i965/vec4: Move perf_debug about register spilling into the visitor.
[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 "brw_context.h"
35 #include "brw_vs.h"
36 #include "brw_util.h"
37 #include "brw_state.h"
38 #include "program/prog_print.h"
39 #include "program/prog_parameter.h"
40
41 #include "util/ralloc.h"
42
43 /**
44 * Decide which set of clip planes should be used when clipping via
45 * gl_Position or gl_ClipVertex.
46 */
47 gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
48 {
49 if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
50 /* There is currently a GLSL vertex shader, so clip according to GLSL
51 * rules, which means compare gl_ClipVertex (or gl_Position, if
52 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
53 * that were stored in EyeUserPlane at the time the clip planes were
54 * specified.
55 */
56 return ctx->Transform.EyeUserPlane;
57 } else {
58 /* Either we are using fixed function or an ARB vertex program. In
59 * either case the clip planes are going to be compared against
60 * gl_Position (which is in clip coordinates) so we have to clip using
61 * _ClipUserPlane, which was transformed into clip coordinates by Mesa
62 * core.
63 */
64 return ctx->Transform._ClipUserPlane;
65 }
66 }
67
68
69 bool
70 brw_vs_prog_data_compare(const void *in_a, const void *in_b)
71 {
72 const struct brw_vs_prog_data *a = in_a;
73 const struct brw_vs_prog_data *b = in_b;
74
75 /* Compare the base structure. */
76 if (!brw_stage_prog_data_compare(&a->base.base, &b->base.base))
77 return false;
78
79 /* Compare the rest of the struct. */
80 const unsigned offset = sizeof(struct brw_stage_prog_data);
81 if (memcmp(((char *) a) + offset, ((char *) b) + offset,
82 sizeof(struct brw_vs_prog_data) - offset)) {
83 return false;
84 }
85
86 return true;
87 }
88
89 bool
90 brw_codegen_vs_prog(struct brw_context *brw,
91 struct gl_shader_program *prog,
92 struct brw_vertex_program *vp,
93 struct brw_vs_prog_key *key)
94 {
95 GLuint program_size;
96 const GLuint *program;
97 struct brw_vs_compile c;
98 struct brw_vs_prog_data prog_data;
99 struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
100 void *mem_ctx;
101 int i;
102 struct gl_shader *vs = NULL;
103
104 if (prog)
105 vs = prog->_LinkedShaders[MESA_SHADER_VERTEX];
106
107 memset(&c, 0, sizeof(c));
108 memcpy(&c.key, key, sizeof(*key));
109 memset(&prog_data, 0, sizeof(prog_data));
110
111 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
112 if (!prog)
113 stage_prog_data->use_alt_mode = true;
114
115 mem_ctx = ralloc_context(NULL);
116
117 c.vp = vp;
118
119 /* Allocate the references to the uniforms that will end up in the
120 * prog_data associated with the compiled program, and which will be freed
121 * by the state cache.
122 */
123 int param_count;
124 if (vs) {
125 /* We add padding around uniform values below vec4 size, with the worst
126 * case being a float value that gets blown up to a vec4, so be
127 * conservative here.
128 */
129 param_count = vs->num_uniform_components * 4;
130
131 } else {
132 param_count = vp->program.Base.Parameters->NumParameters * 4;
133 }
134 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
135 * planes as uniforms.
136 */
137 param_count += c.key.base.nr_userclip_plane_consts * 4;
138
139 stage_prog_data->param =
140 rzalloc_array(NULL, const gl_constant_value *, param_count);
141 stage_prog_data->pull_param =
142 rzalloc_array(NULL, const gl_constant_value *, param_count);
143 stage_prog_data->nr_params = param_count;
144
145 GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
146 prog_data.inputs_read = vp->program.Base.InputsRead;
147
148 if (c.key.copy_edgeflag) {
149 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
150 prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
151 }
152
153 if (brw->gen < 6) {
154 /* Put dummy slots into the VUE for the SF to put the replaced
155 * point sprite coords in. We shouldn't need these dummy slots,
156 * which take up precious URB space, but it would mean that the SF
157 * doesn't get nice aligned pairs of input coords into output
158 * coords, which would be a pain to handle.
159 */
160 for (i = 0; i < 8; i++) {
161 if (c.key.point_coord_replace & (1 << i))
162 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
163 }
164
165 /* if back colors are written, allocate slots for front colors too */
166 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
167 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
168 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
169 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
170 }
171
172 /* In order for legacy clipping to work, we need to populate the clip
173 * distance varying slots whenever clipping is enabled, even if the vertex
174 * shader doesn't write to gl_ClipDistance.
175 */
176 if (c.key.base.userclip_active) {
177 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
178 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
179 }
180
181 brw_compute_vue_map(brw->intelScreen->devinfo,
182 &prog_data.base.vue_map, outputs_written);
183
184 if (0) {
185 _mesa_fprint_program_opt(stderr, &c.vp->program.Base, PROG_PRINT_DEBUG,
186 true);
187 }
188
189 /* Emit GEN4 code.
190 */
191 program = brw_vs_emit(brw, prog, &c, &prog_data, mem_ctx, &program_size);
192 if (program == NULL) {
193 ralloc_free(mem_ctx);
194 return false;
195 }
196
197 /* Scratch space is used for register spilling */
198 if (c.base.last_scratch) {
199 prog_data.base.base.total_scratch
200 = brw_get_scratch_size(c.base.last_scratch*REG_SIZE);
201
202 brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
203 prog_data.base.base.total_scratch *
204 brw->max_vs_threads);
205 }
206
207 brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
208 &c.key, sizeof(c.key),
209 program, program_size,
210 &prog_data, sizeof(prog_data),
211 &brw->vs.base.prog_offset, &brw->vs.prog_data);
212 ralloc_free(mem_ctx);
213
214 return true;
215 }
216
217 static bool
218 key_debug(struct brw_context *brw, const char *name, int a, int b)
219 {
220 if (a != b) {
221 perf_debug(" %s %d->%d\n", name, a, b);
222 return true;
223 }
224 return false;
225 }
226
227 void
228 brw_vs_debug_recompile(struct brw_context *brw,
229 struct gl_shader_program *prog,
230 const struct brw_vs_prog_key *key)
231 {
232 struct brw_cache_item *c = NULL;
233 const struct brw_vs_prog_key *old_key = NULL;
234 bool found = false;
235
236 perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
237
238 for (unsigned int i = 0; i < brw->cache.size; i++) {
239 for (c = brw->cache.items[i]; c; c = c->next) {
240 if (c->cache_id == BRW_CACHE_VS_PROG) {
241 old_key = c->key;
242
243 if (old_key->base.program_string_id == key->base.program_string_id)
244 break;
245 }
246 }
247 if (c)
248 break;
249 }
250
251 if (!c) {
252 perf_debug(" Didn't find previous compile in the shader cache for "
253 "debug\n");
254 return;
255 }
256
257 for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
258 found |= key_debug(brw, "Vertex attrib w/a flags",
259 old_key->gl_attrib_wa_flags[i],
260 key->gl_attrib_wa_flags[i]);
261 }
262
263 found |= key_debug(brw, "user clip flags",
264 old_key->base.userclip_active, key->base.userclip_active);
265
266 found |= key_debug(brw, "user clipping planes as push constants",
267 old_key->base.nr_userclip_plane_consts,
268 key->base.nr_userclip_plane_consts);
269
270 found |= key_debug(brw, "copy edgeflag",
271 old_key->copy_edgeflag, key->copy_edgeflag);
272 found |= key_debug(brw, "PointCoord replace",
273 old_key->point_coord_replace, key->point_coord_replace);
274 found |= key_debug(brw, "vertex color clamping",
275 old_key->clamp_vertex_color, key->clamp_vertex_color);
276
277 found |= brw_debug_recompile_sampler_key(brw, &old_key->base.tex,
278 &key->base.tex);
279
280 if (!found) {
281 perf_debug(" Something else\n");
282 }
283 }
284
285
286 void
287 brw_setup_vue_key_clip_info(struct brw_context *brw,
288 struct brw_vue_prog_key *key,
289 bool program_uses_clip_distance)
290 {
291 struct gl_context *ctx = &brw->ctx;
292
293 key->userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
294 if (key->userclip_active && !program_uses_clip_distance) {
295 key->nr_userclip_plane_consts
296 = _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
297 }
298 }
299
300 static bool
301 brw_vs_state_dirty(struct brw_context *brw)
302 {
303 return brw_state_dirty(brw,
304 _NEW_BUFFERS |
305 _NEW_LIGHT |
306 _NEW_POINT |
307 _NEW_POLYGON |
308 _NEW_TEXTURE |
309 _NEW_TRANSFORM,
310 BRW_NEW_VERTEX_PROGRAM |
311 BRW_NEW_VS_ATTRIB_WORKAROUNDS);
312 }
313
314 static void
315 brw_vs_populate_key(struct brw_context *brw,
316 struct brw_vs_prog_key *key)
317 {
318 struct gl_context *ctx = &brw->ctx;
319 /* BRW_NEW_VERTEX_PROGRAM */
320 struct brw_vertex_program *vp =
321 (struct brw_vertex_program *)brw->vertex_program;
322 struct gl_program *prog = (struct gl_program *) brw->vertex_program;
323 int i;
324
325 memset(key, 0, sizeof(*key));
326
327 /* Just upload the program verbatim for now. Always send it all
328 * the inputs it asks for, whether they are varying or not.
329 */
330 key->base.program_string_id = vp->id;
331 brw_setup_vue_key_clip_info(brw, &key->base,
332 vp->program.Base.UsesClipDistanceOut);
333
334 /* _NEW_POLYGON */
335 if (brw->gen < 6) {
336 key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
337 ctx->Polygon.BackMode != GL_FILL);
338 }
339
340 if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
341 VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
342 /* _NEW_LIGHT | _NEW_BUFFERS */
343 key->clamp_vertex_color = ctx->Light._ClampVertexColor;
344 }
345
346 /* _NEW_POINT */
347 if (brw->gen < 6 && ctx->Point.PointSprite) {
348 for (i = 0; i < 8; i++) {
349 if (ctx->Point.CoordReplace[i])
350 key->point_coord_replace |= (1 << i);
351 }
352 }
353
354 /* _NEW_TEXTURE */
355 brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
356 &key->base.tex);
357
358 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
359 memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
360 sizeof(brw->vb.attrib_wa_flags));
361 }
362
363 void
364 brw_upload_vs_prog(struct brw_context *brw)
365 {
366 struct gl_context *ctx = &brw->ctx;
367 struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
368 struct brw_vs_prog_key key;
369 /* BRW_NEW_VERTEX_PROGRAM */
370 struct brw_vertex_program *vp =
371 (struct brw_vertex_program *)brw->vertex_program;
372
373 if (!brw_vs_state_dirty(brw))
374 return;
375
376 brw_vs_populate_key(brw, &key);
377
378 if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
379 &key, sizeof(key),
380 &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
381 bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
382 vp, &key);
383 (void) success;
384 assert(success);
385 }
386 brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
387
388 if (memcmp(&brw->vs.prog_data->base.vue_map, &brw->vue_map_geom_out,
389 sizeof(brw->vue_map_geom_out)) != 0) {
390 brw->vue_map_vs = brw->vs.prog_data->base.vue_map;
391 brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_VS;
392 if (brw->gen < 6) {
393 /* No geometry shader support, so the VS VUE map is the VUE map for
394 * the output of the "geometry" portion of the pipeline.
395 */
396 brw->vue_map_geom_out = brw->vue_map_vs;
397 brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_GEOM_OUT;
398 }
399 }
400 }
401
402 bool
403 brw_vs_precompile(struct gl_context *ctx,
404 struct gl_shader_program *shader_prog,
405 struct gl_program *prog)
406 {
407 struct brw_context *brw = brw_context(ctx);
408 struct brw_vs_prog_key key;
409 uint32_t old_prog_offset = brw->vs.base.prog_offset;
410 struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
411 bool success;
412
413 struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
414 struct brw_vertex_program *bvp = brw_vertex_program(vp);
415
416 memset(&key, 0, sizeof(key));
417
418 brw_vue_setup_prog_key_for_precompile(ctx, &key.base, bvp->id, &vp->Base);
419 key.clamp_vertex_color =
420 (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
421 VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
422
423 success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
424
425 brw->vs.base.prog_offset = old_prog_offset;
426 brw->vs.prog_data = old_prog_data;
427
428 return success;
429 }