i965: Delete intel_context entirely.
[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 (http://www.tungstengraphics.com) 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 <keith@tungstengraphics.com>
30 */
31
32 #include "brw_context.h"
33 #include "brw_wm.h"
34 #include "brw_state.h"
35 #include "main/formats.h"
36 #include "main/fbobject.h"
37 #include "main/samplerobj.h"
38 #include "program/prog_parameter.h"
39
40 #include "glsl/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(struct brw_context *brw,
48 bool shade_model_flat,
49 const struct gl_fragment_program *fprog)
50 {
51 unsigned barycentric_interp_modes = 0;
52 int attr;
53
54 /* Loop through all fragment shader inputs to figure out what interpolation
55 * modes are in use, and set the appropriate bits in
56 * barycentric_interp_modes.
57 */
58 for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
59 enum glsl_interp_qualifier interp_qualifier =
60 fprog->InterpQualifier[attr];
61 bool is_centroid = fprog->IsCentroid & BITFIELD64_BIT(attr);
62 bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
63
64 /* Ignore unused inputs. */
65 if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
66 continue;
67
68 /* Ignore WPOS and FACE, because they don't require interpolation. */
69 if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
70 continue;
71
72 /* Determine the set (or sets) of barycentric coordinates needed to
73 * interpolate this variable. Note that when
74 * brw->needs_unlit_centroid_workaround is set, centroid interpolation
75 * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
76 * for lit pixels, so we need both sets of barycentric coordinates.
77 */
78 if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
79 if (is_centroid) {
80 barycentric_interp_modes |=
81 1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
82 }
83 if (!is_centroid || brw->needs_unlit_centroid_workaround) {
84 barycentric_interp_modes |=
85 1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
86 }
87 } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
88 (!(shade_model_flat && is_gl_Color) &&
89 interp_qualifier == INTERP_QUALIFIER_NONE)) {
90 if (is_centroid) {
91 barycentric_interp_modes |=
92 1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
93 }
94 if (!is_centroid || brw->needs_unlit_centroid_workaround) {
95 barycentric_interp_modes |=
96 1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
97 }
98 }
99 }
100
101 return barycentric_interp_modes;
102 }
103
104 bool
105 brw_wm_prog_data_compare(const void *in_a, const void *in_b,
106 int aux_size, const void *in_key)
107 {
108 const struct brw_wm_prog_data *a = in_a;
109 const struct brw_wm_prog_data *b = in_b;
110
111 /* Compare all the struct up to the pointers. */
112 if (memcmp(a, b, offsetof(struct brw_wm_prog_data, param)))
113 return false;
114
115 if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
116 return false;
117
118 if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
119 return false;
120
121 return true;
122 }
123
124 void
125 brw_wm_prog_data_free(const void *in_prog_data)
126 {
127 const struct brw_wm_prog_data *prog_data = in_prog_data;
128
129 ralloc_free((void *)prog_data->param);
130 ralloc_free((void *)prog_data->pull_param);
131 }
132
133 /**
134 * All Mesa program -> GPU code generation goes through this function.
135 * Depending on the instructions used (i.e. flow control instructions)
136 * we'll use one of two code generators.
137 */
138 bool do_wm_prog(struct brw_context *brw,
139 struct gl_shader_program *prog,
140 struct brw_fragment_program *fp,
141 struct brw_wm_prog_key *key)
142 {
143 struct brw_wm_compile *c;
144 const GLuint *program;
145 struct gl_shader *fs = NULL;
146 GLuint program_size;
147
148 if (prog)
149 fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
150
151 c = rzalloc(NULL, struct brw_wm_compile);
152
153 /* Allocate the references to the uniforms that will end up in the
154 * prog_data associated with the compiled program, and which will be freed
155 * by the state cache.
156 */
157 int param_count;
158 if (fs) {
159 param_count = fs->num_uniform_components;
160 } else {
161 param_count = fp->program.Base.Parameters->NumParameters * 4;
162 }
163 /* The backend also sometimes adds params for texture size. */
164 param_count += 2 * BRW_MAX_TEX_UNIT;
165 c->prog_data.param = rzalloc_array(NULL, const float *, param_count);
166 c->prog_data.pull_param = rzalloc_array(NULL, const float *, param_count);
167
168 memcpy(&c->key, key, sizeof(*key));
169
170 c->prog_data.barycentric_interp_modes =
171 brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
172 &fp->program);
173
174 program = brw_wm_fs_emit(brw, c, &fp->program, prog, &program_size);
175 if (program == NULL)
176 return false;
177
178 /* Scratch space is used for register spilling */
179 if (c->last_scratch) {
180 perf_debug("Fragment shader triggered register spilling. "
181 "Try reducing the number of live scalar values to "
182 "improve performance.\n");
183
184 c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
185
186 brw_get_scratch_bo(brw, &brw->wm.scratch_bo,
187 c->prog_data.total_scratch * brw->max_wm_threads);
188 }
189
190 if (unlikely(INTEL_DEBUG & DEBUG_WM))
191 fprintf(stderr, "\n");
192
193 brw_upload_cache(&brw->cache, BRW_WM_PROG,
194 &c->key, sizeof(c->key),
195 program, program_size,
196 &c->prog_data, sizeof(c->prog_data),
197 &brw->wm.prog_offset, &brw->wm.prog_data);
198
199 ralloc_free(c);
200
201 return true;
202 }
203
204 static bool
205 key_debug(struct brw_context *brw, const char *name, int a, int b)
206 {
207 if (a != b) {
208 perf_debug(" %s %d->%d\n", name, a, b);
209 return true;
210 } else {
211 return false;
212 }
213 }
214
215 bool
216 brw_debug_recompile_sampler_key(struct brw_context *brw,
217 const struct brw_sampler_prog_key_data *old_key,
218 const struct brw_sampler_prog_key_data *key)
219 {
220 bool found = false;
221
222 for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
223 found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
224 old_key->swizzles[i], key->swizzles[i]);
225 }
226 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
227 old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
228 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
229 old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
230 found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
231 old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
232 found |= key_debug(brw, "GL_MESA_ycbcr texturing\n",
233 old_key->yuvtex_mask, key->yuvtex_mask);
234 found |= key_debug(brw, "GL_MESA_ycbcr UV swapping\n",
235 old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
236
237 return found;
238 }
239
240 void
241 brw_wm_debug_recompile(struct brw_context *brw,
242 struct gl_shader_program *prog,
243 const struct brw_wm_prog_key *key)
244 {
245 struct brw_cache_item *c = NULL;
246 const struct brw_wm_prog_key *old_key = NULL;
247 bool found = false;
248
249 perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
250
251 for (unsigned int i = 0; i < brw->cache.size; i++) {
252 for (c = brw->cache.items[i]; c; c = c->next) {
253 if (c->cache_id == BRW_WM_PROG) {
254 old_key = c->key;
255
256 if (old_key->program_string_id == key->program_string_id)
257 break;
258 }
259 }
260 if (c)
261 break;
262 }
263
264 if (!c) {
265 perf_debug(" Didn't find previous compile in the shader cache for debug\n");
266 return;
267 }
268
269 found |= key_debug(brw, "alphatest, computed depth, depth test, or "
270 "depth write",
271 old_key->iz_lookup, key->iz_lookup);
272 found |= key_debug(brw, "depth statistics",
273 old_key->stats_wm, key->stats_wm);
274 found |= key_debug(brw, "flat shading",
275 old_key->flat_shade, key->flat_shade);
276 found |= key_debug(brw, "number of color buffers",
277 old_key->nr_color_regions, key->nr_color_regions);
278 found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
279 old_key->replicate_alpha, key->replicate_alpha);
280 found |= key_debug(brw, "rendering to FBO",
281 old_key->render_to_fbo, key->render_to_fbo);
282 found |= key_debug(brw, "fragment color clamping",
283 old_key->clamp_fragment_color, key->clamp_fragment_color);
284 found |= key_debug(brw, "line smoothing",
285 old_key->line_aa, key->line_aa);
286 found |= key_debug(brw, "renderbuffer height",
287 old_key->drawable_height, key->drawable_height);
288 found |= key_debug(brw, "input slots valid",
289 old_key->input_slots_valid, key->input_slots_valid);
290
291 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
292
293 if (!found) {
294 perf_debug(" Something else\n");
295 }
296 }
297
298 void
299 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
300 const struct gl_program *prog,
301 struct brw_sampler_prog_key_data *key)
302 {
303 struct brw_context *brw = brw_context(ctx);
304
305 for (int s = 0; s < MAX_SAMPLERS; s++) {
306 key->swizzles[s] = SWIZZLE_NOOP;
307
308 if (!(prog->SamplersUsed & (1 << s)))
309 continue;
310
311 int unit_id = prog->SamplerUnits[s];
312 const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
313
314 if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
315 const struct gl_texture_object *t = unit->_Current;
316 const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
317 struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
318
319 const bool alpha_depth = t->DepthMode == GL_ALPHA &&
320 (img->_BaseFormat == GL_DEPTH_COMPONENT ||
321 img->_BaseFormat == GL_DEPTH_STENCIL);
322
323 /* Haswell handles texture swizzling as surface format overrides
324 * (except for GL_ALPHA); all other platforms need MOVs in the shader.
325 */
326 if (!brw->is_haswell || alpha_depth)
327 key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
328
329 if (img->InternalFormat == GL_YCBCR_MESA) {
330 key->yuvtex_mask |= 1 << s;
331 if (img->TexFormat == MESA_FORMAT_YCBCR)
332 key->yuvtex_swap_mask |= 1 << s;
333 }
334
335 if (sampler->MinFilter != GL_NEAREST &&
336 sampler->MagFilter != GL_NEAREST) {
337 if (sampler->WrapS == GL_CLAMP)
338 key->gl_clamp_mask[0] |= 1 << s;
339 if (sampler->WrapT == GL_CLAMP)
340 key->gl_clamp_mask[1] |= 1 << s;
341 if (sampler->WrapR == GL_CLAMP)
342 key->gl_clamp_mask[2] |= 1 << s;
343 }
344 }
345 }
346 }
347
348 static void brw_wm_populate_key( struct brw_context *brw,
349 struct brw_wm_prog_key *key )
350 {
351 struct gl_context *ctx = &brw->ctx;
352 /* BRW_NEW_FRAGMENT_PROGRAM */
353 const struct brw_fragment_program *fp =
354 (struct brw_fragment_program *)brw->fragment_program;
355 const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
356 GLuint lookup = 0;
357 GLuint line_aa;
358 bool program_uses_dfdy = fp->program.UsesDFdy;
359
360 memset(key, 0, sizeof(*key));
361
362 /* Build the index for table lookup
363 */
364 if (brw->gen < 6) {
365 /* _NEW_COLOR */
366 if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
367 lookup |= IZ_PS_KILL_ALPHATEST_BIT;
368
369 if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
370 lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
371
372 /* _NEW_DEPTH */
373 if (ctx->Depth.Test)
374 lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
375
376 if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
377 lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
378
379 /* _NEW_STENCIL | _NEW_BUFFERS */
380 if (ctx->Stencil._Enabled) {
381 lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
382
383 if (ctx->Stencil.WriteMask[0] ||
384 ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
385 lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
386 }
387 key->iz_lookup = lookup;
388 }
389
390 line_aa = AA_NEVER;
391
392 /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
393 if (ctx->Line.SmoothFlag) {
394 if (brw->reduced_primitive == GL_LINES) {
395 line_aa = AA_ALWAYS;
396 }
397 else if (brw->reduced_primitive == GL_TRIANGLES) {
398 if (ctx->Polygon.FrontMode == GL_LINE) {
399 line_aa = AA_SOMETIMES;
400
401 if (ctx->Polygon.BackMode == GL_LINE ||
402 (ctx->Polygon.CullFlag &&
403 ctx->Polygon.CullFaceMode == GL_BACK))
404 line_aa = AA_ALWAYS;
405 }
406 else if (ctx->Polygon.BackMode == GL_LINE) {
407 line_aa = AA_SOMETIMES;
408
409 if ((ctx->Polygon.CullFlag &&
410 ctx->Polygon.CullFaceMode == GL_FRONT))
411 line_aa = AA_ALWAYS;
412 }
413 }
414 }
415
416 key->line_aa = line_aa;
417
418 if (brw->gen < 6)
419 key->stats_wm = brw->stats_wm;
420
421 /* _NEW_LIGHT */
422 key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
423
424 /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
425 key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
426
427 /* _NEW_TEXTURE */
428 brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
429
430 /* _NEW_BUFFERS */
431 /*
432 * Include the draw buffer origin and height so that we can calculate
433 * fragment position values relative to the bottom left of the drawable,
434 * from the incoming screen origin relative position we get as part of our
435 * payload.
436 *
437 * This is only needed for the WM_WPOSXY opcode when the fragment program
438 * uses the gl_FragCoord input.
439 *
440 * We could avoid recompiling by including this as a constant referenced by
441 * our program, but if we were to do that it would also be nice to handle
442 * getting that constant updated at batchbuffer submit time (when we
443 * hold the lock and know where the buffer really is) rather than at emit
444 * time when we don't hold the lock and are just guessing. We could also
445 * just avoid using this as key data if the program doesn't use
446 * fragment.position.
447 *
448 * For DRI2 the origin_x/y will always be (0,0) but we still need the
449 * drawable height in order to invert the Y axis.
450 */
451 if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
452 key->drawable_height = ctx->DrawBuffer->Height;
453 }
454
455 if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
456 key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
457 }
458
459 /* _NEW_BUFFERS */
460 key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
461
462 /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
463 key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
464 (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
465
466 /* BRW_NEW_VUE_MAP_GEOM_OUT */
467 if (brw->gen < 6)
468 key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
469
470 /* The unique fragment program ID */
471 key->program_string_id = fp->id;
472 }
473
474
475 static void
476 brw_upload_wm_prog(struct brw_context *brw)
477 {
478 struct gl_context *ctx = &brw->ctx;
479 struct brw_wm_prog_key key;
480 struct brw_fragment_program *fp = (struct brw_fragment_program *)
481 brw->fragment_program;
482
483 brw_wm_populate_key(brw, &key);
484
485 if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
486 &key, sizeof(key),
487 &brw->wm.prog_offset, &brw->wm.prog_data)) {
488 bool success = do_wm_prog(brw, ctx->Shader._CurrentFragmentProgram, fp,
489 &key);
490 (void) success;
491 assert(success);
492 }
493 }
494
495
496 const struct brw_tracked_state brw_wm_prog = {
497 .dirty = {
498 .mesa = (_NEW_COLOR |
499 _NEW_DEPTH |
500 _NEW_STENCIL |
501 _NEW_POLYGON |
502 _NEW_LINE |
503 _NEW_LIGHT |
504 _NEW_FRAG_CLAMP |
505 _NEW_BUFFERS |
506 _NEW_TEXTURE |
507 _NEW_MULTISAMPLE),
508 .brw = (BRW_NEW_FRAGMENT_PROGRAM |
509 BRW_NEW_REDUCED_PRIMITIVE |
510 BRW_NEW_VUE_MAP_GEOM_OUT |
511 BRW_NEW_STATS_WM)
512 },
513 .emit = brw_upload_wm_prog
514 };
515