Merge remote-tracking branch 'mesa-public/master' into vulkan
[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 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keithw@vmware.com>
30 */
31
32 #include "brw_context.h"
33 #include "brw_wm.h"
34 #include "brw_state.h"
35 #include "main/enums.h"
36 #include "main/formats.h"
37 #include "main/fbobject.h"
38 #include "main/samplerobj.h"
39 #include "main/framebuffer.h"
40 #include "program/prog_parameter.h"
41 #include "program/program.h"
42 #include "intel_mipmap_tree.h"
43
44 #include "util/ralloc.h"
45
46 /**
47 * Return a bitfield where bit n is set if barycentric interpolation mode n
48 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
49 */
50 unsigned
51 brw_compute_barycentric_interp_modes(struct brw_context *brw,
52 bool shade_model_flat,
53 bool persample_shading,
54 const struct gl_fragment_program *fprog)
55 {
56 unsigned barycentric_interp_modes = 0;
57 int attr;
58
59 /* Loop through all fragment shader inputs to figure out what interpolation
60 * modes are in use, and set the appropriate bits in
61 * barycentric_interp_modes.
62 */
63 for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
64 enum glsl_interp_qualifier interp_qualifier =
65 fprog->InterpQualifier[attr];
66 bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
67 !persample_shading;
68 bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
69 persample_shading;
70 bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
71
72 /* Ignore unused inputs. */
73 if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
74 continue;
75
76 /* Ignore WPOS and FACE, because they don't require interpolation. */
77 if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
78 continue;
79
80 /* Determine the set (or sets) of barycentric coordinates needed to
81 * interpolate this variable. Note that when
82 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
83 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
84 * for lit pixels, so we need both sets of barycentric coordinates.
85 */
86 if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
87 if (is_centroid) {
88 barycentric_interp_modes |=
89 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
90 } else if (is_sample) {
91 barycentric_interp_modes |=
92 1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
93 }
94 if ((!is_centroid && !is_sample) ||
95 brw->needs_unlit_centroid_workaround) {
96 barycentric_interp_modes |=
97 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
98 }
99 } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
100 (!(shade_model_flat && is_gl_Color) &&
101 interp_qualifier == INTERP_QUALIFIER_NONE)) {
102 if (is_centroid) {
103 barycentric_interp_modes |=
104 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
105 } else if (is_sample) {
106 barycentric_interp_modes |=
107 1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
108 }
109 if ((!is_centroid && !is_sample) ||
110 brw->needs_unlit_centroid_workaround) {
111 barycentric_interp_modes |=
112 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
113 }
114 }
115 }
116
117 return barycentric_interp_modes;
118 }
119
120 static uint8_t
121 computed_depth_mode(struct gl_fragment_program *fp)
122 {
123 if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
124 switch (fp->FragDepthLayout) {
125 case FRAG_DEPTH_LAYOUT_NONE:
126 case FRAG_DEPTH_LAYOUT_ANY:
127 return BRW_PSCDEPTH_ON;
128 case FRAG_DEPTH_LAYOUT_GREATER:
129 return BRW_PSCDEPTH_ON_GE;
130 case FRAG_DEPTH_LAYOUT_LESS:
131 return BRW_PSCDEPTH_ON_LE;
132 case FRAG_DEPTH_LAYOUT_UNCHANGED:
133 return BRW_PSCDEPTH_OFF;
134 }
135 }
136 return BRW_PSCDEPTH_OFF;
137 }
138
139 bool
140 brw_wm_prog_data_compare(const void *in_a, const void *in_b)
141 {
142 const struct brw_wm_prog_data *a = in_a;
143 const struct brw_wm_prog_data *b = in_b;
144
145 /* Compare the base structure. */
146 if (!brw_stage_prog_data_compare(&a->base, &b->base))
147 return false;
148
149 /* Compare the rest of the structure. */
150 const unsigned offset = sizeof(struct brw_stage_prog_data);
151 if (memcmp(((char *) a) + offset, ((char *) b) + offset,
152 sizeof(struct brw_wm_prog_data) - offset))
153 return false;
154
155 return true;
156 }
157
158 /**
159 * All Mesa program -> GPU code generation goes through this function.
160 * Depending on the instructions used (i.e. flow control instructions)
161 * we'll use one of two code generators.
162 */
163 bool
164 brw_codegen_wm_prog(struct brw_context *brw,
165 struct gl_shader_program *prog,
166 struct brw_fragment_program *fp,
167 struct brw_wm_prog_key *key)
168 {
169 struct gl_context *ctx = &brw->ctx;
170 void *mem_ctx = ralloc_context(NULL);
171 struct brw_wm_prog_data prog_data;
172 const GLuint *program;
173 struct gl_shader *fs = NULL;
174 GLuint program_size;
175
176 if (prog)
177 fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
178
179 memset(&prog_data, 0, sizeof(prog_data));
180 /* key->alpha_test_func means simulating alpha testing via discards,
181 * so the shader definitely kills pixels.
182 */
183 prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
184
185 prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
186
187 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
188 if (!prog)
189 prog_data.base.use_alt_mode = true;
190
191 /* Allocate the references to the uniforms that will end up in the
192 * prog_data associated with the compiled program, and which will be freed
193 * by the state cache.
194 */
195 int param_count;
196 if (fs) {
197 param_count = fs->num_uniform_components;
198 } else {
199 param_count = fp->program.Base.Parameters->NumParameters * 4;
200 }
201 /* The backend also sometimes adds params for texture size. */
202 param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
203 prog_data.base.param =
204 rzalloc_array(NULL, const gl_constant_value *, param_count);
205 prog_data.base.pull_param =
206 rzalloc_array(NULL, const gl_constant_value *, param_count);
207 prog_data.base.nr_params = param_count;
208
209 prog_data.barycentric_interp_modes =
210 brw_compute_barycentric_interp_modes(brw, key->flat_shade,
211 key->persample_shading,
212 &fp->program);
213
214 program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
215 &fp->program, prog, &program_size);
216 if (program == NULL) {
217 ralloc_free(mem_ctx);
218 return false;
219 }
220
221 if (prog_data.base.total_scratch) {
222 brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
223 prog_data.base.total_scratch * brw->max_wm_threads);
224 }
225
226 if (unlikely(INTEL_DEBUG & DEBUG_WM))
227 fprintf(stderr, "\n");
228
229 brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
230 key, sizeof(struct brw_wm_prog_key),
231 program, program_size,
232 &prog_data, sizeof(prog_data),
233 &brw->wm.base.prog_offset, &brw->wm.prog_data);
234
235 ralloc_free(mem_ctx);
236
237 return true;
238 }
239
240 static bool
241 key_debug(struct brw_context *brw, const char *name, int a, int b)
242 {
243 if (a != b) {
244 perf_debug(" %s %d->%d\n", name, a, b);
245 return true;
246 } else {
247 return false;
248 }
249 }
250
251 bool
252 brw_debug_recompile_sampler_key(struct brw_context *brw,
253 const struct brw_sampler_prog_key_data *old_key,
254 const struct brw_sampler_prog_key_data *key)
255 {
256 bool found = false;
257
258 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
259 found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
260 old_key->swizzles[i], key->swizzles[i]);
261 }
262 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
263 old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
264 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
265 old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
266 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
267 old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
268 found |= key_debug(brw, "gather channel quirk on any texture unit",
269 old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
270 found |= key_debug(brw, "compressed multisample layout",
271 old_key->compressed_multisample_layout_mask,
272 key->compressed_multisample_layout_mask);
273
274 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
275 found |= key_debug(brw, "textureGather workarounds",
276 old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
277 }
278
279 return found;
280 }
281
282 void
283 brw_wm_debug_recompile(struct brw_context *brw,
284 struct gl_shader_program *prog,
285 const struct brw_wm_prog_key *key)
286 {
287 struct brw_cache_item *c = NULL;
288 const struct brw_wm_prog_key *old_key = NULL;
289 bool found = false;
290
291 perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
292
293 for (unsigned int i = 0; i < brw->cache.size; i++) {
294 for (c = brw->cache.items[i]; c; c = c->next) {
295 if (c->cache_id == BRW_CACHE_FS_PROG) {
296 old_key = c->key;
297
298 if (old_key->program_string_id == key->program_string_id)
299 break;
300 }
301 }
302 if (c)
303 break;
304 }
305
306 if (!c) {
307 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
308 return;
309 }
310
311 found |= key_debug(brw, "alphatest, computed depth, depth test, or "
312 "depth write",
313 old_key->iz_lookup, key->iz_lookup);
314 found |= key_debug(brw, "depth statistics",
315 old_key->stats_wm, key->stats_wm);
316 found |= key_debug(brw, "flat shading",
317 old_key->flat_shade, key->flat_shade);
318 found |= key_debug(brw, "per-sample shading",
319 old_key->persample_shading, key->persample_shading);
320 found |= key_debug(brw, "per-sample shading and 2x MSAA",
321 old_key->persample_2x, key->persample_2x);
322 found |= key_debug(brw, "number of color buffers",
323 old_key->nr_color_regions, key->nr_color_regions);
324 found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
325 old_key->replicate_alpha, key->replicate_alpha);
326 found |= key_debug(brw, "rendering to FBO",
327 old_key->render_to_fbo, key->render_to_fbo);
328 found |= key_debug(brw, "fragment color clamping",
329 old_key->clamp_fragment_color, key->clamp_fragment_color);
330 found |= key_debug(brw, "line smoothing",
331 old_key->line_aa, key->line_aa);
332 found |= key_debug(brw, "renderbuffer height",
333 old_key->drawable_height, key->drawable_height);
334 found |= key_debug(brw, "input slots valid",
335 old_key->input_slots_valid, key->input_slots_valid);
336 found |= key_debug(brw, "mrt alpha test function",
337 old_key->alpha_test_func, key->alpha_test_func);
338 found |= key_debug(brw, "mrt alpha test reference value",
339 old_key->alpha_test_ref, key->alpha_test_ref);
340
341 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
342
343 if (!found) {
344 perf_debug(" Something else\n");
345 }
346 }
347
348 static uint8_t
349 gen6_gather_workaround(GLenum internalformat)
350 {
351 switch (internalformat) {
352 case GL_R8I: return WA_SIGN | WA_8BIT;
353 case GL_R8UI: return WA_8BIT;
354 case GL_R16I: return WA_SIGN | WA_16BIT;
355 case GL_R16UI: return WA_16BIT;
356 /* note that even though GL_R32I and GL_R32UI have format overrides
357 * in the surface state, there is no shader w/a required */
358 default: return 0;
359 }
360 }
361
362 void
363 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
364 const struct gl_program *prog,
365 unsigned sampler_count,
366 struct brw_sampler_prog_key_data *key)
367 {
368 struct brw_context *brw = brw_context(ctx);
369
370 for (int s = 0; s < sampler_count; s++) {
371 key->swizzles[s] = SWIZZLE_NOOP;
372
373 if (!(prog->SamplersUsed & (1 << s)))
374 continue;
375
376 int unit_id = prog->SamplerUnits[s];
377 const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
378
379 if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
380 const struct gl_texture_object *t = unit->_Current;
381 const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
382 struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
383
384 const bool alpha_depth = t->DepthMode == GL_ALPHA &&
385 (img->_BaseFormat == GL_DEPTH_COMPONENT ||
386 img->_BaseFormat == GL_DEPTH_STENCIL);
387
388 /* Haswell handles texture swizzling as surface format overrides
389 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
390 */
391 if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
392 key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
393
394 if (brw->gen < 8 &&
395 sampler->MinFilter != GL_NEAREST &&
396 sampler->MagFilter != GL_NEAREST) {
397 if (sampler->WrapS == GL_CLAMP)
398 key->gl_clamp_mask[0] |= 1 << s;
399 if (sampler->WrapT == GL_CLAMP)
400 key->gl_clamp_mask[1] |= 1 << s;
401 if (sampler->WrapR == GL_CLAMP)
402 key->gl_clamp_mask[2] |= 1 << s;
403 }
404
405 /* gather4's channel select for green from RG32F is broken;
406 * requires a shader w/a on IVB; fixable with just SCS on HSW. */
407 if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
408 if (img->InternalFormat == GL_RG32F)
409 key->gather_channel_quirk_mask |= 1 << s;
410 }
411
412 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
413 * UNORM/FLOAT instead and fix it in the shader.
414 */
415 if (brw->gen == 6 && prog->UsesGather) {
416 key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
417 }
418
419 /* If this is a multisample sampler, and uses the CMS MSAA layout,
420 * then we need to emit slightly different code to first sample the
421 * MCS surface.
422 */
423 struct intel_texture_object *intel_tex =
424 intel_texture_object((struct gl_texture_object *)t);
425
426 if (brw->gen >= 7 &&
427 intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
428 key->compressed_multisample_layout_mask |= 1 << s;
429 }
430 }
431 }
432 }
433
434 static bool
435 brw_wm_state_dirty (struct brw_context *brw)
436 {
437 return brw_state_dirty(brw,
438 _NEW_BUFFERS |
439 _NEW_COLOR |
440 _NEW_DEPTH |
441 _NEW_FRAG_CLAMP |
442 _NEW_HINT |
443 _NEW_LIGHT |
444 _NEW_LINE |
445 _NEW_MULTISAMPLE |
446 _NEW_POLYGON |
447 _NEW_STENCIL |
448 _NEW_TEXTURE,
449 BRW_NEW_FRAGMENT_PROGRAM |
450 BRW_NEW_REDUCED_PRIMITIVE |
451 BRW_NEW_STATS_WM |
452 BRW_NEW_VUE_MAP_GEOM_OUT);
453 }
454
455 static void brw_wm_populate_key( struct brw_context *brw,
456 struct brw_wm_prog_key *key )
457 {
458 struct gl_context *ctx = &brw->ctx;
459 /* BRW_NEW_FRAGMENT_PROGRAM */
460 const struct brw_fragment_program *fp =
461 (struct brw_fragment_program *)brw->fragment_program;
462 const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
463 GLuint lookup = 0;
464 GLuint line_aa;
465 bool program_uses_dfdy = fp->program.UsesDFdy;
466 const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
467
468 memset(key, 0, sizeof(*key));
469
470 /* Build the index for table lookup
471 */
472 if (brw->gen < 6) {
473 /* _NEW_COLOR */
474 if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
475 lookup |= IZ_PS_KILL_ALPHATEST_BIT;
476
477 if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
478 lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
479
480 /* _NEW_DEPTH */
481 if (ctx->Depth.Test)
482 lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
483
484 if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
485 lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
486
487 /* _NEW_STENCIL | _NEW_BUFFERS */
488 if (ctx->Stencil._Enabled) {
489 lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
490
491 if (ctx->Stencil.WriteMask[0] ||
492 ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
493 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
494 }
495 key->iz_lookup = lookup;
496 }
497
498 line_aa = AA_NEVER;
499
500 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
501 if (ctx->Line.SmoothFlag) {
502 if (brw->reduced_primitive == GL_LINES) {
503 line_aa = AA_ALWAYS;
504 }
505 else if (brw->reduced_primitive == GL_TRIANGLES) {
506 if (ctx->Polygon.FrontMode == GL_LINE) {
507 line_aa = AA_SOMETIMES;
508
509 if (ctx->Polygon.BackMode == GL_LINE ||
510 (ctx->Polygon.CullFlag &&
511 ctx->Polygon.CullFaceMode == GL_BACK))
512 line_aa = AA_ALWAYS;
513 }
514 else if (ctx->Polygon.BackMode == GL_LINE) {
515 line_aa = AA_SOMETIMES;
516
517 if ((ctx->Polygon.CullFlag &&
518 ctx->Polygon.CullFaceMode == GL_FRONT))
519 line_aa = AA_ALWAYS;
520 }
521 }
522 }
523
524 key->line_aa = line_aa;
525
526 /* _NEW_HINT */
527 key->high_quality_derivatives =
528 ctx->Hint.FragmentShaderDerivative == GL_NICEST;
529
530 if (brw->gen < 6)
531 key->stats_wm = brw->stats_wm;
532
533 /* _NEW_LIGHT */
534 key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
535
536 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
537 key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
538
539 /* _NEW_TEXTURE */
540 brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
541 &key->tex);
542
543 /* _NEW_BUFFERS */
544 /*
545 * Include the draw buffer origin and height so that we can calculate
546 * fragment position values relative to the bottom left of the drawable,
547 * from the incoming screen origin relative position we get as part of our
548 * payload.
549 *
550 * This is only needed for the WM_WPOSXY opcode when the fragment program
551 * uses the gl_FragCoord input.
552 *
553 * We could avoid recompiling by including this as a constant referenced by
554 * our program, but if we were to do that it would also be nice to handle
555 * getting that constant updated at batchbuffer submit time (when we
556 * hold the lock and know where the buffer really is) rather than at emit
557 * time when we don't hold the lock and are just guessing. We could also
558 * just avoid using this as key data if the program doesn't use
559 * fragment.position.
560 *
561 * For DRI2 the origin_x/y will always be (0,0) but we still need the
562 * drawable height in order to invert the Y axis.
563 */
564 if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
565 key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
566 }
567
568 if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
569 key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
570 }
571
572 /* _NEW_BUFFERS */
573 key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
574
575 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
576 key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
577 (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
578
579 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
580 /* Ignore sample qualifier while computing this flag. */
581 key->persample_shading =
582 _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
583 if (key->persample_shading)
584 key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
585
586 key->compute_pos_offset =
587 _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
588 fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
589
590 key->compute_sample_id =
591 multisample_fbo &&
592 ctx->Multisample.Enabled &&
593 (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
594
595 /* BRW_NEW_VUE_MAP_GEOM_OUT */
596 if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
597 BRW_FS_VARYING_INPUT_MASK) > 16)
598 key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
599
600
601 /* _NEW_COLOR | _NEW_BUFFERS */
602 /* Pre-gen6, the hardware alpha test always used each render
603 * target's alpha to do alpha test, as opposed to render target 0's alpha
604 * like GL requires. Fix that by building the alpha test into the
605 * shader, and we'll skip enabling the fixed function alpha test.
606 */
607 if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 && ctx->Color.AlphaEnabled) {
608 key->alpha_test_func = ctx->Color.AlphaFunc;
609 key->alpha_test_ref = ctx->Color.AlphaRef;
610 }
611
612 /* The unique fragment program ID */
613 key->program_string_id = fp->id;
614 }
615
616 void
617 brw_upload_wm_prog(struct brw_context *brw)
618 {
619 struct gl_context *ctx = &brw->ctx;
620 struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
621 struct brw_wm_prog_key key;
622 struct brw_fragment_program *fp = (struct brw_fragment_program *)
623 brw->fragment_program;
624
625 if (!brw_wm_state_dirty(brw))
626 return;
627
628 brw_wm_populate_key(brw, &key);
629
630 if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
631 &key, sizeof(key),
632 &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
633 bool success = brw_codegen_wm_prog(brw, current, fp, &key);
634 (void) success;
635 assert(success);
636 }
637 brw->wm.base.prog_data = &brw->wm.prog_data->base;
638 }