gallium: edgeflags change fixes
[mesa.git] / src / mesa / state_tracker / st_program.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * 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, sub license, 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 portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 * Brian Paul
31 */
32
33
34 #include "main/imports.h"
35 #include "main/mtypes.h"
36 #include "shader/prog_print.h"
37 #include "shader/programopt.h"
38
39 #include "pipe/p_context.h"
40 #include "pipe/p_defines.h"
41 #include "pipe/p_shader_tokens.h"
42 #include "draw/draw_context.h"
43 #include "tgsi/tgsi_dump.h"
44
45 #include "st_debug.h"
46 #include "st_context.h"
47 #include "st_atom.h"
48 #include "st_program.h"
49 #include "st_mesa_to_tgsi.h"
50 #include "cso_cache/cso_context.h"
51
52
53
54 /**
55 * Clean out any old compilations:
56 */
57 void
58 st_vp_release_varients( struct st_context *st,
59 struct st_vertex_program *stvp )
60 {
61 struct st_vp_varient *vpv;
62
63 for (vpv = stvp->varients; vpv; ) {
64 struct st_vp_varient *next = vpv->next;
65
66 if (vpv->driver_shader)
67 cso_delete_vertex_shader(st->cso_context, vpv->driver_shader);
68
69 if (vpv->draw_shader)
70 draw_delete_vertex_shader( st->draw, vpv->draw_shader );
71
72 if (vpv->state.tokens)
73 st_free_tokens(vpv->state.tokens);
74
75 FREE( vpv );
76
77 vpv = next;
78 }
79
80 stvp->varients = NULL;
81 }
82
83
84
85
86 /**
87 * Translate a Mesa vertex shader into a TGSI shader.
88 * \param outputMapping to map vertex program output registers (VERT_RESULT_x)
89 * to TGSI output slots
90 * \param tokensOut destination for TGSI tokens
91 * \return pointer to cached pipe_shader object.
92 */
93 void
94 st_prepare_vertex_program(struct st_context *st,
95 struct st_vertex_program *stvp)
96 {
97 GLuint attr;
98
99 stvp->num_inputs = 0;
100 stvp->num_outputs = 0;
101
102 if (stvp->Base.IsPositionInvariant)
103 _mesa_insert_mvp_code(st->ctx, &stvp->Base);
104
105 /*
106 * Determine number of inputs, the mappings between VERT_ATTRIB_x
107 * and TGSI generic input indexes, plus input attrib semantic info.
108 */
109 for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
110 if (stvp->Base.Base.InputsRead & (1 << attr)) {
111 stvp->input_to_index[attr] = stvp->num_inputs;
112 stvp->index_to_input[stvp->num_inputs] = attr;
113 stvp->num_inputs++;
114 }
115 }
116 /* bit of a hack, presetup potentially unused edgeflag input */
117 stvp->input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
118 stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
119
120 /* Compute mapping of vertex program outputs to slots.
121 */
122 for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
123 if ((stvp->Base.Base.OutputsWritten & (1 << attr)) == 0) {
124 stvp->result_to_output[attr] = ~0;
125 }
126 else {
127 unsigned slot = stvp->num_outputs++;
128
129 stvp->result_to_output[attr] = slot;
130
131 switch (attr) {
132 case VERT_RESULT_HPOS:
133 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
134 stvp->output_semantic_index[slot] = 0;
135 break;
136 case VERT_RESULT_COL0:
137 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
138 stvp->output_semantic_index[slot] = 0;
139 break;
140 case VERT_RESULT_COL1:
141 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
142 stvp->output_semantic_index[slot] = 1;
143 break;
144 case VERT_RESULT_BFC0:
145 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
146 stvp->output_semantic_index[slot] = 0;
147 break;
148 case VERT_RESULT_BFC1:
149 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
150 stvp->output_semantic_index[slot] = 1;
151 break;
152 case VERT_RESULT_FOGC:
153 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
154 stvp->output_semantic_index[slot] = 0;
155 break;
156 case VERT_RESULT_PSIZ:
157 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
158 stvp->output_semantic_index[slot] = 0;
159 break;
160 case VERT_RESULT_EDGE:
161 assert(0);
162 break;
163
164 case VERT_RESULT_TEX0:
165 case VERT_RESULT_TEX1:
166 case VERT_RESULT_TEX2:
167 case VERT_RESULT_TEX3:
168 case VERT_RESULT_TEX4:
169 case VERT_RESULT_TEX5:
170 case VERT_RESULT_TEX6:
171 case VERT_RESULT_TEX7:
172 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
173 stvp->output_semantic_index[slot] = attr - VERT_RESULT_TEX0;
174 break;
175
176 case VERT_RESULT_VAR0:
177 default:
178 assert(attr < VERT_RESULT_MAX);
179 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
180 stvp->output_semantic_index[slot] = (FRAG_ATTRIB_VAR0 -
181 FRAG_ATTRIB_TEX0 +
182 attr -
183 VERT_RESULT_VAR0);
184 break;
185 }
186 }
187 }
188 }
189
190
191 struct st_vp_varient *
192 st_translate_vertex_program(struct st_context *st,
193 struct st_vertex_program *stvp,
194 const struct st_vp_varient_key *key)
195 {
196 struct st_vp_varient *vpv = CALLOC_STRUCT(st_vp_varient);
197 struct pipe_context *pipe = st->pipe;
198 struct ureg_program *ureg;
199 enum pipe_error error;
200
201 ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
202 if (ureg == NULL)
203 return NULL;
204
205 vpv->num_inputs = stvp->num_inputs;
206
207 error =
208 st_translate_mesa_program(st->ctx,
209 TGSI_PROCESSOR_VERTEX,
210 ureg,
211 &stvp->Base.Base,
212 /* inputs */
213 stvp->num_inputs,
214 stvp->input_to_index,
215 NULL, /* input semantic name */
216 NULL, /* input semantic index */
217 NULL,
218 /* outputs */
219 stvp->num_outputs,
220 stvp->result_to_output,
221 stvp->output_semantic_name,
222 stvp->output_semantic_index );
223
224 if (error)
225 goto fail;
226
227 /* Edgeflags will be the last input:
228 */
229 if (key->passthrough_edgeflags) {
230 ureg_MOV( ureg,
231 ureg_DECL_output( ureg, TGSI_SEMANTIC_EDGEFLAG, 0 ),
232 ureg_DECL_vs_input( ureg, vpv->num_inputs ));
233 vpv->num_inputs++;
234 }
235
236 vpv->state.tokens = ureg_get_tokens( ureg, NULL );
237 ureg_destroy( ureg );
238
239 vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->state);
240
241 if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
242 _mesa_print_program(&stvp->Base.Base);
243 debug_printf("\n");
244 }
245
246 if (ST_DEBUG & DEBUG_TGSI) {
247 tgsi_dump( vpv->state.tokens, 0 );
248 debug_printf("\n");
249 }
250
251 return vpv;
252
253 fail:
254 ureg_destroy( ureg );
255 return NULL;
256 }
257
258
259
260 /**
261 * Translate a Mesa fragment shader into a TGSI shader.
262 * \param inputMapping to map fragment program input registers to TGSI
263 * input slots
264 * \return pointer to cached pipe_shader object.
265 */
266 void
267 st_translate_fragment_program(struct st_context *st,
268 struct st_fragment_program *stfp,
269 const GLuint inputMapping[])
270 {
271 struct pipe_context *pipe = st->pipe;
272 GLuint outputMapping[FRAG_RESULT_MAX];
273 GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
274 GLuint interpMode[16]; /* XXX size? */
275 GLuint attr;
276 enum pipe_error error;
277 const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
278 struct ureg_program *ureg;
279 GLuint vslot = 0;
280
281 uint fs_num_inputs = 0;
282
283 ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
284 ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
285 uint fs_num_outputs = 0;
286
287 /* which vertex output goes to the first fragment input: */
288 if (inputsRead & FRAG_BIT_WPOS)
289 vslot = 0;
290 else
291 vslot = 1;
292
293 /*
294 * Convert Mesa program inputs to TGSI input register semantics.
295 */
296 for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
297 if (inputsRead & (1 << attr)) {
298 const GLuint slot = fs_num_inputs;
299
300 defaultInputMapping[attr] = slot;
301
302 stfp->input_map[slot] = vslot++;
303
304 fs_num_inputs++;
305
306 switch (attr) {
307 case FRAG_ATTRIB_WPOS:
308 stfp->input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
309 stfp->input_semantic_index[slot] = 0;
310 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
311 break;
312 case FRAG_ATTRIB_COL0:
313 stfp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
314 stfp->input_semantic_index[slot] = 0;
315 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
316 break;
317 case FRAG_ATTRIB_COL1:
318 stfp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
319 stfp->input_semantic_index[slot] = 1;
320 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
321 break;
322 case FRAG_ATTRIB_FOGC:
323 stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
324 stfp->input_semantic_index[slot] = 0;
325 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
326 break;
327 case FRAG_ATTRIB_FACE:
328 stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
329 stfp->input_semantic_index[slot] = 0;
330 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
331 break;
332
333 /* In most cases, there is nothing special about these
334 * inputs, so adopt a convention to use the generic
335 * semantic name and the mesa FRAG_ATTRIB_ number as the
336 * index.
337 *
338 * All that is required is that the vertex shader labels
339 * its own outputs similarly, and that the vertex shader
340 * generates at least every output required by the
341 * fragment shader plus fixed-function hardware (such as
342 * BFC).
343 *
344 * There is no requirement that semantic indexes start at
345 * zero or be restricted to a particular range -- nobody
346 * should be building tables based on semantic index.
347 */
348 case FRAG_ATTRIB_TEX0:
349 case FRAG_ATTRIB_TEX1:
350 case FRAG_ATTRIB_TEX2:
351 case FRAG_ATTRIB_TEX3:
352 case FRAG_ATTRIB_TEX4:
353 case FRAG_ATTRIB_TEX5:
354 case FRAG_ATTRIB_TEX6:
355 case FRAG_ATTRIB_TEX7:
356 case FRAG_ATTRIB_PNTC:
357 case FRAG_ATTRIB_VAR0:
358 default:
359 /* Actually, let's try and zero-base this just for
360 * readability of the generated TGSI.
361 */
362 assert(attr >= FRAG_ATTRIB_TEX0);
363 stfp->input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0);
364 stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
365 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
366 break;
367 }
368 }
369 }
370
371 /*
372 * Semantics and mapping for outputs
373 */
374 {
375 uint numColors = 0;
376 GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten;
377
378 /* if z is written, emit that first */
379 if (outputsWritten & (1 << FRAG_RESULT_DEPTH)) {
380 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
381 fs_output_semantic_index[fs_num_outputs] = 0;
382 outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
383 fs_num_outputs++;
384 outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
385 }
386
387 /* handle remaning outputs (color) */
388 for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
389 if (outputsWritten & (1 << attr)) {
390 switch (attr) {
391 case FRAG_RESULT_DEPTH:
392 /* handled above */
393 assert(0);
394 break;
395 default:
396 assert(attr == FRAG_RESULT_COLOR ||
397 (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
398 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
399 fs_output_semantic_index[fs_num_outputs] = numColors;
400 outputMapping[attr] = fs_num_outputs;
401 numColors++;
402 break;
403 }
404
405 fs_num_outputs++;
406 }
407 }
408 }
409
410 if (!inputMapping)
411 inputMapping = defaultInputMapping;
412
413 ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
414 if (ureg == NULL)
415 return;
416
417
418 error =
419 st_translate_mesa_program(st->ctx,
420 TGSI_PROCESSOR_FRAGMENT,
421 ureg,
422 &stfp->Base.Base,
423 /* inputs */
424 fs_num_inputs,
425 inputMapping,
426 stfp->input_semantic_name,
427 stfp->input_semantic_index,
428 interpMode,
429 /* outputs */
430 fs_num_outputs,
431 outputMapping,
432 fs_output_semantic_name,
433 fs_output_semantic_index );
434
435 stfp->state.tokens = ureg_get_tokens( ureg, NULL );
436 ureg_destroy( ureg );
437 stfp->driver_shader = pipe->create_fs_state(pipe, &stfp->state);
438
439 if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
440 _mesa_print_program(&stfp->Base.Base);
441 debug_printf("\n");
442 }
443
444 if (ST_DEBUG & DEBUG_TGSI) {
445 tgsi_dump( stfp->state.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
446 debug_printf("\n");
447 }
448 }
449
450
451 /**
452 * Debug- print current shader text
453 */
454 void
455 st_print_shaders(GLcontext *ctx)
456 {
457 struct gl_shader_program *shProg = ctx->Shader.CurrentProgram;
458 if (shProg) {
459 GLuint i;
460 for (i = 0; i < shProg->NumShaders; i++) {
461 printf("GLSL shader %u of %u:\n", i, shProg->NumShaders);
462 printf("%s\n", shProg->Shaders[i]->Source);
463 }
464 }
465 }