Merge remote branch 'main/master' into radeon-rewrite
[mesa.git] / src / mesa / state_tracker / st_atom_shader.c
1 /**************************************************************************
2 *
3 * Copyright 2003 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 /**
29 * State validation for vertex/fragment shaders.
30 * Note that we have to delay most vertex/fragment shader translation
31 * until rendering time since the linkage between the vertex outputs and
32 * fragment inputs can vary depending on the pairing of shaders.
33 *
34 * Authors:
35 * Brian Paul
36 */
37
38
39
40 #include "main/imports.h"
41 #include "main/mtypes.h"
42 #include "main/macros.h"
43 #include "shader/program.h"
44
45 #include "pipe/p_context.h"
46 #include "pipe/p_shader_tokens.h"
47
48 #include "util/u_simple_shaders.h"
49
50 #include "cso_cache/cso_context.h"
51
52 #include "st_context.h"
53 #include "st_atom.h"
54 #include "st_program.h"
55 #include "st_atom_shader.h"
56 #include "st_mesa_to_tgsi.h"
57
58
59 /**
60 * This represents a vertex program, especially translated to match
61 * the inputs of a particular fragment shader.
62 */
63 struct translated_vertex_program
64 {
65 struct st_vertex_program *master;
66
67 /** The fragment shader "signature" this vertex shader is meant for: */
68 GLbitfield frag_inputs;
69
70 /** Compared against master vertex program's serialNo: */
71 GLuint serialNo;
72
73 /** Maps VERT_RESULT_x to slot */
74 GLuint output_to_slot[VERT_RESULT_MAX];
75 ubyte output_to_semantic_name[VERT_RESULT_MAX];
76 ubyte output_to_semantic_index[VERT_RESULT_MAX];
77
78 /** Pointer to the translated vertex program */
79 struct st_vertex_program *vp;
80
81 struct translated_vertex_program *next; /**< next in linked list */
82 };
83
84
85
86 /**
87 * Given a vertex program output attribute, return the corresponding
88 * fragment program input attribute.
89 * \return -1 for vertex outputs that have no corresponding fragment input
90 */
91 static GLint
92 vp_out_to_fp_in(GLuint vertResult)
93 {
94 if (vertResult >= VERT_RESULT_TEX0 &&
95 vertResult < VERT_RESULT_TEX0 + MAX_TEXTURE_COORD_UNITS)
96 return FRAG_ATTRIB_TEX0 + (vertResult - VERT_RESULT_TEX0);
97
98 if (vertResult >= VERT_RESULT_VAR0 &&
99 vertResult < VERT_RESULT_VAR0 + MAX_VARYING)
100 return FRAG_ATTRIB_VAR0 + (vertResult - VERT_RESULT_VAR0);
101
102 switch (vertResult) {
103 case VERT_RESULT_HPOS:
104 return FRAG_ATTRIB_WPOS;
105 case VERT_RESULT_COL0:
106 return FRAG_ATTRIB_COL0;
107 case VERT_RESULT_COL1:
108 return FRAG_ATTRIB_COL1;
109 case VERT_RESULT_FOGC:
110 return FRAG_ATTRIB_FOGC;
111 default:
112 /* Back-face colors, edge flags, etc */
113 return -1;
114 }
115 }
116
117
118 /**
119 * Find a translated vertex program that corresponds to stvp and
120 * has outputs matched to stfp's inputs.
121 * This performs vertex and fragment translation (to TGSI) when needed.
122 */
123 static struct translated_vertex_program *
124 find_translated_vp(struct st_context *st,
125 struct st_vertex_program *stvp,
126 struct st_fragment_program *stfp)
127 {
128 static const GLuint UNUSED = ~0;
129 struct translated_vertex_program *xvp;
130 const GLbitfield fragInputsRead = stfp->Base.Base.InputsRead;
131
132 /*
133 * Translate fragment program if needed.
134 */
135 if (!stfp->state.tokens) {
136 GLuint inAttr, numIn = 0;
137
138 for (inAttr = 0; inAttr < FRAG_ATTRIB_MAX; inAttr++) {
139 if (fragInputsRead & (1 << inAttr)) {
140 stfp->input_to_slot[inAttr] = numIn;
141 numIn++;
142 }
143 else {
144 stfp->input_to_slot[inAttr] = UNUSED;
145 }
146 }
147
148 stfp->num_input_slots = numIn;
149
150 assert(stfp->Base.Base.NumInstructions > 1);
151
152 st_translate_fragment_program(st, stfp, stfp->input_to_slot);
153 }
154
155
156 /* See if we've got a translated vertex program whose outputs match
157 * the fragment program's inputs.
158 * XXX This could be a hash lookup, using InputsRead as the key.
159 */
160 for (xvp = stfp->vertex_programs; xvp; xvp = xvp->next) {
161 if (xvp->master == stvp && xvp->frag_inputs == fragInputsRead) {
162 break;
163 }
164 }
165
166 /* No? Allocate translated vp object now */
167 if (!xvp) {
168 xvp = ST_CALLOC_STRUCT(translated_vertex_program);
169 xvp->frag_inputs = fragInputsRead;
170 xvp->master = stvp;
171
172 xvp->next = stfp->vertex_programs;
173 stfp->vertex_programs = xvp;
174 }
175
176 /* See if we need to translate vertex program to TGSI form */
177 if (xvp->serialNo != stvp->serialNo) {
178 GLuint outAttr, dummySlot;
179 const GLbitfield outputsWritten = stvp->Base.Base.OutputsWritten;
180 GLuint numVpOuts = 0;
181 GLboolean emitPntSize = GL_FALSE, emitBFC0 = GL_FALSE, emitBFC1 = GL_FALSE;
182 GLint maxGeneric;
183
184 /* Compute mapping of vertex program outputs to slots, which depends
185 * on the fragment program's input->slot mapping.
186 */
187 for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
188 /* set defaults: */
189 xvp->output_to_slot[outAttr] = UNUSED;
190 xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_COUNT;
191 xvp->output_to_semantic_index[outAttr] = 99;
192
193 if (outAttr == VERT_RESULT_HPOS) {
194 /* always put xformed position into slot zero */
195 xvp->output_to_slot[VERT_RESULT_HPOS] = 0;
196 xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_POSITION;
197 xvp->output_to_semantic_index[outAttr] = 0;
198 numVpOuts++;
199 }
200 else if (outputsWritten & (1 << outAttr)) {
201 /* see if the frag prog wants this vert output */
202 GLint fpInAttrib = vp_out_to_fp_in(outAttr);
203 if (fpInAttrib >= 0) {
204 GLuint fpInSlot = stfp->input_to_slot[fpInAttrib];
205 if (fpInSlot != ~0) {
206 /* match this vp output to the fp input */
207 GLuint vpOutSlot = stfp->input_map[fpInSlot];
208 xvp->output_to_slot[outAttr] = vpOutSlot;
209 xvp->output_to_semantic_name[outAttr] = stfp->input_semantic_name[fpInSlot];
210 xvp->output_to_semantic_index[outAttr] = stfp->input_semantic_index[fpInSlot];
211 numVpOuts++;
212 }
213 }
214 else if (outAttr == VERT_RESULT_PSIZ)
215 emitPntSize = GL_TRUE;
216 else if (outAttr == VERT_RESULT_BFC0)
217 emitBFC0 = GL_TRUE;
218 else if (outAttr == VERT_RESULT_BFC1)
219 emitBFC1 = GL_TRUE;
220 }
221 #if 0 /*debug*/
222 printf("assign vp output_to_slot[%d] = %d\n", outAttr,
223 xvp->output_to_slot[outAttr]);
224 #endif
225 }
226
227 /* must do these last */
228 if (emitPntSize) {
229 xvp->output_to_slot[VERT_RESULT_PSIZ] = numVpOuts++;
230 xvp->output_to_semantic_name[VERT_RESULT_PSIZ] = TGSI_SEMANTIC_PSIZE;
231 xvp->output_to_semantic_index[VERT_RESULT_PSIZ] = 0;
232 }
233 if (emitBFC0) {
234 xvp->output_to_slot[VERT_RESULT_BFC0] = numVpOuts++;
235 xvp->output_to_semantic_name[VERT_RESULT_BFC0] = TGSI_SEMANTIC_COLOR;
236 xvp->output_to_semantic_index[VERT_RESULT_BFC0] = 0;
237 }
238 if (emitBFC1) {
239 xvp->output_to_slot[VERT_RESULT_BFC1] = numVpOuts++;
240 xvp->output_to_semantic_name[VERT_RESULT_BFC0] = TGSI_SEMANTIC_COLOR;
241 xvp->output_to_semantic_index[VERT_RESULT_BFC0] = 1;
242 }
243
244 /* Unneeded vertex program outputs will go to this slot.
245 * We could use this info to do dead code elimination in the
246 * vertex program.
247 */
248 dummySlot = numVpOuts;
249
250 /* find max GENERIC slot index */
251 maxGeneric = -1;
252 for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
253 if (xvp->output_to_semantic_name[outAttr] == TGSI_SEMANTIC_GENERIC) {
254 maxGeneric = MAX2(maxGeneric,
255 xvp->output_to_semantic_index[outAttr]);
256 }
257 }
258
259 /* Map vert program outputs that aren't used to the dummy slot
260 * (and an unused generic attribute slot).
261 */
262 for (outAttr = 0; outAttr < VERT_RESULT_MAX; outAttr++) {
263 if (outputsWritten & (1 << outAttr)) {
264 if (xvp->output_to_slot[outAttr] == UNUSED) {
265 xvp->output_to_slot[outAttr] = dummySlot;
266 xvp->output_to_semantic_name[outAttr] = TGSI_SEMANTIC_GENERIC;
267 xvp->output_to_semantic_index[outAttr] = maxGeneric + 1;
268 }
269 }
270
271 #if 0 /*debug*/
272 printf("vp output_to_slot[%d] = %d\n", outAttr,
273 xvp->output_to_slot[outAttr]);
274 #endif
275 }
276
277 assert(stvp->Base.Base.NumInstructions > 1);
278
279 st_translate_vertex_program(st, stvp, xvp->output_to_slot,
280 xvp->output_to_semantic_name,
281 xvp->output_to_semantic_index);
282
283 xvp->vp = stvp;
284
285 /* translated VP is up to date now */
286 xvp->serialNo = stvp->serialNo;
287 }
288
289 return xvp;
290 }
291
292
293 void
294 st_free_translated_vertex_programs(struct st_context *st,
295 struct translated_vertex_program *xvp)
296 {
297 struct translated_vertex_program *next;
298
299 while (xvp) {
300 next = xvp->next;
301 _mesa_free(xvp);
302 xvp = next;
303 }
304 }
305
306
307 static void *
308 get_passthrough_fs(struct st_context *st)
309 {
310 if (!st->passthrough_fs) {
311 st->passthrough_fs =
312 util_make_fragment_passthrough_shader(st->pipe);
313 }
314
315 return st->passthrough_fs;
316 }
317
318
319 static void
320 update_linkage( struct st_context *st )
321 {
322 struct st_vertex_program *stvp;
323 struct st_fragment_program *stfp;
324 struct translated_vertex_program *xvp;
325
326 /* find active shader and params -- Should be covered by
327 * ST_NEW_VERTEX_PROGRAM
328 */
329 assert(st->ctx->VertexProgram._Current);
330 stvp = st_vertex_program(st->ctx->VertexProgram._Current);
331 assert(stvp->Base.Base.Target == GL_VERTEX_PROGRAM_ARB);
332
333 assert(st->ctx->FragmentProgram._Current);
334 stfp = st_fragment_program(st->ctx->FragmentProgram._Current);
335 assert(stfp->Base.Base.Target == GL_FRAGMENT_PROGRAM_ARB);
336
337 xvp = find_translated_vp(st, stvp, stfp);
338
339 st_reference_vertprog(st, &st->vp, stvp);
340 st_reference_fragprog(st, &st->fp, stfp);
341
342 cso_set_vertex_shader_handle(st->cso_context, stvp->driver_shader);
343
344 if (st->missing_textures) {
345 /* use a pass-through frag shader that uses no textures */
346 void *fs = get_passthrough_fs(st);
347 cso_set_fragment_shader_handle(st->cso_context, fs);
348 }
349 else {
350 cso_set_fragment_shader_handle(st->cso_context, stfp->driver_shader);
351 }
352
353 st->vertex_result_to_slot = xvp->output_to_slot;
354 }
355
356
357 const struct st_tracked_state st_update_shader = {
358 "st_update_shader", /* name */
359 { /* dirty */
360 0, /* mesa */
361 ST_NEW_VERTEX_PROGRAM | ST_NEW_FRAGMENT_PROGRAM /* st */
362 },
363 update_linkage /* update */
364 };