Merge branch 'gallium-new-formats'
[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_program.h"
48 #include "st_mesa_to_tgsi.h"
49 #include "cso_cache/cso_context.h"
50
51
52
53 /**
54 * Clean out any old compilations:
55 */
56 void
57 st_vp_release_varients( struct st_context *st,
58 struct st_vertex_program *stvp )
59 {
60 struct st_vp_varient *vpv;
61
62 for (vpv = stvp->varients; vpv; ) {
63 struct st_vp_varient *next = vpv->next;
64
65 if (vpv->driver_shader)
66 cso_delete_vertex_shader(st->cso_context, vpv->driver_shader);
67
68 if (vpv->draw_shader)
69 draw_delete_vertex_shader( st->draw, vpv->draw_shader );
70
71 if (vpv->tgsi.tokens)
72 st_free_tokens(vpv->tgsi.tokens);
73
74 FREE( vpv );
75
76 vpv = next;
77 }
78
79 stvp->varients = NULL;
80 }
81
82
83
84
85 /**
86 * Translate a Mesa vertex shader into a TGSI shader.
87 * \param outputMapping to map vertex program output registers (VERT_RESULT_x)
88 * to TGSI output slots
89 * \param tokensOut destination for TGSI tokens
90 * \return pointer to cached pipe_shader object.
91 */
92 void
93 st_prepare_vertex_program(struct st_context *st,
94 struct st_vertex_program *stvp)
95 {
96 GLuint attr;
97
98 stvp->num_inputs = 0;
99 stvp->num_outputs = 0;
100
101 if (stvp->Base.IsPositionInvariant)
102 _mesa_insert_mvp_code(st->ctx, &stvp->Base);
103
104 assert(stvp->Base.Base.NumInstructions > 1);
105
106 /*
107 * Determine number of inputs, the mappings between VERT_ATTRIB_x
108 * and TGSI generic input indexes, plus input attrib semantic info.
109 */
110 for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
111 if (stvp->Base.Base.InputsRead & (1 << attr)) {
112 stvp->input_to_index[attr] = stvp->num_inputs;
113 stvp->index_to_input[stvp->num_inputs] = attr;
114 stvp->num_inputs++;
115 }
116 }
117 /* bit of a hack, presetup potentially unused edgeflag input */
118 stvp->input_to_index[VERT_ATTRIB_EDGEFLAG] = stvp->num_inputs;
119 stvp->index_to_input[stvp->num_inputs] = VERT_ATTRIB_EDGEFLAG;
120
121 /* Compute mapping of vertex program outputs to slots.
122 */
123 for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
124 if ((stvp->Base.Base.OutputsWritten & BITFIELD64_BIT(attr)) == 0) {
125 stvp->result_to_output[attr] = ~0;
126 }
127 else {
128 unsigned slot = stvp->num_outputs++;
129
130 stvp->result_to_output[attr] = slot;
131
132 switch (attr) {
133 case VERT_RESULT_HPOS:
134 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
135 stvp->output_semantic_index[slot] = 0;
136 break;
137 case VERT_RESULT_COL0:
138 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
139 stvp->output_semantic_index[slot] = 0;
140 break;
141 case VERT_RESULT_COL1:
142 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
143 stvp->output_semantic_index[slot] = 1;
144 break;
145 case VERT_RESULT_BFC0:
146 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
147 stvp->output_semantic_index[slot] = 0;
148 break;
149 case VERT_RESULT_BFC1:
150 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
151 stvp->output_semantic_index[slot] = 1;
152 break;
153 case VERT_RESULT_FOGC:
154 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
155 stvp->output_semantic_index[slot] = 0;
156 break;
157 case VERT_RESULT_PSIZ:
158 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
159 stvp->output_semantic_index[slot] = 0;
160 break;
161 case VERT_RESULT_EDGE:
162 assert(0);
163 break;
164
165 case VERT_RESULT_TEX0:
166 case VERT_RESULT_TEX1:
167 case VERT_RESULT_TEX2:
168 case VERT_RESULT_TEX3:
169 case VERT_RESULT_TEX4:
170 case VERT_RESULT_TEX5:
171 case VERT_RESULT_TEX6:
172 case VERT_RESULT_TEX7:
173 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
174 stvp->output_semantic_index[slot] = attr - VERT_RESULT_TEX0;
175 break;
176
177 case VERT_RESULT_VAR0:
178 default:
179 assert(attr < VERT_RESULT_MAX);
180 stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
181 stvp->output_semantic_index[slot] = (FRAG_ATTRIB_VAR0 -
182 FRAG_ATTRIB_TEX0 +
183 attr -
184 VERT_RESULT_VAR0);
185 break;
186 }
187 }
188 }
189 /* similar hack to above, presetup potentially unused edgeflag output */
190 stvp->result_to_output[VERT_RESULT_EDGE] = stvp->num_outputs;
191 stvp->output_semantic_name[stvp->num_outputs] = TGSI_SEMANTIC_EDGEFLAG;
192 stvp->output_semantic_index[stvp->num_outputs] = 0;
193 }
194
195
196 struct st_vp_varient *
197 st_translate_vertex_program(struct st_context *st,
198 struct st_vertex_program *stvp,
199 const struct st_vp_varient_key *key)
200 {
201 struct st_vp_varient *vpv = CALLOC_STRUCT(st_vp_varient);
202 struct pipe_context *pipe = st->pipe;
203 struct ureg_program *ureg;
204 enum pipe_error error;
205 unsigned num_outputs;
206
207 ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
208 if (ureg == NULL) {
209 FREE(vpv);
210 return NULL;
211 }
212
213 vpv->num_inputs = stvp->num_inputs;
214 num_outputs = stvp->num_outputs;
215 if (key->passthrough_edgeflags) {
216 vpv->num_inputs++;
217 num_outputs++;
218 }
219
220 error =
221 st_translate_mesa_program(st->ctx,
222 TGSI_PROCESSOR_VERTEX,
223 ureg,
224 &stvp->Base.Base,
225 /* inputs */
226 vpv->num_inputs,
227 stvp->input_to_index,
228 NULL, /* input semantic name */
229 NULL, /* input semantic index */
230 NULL,
231 /* outputs */
232 num_outputs,
233 stvp->result_to_output,
234 stvp->output_semantic_name,
235 stvp->output_semantic_index,
236 key->passthrough_edgeflags );
237
238 if (error)
239 goto fail;
240
241 vpv->tgsi.tokens = ureg_get_tokens( ureg, NULL );
242 if (!vpv->tgsi.tokens)
243 goto fail;
244
245 ureg_destroy( ureg );
246
247 vpv->driver_shader = pipe->create_vs_state(pipe, &vpv->tgsi);
248
249 if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
250 _mesa_print_program(&stvp->Base.Base);
251 debug_printf("\n");
252 }
253
254 if (ST_DEBUG & DEBUG_TGSI) {
255 tgsi_dump( vpv->tgsi.tokens, 0 );
256 debug_printf("\n");
257 }
258
259 return vpv;
260
261 fail:
262 debug_printf("%s: failed to translate Mesa program:\n", __FUNCTION__);
263 _mesa_print_program(&stvp->Base.Base);
264 debug_assert(0);
265
266 ureg_destroy( ureg );
267 return NULL;
268 }
269
270
271
272 /**
273 * Translate a Mesa fragment shader into a TGSI shader.
274 * \return pointer to cached pipe_shader object.
275 */
276 void
277 st_translate_fragment_program(struct st_context *st,
278 struct st_fragment_program *stfp )
279 {
280 struct pipe_context *pipe = st->pipe;
281 GLuint outputMapping[FRAG_RESULT_MAX];
282 GLuint inputMapping[FRAG_ATTRIB_MAX];
283 GLuint interpMode[PIPE_MAX_SHADER_INPUTS]; /* XXX size? */
284 GLuint attr;
285 enum pipe_error error;
286 const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
287 struct ureg_program *ureg;
288
289 ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS];
290 ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
291 uint fs_num_inputs = 0;
292
293 ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
294 ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
295 uint fs_num_outputs = 0;
296
297 /*
298 * Convert Mesa program inputs to TGSI input register semantics.
299 */
300 for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
301 if (inputsRead & (1 << attr)) {
302 const GLuint slot = fs_num_inputs++;
303
304 inputMapping[attr] = slot;
305
306 switch (attr) {
307 case FRAG_ATTRIB_WPOS:
308 input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
309 input_semantic_index[slot] = 0;
310 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
311 break;
312 case FRAG_ATTRIB_COL0:
313 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
314 input_semantic_index[slot] = 0;
315 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
316 break;
317 case FRAG_ATTRIB_COL1:
318 input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
319 input_semantic_index[slot] = 1;
320 interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
321 break;
322 case FRAG_ATTRIB_FOGC:
323 input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
324 input_semantic_index[slot] = 0;
325 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
326 break;
327 case FRAG_ATTRIB_FACE:
328 input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
329 input_semantic_index[slot] = 0;
330 interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
331 break;
332 case FRAG_ATTRIB_PNTC:
333 /* This is a hack. We really need a new semantic label for
334 * point coord. The draw module needs to know which fragment
335 * shader input is the point coord attribute so that it can set
336 * up the right vertex attribute values.
337 */
338 input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
339 input_semantic_index[slot] = 0;
340 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
341 break;
342
343 /* In most cases, there is nothing special about these
344 * inputs, so adopt a convention to use the generic
345 * semantic name and the mesa FRAG_ATTRIB_ number as the
346 * index.
347 *
348 * All that is required is that the vertex shader labels
349 * its own outputs similarly, and that the vertex shader
350 * generates at least every output required by the
351 * fragment shader plus fixed-function hardware (such as
352 * BFC).
353 *
354 * There is no requirement that semantic indexes start at
355 * zero or be restricted to a particular range -- nobody
356 * should be building tables based on semantic index.
357 */
358 case FRAG_ATTRIB_TEX0:
359 case FRAG_ATTRIB_TEX1:
360 case FRAG_ATTRIB_TEX2:
361 case FRAG_ATTRIB_TEX3:
362 case FRAG_ATTRIB_TEX4:
363 case FRAG_ATTRIB_TEX5:
364 case FRAG_ATTRIB_TEX6:
365 case FRAG_ATTRIB_TEX7:
366 case FRAG_ATTRIB_VAR0:
367 default:
368 /* Actually, let's try and zero-base this just for
369 * readability of the generated TGSI.
370 */
371 assert(attr >= FRAG_ATTRIB_TEX0);
372 input_semantic_index[slot] = (attr - FRAG_ATTRIB_TEX0);
373 input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
374 interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
375 break;
376 }
377 }
378 else {
379 inputMapping[attr] = -1;
380 }
381 }
382
383 /*
384 * Semantics and mapping for outputs
385 */
386 {
387 uint numColors = 0;
388 GLbitfield64 outputsWritten = stfp->Base.Base.OutputsWritten;
389
390 /* if z is written, emit that first */
391 if (outputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
392 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
393 fs_output_semantic_index[fs_num_outputs] = 0;
394 outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
395 fs_num_outputs++;
396 outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
397 }
398
399 /* handle remaning outputs (color) */
400 for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
401 if (outputsWritten & BITFIELD64_BIT(attr)) {
402 switch (attr) {
403 case FRAG_RESULT_DEPTH:
404 /* handled above */
405 assert(0);
406 break;
407 default:
408 assert(attr == FRAG_RESULT_COLOR ||
409 (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
410 fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
411 fs_output_semantic_index[fs_num_outputs] = numColors;
412 outputMapping[attr] = fs_num_outputs;
413 numColors++;
414 break;
415 }
416
417 fs_num_outputs++;
418 }
419 }
420 }
421
422 ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
423 if (ureg == NULL)
424 return;
425
426
427 error =
428 st_translate_mesa_program(st->ctx,
429 TGSI_PROCESSOR_FRAGMENT,
430 ureg,
431 &stfp->Base.Base,
432 /* inputs */
433 fs_num_inputs,
434 inputMapping,
435 input_semantic_name,
436 input_semantic_index,
437 interpMode,
438 /* outputs */
439 fs_num_outputs,
440 outputMapping,
441 fs_output_semantic_name,
442 fs_output_semantic_index, FALSE );
443
444 stfp->tgsi.tokens = ureg_get_tokens( ureg, NULL );
445 ureg_destroy( ureg );
446 stfp->driver_shader = pipe->create_fs_state(pipe, &stfp->tgsi);
447
448 if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
449 _mesa_print_program(&stfp->Base.Base);
450 debug_printf("\n");
451 }
452
453 if (ST_DEBUG & DEBUG_TGSI) {
454 tgsi_dump( stfp->tgsi.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
455 debug_printf("\n");
456 }
457 }
458
459
460 /**
461 * Debug- print current shader text
462 */
463 void
464 st_print_shaders(GLcontext *ctx)
465 {
466 struct gl_shader_program *shProg = ctx->Shader.CurrentProgram;
467 if (shProg) {
468 GLuint i;
469 for (i = 0; i < shProg->NumShaders; i++) {
470 printf("GLSL shader %u of %u:\n", i, shProg->NumShaders);
471 printf("%s\n", shProg->Shaders[i]->Source);
472 }
473 }
474 }