
396 const struct gl_program *prog,
397 unsigned sampler_count,
398 struct brw_sampler_prog_key_data *key)
399 {
400 struct brw_context *brw = brw_context(ctx);
401
402 for (int s = 0; s < sampler_count; s++) {
403 key->swizzles[s] = SWIZZLE_NOOP;
404
405 if (!(prog->SamplersUsed & (1 << s)))
406 continue;
407
408 int unit_id = prog->SamplerUnits[s];
409 const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
410
411 if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
412 const struct gl_texture_object *t = unit->_Current;
413 const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
414 struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
415
416 const bool alpha_depth = t->DepthMode == GL_ALPHA &&
417 (img->_BaseFormat == GL_DEPTH_COMPONENT ||
418 img->_BaseFormat == GL_DEPTH_STENCIL);
419
420 /* Haswell handles texture swizzling as surface format overrides
421 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
422 */
423 if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
424 key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
425
426 if (brw->gen < 8 &&
427 sampler->MinFilter != GL_NEAREST &&
428 sampler->MagFilter != GL_NEAREST) {
429 if (sampler->WrapS == GL_CLAMP)
430 key->gl_clamp_mask[0] |= 1 << s;
431 if (sampler->WrapT == GL_CLAMP)
432 key->gl_clamp_mask[1] |= 1 << s;
433 if (sampler->WrapR == GL_CLAMP)
434 key->gl_clamp_mask[2] |= 1 << s;
435 }
436
437 /* gather4's channel select for green from RG32F is broken; requires
438 * a shader w/a on IVB; fixable with just SCS on HSW.
439 */
440 if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
441 if (img->InternalFormat == GL_RG32F)
442 key->gather_channel_quirk_mask |= 1 << s;
443 }
444
445 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
446 * UNORM/FLOAT instead and fix it in the shader.
447 */
448 if (brw->gen == 6 && prog->UsesGather) {
449 key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
450 }
451
452 /* If this is a multisample sampler, and uses the CMS MSAA layout,
453 * then we need to emit slightly different code to first sample the
454 * MCS surface.
455 */
456 struct intel_texture_object *intel_tex =
457 intel_texture_object((struct gl_texture_object *)t);
458
459 if (brw->gen >= 7 &&
460 intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
461 key->compressed_multisample_layout_mask |= 1 << s;
462 }
463 }
464 }
465 }
466
467 static bool
468 brw_wm_state_dirty (struct brw_context *brw)
469 {
470 return brw_state_dirty(brw,
471 _NEW_BUFFERS |
472 _NEW_COLOR |
473 _NEW_DEPTH |
474 _NEW_FRAG_CLAMP |
475 _NEW_HINT |
476 _NEW_LIGHT |
477 _NEW_LINE |
478 _NEW_MULTISAMPLE |
479 _NEW_POLYGON |
480 _NEW_STENCIL |
481 _NEW_TEXTURE,
482 BRW_NEW_FRAGMENT_PROGRAM |
483 BRW_NEW_REDUCED_PRIMITIVE |
484 BRW_NEW_STATS_WM |
485 BRW_NEW_VUE_MAP_GEOM_OUT);
486 }
487
488 static void
489 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
490 {
491 struct gl_context *ctx = &brw->ctx;
492 /* BRW_NEW_FRAGMENT_PROGRAM */
493 const struct brw_fragment_program *fp =
494 (struct brw_fragment_program *) brw->fragment_program;
495 const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
496 GLuint lookup = 0;
497 GLuint line_aa;
498 bool program_uses_dfdy = fp->program.UsesDFdy;
499 const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
500
501 memset(key, 0, sizeof(*key));
502
503 /* Build the index for table lookup
504 */
505 if (brw->gen < 6) {
506 /* _NEW_COLOR */
507 if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
508 lookup |= IZ_PS_KILL_ALPHATEST_BIT;
509
510 if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
511 lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
512
513 /* _NEW_DEPTH */
514 if (ctx->Depth.Test)
515 lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
516
517 if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
518 lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
519
520 /* _NEW_STENCIL | _NEW_BUFFERS */
521 if (ctx->Stencil._Enabled) {
522 lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
523
524 if (ctx->Stencil.WriteMask[0] ||
525 ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
526 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
527 }
528 key->iz_lookup = lookup;
529 }
530
531 line_aa = AA_NEVER;
532
533 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
534 if (ctx->Line.SmoothFlag) {
535 if (brw->reduced_primitive == GL_LINES) {
536 line_aa = AA_ALWAYS;
537 }
538 else if (brw->reduced_primitive == GL_TRIANGLES) {
539 if (ctx->Polygon.FrontMode == GL_LINE) {
540 line_aa = AA_SOMETIMES;
541
542 if (ctx->Polygon.BackMode == GL_LINE ||
543 (ctx->Polygon.CullFlag &&
544 ctx->Polygon.CullFaceMode == GL_BACK))
545 line_aa = AA_ALWAYS;
546 }
547 else if (ctx->Polygon.BackMode == GL_LINE) {
548 line_aa = AA_SOMETIMES;
549
550 if ((ctx->Polygon.CullFlag &&
551 ctx->Polygon.CullFaceMode == GL_FRONT))
552 line_aa = AA_ALWAYS;
553 }
554 }
555 }
556
557 key->line_aa = line_aa;
558
559 /* _NEW_HINT */
560 key->high_quality_derivatives =
561 ctx->Hint.FragmentShaderDerivative == GL_NICEST;
562
563 if (brw->gen < 6)
564 key->stats_wm = brw->stats_wm;
565
566 /* _NEW_LIGHT */
567 key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
568
569 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
570 key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
571
572 /* _NEW_TEXTURE */
573 brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
574 &key->tex);
575
576 /* _NEW_BUFFERS */
577 /*
578 * Include the draw buffer origin and height so that we can calculate
579 * fragment position values relative to the bottom left of the drawable,
580 * from the incoming screen origin relative position we get as part of our
581 * payload.
582 *
583 * This is only needed for the WM_WPOSXY opcode when the fragment program
584 * uses the gl_FragCoord input.
585 *
586 * We could avoid recompiling by including this as a constant referenced by
587 * our program, but if we were to do that it would also be nice to handle
588 * getting that constant updated at batchbuffer submit time (when we
589 * hold the lock and know where the buffer really is) rather than at emit
590 * time when we don't hold the lock and are just guessing. We could also
591 * just avoid using this as key data if the program doesn't use
592 * fragment.position.
593 *
594 * For DRI2 the origin_x/y will always be (0,0) but we still need the
595 * drawable height in order to invert the Y axis.
596 */
597 if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
598 key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
599 }
600
601 if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
602 key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
603 }
604
605 /* _NEW_BUFFERS */
606 key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
607
608 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
609 key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
610 (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
611
612 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
613 /* Ignore sample qualifier while computing this flag. */
614 key->persample_shading =
615 _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
616 if (key->persample_shading)
617 key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
618
619 key->compute_pos_offset =
620 _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
621 fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
622
623 key->compute_sample_id =
624 multisample_fbo &&
625 ctx->Multisample.Enabled &&
626 (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
627
628 /* BRW_NEW_VUE_MAP_GEOM_OUT */
629 if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
630 BRW_FS_VARYING_INPUT_MASK) > 16)
631 key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
632
633
634 /* _NEW_COLOR | _NEW_BUFFERS */
635 /* Pre-gen6, the hardware alpha test always used each render
636 * target's alpha to do alpha test, as opposed to render target 0's alpha
637 * like GL requires. Fix that by building the alpha test into the
638 * shader, and we'll skip enabling the fixed function alpha test.
639 */
640 if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
641 ctx->Color.AlphaEnabled) {
642 key->alpha_test_func = ctx->Color.AlphaFunc;
643 key->alpha_test_ref = ctx->Color.AlphaRef;
644 }
645
646 /* The unique fragment program ID */
647 key->program_string_id = fp->id;
648 }
649
650 void
651 brw_upload_wm_prog(struct brw_context *brw)
652 {
653 struct gl_context *ctx = &brw->ctx;
654 struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
655 struct brw_wm_prog_key key;
656 struct brw_fragment_program *fp = (struct brw_fragment_program *)
657 brw->fragment_program;
658
659 if (!brw_wm_state_dirty(brw))
660 return;
661
662 brw_wm_populate_key(brw, &key);
663
664 if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
665 &key, sizeof(key),
666 &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
667 bool success = brw_codegen_wm_prog(brw, current, fp, &key);
668 (void) success;
669 assert(success);
670 }
671 brw->wm.base.prog_data = &brw->wm.prog_data->base;
672 }
673
674 bool
675 brw_fs_precompile(struct gl_context *ctx,
676 struct gl_shader_program *shader_prog,
677 struct gl_program *prog)
678 {
679 struct brw_context *brw = brw_context(ctx);
680 struct brw_wm_prog_key key;
681
682 struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
683 struct brw_fragment_program *bfp = brw_fragment_program(fp);
684 bool program_uses_dfdy = fp->UsesDFdy;
685
686 memset(&key, 0, sizeof(key));
687
688 if (brw->gen < 6) {
689 if (fp->UsesKill)
690 key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
691
692 if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
693 key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
694
695 /* Just assume depth testing. */
696 key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
697 key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
698 }
699
700 if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
701 BRW_FS_VARYING_INPUT_MASK) > 16)
702 key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
703
704 brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
705
706 if (fp->Base.InputsRead & VARYING_BIT_POS) {
707 key.drawable_height = ctx->DrawBuffer->Height;
708 }
709
710 key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
711 ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
712 BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
713
714 if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
715 key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
716 key.nr_color_regions > 1;
717 }
718
719 key.program_string_id = bfp->id;
720
721 uint32_t old_prog_offset = brw->wm.base.prog_offset;
722 struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
723
724 bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
725
726 brw->wm.base.prog_offset = old_prog_offset;
727 brw->wm.prog_data = old_prog_data;
728
729 return success;
730 }