2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Joakim Sindholt <opensource@zhasha.com>
4 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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:
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
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. */
25 #include "util/format/u_format.h"
26 #include "util/u_math.h"
27 #include "util/u_memory.h"
29 #include "tgsi/tgsi_dump.h"
30 #include "tgsi/tgsi_ureg.h"
33 #include "r300_context.h"
34 #include "r300_emit.h"
35 #include "r300_screen.h"
38 #include "r300_texture.h"
39 #include "r300_tgsi_to_rc.h"
41 #include "compiler/radeon_compiler.h"
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)
50 r300_shader_semantics_reset(fs_inputs);
52 for (i = 0; i < info->num_inputs; i++) {
53 index = info->input_semantic_index[i];
55 switch (info->input_semantic_name[i]) {
56 case TGSI_SEMANTIC_COLOR:
57 assert(index < ATTR_COLOR_COUNT);
58 fs_inputs->color[index] = i;
61 case TGSI_SEMANTIC_PCOORD:
62 fs_inputs->pcoord = i;
65 case TGSI_SEMANTIC_TEXCOORD:
66 assert(index < ATTR_TEXCOORD_COUNT);
67 fs_inputs->texcoord[index] = i;
70 case TGSI_SEMANTIC_GENERIC:
71 assert(index < ATTR_GENERIC_COUNT);
72 fs_inputs->generic[index] = i;
75 case TGSI_SEMANTIC_FOG:
80 case TGSI_SEMANTIC_POSITION:
85 case TGSI_SEMANTIC_FACE:
91 fprintf(stderr, "r300: FP: Unknown input semantic: %i\n",
92 info->input_semantic_name[i]);
97 static void find_output_registers(struct r300_fragment_program_compiler * compiler,
98 struct r300_fragment_shader_code *shader)
102 /* Mark the outputs as not present initially */
103 compiler->OutputColor[0] = shader->info.num_outputs;
104 compiler->OutputColor[1] = shader->info.num_outputs;
105 compiler->OutputColor[2] = shader->info.num_outputs;
106 compiler->OutputColor[3] = shader->info.num_outputs;
107 compiler->OutputDepth = shader->info.num_outputs;
109 /* Now see where they really are. */
110 for(i = 0; i < shader->info.num_outputs; ++i) {
111 switch(shader->info.output_semantic_name[i]) {
112 case TGSI_SEMANTIC_COLOR:
113 compiler->OutputColor[shader->info.output_semantic_index[i]] = i;
115 case TGSI_SEMANTIC_POSITION:
116 compiler->OutputDepth = i;
122 static void allocate_hardware_inputs(
123 struct r300_fragment_program_compiler * c,
124 void (*allocate)(void * data, unsigned input, unsigned hwreg),
127 struct r300_shader_semantics* inputs =
128 (struct r300_shader_semantics*)c->UserData;
131 /* Allocate input registers. */
132 for (i = 0; i < ATTR_COLOR_COUNT; i++) {
133 if (inputs->color[i] != ATTR_UNUSED) {
134 allocate(mydata, inputs->color[i], reg++);
137 if (inputs->face != ATTR_UNUSED) {
138 allocate(mydata, inputs->face, reg++);
140 for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
141 if (inputs->generic[i] != ATTR_UNUSED) {
142 allocate(mydata, inputs->generic[i], reg++);
145 for (i = 0; i < ATTR_TEXCOORD_COUNT; i++) {
146 if (inputs->texcoord[i] != ATTR_UNUSED) {
147 allocate(mydata, inputs->texcoord[i], reg++);
150 if (inputs->pcoord != ATTR_UNUSED) {
151 allocate(mydata, inputs->pcoord, reg++);
153 if (inputs->fog != ATTR_UNUSED) {
154 allocate(mydata, inputs->fog, reg++);
156 if (inputs->wpos != ATTR_UNUSED) {
157 allocate(mydata, inputs->wpos, reg++);
161 void r300_fragment_program_get_external_state(
162 struct r300_context* r300,
163 struct r300_fragment_program_external_state* state)
165 struct r300_textures_state *texstate = r300->textures_state.state;
168 state->alpha_to_one = r300->alpha_to_one && r300->msaa_enable;
170 for (i = 0; i < texstate->sampler_state_count; i++) {
171 struct r300_sampler_state *s = texstate->sampler_states[i];
172 struct r300_sampler_view *v = texstate->sampler_views[i];
173 struct r300_resource *t;
179 t = r300_resource(v->base.texture);
181 if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
182 state->unit[i].compare_mode_enabled = 1;
184 /* Fortunately, no need to translate this. */
185 state->unit[i].texture_compare_func = s->state.compare_func;
188 /* Pass texture swizzling to the compiler, some lowering passes need it. */
189 if (state->unit[i].compare_mode_enabled) {
190 state->unit[i].texture_swizzle =
191 RC_MAKE_SWIZZLE(v->swizzle[0], v->swizzle[1],
192 v->swizzle[2], v->swizzle[3]);
195 /* XXX this should probably take into account STR, not just S. */
196 if (t->tex.is_npot) {
197 switch (s->state.wrap_s) {
198 case PIPE_TEX_WRAP_REPEAT:
199 state->unit[i].wrap_mode = RC_WRAP_REPEAT;
202 case PIPE_TEX_WRAP_MIRROR_REPEAT:
203 state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT;
206 case PIPE_TEX_WRAP_MIRROR_CLAMP:
207 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
208 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
209 state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP;
213 state->unit[i].wrap_mode = RC_WRAP_NONE;
216 if (t->b.target == PIPE_TEXTURE_3D)
217 state->unit[i].clamp_and_scale_before_fetch = true;
222 static void r300_translate_fragment_shader(
223 struct r300_context* r300,
224 struct r300_fragment_shader_code* shader,
225 const struct tgsi_token *tokens);
227 static void r300_dummy_fragment_shader(
228 struct r300_context* r300,
229 struct r300_fragment_shader_code* shader)
231 struct pipe_shader_state state;
232 struct ureg_program *ureg;
236 /* Make a simple fragment shader which outputs (0, 0, 0, 1) */
237 ureg = ureg_create(PIPE_SHADER_FRAGMENT);
238 out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
239 imm = ureg_imm4f(ureg, 0, 0, 0, 1);
241 ureg_MOV(ureg, out, imm);
244 state.tokens = ureg_finalize(ureg);
246 shader->dummy = true;
247 r300_translate_fragment_shader(r300, shader, state.tokens);
252 static void r300_emit_fs_code_to_buffer(
253 struct r300_context *r300,
254 struct r300_fragment_shader_code *shader)
256 struct rX00_fragment_program_code *generic_code = &shader->code;
257 unsigned imm_count = shader->immediates_count;
258 unsigned imm_first = shader->externals_count;
259 unsigned imm_end = generic_code->constants.Count;
260 struct rc_constant *constants = generic_code->constants.Constants;
264 if (r300->screen->caps.is_r500) {
265 struct r500_fragment_program_code *code = &generic_code->code.r500;
267 shader->cb_code_size = 19 +
268 ((code->inst_end + 1) * 6) +
270 code->int_constant_count * 2;
272 NEW_CB(shader->cb_code, shader->cb_code_size);
273 OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
274 OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx);
275 OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl);
276 for(i = 0; i < code->int_constant_count; i++){
277 OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4),
278 code->int_constants[i]);
280 OUT_CB_REG(R500_US_CODE_RANGE,
281 R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
282 OUT_CB_REG(R500_US_CODE_OFFSET, 0);
283 OUT_CB_REG(R500_US_CODE_ADDR,
284 R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end));
286 OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR);
287 OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6);
288 for (i = 0; i <= code->inst_end; i++) {
289 OUT_CB(code->inst[i].inst0);
290 OUT_CB(code->inst[i].inst1);
291 OUT_CB(code->inst[i].inst2);
292 OUT_CB(code->inst[i].inst3);
293 OUT_CB(code->inst[i].inst4);
294 OUT_CB(code->inst[i].inst5);
297 /* Emit immediates. */
299 for(i = imm_first; i < imm_end; ++i) {
300 if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
301 const float *data = constants[i].u.Immediate;
303 OUT_CB_REG(R500_GA_US_VECTOR_INDEX,
304 R500_GA_US_VECTOR_INDEX_TYPE_CONST |
305 (i & R500_GA_US_VECTOR_INDEX_MASK));
306 OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
307 OUT_CB_TABLE(data, 4);
312 struct r300_fragment_program_code *code = &generic_code->code.r300;
313 unsigned int alu_length = code->alu.length;
314 unsigned int alu_iterations = ((alu_length - 1) / 64) + 1;
315 unsigned int tex_length = code->tex.length;
316 unsigned int tex_iterations =
317 tex_length > 0 ? ((tex_length - 1) / 32) + 1 : 0;
318 unsigned int iterations =
319 alu_iterations > tex_iterations ? alu_iterations : tex_iterations;
320 unsigned int bank = 0;
322 shader->cb_code_size = 15 +
323 /* R400_US_CODE_BANK */
324 (r300->screen->caps.is_r400 ? 2 * (iterations + 1): 0) +
325 /* R400_US_CODE_EXT */
326 (r300->screen->caps.is_r400 ? 2 : 0) +
327 /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0, R400_US_ALU_EXT_ADDR_0 */
328 (code->r390_mode ? (5 * alu_iterations) : 4) +
329 /* R400_US_ALU_EXT_ADDR_[0-63] */
330 (code->r390_mode ? (code->alu.length) : 0) +
331 /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0 */
332 code->alu.length * 4 +
333 /* R300_US_TEX_INST_0, R300_US_TEX_INST_[0-31] */
334 (code->tex.length > 0 ? code->tex.length + tex_iterations : 0) +
337 NEW_CB(shader->cb_code, shader->cb_code_size);
339 OUT_CB_REG(R300_US_CONFIG, code->config);
340 OUT_CB_REG(R300_US_PIXSIZE, code->pixsize);
341 OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset);
343 if (code->r390_mode) {
344 OUT_CB_REG(R400_US_CODE_EXT, code->r400_code_offset_ext);
345 } else if (r300->screen->caps.is_r400) {
346 /* This register appears to affect shaders even if r390_mode is
347 * disabled, so it needs to be set to 0 for shaders that
348 * don't use r390_mode. */
349 OUT_CB_REG(R400_US_CODE_EXT, 0);
352 OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4);
353 OUT_CB_TABLE(code->code_addr, 4);
356 unsigned int bank_alu_length = (alu_length < 64 ? alu_length : 64);
357 unsigned int bank_alu_offset = bank * 64;
358 unsigned int bank_tex_length = (tex_length < 32 ? tex_length : 32);
359 unsigned int bank_tex_offset = bank * 32;
361 if (r300->screen->caps.is_r400) {
362 OUT_CB_REG(R400_US_CODE_BANK, code->r390_mode ?
363 (bank << R400_BANK_SHIFT) | R400_R390_MODE_ENABLE : 0);//2
366 if (bank_alu_length > 0) {
367 OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, bank_alu_length);
368 for (i = 0; i < bank_alu_length; i++)
369 OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_inst);
371 OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, bank_alu_length);
372 for (i = 0; i < bank_alu_length; i++)
373 OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_addr);
375 OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, bank_alu_length);
376 for (i = 0; i < bank_alu_length; i++)
377 OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_inst);
379 OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, bank_alu_length);
380 for (i = 0; i < bank_alu_length; i++)
381 OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_addr);
383 if (code->r390_mode) {
384 OUT_CB_REG_SEQ(R400_US_ALU_EXT_ADDR_0, bank_alu_length);
385 for (i = 0; i < bank_alu_length; i++)
386 OUT_CB(code->alu.inst[i + bank_alu_offset].r400_ext_addr);
390 if (bank_tex_length > 0) {
391 OUT_CB_REG_SEQ(R300_US_TEX_INST_0, bank_tex_length);
392 OUT_CB_TABLE(code->tex.inst + bank_tex_offset, bank_tex_length);
395 alu_length -= bank_alu_length;
396 tex_length -= bank_tex_length;
398 } while(code->r390_mode && (alu_length > 0 || tex_length > 0));
400 /* R400_US_CODE_BANK needs to be reset to 0, otherwise some shaders
401 * will be rendered incorrectly. */
402 if (r300->screen->caps.is_r400) {
403 OUT_CB_REG(R400_US_CODE_BANK,
404 code->r390_mode ? R400_R390_MODE_ENABLE : 0);
407 /* Emit immediates. */
409 for(i = imm_first; i < imm_end; ++i) {
410 if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
411 const float *data = constants[i].u.Immediate;
413 OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
414 OUT_CB(pack_float24(data[0]));
415 OUT_CB(pack_float24(data[1]));
416 OUT_CB(pack_float24(data[2]));
417 OUT_CB(pack_float24(data[3]));
423 OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src);
424 OUT_CB_REG(R300_US_W_FMT, shader->us_out_w);
428 static void r300_translate_fragment_shader(
429 struct r300_context* r300,
430 struct r300_fragment_shader_code* shader,
431 const struct tgsi_token *tokens)
433 struct r300_fragment_program_compiler compiler;
434 struct tgsi_to_rc ttr;
438 tgsi_scan_shader(tokens, &shader->info);
439 r300_shader_read_fs_inputs(&shader->info, &shader->inputs);
441 wpos = shader->inputs.wpos;
442 face = shader->inputs.face;
444 /* Setup the compiler. */
445 memset(&compiler, 0, sizeof(compiler));
446 rc_init(&compiler.Base, &r300->fs_regalloc_state);
447 DBG_ON(r300, DBG_FP) ? compiler.Base.Debug |= RC_DBG_LOG : 0;
449 compiler.code = &shader->code;
450 compiler.state = shader->compare_state;
452 compiler.Base.debug = &r300->context.debug;
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;
469 find_output_registers(&compiler, shader);
472 shader->info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS];
474 if (compiler.Base.Debug & RC_DBG_LOG) {
475 DBG(r300, DBG_FP, "r300: Initial fragment program\n");
476 tgsi_dump(tokens, 0);
479 /* Translate TGSI to our internal representation */
480 ttr.compiler = &compiler.Base;
481 ttr.info = &shader->info;
483 r300_tgsi_to_rc(&ttr, tokens);
486 fprintf(stderr, "r300 FP: Cannot translate a shader. "
487 "Using a dummy shader instead.\n");
488 r300_dummy_fragment_shader(r300, shader);
492 if (!r300->screen->caps.is_r500 ||
493 compiler.Base.Program.Constants.Count > 200) {
494 compiler.Base.remove_unused_constants = true;
498 * Transform the program to support WPOS.
500 * Introduce a small fragment at the start of the program that will be
501 * the only code that directly reads the WPOS input.
502 * All other code pieces that reference that input will be rewritten
503 * to read from a newly allocated temporary. */
504 if (wpos != ATTR_UNUSED) {
505 /* Moving the input to some other reg is not really necessary. */
506 rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, true);
509 if (face != ATTR_UNUSED) {
510 rc_transform_fragment_face(&compiler.Base, face);
513 /* Invoke the compiler */
514 r3xx_compile_fragment_program(&compiler);
516 if (compiler.Base.Error) {
517 fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader"
518 " instead.\n", compiler.Base.ErrorMsg);
521 fprintf(stderr, "r300 FP: Cannot compile the dummy shader! "
526 rc_destroy(&compiler.Base);
527 r300_dummy_fragment_shader(r300, shader);
531 /* Shaders with zero instructions are invalid,
532 * use the dummy shader instead. */
533 if (shader->code.code.r500.inst_end == -1) {
534 rc_destroy(&compiler.Base);
535 r300_dummy_fragment_shader(r300, shader);
539 /* Initialize numbers of constants for each type. */
540 shader->externals_count = 0;
542 i < shader->code.constants.Count &&
543 shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) {
544 shader->externals_count = i+1;
546 shader->immediates_count = 0;
547 shader->rc_state_count = 0;
549 for (i = shader->externals_count; i < shader->code.constants.Count; i++) {
550 switch (shader->code.constants.Constants[i].Type) {
551 case RC_CONSTANT_IMMEDIATE:
552 ++shader->immediates_count;
554 case RC_CONSTANT_STATE:
555 ++shader->rc_state_count;
562 /* Setup shader depth output. */
563 if (shader->code.writes_depth) {
564 shader->fg_depth_src = R300_FG_DEPTH_SRC_SHADER;
565 shader->us_out_w = R300_W_FMT_W24 | R300_W_SRC_US;
567 shader->fg_depth_src = R300_FG_DEPTH_SRC_SCAN;
568 shader->us_out_w = R300_W_FMT_W0 | R300_W_SRC_US;
571 /* And, finally... */
572 rc_destroy(&compiler.Base);
574 /* Build the command buffer. */
575 r300_emit_fs_code_to_buffer(r300, shader);
578 bool r300_pick_fragment_shader(struct r300_context *r300,
579 struct r300_fragment_shader* fs,
580 struct r300_fragment_program_external_state *state)
582 struct r300_fragment_shader_code* ptr;
585 /* Build the fragment shader for the first time. */
586 fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code);
588 memcpy(&fs->shader->compare_state, state, sizeof(*state));
589 r300_translate_fragment_shader(r300, fs->shader, fs->state.tokens);
593 /* Check if the currently-bound shader has been compiled
594 * with the texture-compare state we need. */
595 if (memcmp(&fs->shader->compare_state, state, sizeof(*state)) != 0) {
596 /* Search for the right shader. */
599 if (memcmp(&ptr->compare_state, state, sizeof(*state)) == 0) {
600 if (fs->shader != ptr) {
604 /* The currently-bound one is OK. */
610 /* Not found, gotta compile a new one. */
611 ptr = CALLOC_STRUCT(r300_fragment_shader_code);
612 ptr->next = fs->first;
613 fs->first = fs->shader = ptr;
615 memcpy(&ptr->compare_state, state, sizeof(*state));
616 r300_translate_fragment_shader(r300, ptr, fs->state.tokens);