Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4358 | Serge | 1 | /* |
2 | * Copyright 2008 Corbin Simpson |
||
3 | * Joakim Sindholt |
||
4 | * Copyright 2009 Marek Olšák |
||
5 | * |
||
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
7 | * copy of this software and associated documentation files (the "Software"), |
||
8 | * to deal in the Software without restriction, including without limitation |
||
9 | * on the rights to use, copy, modify, merge, publish, distribute, sub |
||
10 | * license, and/or sell copies of the Software, and to permit persons to whom |
||
11 | * the Software is furnished to do so, subject to the following conditions: |
||
12 | * |
||
13 | * The above copyright notice and this permission notice (including the next |
||
14 | * paragraph) shall be included in all copies or substantial portions of the |
||
15 | * Software. |
||
16 | * |
||
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
18 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL |
||
20 | * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, |
||
21 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
||
22 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
||
23 | * USE OR OTHER DEALINGS IN THE SOFTWARE. */ |
||
24 | |||
25 | #include "util/u_format.h" |
||
26 | #include "util/u_math.h" |
||
27 | #include "util/u_memory.h" |
||
28 | |||
29 | #include "tgsi/tgsi_dump.h" |
||
30 | #include "tgsi/tgsi_ureg.h" |
||
31 | |||
32 | #include "r300_cb.h" |
||
33 | #include "r300_context.h" |
||
34 | #include "r300_emit.h" |
||
35 | #include "r300_screen.h" |
||
36 | #include "r300_fs.h" |
||
37 | #include "r300_reg.h" |
||
38 | #include "r300_texture.h" |
||
39 | #include "r300_tgsi_to_rc.h" |
||
40 | |||
41 | #include "compiler/radeon_compiler.h" |
||
42 | |||
43 | /* Convert info about FS input semantics to r300_shader_semantics. */ |
||
44 | void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, |
||
45 | struct r300_shader_semantics* fs_inputs) |
||
46 | { |
||
47 | int i; |
||
48 | unsigned index; |
||
49 | |||
50 | r300_shader_semantics_reset(fs_inputs); |
||
51 | |||
52 | for (i = 0; i < info->num_inputs; i++) { |
||
53 | index = info->input_semantic_index[i]; |
||
54 | |||
55 | switch (info->input_semantic_name[i]) { |
||
56 | case TGSI_SEMANTIC_COLOR: |
||
57 | assert(index < ATTR_COLOR_COUNT); |
||
58 | fs_inputs->color[index] = i; |
||
59 | break; |
||
60 | |||
61 | case TGSI_SEMANTIC_GENERIC: |
||
62 | assert(index < ATTR_GENERIC_COUNT); |
||
63 | fs_inputs->generic[index] = i; |
||
64 | break; |
||
65 | |||
66 | case TGSI_SEMANTIC_FOG: |
||
67 | assert(index == 0); |
||
68 | fs_inputs->fog = i; |
||
69 | break; |
||
70 | |||
71 | case TGSI_SEMANTIC_POSITION: |
||
72 | assert(index == 0); |
||
73 | fs_inputs->wpos = i; |
||
74 | break; |
||
75 | |||
76 | case TGSI_SEMANTIC_FACE: |
||
77 | assert(index == 0); |
||
78 | fs_inputs->face = i; |
||
79 | break; |
||
80 | |||
81 | default: |
||
82 | fprintf(stderr, "r300: FP: Unknown input semantic: %i\n", |
||
83 | info->input_semantic_name[i]); |
||
84 | } |
||
85 | } |
||
86 | } |
||
87 | |||
88 | static void find_output_registers(struct r300_fragment_program_compiler * compiler, |
||
89 | struct r300_fragment_shader_code *shader) |
||
90 | { |
||
91 | unsigned i, colorbuf_count = 0; |
||
92 | |||
93 | /* Mark the outputs as not present initially */ |
||
94 | compiler->OutputColor[0] = shader->info.num_outputs; |
||
95 | compiler->OutputColor[1] = shader->info.num_outputs; |
||
96 | compiler->OutputColor[2] = shader->info.num_outputs; |
||
97 | compiler->OutputColor[3] = shader->info.num_outputs; |
||
98 | compiler->OutputDepth = shader->info.num_outputs; |
||
99 | |||
100 | /* Now see where they really are. */ |
||
101 | for(i = 0; i < shader->info.num_outputs; ++i) { |
||
102 | switch(shader->info.output_semantic_name[i]) { |
||
103 | case TGSI_SEMANTIC_COLOR: |
||
104 | compiler->OutputColor[colorbuf_count] = i; |
||
105 | colorbuf_count++; |
||
106 | break; |
||
107 | case TGSI_SEMANTIC_POSITION: |
||
108 | compiler->OutputDepth = i; |
||
109 | break; |
||
110 | } |
||
111 | } |
||
112 | } |
||
113 | |||
114 | static void allocate_hardware_inputs( |
||
115 | struct r300_fragment_program_compiler * c, |
||
116 | void (*allocate)(void * data, unsigned input, unsigned hwreg), |
||
117 | void * mydata) |
||
118 | { |
||
119 | struct r300_shader_semantics* inputs = |
||
120 | (struct r300_shader_semantics*)c->UserData; |
||
121 | int i, reg = 0; |
||
122 | |||
123 | /* Allocate input registers. */ |
||
124 | for (i = 0; i < ATTR_COLOR_COUNT; i++) { |
||
125 | if (inputs->color[i] != ATTR_UNUSED) { |
||
126 | allocate(mydata, inputs->color[i], reg++); |
||
127 | } |
||
128 | } |
||
129 | if (inputs->face != ATTR_UNUSED) { |
||
130 | allocate(mydata, inputs->face, reg++); |
||
131 | } |
||
132 | for (i = 0; i < ATTR_GENERIC_COUNT; i++) { |
||
133 | if (inputs->generic[i] != ATTR_UNUSED) { |
||
134 | allocate(mydata, inputs->generic[i], reg++); |
||
135 | } |
||
136 | } |
||
137 | if (inputs->fog != ATTR_UNUSED) { |
||
138 | allocate(mydata, inputs->fog, reg++); |
||
139 | } |
||
140 | if (inputs->wpos != ATTR_UNUSED) { |
||
141 | allocate(mydata, inputs->wpos, reg++); |
||
142 | } |
||
143 | } |
||
144 | |||
145 | static void get_external_state( |
||
146 | struct r300_context* r300, |
||
147 | struct r300_fragment_program_external_state* state) |
||
148 | { |
||
149 | struct r300_textures_state *texstate = r300->textures_state.state; |
||
150 | unsigned i; |
||
151 | |||
152 | state->alpha_to_one = r300->alpha_to_one && r300->msaa_enable; |
||
153 | |||
154 | for (i = 0; i < texstate->sampler_state_count; i++) { |
||
155 | struct r300_sampler_state *s = texstate->sampler_states[i]; |
||
156 | struct r300_sampler_view *v = texstate->sampler_views[i]; |
||
157 | struct r300_resource *t; |
||
158 | |||
159 | if (!s || !v) { |
||
160 | continue; |
||
161 | } |
||
162 | |||
163 | t = r300_resource(v->base.texture); |
||
164 | |||
165 | if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { |
||
166 | state->unit[i].compare_mode_enabled = 1; |
||
167 | |||
168 | /* Fortunately, no need to translate this. */ |
||
169 | state->unit[i].texture_compare_func = s->state.compare_func; |
||
170 | } |
||
171 | |||
172 | state->unit[i].non_normalized_coords = !s->state.normalized_coords; |
||
173 | state->unit[i].convert_unorm_to_snorm = |
||
174 | v->base.format == PIPE_FORMAT_RGTC1_SNORM || |
||
175 | v->base.format == PIPE_FORMAT_LATC1_SNORM; |
||
176 | |||
177 | /* Pass texture swizzling to the compiler, some lowering passes need it. */ |
||
178 | if (v->base.format == PIPE_FORMAT_RGTC1_SNORM || |
||
179 | v->base.format == PIPE_FORMAT_LATC1_SNORM) { |
||
180 | unsigned char swizzle[4]; |
||
181 | |||
182 | util_format_compose_swizzles( |
||
183 | util_format_description(v->base.format)->swizzle, |
||
184 | v->swizzle, |
||
185 | swizzle); |
||
186 | |||
187 | state->unit[i].texture_swizzle = |
||
188 | RC_MAKE_SWIZZLE(swizzle[0], swizzle[1], |
||
189 | swizzle[2], swizzle[3]); |
||
190 | } else if (state->unit[i].compare_mode_enabled) { |
||
191 | state->unit[i].texture_swizzle = |
||
192 | RC_MAKE_SWIZZLE(v->swizzle[0], v->swizzle[1], |
||
193 | v->swizzle[2], v->swizzle[3]); |
||
194 | } |
||
195 | |||
196 | /* XXX this should probably take into account STR, not just S. */ |
||
197 | if (t->tex.is_npot) { |
||
198 | switch (s->state.wrap_s) { |
||
199 | case PIPE_TEX_WRAP_REPEAT: |
||
200 | state->unit[i].wrap_mode = RC_WRAP_REPEAT; |
||
201 | break; |
||
202 | |||
203 | case PIPE_TEX_WRAP_MIRROR_REPEAT: |
||
204 | state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT; |
||
205 | break; |
||
206 | |||
207 | case PIPE_TEX_WRAP_MIRROR_CLAMP: |
||
208 | case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: |
||
209 | case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: |
||
210 | state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP; |
||
211 | break; |
||
212 | |||
213 | default: |
||
214 | state->unit[i].wrap_mode = RC_WRAP_NONE; |
||
215 | } |
||
216 | |||
217 | if (t->b.b.target == PIPE_TEXTURE_3D) |
||
218 | state->unit[i].clamp_and_scale_before_fetch = TRUE; |
||
219 | } |
||
220 | } |
||
221 | } |
||
222 | |||
223 | static void r300_translate_fragment_shader( |
||
224 | struct r300_context* r300, |
||
225 | struct r300_fragment_shader_code* shader, |
||
226 | const struct tgsi_token *tokens); |
||
227 | |||
228 | static void r300_dummy_fragment_shader( |
||
229 | struct r300_context* r300, |
||
230 | struct r300_fragment_shader_code* shader) |
||
231 | { |
||
232 | struct pipe_shader_state state; |
||
233 | struct ureg_program *ureg; |
||
234 | struct ureg_dst out; |
||
235 | struct ureg_src imm; |
||
236 | |||
237 | /* Make a simple fragment shader which outputs (0, 0, 0, 1) */ |
||
238 | ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT); |
||
239 | out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0); |
||
240 | imm = ureg_imm4f(ureg, 0, 0, 0, 1); |
||
241 | |||
242 | ureg_MOV(ureg, out, imm); |
||
243 | ureg_END(ureg); |
||
244 | |||
245 | state.tokens = ureg_finalize(ureg); |
||
246 | |||
247 | shader->dummy = TRUE; |
||
248 | r300_translate_fragment_shader(r300, shader, state.tokens); |
||
249 | |||
250 | ureg_destroy(ureg); |
||
251 | } |
||
252 | |||
253 | static void r300_emit_fs_code_to_buffer( |
||
254 | struct r300_context *r300, |
||
255 | struct r300_fragment_shader_code *shader) |
||
256 | { |
||
257 | struct rX00_fragment_program_code *generic_code = &shader->code; |
||
258 | unsigned imm_count = shader->immediates_count; |
||
259 | unsigned imm_first = shader->externals_count; |
||
260 | unsigned imm_end = generic_code->constants.Count; |
||
261 | struct rc_constant *constants = generic_code->constants.Constants; |
||
262 | unsigned i; |
||
263 | CB_LOCALS; |
||
264 | |||
265 | if (r300->screen->caps.is_r500) { |
||
266 | struct r500_fragment_program_code *code = &generic_code->code.r500; |
||
267 | |||
268 | shader->cb_code_size = 19 + |
||
269 | ((code->inst_end + 1) * 6) + |
||
270 | imm_count * 7 + |
||
271 | code->int_constant_count * 2; |
||
272 | |||
273 | NEW_CB(shader->cb_code, shader->cb_code_size); |
||
274 | OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); |
||
275 | OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx); |
||
276 | OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl); |
||
277 | for(i = 0; i < code->int_constant_count; i++){ |
||
278 | OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4), |
||
279 | code->int_constants[i]); |
||
280 | } |
||
281 | OUT_CB_REG(R500_US_CODE_RANGE, |
||
282 | R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); |
||
283 | OUT_CB_REG(R500_US_CODE_OFFSET, 0); |
||
284 | OUT_CB_REG(R500_US_CODE_ADDR, |
||
285 | R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); |
||
286 | |||
287 | OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); |
||
288 | OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); |
||
289 | for (i = 0; i <= code->inst_end; i++) { |
||
290 | OUT_CB(code->inst[i].inst0); |
||
291 | OUT_CB(code->inst[i].inst1); |
||
292 | OUT_CB(code->inst[i].inst2); |
||
293 | OUT_CB(code->inst[i].inst3); |
||
294 | OUT_CB(code->inst[i].inst4); |
||
295 | OUT_CB(code->inst[i].inst5); |
||
296 | } |
||
297 | |||
298 | /* Emit immediates. */ |
||
299 | if (imm_count) { |
||
300 | for(i = imm_first; i < imm_end; ++i) { |
||
301 | if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { |
||
302 | const float *data = constants[i].u.Immediate; |
||
303 | |||
304 | OUT_CB_REG(R500_GA_US_VECTOR_INDEX, |
||
305 | R500_GA_US_VECTOR_INDEX_TYPE_CONST | |
||
306 | (i & R500_GA_US_VECTOR_INDEX_MASK)); |
||
307 | OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4); |
||
308 | OUT_CB_TABLE(data, 4); |
||
309 | } |
||
310 | } |
||
311 | } |
||
312 | } else { /* r300 */ |
||
313 | struct r300_fragment_program_code *code = &generic_code->code.r300; |
||
314 | unsigned int alu_length = code->alu.length; |
||
315 | unsigned int alu_iterations = ((alu_length - 1) / 64) + 1; |
||
316 | unsigned int tex_length = code->tex.length; |
||
317 | unsigned int tex_iterations = |
||
318 | tex_length > 0 ? ((tex_length - 1) / 32) + 1 : 0; |
||
319 | unsigned int iterations = |
||
320 | alu_iterations > tex_iterations ? alu_iterations : tex_iterations; |
||
321 | unsigned int bank = 0; |
||
322 | |||
323 | shader->cb_code_size = 15 + |
||
324 | /* R400_US_CODE_BANK */ |
||
325 | (r300->screen->caps.is_r400 ? 2 * (iterations + 1): 0) + |
||
326 | /* R400_US_CODE_EXT */ |
||
327 | (r300->screen->caps.is_r400 ? 2 : 0) + |
||
328 | /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0, R400_US_ALU_EXT_ADDR_0 */ |
||
329 | (code->r390_mode ? (5 * alu_iterations) : 4) + |
||
330 | /* R400_US_ALU_EXT_ADDR_[0-63] */ |
||
331 | (code->r390_mode ? (code->alu.length) : 0) + |
||
332 | /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0 */ |
||
333 | code->alu.length * 4 + |
||
334 | /* R300_US_TEX_INST_0, R300_US_TEX_INST_[0-31] */ |
||
335 | (code->tex.length > 0 ? code->tex.length + tex_iterations : 0) + |
||
336 | imm_count * 5; |
||
337 | |||
338 | NEW_CB(shader->cb_code, shader->cb_code_size); |
||
339 | |||
340 | OUT_CB_REG(R300_US_CONFIG, code->config); |
||
341 | OUT_CB_REG(R300_US_PIXSIZE, code->pixsize); |
||
342 | OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset); |
||
343 | |||
344 | if (code->r390_mode) { |
||
345 | OUT_CB_REG(R400_US_CODE_EXT, code->r400_code_offset_ext); |
||
346 | } else if (r300->screen->caps.is_r400) { |
||
347 | /* This register appears to affect shaders even if r390_mode is |
||
348 | * disabled, so it needs to be set to 0 for shaders that |
||
349 | * don't use r390_mode. */ |
||
350 | OUT_CB_REG(R400_US_CODE_EXT, 0); |
||
351 | } |
||
352 | |||
353 | OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4); |
||
354 | OUT_CB_TABLE(code->code_addr, 4); |
||
355 | |||
356 | do { |
||
357 | unsigned int bank_alu_length = (alu_length < 64 ? alu_length : 64); |
||
358 | unsigned int bank_alu_offset = bank * 64; |
||
359 | unsigned int bank_tex_length = (tex_length < 32 ? tex_length : 32); |
||
360 | unsigned int bank_tex_offset = bank * 32; |
||
361 | |||
362 | if (r300->screen->caps.is_r400) { |
||
363 | OUT_CB_REG(R400_US_CODE_BANK, code->r390_mode ? |
||
364 | (bank << R400_BANK_SHIFT) | R400_R390_MODE_ENABLE : 0);//2 |
||
365 | } |
||
366 | |||
367 | if (bank_alu_length > 0) { |
||
368 | OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, bank_alu_length); |
||
369 | for (i = 0; i < bank_alu_length; i++) |
||
370 | OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_inst); |
||
371 | |||
372 | OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, bank_alu_length); |
||
373 | for (i = 0; i < bank_alu_length; i++) |
||
374 | OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_addr); |
||
375 | |||
376 | OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, bank_alu_length); |
||
377 | for (i = 0; i < bank_alu_length; i++) |
||
378 | OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_inst); |
||
379 | |||
380 | OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, bank_alu_length); |
||
381 | for (i = 0; i < bank_alu_length; i++) |
||
382 | OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_addr); |
||
383 | |||
384 | if (code->r390_mode) { |
||
385 | OUT_CB_REG_SEQ(R400_US_ALU_EXT_ADDR_0, bank_alu_length); |
||
386 | for (i = 0; i < bank_alu_length; i++) |
||
387 | OUT_CB(code->alu.inst[i + bank_alu_offset].r400_ext_addr); |
||
388 | } |
||
389 | } |
||
390 | |||
391 | if (bank_tex_length > 0) { |
||
392 | OUT_CB_REG_SEQ(R300_US_TEX_INST_0, bank_tex_length); |
||
393 | OUT_CB_TABLE(code->tex.inst + bank_tex_offset, bank_tex_length); |
||
394 | } |
||
395 | |||
396 | alu_length -= bank_alu_length; |
||
397 | tex_length -= bank_tex_length; |
||
398 | bank++; |
||
399 | } while(code->r390_mode && (alu_length > 0 || tex_length > 0)); |
||
400 | |||
401 | /* R400_US_CODE_BANK needs to be reset to 0, otherwise some shaders |
||
402 | * will be rendered incorrectly. */ |
||
403 | if (r300->screen->caps.is_r400) { |
||
404 | OUT_CB_REG(R400_US_CODE_BANK, |
||
405 | code->r390_mode ? R400_R390_MODE_ENABLE : 0); |
||
406 | } |
||
407 | |||
408 | /* Emit immediates. */ |
||
409 | if (imm_count) { |
||
410 | for(i = imm_first; i < imm_end; ++i) { |
||
411 | if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { |
||
412 | const float *data = constants[i].u.Immediate; |
||
413 | |||
414 | OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); |
||
415 | OUT_CB(pack_float24(data[0])); |
||
416 | OUT_CB(pack_float24(data[1])); |
||
417 | OUT_CB(pack_float24(data[2])); |
||
418 | OUT_CB(pack_float24(data[3])); |
||
419 | } |
||
420 | } |
||
421 | } |
||
422 | } |
||
423 | |||
424 | OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src); |
||
425 | OUT_CB_REG(R300_US_W_FMT, shader->us_out_w); |
||
426 | END_CB; |
||
427 | } |
||
428 | |||
429 | static void r300_translate_fragment_shader( |
||
430 | struct r300_context* r300, |
||
431 | struct r300_fragment_shader_code* shader, |
||
432 | const struct tgsi_token *tokens) |
||
433 | { |
||
434 | struct r300_fragment_program_compiler compiler; |
||
435 | struct tgsi_to_rc ttr; |
||
436 | int wpos, face; |
||
437 | unsigned i; |
||
438 | |||
439 | tgsi_scan_shader(tokens, &shader->info); |
||
440 | r300_shader_read_fs_inputs(&shader->info, &shader->inputs); |
||
441 | |||
442 | wpos = shader->inputs.wpos; |
||
443 | face = shader->inputs.face; |
||
444 | |||
445 | /* Setup the compiler. */ |
||
446 | memset(&compiler, 0, sizeof(compiler)); |
||
447 | rc_init(&compiler.Base, &r300->fs_regalloc_state); |
||
448 | DBG_ON(r300, DBG_FP) ? compiler.Base.Debug |= RC_DBG_LOG : 0; |
||
449 | DBG_ON(r300, DBG_P_STAT) ? compiler.Base.Debug |= RC_DBG_STATS : 0; |
||
450 | |||
451 | compiler.code = &shader->code; |
||
452 | compiler.state = shader->compare_state; |
||
453 | compiler.Base.is_r500 = r300->screen->caps.is_r500; |
||
454 | compiler.Base.is_r400 = r300->screen->caps.is_r400; |
||
455 | compiler.Base.disable_optimizations = DBG_ON(r300, DBG_NO_OPT); |
||
456 | compiler.Base.has_half_swizzles = TRUE; |
||
457 | compiler.Base.has_presub = TRUE; |
||
458 | compiler.Base.has_omod = TRUE; |
||
459 | compiler.Base.max_temp_regs = |
||
460 | compiler.Base.is_r500 ? 128 : (compiler.Base.is_r400 ? 64 : 32); |
||
461 | compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32; |
||
462 | compiler.Base.max_alu_insts = |
||
463 | (compiler.Base.is_r500 || compiler.Base.is_r400) ? 512 : 64; |
||
464 | compiler.Base.max_tex_insts = |
||
465 | (compiler.Base.is_r500 || compiler.Base.is_r400) ? 512 : 32; |
||
466 | compiler.AllocateHwInputs = &allocate_hardware_inputs; |
||
467 | compiler.UserData = &shader->inputs; |
||
468 | |||
469 | find_output_registers(&compiler, shader); |
||
470 | |||
471 | shader->write_all = FALSE; |
||
472 | for (i = 0; i < shader->info.num_properties; i++) { |
||
473 | if (shader->info.properties[i].name == TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS) { |
||
474 | shader->write_all = TRUE; |
||
475 | } |
||
476 | } |
||
477 | |||
478 | if (compiler.Base.Debug & RC_DBG_LOG) { |
||
479 | DBG(r300, DBG_FP, "r300: Initial fragment program\n"); |
||
480 | tgsi_dump(tokens, 0); |
||
481 | } |
||
482 | |||
483 | /* Translate TGSI to our internal representation */ |
||
484 | ttr.compiler = &compiler.Base; |
||
485 | ttr.info = &shader->info; |
||
486 | ttr.use_half_swizzles = TRUE; |
||
487 | |||
488 | r300_tgsi_to_rc(&ttr, tokens); |
||
489 | |||
490 | if (ttr.error) { |
||
491 | fprintf(stderr, "r300 FP: Cannot translate a shader. " |
||
492 | "Using a dummy shader instead.\n"); |
||
493 | r300_dummy_fragment_shader(r300, shader); |
||
494 | return; |
||
495 | } |
||
496 | |||
497 | if (!r300->screen->caps.is_r500 || |
||
498 | compiler.Base.Program.Constants.Count > 200) { |
||
499 | compiler.Base.remove_unused_constants = TRUE; |
||
500 | } |
||
501 | |||
502 | /** |
||
503 | * Transform the program to support WPOS. |
||
504 | * |
||
505 | * Introduce a small fragment at the start of the program that will be |
||
506 | * the only code that directly reads the WPOS input. |
||
507 | * All other code pieces that reference that input will be rewritten |
||
508 | * to read from a newly allocated temporary. */ |
||
509 | if (wpos != ATTR_UNUSED) { |
||
510 | /* Moving the input to some other reg is not really necessary. */ |
||
511 | rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE); |
||
512 | } |
||
513 | |||
514 | if (face != ATTR_UNUSED) { |
||
515 | rc_transform_fragment_face(&compiler.Base, face); |
||
516 | } |
||
517 | |||
518 | /* Invoke the compiler */ |
||
519 | r3xx_compile_fragment_program(&compiler); |
||
520 | |||
521 | if (compiler.Base.Error) { |
||
522 | fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader" |
||
523 | " instead.\n", compiler.Base.ErrorMsg); |
||
524 | |||
525 | if (shader->dummy) { |
||
526 | fprintf(stderr, "r300 FP: Cannot compile the dummy shader! " |
||
527 | "Giving up...\n"); |
||
528 | abort(); |
||
529 | } |
||
530 | |||
531 | rc_destroy(&compiler.Base); |
||
532 | r300_dummy_fragment_shader(r300, shader); |
||
533 | return; |
||
534 | } |
||
535 | |||
536 | /* Shaders with zero instructions are invalid, |
||
537 | * use the dummy shader instead. */ |
||
538 | if (shader->code.code.r500.inst_end == -1) { |
||
539 | rc_destroy(&compiler.Base); |
||
540 | r300_dummy_fragment_shader(r300, shader); |
||
541 | return; |
||
542 | } |
||
543 | |||
544 | /* Initialize numbers of constants for each type. */ |
||
545 | shader->externals_count = 0; |
||
546 | for (i = 0; |
||
547 | i < shader->code.constants.Count && |
||
548 | shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { |
||
549 | shader->externals_count = i+1; |
||
550 | } |
||
551 | shader->immediates_count = 0; |
||
552 | shader->rc_state_count = 0; |
||
553 | |||
554 | for (i = shader->externals_count; i < shader->code.constants.Count; i++) { |
||
555 | switch (shader->code.constants.Constants[i].Type) { |
||
556 | case RC_CONSTANT_IMMEDIATE: |
||
557 | ++shader->immediates_count; |
||
558 | break; |
||
559 | case RC_CONSTANT_STATE: |
||
560 | ++shader->rc_state_count; |
||
561 | break; |
||
562 | default: |
||
563 | assert(0); |
||
564 | } |
||
565 | } |
||
566 | |||
567 | /* Setup shader depth output. */ |
||
568 | if (shader->code.writes_depth) { |
||
569 | shader->fg_depth_src = R300_FG_DEPTH_SRC_SHADER; |
||
570 | shader->us_out_w = R300_W_FMT_W24 | R300_W_SRC_US; |
||
571 | } else { |
||
572 | shader->fg_depth_src = R300_FG_DEPTH_SRC_SCAN; |
||
573 | shader->us_out_w = R300_W_FMT_W0 | R300_W_SRC_US; |
||
574 | } |
||
575 | |||
576 | /* And, finally... */ |
||
577 | rc_destroy(&compiler.Base); |
||
578 | |||
579 | /* Build the command buffer. */ |
||
580 | r300_emit_fs_code_to_buffer(r300, shader); |
||
581 | } |
||
582 | |||
583 | boolean r300_pick_fragment_shader(struct r300_context* r300) |
||
584 | { |
||
585 | struct r300_fragment_shader* fs = r300_fs(r300); |
||
586 | struct r300_fragment_program_external_state state = {{{ 0 }}}; |
||
587 | struct r300_fragment_shader_code* ptr; |
||
588 | |||
589 | get_external_state(r300, &state); |
||
590 | |||
591 | if (!fs->first) { |
||
592 | /* Build the fragment shader for the first time. */ |
||
593 | fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code); |
||
594 | |||
595 | memcpy(&fs->shader->compare_state, &state, |
||
596 | sizeof(struct r300_fragment_program_external_state)); |
||
597 | r300_translate_fragment_shader(r300, fs->shader, fs->state.tokens); |
||
598 | return TRUE; |
||
599 | |||
600 | } else { |
||
601 | /* Check if the currently-bound shader has been compiled |
||
602 | * with the texture-compare state we need. */ |
||
603 | if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) { |
||
604 | /* Search for the right shader. */ |
||
605 | ptr = fs->first; |
||
606 | while (ptr) { |
||
607 | if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) { |
||
608 | if (fs->shader != ptr) { |
||
609 | fs->shader = ptr; |
||
610 | return TRUE; |
||
611 | } |
||
612 | /* The currently-bound one is OK. */ |
||
613 | return FALSE; |
||
614 | } |
||
615 | ptr = ptr->next; |
||
616 | } |
||
617 | |||
618 | /* Not found, gotta compile a new one. */ |
||
619 | ptr = CALLOC_STRUCT(r300_fragment_shader_code); |
||
620 | ptr->next = fs->first; |
||
621 | fs->first = fs->shader = ptr; |
||
622 | |||
623 | ptr->compare_state = state; |
||
624 | r300_translate_fragment_shader(r300, ptr, fs->state.tokens); |
||
625 | return TRUE; |
||
626 | } |
||
627 | } |
||
628 | |||
629 | return FALSE; |
||
630 | }>>>>>>>>>><>>>>=>>>>>>>>> |