i965/fs: Lower 32x32 bit multiplication on BXT.
[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 "main/enums.h"
30 #include "main/formats.h"
31 #include "main/fbobject.h"
32 #include "main/samplerobj.h"
33 #include "main/framebuffer.h"
34 #include "program/prog_parameter.h"
35 #include "program/program.h"
36 #include "intel_mipmap_tree.h"
37
38 #include "util/ralloc.h"
39
40 /**
41 * Return a bitfield where bit n is set if barycentric interpolation mode n
42 * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
43 */
44 static unsigned
45 brw_compute_barycentric_interp_modes(struct brw_context *brw,
46 bool shade_model_flat,
47 bool persample_shading,
48 const struct gl_fragment_program *fprog)
49 {
50 unsigned barycentric_interp_modes = 0;
51 int attr;
52
53 /* Loop through all fragment shader inputs to figure out what interpolation
54 * modes are in use, and set the appropriate bits in
55 * barycentric_interp_modes.
56 */
57 for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
58 enum glsl_interp_qualifier interp_qualifier =
59 fprog->InterpQualifier[attr];
60 bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
61 !persample_shading;
62 bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
63 persample_shading;
64 bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
65
66 /* Ignore unused inputs. */
67 if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
68 continue;
69
70 /* Ignore WPOS and FACE, because they don't require interpolation. */
71 if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
72 continue;
73
74 /* Determine the set (or sets) of barycentric coordinates needed to
75 * interpolate this variable. Note that when
76 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
77 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
78 * for lit pixels, so we need both sets of barycentric coordinates.
79 */
80 if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
81 if (is_centroid) {
82 barycentric_interp_modes |=
83 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
84 } else if (is_sample) {
85 barycentric_interp_modes |=
86 1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
87 }
88 if ((!is_centroid && !is_sample) ||
89 brw->needs_unlit_centroid_workaround) {
90 barycentric_interp_modes |=
91 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
92 }
93 } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
94 (!(shade_model_flat && is_gl_Color) &&
95 interp_qualifier == INTERP_QUALIFIER_NONE)) {
96 if (is_centroid) {
97 barycentric_interp_modes |=
98 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
99 } else if (is_sample) {
100 barycentric_interp_modes |=
101 1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
102 }
103 if ((!is_centroid && !is_sample) ||
104 brw->needs_unlit_centroid_workaround) {
105 barycentric_interp_modes |=
106 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
107 }
108 }
109 }
110
111 return barycentric_interp_modes;
112 }
113
114 static uint8_t
115 computed_depth_mode(struct gl_fragment_program *fp)
116 {
117 if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
118 switch (fp->FragDepthLayout) {
119 case FRAG_DEPTH_LAYOUT_NONE:
120 case FRAG_DEPTH_LAYOUT_ANY:
121 return BRW_PSCDEPTH_ON;
122 case FRAG_DEPTH_LAYOUT_GREATER:
123 return BRW_PSCDEPTH_ON_GE;
124 case FRAG_DEPTH_LAYOUT_LESS:
125 return BRW_PSCDEPTH_ON_LE;
126 case FRAG_DEPTH_LAYOUT_UNCHANGED:
127 return BRW_PSCDEPTH_OFF;
128 }
129 }
130 return BRW_PSCDEPTH_OFF;
131 }
132
133 bool
134 brw_wm_prog_data_compare(const void *in_a, const void *in_b)
135 {
136 const struct brw_wm_prog_data *a = in_a;
137 const struct brw_wm_prog_data *b = in_b;
138
139 /* Compare the base structure. */
140 if (!brw_stage_prog_data_compare(&a->base, &b->base))
141 return false;
142
143 /* Compare the rest of the structure. */
144 const unsigned offset = sizeof(struct brw_stage_prog_data);
145 if (memcmp(((char *) a) + offset, ((char *) b) + offset,
146 sizeof(struct brw_wm_prog_data) - offset))
147 return false;
148
149 return true;
150 }
151
152 /**
153 * All Mesa program -> GPU code generation goes through this function.
154 * Depending on the instructions used (i.e. flow control instructions)
155 * we'll use one of two code generators.
156 */
157 bool
158 brw_codegen_wm_prog(struct brw_context *brw,
159 struct gl_shader_program *prog,
160 struct brw_fragment_program *fp,
161 struct brw_wm_prog_key *key)
162 {
163 struct gl_context *ctx = &brw->ctx;
164 void *mem_ctx = ralloc_context(NULL);
165 struct brw_wm_prog_data prog_data;
166 const GLuint *program;
167 struct gl_shader *fs = NULL;
168 GLuint program_size;
169
170 if (prog)
171 fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
172
173 memset(&prog_data, 0, sizeof(prog_data));
174 /* key->alpha_test_func means simulating alpha testing via discards,
175 * so the shader definitely kills pixels.
176 */
177 prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
178 prog_data.uses_omask =
179 fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
180 prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
181
182 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
183 if (!prog)
184 prog_data.base.use_alt_mode = true;
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;
191 if (fs) {
192 param_count = fs->num_uniform_components;
193 } else {
194 param_count = fp->program.Base.Parameters->NumParameters * 4;
195 }
196 /* The backend also sometimes adds params for texture size. */
197 param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
198 prog_data.base.param =
199 rzalloc_array(NULL, const gl_constant_value *, param_count);
200 prog_data.base.pull_param =
201 rzalloc_array(NULL, const gl_constant_value *, param_count);
202 prog_data.base.nr_params = param_count;
203
204 prog_data.barycentric_interp_modes =
205 brw_compute_barycentric_interp_modes(brw, key->flat_shade,
206 key->persample_shading,
207 &fp->program);
208
209 program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
210 &fp->program, prog, &program_size);
211 if (program == NULL) {
212 ralloc_free(mem_ctx);
213 return false;
214 }
215
216 if (prog_data.base.total_scratch) {
217 brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
218 prog_data.base.total_scratch * brw->max_wm_threads);
219 }
220
221 if (unlikely(INTEL_DEBUG & DEBUG_WM))
222 fprintf(stderr, "\n");
223
224 brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
225 key, sizeof(struct brw_wm_prog_key),
226 program, program_size,
227 &prog_data, sizeof(prog_data),
228 &brw->wm.base.prog_offset, &brw->wm.prog_data);
229
230 ralloc_free(mem_ctx);
231
232 return true;
233 }
234
235 static bool
236 key_debug(struct brw_context *brw, const char *name, int a, int b)
237 {
238 if (a != b) {
239 perf_debug(" %s %d->%d\n", name, a, b);
240 return true;
241 } else {
242 return false;
243 }
244 }
245
246 bool
247 brw_debug_recompile_sampler_key(struct brw_context *brw,
248 const struct brw_sampler_prog_key_data *old_key,
249 const struct brw_sampler_prog_key_data *key)
250 {
251 bool found = false;
252
253 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
254 found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
255 old_key->swizzles[i], key->swizzles[i]);
256 }
257 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
258 old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
259 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
260 old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
261 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
262 old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
263 found |= key_debug(brw, "gather channel quirk on any texture unit",
264 old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
265 found |= key_debug(brw, "compressed multisample layout",
266 old_key->compressed_multisample_layout_mask,
267 key->compressed_multisample_layout_mask);
268
269 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
270 found |= key_debug(brw, "textureGather workarounds",
271 old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
272 }
273
274 return found;
275 }
276
277 void
278 brw_wm_debug_recompile(struct brw_context *brw,
279 struct gl_shader_program *prog,
280 const struct brw_wm_prog_key *key)
281 {
282 struct brw_cache_item *c = NULL;
283 const struct brw_wm_prog_key *old_key = NULL;
284 bool found = false;
285
286 perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
287
288 for (unsigned int i = 0; i < brw->cache.size; i++) {
289 for (c = brw->cache.items[i]; c; c = c->next) {
290 if (c->cache_id == BRW_CACHE_FS_PROG) {
291 old_key = c->key;
292
293 if (old_key->program_string_id == key->program_string_id)
294 break;
295 }
296 }
297 if (c)
298 break;
299 }
300
301 if (!c) {
302 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
303 return;
304 }
305
306 found |= key_debug(brw, "alphatest, computed depth, depth test, or "
307 "depth write",
308 old_key->iz_lookup, key->iz_lookup);
309 found |= key_debug(brw, "depth statistics",
310 old_key->stats_wm, key->stats_wm);
311 found |= key_debug(brw, "flat shading",
312 old_key->flat_shade, key->flat_shade);
313 found |= key_debug(brw, "per-sample shading",
314 old_key->persample_shading, key->persample_shading);
315 found |= key_debug(brw, "per-sample shading and 2x MSAA",
316 old_key->persample_2x, key->persample_2x);
317 found |= key_debug(brw, "number of color buffers",
318 old_key->nr_color_regions, key->nr_color_regions);
319 found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
320 old_key->replicate_alpha, key->replicate_alpha);
321 found |= key_debug(brw, "rendering to FBO",
322 old_key->render_to_fbo, key->render_to_fbo);
323 found |= key_debug(brw, "fragment color clamping",
324 old_key->clamp_fragment_color, key->clamp_fragment_color);
325 found |= key_debug(brw, "line smoothing",
326 old_key->line_aa, key->line_aa);
327 found |= key_debug(brw, "renderbuffer height",
328 old_key->drawable_height, key->drawable_height);
329 found |= key_debug(brw, "input slots valid",
330 old_key->input_slots_valid, key->input_slots_valid);
331 found |= key_debug(brw, "mrt alpha test function",
332 old_key->alpha_test_func, key->alpha_test_func);
333 found |= key_debug(brw, "mrt alpha test reference value",
334 old_key->alpha_test_ref, key->alpha_test_ref);
335
336 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
337
338 if (!found) {
339 perf_debug(" Something else\n");
340 }
341 }
342
343 static uint8_t
344 gen6_gather_workaround(GLenum internalformat)
345 {
346 switch (internalformat) {
347 case GL_R8I: return WA_SIGN | WA_8BIT;
348 case GL_R8UI: return WA_8BIT;
349 case GL_R16I: return WA_SIGN | WA_16BIT;
350 case GL_R16UI: return WA_16BIT;
351 default:
352 /* Note that even though GL_R32I and GL_R32UI have format overrides in
353 * the surface state, there is no shader w/a required.
354 */
355 return 0;
356 }
357 }
358
359 void
360 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
361 const struct gl_program *prog,
362 unsigned sampler_count,
363 struct brw_sampler_prog_key_data *key)
364 {
365 struct brw_context *brw = brw_context(ctx);
366
367 for (int s = 0; s < sampler_count; s++) {
368 key->swizzles[s] = SWIZZLE_NOOP;
369
370 if (!(prog->SamplersUsed & (1 << s)))
371 continue;
372
373 int unit_id = prog->SamplerUnits[s];
374 const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
375
376 if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
377 const struct gl_texture_object *t = unit->_Current;
378 const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
379 struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
380
381 const bool alpha_depth = t->DepthMode == GL_ALPHA &&
382 (img->_BaseFormat == GL_DEPTH_COMPONENT ||
383 img->_BaseFormat == GL_DEPTH_STENCIL);
384
385 /* Haswell handles texture swizzling as surface format overrides
386 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
387 */
388 if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
389 key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
390
391 if (brw->gen < 8 &&
392 sampler->MinFilter != GL_NEAREST &&
393 sampler->MagFilter != GL_NEAREST) {
394 if (sampler->WrapS == GL_CLAMP)
395 key->gl_clamp_mask[0] |= 1 << s;
396 if (sampler->WrapT == GL_CLAMP)
397 key->gl_clamp_mask[1] |= 1 << s;
398 if (sampler->WrapR == GL_CLAMP)
399 key->gl_clamp_mask[2] |= 1 << s;
400 }
401
402 /* gather4's channel select for green from RG32F is broken; requires
403 * a shader w/a on IVB; fixable with just SCS on HSW.
404 */
405 if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
406 if (img->InternalFormat == GL_RG32F)
407 key->gather_channel_quirk_mask |= 1 << s;
408 }
409
410 /* Gen6's gather4 is broken for UINT/SINT; we treat them as
411 * UNORM/FLOAT instead and fix it in the shader.
412 */
413 if (brw->gen == 6 && prog->UsesGather) {
414 key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
415 }
416
417 /* If this is a multisample sampler, and uses the CMS MSAA layout,
418 * then we need to emit slightly different code to first sample the
419 * MCS surface.
420 */
421 struct intel_texture_object *intel_tex =
422 intel_texture_object((struct gl_texture_object *)t);
423
424 if (brw->gen >= 7 &&
425 intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
426 key->compressed_multisample_layout_mask |= 1 << s;
427 }
428 }
429 }
430 }
431
432 static bool
433 brw_wm_state_dirty (struct brw_context *brw)
434 {
435 return brw_state_dirty(brw,
436 _NEW_BUFFERS |
437 _NEW_COLOR |
438 _NEW_DEPTH |
439 _NEW_FRAG_CLAMP |
440 _NEW_HINT |
441 _NEW_LIGHT |
442 _NEW_LINE |
443 _NEW_MULTISAMPLE |
444 _NEW_POLYGON |
445 _NEW_STENCIL |
446 _NEW_TEXTURE,
447 BRW_NEW_FRAGMENT_PROGRAM |
448 BRW_NEW_REDUCED_PRIMITIVE |
449 BRW_NEW_STATS_WM |
450 BRW_NEW_VUE_MAP_GEOM_OUT);
451 }
452
453 static void
454 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
455 {
456 struct gl_context *ctx = &brw->ctx;
457 /* BRW_NEW_FRAGMENT_PROGRAM */
458 const struct brw_fragment_program *fp =
459 (struct brw_fragment_program *) brw->fragment_program;
460 const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
461 GLuint lookup = 0;
462 GLuint line_aa;
463 bool program_uses_dfdy = fp->program.UsesDFdy;
464 const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
465
466 memset(key, 0, sizeof(*key));
467
468 /* Build the index for table lookup
469 */
470 if (brw->gen < 6) {
471 /* _NEW_COLOR */
472 if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
473 lookup |= IZ_PS_KILL_ALPHATEST_BIT;
474
475 if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
476 lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
477
478 /* _NEW_DEPTH */
479 if (ctx->Depth.Test)
480 lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
481
482 if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
483 lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
484
485 /* _NEW_STENCIL | _NEW_BUFFERS */
486 if (ctx->Stencil._Enabled) {
487 lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
488
489 if (ctx->Stencil.WriteMask[0] ||
490 ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
491 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
492 }
493 key->iz_lookup = lookup;
494 }
495
496 line_aa = AA_NEVER;
497
498 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
499 if (ctx->Line.SmoothFlag) {
500 if (brw->reduced_primitive == GL_LINES) {
501 line_aa = AA_ALWAYS;
502 }
503 else if (brw->reduced_primitive == GL_TRIANGLES) {
504 if (ctx->Polygon.FrontMode == GL_LINE) {
505 line_aa = AA_SOMETIMES;
506
507 if (ctx->Polygon.BackMode == GL_LINE ||
508 (ctx->Polygon.CullFlag &&
509 ctx->Polygon.CullFaceMode == GL_BACK))
510 line_aa = AA_ALWAYS;
511 }
512 else if (ctx->Polygon.BackMode == GL_LINE) {
513 line_aa = AA_SOMETIMES;
514
515 if ((ctx->Polygon.CullFlag &&
516 ctx->Polygon.CullFaceMode == GL_FRONT))
517 line_aa = AA_ALWAYS;
518 }
519 }
520 }
521
522 key->line_aa = line_aa;
523
524 /* _NEW_HINT */
525 key->high_quality_derivatives =
526 ctx->Hint.FragmentShaderDerivative == GL_NICEST;
527
528 if (brw->gen < 6)
529 key->stats_wm = brw->stats_wm;
530
531 /* _NEW_LIGHT */
532 key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
533
534 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
535 key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
536
537 /* _NEW_TEXTURE */
538 brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
539 &key->tex);
540
541 /* _NEW_BUFFERS */
542 /*
543 * Include the draw buffer origin and height so that we can calculate
544 * fragment position values relative to the bottom left of the drawable,
545 * from the incoming screen origin relative position we get as part of our
546 * payload.
547 *
548 * This is only needed for the WM_WPOSXY opcode when the fragment program
549 * uses the gl_FragCoord input.
550 *
551 * We could avoid recompiling by including this as a constant referenced by
552 * our program, but if we were to do that it would also be nice to handle
553 * getting that constant updated at batchbuffer submit time (when we
554 * hold the lock and know where the buffer really is) rather than at emit
555 * time when we don't hold the lock and are just guessing. We could also
556 * just avoid using this as key data if the program doesn't use
557 * fragment.position.
558 *
559 * For DRI2 the origin_x/y will always be (0,0) but we still need the
560 * drawable height in order to invert the Y axis.
561 */
562 if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
563 key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
564 }
565
566 if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
567 key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
568 }
569
570 /* _NEW_BUFFERS */
571 key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
572
573 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
574 key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
575 (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
576
577 /* _NEW_BUFFERS _NEW_MULTISAMPLE */
578 /* Ignore sample qualifier while computing this flag. */
579 key->persample_shading =
580 _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
581 if (key->persample_shading)
582 key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
583
584 key->compute_pos_offset =
585 _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
586 fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
587
588 key->compute_sample_id =
589 multisample_fbo &&
590 ctx->Multisample.Enabled &&
591 (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
592
593 /* BRW_NEW_VUE_MAP_GEOM_OUT */
594 if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
595 BRW_FS_VARYING_INPUT_MASK) > 16)
596 key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
597
598
599 /* _NEW_COLOR | _NEW_BUFFERS */
600 /* Pre-gen6, the hardware alpha test always used each render
601 * target's alpha to do alpha test, as opposed to render target 0's alpha
602 * like GL requires. Fix that by building the alpha test into the
603 * shader, and we'll skip enabling the fixed function alpha test.
604 */
605 if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
606 ctx->Color.AlphaEnabled) {
607 key->alpha_test_func = ctx->Color.AlphaFunc;
608 key->alpha_test_ref = ctx->Color.AlphaRef;
609 }
610
611 /* The unique fragment program ID */
612 key->program_string_id = fp->id;
613 }
614
615 void
616 brw_upload_wm_prog(struct brw_context *brw)
617 {
618 struct gl_context *ctx = &brw->ctx;
619 struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
620 struct brw_wm_prog_key key;
621 struct brw_fragment_program *fp = (struct brw_fragment_program *)
622 brw->fragment_program;
623
624 if (!brw_wm_state_dirty(brw))
625 return;
626
627 brw_wm_populate_key(brw, &key);
628
629 if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
630 &key, sizeof(key),
631 &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
632 bool success = brw_codegen_wm_prog(brw, current, fp, &key);
633 (void) success;
634 assert(success);
635 }
636 brw->wm.base.prog_data = &brw->wm.prog_data->base;
637 }