r300_fs.c revision 56ba7e913fef0ea2b1bead582108f9ab3ab8263d
1/*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 *                Joakim Sindholt <opensource@zhasha.com>
4 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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_math.h"
26#include "util/u_memory.h"
27
28#include "tgsi/tgsi_dump.h"
29#include "tgsi/tgsi_ureg.h"
30
31#include "r300_cb.h"
32#include "r300_context.h"
33#include "r300_emit.h"
34#include "r300_screen.h"
35#include "r300_fs.h"
36#include "r300_reg.h"
37#include "r300_tgsi_to_rc.h"
38
39#include "radeon_code.h"
40#include "radeon_compiler.h"
41
42/* Convert info about FS input semantics to r300_shader_semantics. */
43void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
44                                struct r300_shader_semantics* fs_inputs)
45{
46    int i;
47    unsigned index;
48
49    r300_shader_semantics_reset(fs_inputs);
50
51    for (i = 0; i < info->num_inputs; i++) {
52        index = info->input_semantic_index[i];
53
54        switch (info->input_semantic_name[i]) {
55            case TGSI_SEMANTIC_COLOR:
56                assert(index < ATTR_COLOR_COUNT);
57                fs_inputs->color[index] = i;
58                break;
59
60            case TGSI_SEMANTIC_GENERIC:
61                assert(index < ATTR_GENERIC_COUNT);
62                fs_inputs->generic[index] = i;
63                break;
64
65            case TGSI_SEMANTIC_FOG:
66                assert(index == 0);
67                fs_inputs->fog = i;
68                break;
69
70            case TGSI_SEMANTIC_POSITION:
71                assert(index == 0);
72                fs_inputs->wpos = i;
73                break;
74
75            case TGSI_SEMANTIC_FACE:
76                assert(index == 0);
77                fs_inputs->face = i;
78                break;
79
80            default:
81                fprintf(stderr, "r300: FP: Unknown input semantic: %i\n",
82                        info->input_semantic_name[i]);
83        }
84    }
85}
86
87static void find_output_registers(struct r300_fragment_program_compiler * compiler,
88                                  struct r300_fragment_shader_code *shader)
89{
90    unsigned i, colorbuf_count = 0;
91
92    /* Mark the outputs as not present initially */
93    compiler->OutputColor[0] = shader->info.num_outputs;
94    compiler->OutputColor[1] = shader->info.num_outputs;
95    compiler->OutputColor[2] = shader->info.num_outputs;
96    compiler->OutputColor[3] = shader->info.num_outputs;
97    compiler->OutputDepth = shader->info.num_outputs;
98
99    /* Now see where they really are. */
100    for(i = 0; i < shader->info.num_outputs; ++i) {
101        switch(shader->info.output_semantic_name[i]) {
102            case TGSI_SEMANTIC_COLOR:
103                compiler->OutputColor[colorbuf_count] = i;
104                colorbuf_count++;
105                break;
106            case TGSI_SEMANTIC_POSITION:
107                compiler->OutputDepth = i;
108                break;
109        }
110    }
111}
112
113static void allocate_hardware_inputs(
114    struct r300_fragment_program_compiler * c,
115    void (*allocate)(void * data, unsigned input, unsigned hwreg),
116    void * mydata)
117{
118    struct r300_shader_semantics* inputs =
119        (struct r300_shader_semantics*)c->UserData;
120    int i, reg = 0;
121
122    /* Allocate input registers. */
123    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
124        if (inputs->color[i] != ATTR_UNUSED) {
125            allocate(mydata, inputs->color[i], reg++);
126        }
127    }
128    if (inputs->face != ATTR_UNUSED) {
129        allocate(mydata, inputs->face, reg++);
130    }
131    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
132        if (inputs->generic[i] != ATTR_UNUSED) {
133            allocate(mydata, inputs->generic[i], reg++);
134        }
135    }
136    if (inputs->fog != ATTR_UNUSED) {
137        allocate(mydata, inputs->fog, reg++);
138    }
139    if (inputs->wpos != ATTR_UNUSED) {
140        allocate(mydata, inputs->wpos, reg++);
141    }
142}
143
144static void get_external_state(
145    struct r300_context* r300,
146    struct r300_fragment_program_external_state* state)
147{
148    struct r300_textures_state *texstate = r300->textures_state.state;
149    unsigned i;
150    unsigned char *swizzle;
151
152    for (i = 0; i < texstate->sampler_state_count; i++) {
153        struct r300_sampler_state *s = texstate->sampler_states[i];
154        struct r300_sampler_view *v = texstate->sampler_views[i];
155        struct r300_resource *t;
156
157        if (!s || !v) {
158            continue;
159        }
160
161        t = r300_resource(texstate->sampler_views[i]->base.texture);
162
163        if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
164            state->unit[i].compare_mode_enabled = 1;
165
166            /* Pass depth texture swizzling to the compiler. */
167            if (texstate->sampler_views[i]) {
168                swizzle = texstate->sampler_views[i]->swizzle;
169
170                state->unit[i].depth_texture_swizzle =
171                    RC_MAKE_SWIZZLE(swizzle[0], swizzle[1],
172                                    swizzle[2], swizzle[3]);
173            } else {
174                state->unit[i].depth_texture_swizzle = RC_SWIZZLE_XYZW;
175            }
176
177            /* Fortunately, no need to translate this. */
178            state->unit[i].texture_compare_func = s->state.compare_func;
179        }
180
181        state->unit[i].non_normalized_coords = !s->state.normalized_coords;
182
183        /* XXX this should probably take into account STR, not just S. */
184        if (t->tex.is_npot) {
185            switch (s->state.wrap_s) {
186            case PIPE_TEX_WRAP_REPEAT:
187                state->unit[i].wrap_mode = RC_WRAP_REPEAT;
188                break;
189
190            case PIPE_TEX_WRAP_MIRROR_REPEAT:
191                state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT;
192                break;
193
194            case PIPE_TEX_WRAP_MIRROR_CLAMP:
195            case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
196            case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
197                state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP;
198                break;
199
200            default:
201                state->unit[i].wrap_mode = RC_WRAP_NONE;
202            }
203
204            if (t->b.b.b.target == PIPE_TEXTURE_3D)
205                state->unit[i].clamp_and_scale_before_fetch = TRUE;
206        }
207    }
208}
209
210static void r300_translate_fragment_shader(
211    struct r300_context* r300,
212    struct r300_fragment_shader_code* shader,
213    const struct tgsi_token *tokens);
214
215static void r300_dummy_fragment_shader(
216    struct r300_context* r300,
217    struct r300_fragment_shader_code* shader)
218{
219    struct pipe_shader_state state;
220    struct ureg_program *ureg;
221    struct ureg_dst out;
222    struct ureg_src imm;
223
224    /* Make a simple fragment shader which outputs (0, 0, 0, 1) */
225    ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
226    out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
227    imm = ureg_imm4f(ureg, 0, 0, 0, 1);
228
229    ureg_MOV(ureg, out, imm);
230    ureg_END(ureg);
231
232    state.tokens = ureg_finalize(ureg);
233
234    shader->dummy = TRUE;
235    r300_translate_fragment_shader(r300, shader, state.tokens);
236
237    ureg_destroy(ureg);
238}
239
240static void r300_emit_fs_code_to_buffer(
241    struct r300_context *r300,
242    struct r300_fragment_shader_code *shader)
243{
244    struct rX00_fragment_program_code *generic_code = &shader->code;
245    unsigned imm_count = shader->immediates_count;
246    unsigned imm_first = shader->externals_count;
247    unsigned imm_end = generic_code->constants.Count;
248    struct rc_constant *constants = generic_code->constants.Constants;
249    unsigned i;
250    CB_LOCALS;
251
252    if (r300->screen->caps.is_r500) {
253        struct r500_fragment_program_code *code = &generic_code->code.r500;
254
255        shader->cb_code_size = 19 +
256                               ((code->inst_end + 1) * 6) +
257                               imm_count * 7 +
258                               code->int_constant_count * 2;
259
260        NEW_CB(shader->cb_code, shader->cb_code_size);
261        OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
262        OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx);
263        OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl);
264        for(i = 0; i < code->int_constant_count; i++){
265                OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4),
266                                                code->int_constants[i]);
267        }
268        OUT_CB_REG(R500_US_CODE_RANGE,
269                   R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
270        OUT_CB_REG(R500_US_CODE_OFFSET, 0);
271        OUT_CB_REG(R500_US_CODE_ADDR,
272                   R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end));
273
274        OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR);
275        OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6);
276        for (i = 0; i <= code->inst_end; i++) {
277            OUT_CB(code->inst[i].inst0);
278            OUT_CB(code->inst[i].inst1);
279            OUT_CB(code->inst[i].inst2);
280            OUT_CB(code->inst[i].inst3);
281            OUT_CB(code->inst[i].inst4);
282            OUT_CB(code->inst[i].inst5);
283        }
284
285        /* Emit immediates. */
286        if (imm_count) {
287            for(i = imm_first; i < imm_end; ++i) {
288                if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
289                    const float *data = constants[i].u.Immediate;
290
291                    OUT_CB_REG(R500_GA_US_VECTOR_INDEX,
292                               R500_GA_US_VECTOR_INDEX_TYPE_CONST |
293                               (i & R500_GA_US_VECTOR_INDEX_MASK));
294                    OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
295                    OUT_CB_TABLE(data, 4);
296                }
297            }
298        }
299    } else { /* r300 */
300        struct r300_fragment_program_code *code = &generic_code->code.r300;
301        unsigned int alu_length = code->alu.length;
302        unsigned int alu_iterations = ((alu_length - 1) / 64) + 1;
303        unsigned int tex_length = code->tex.length;
304        unsigned int tex_iterations =
305            tex_length > 0 ? ((tex_length - 1) / 32) + 1 : 0;
306        unsigned int iterations =
307            alu_iterations > tex_iterations ? alu_iterations : tex_iterations;
308        unsigned int bank = 0;
309
310        shader->cb_code_size = 15 +
311            /* R400_US_CODE_BANK */
312            (r300->screen->caps.is_r400 ? 2 * (iterations + 1): 0) +
313            /* R400_US_CODE_EXT */
314            (r300->screen->caps.is_r400 ? 2 : 0) +
315            /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0, R400_US_ALU_EXT_ADDR_0 */
316            (code->r390_mode ? (5 * alu_iterations) : 4) +
317            /* R400_US_ALU_EXT_ADDR_[0-63] */
318            (code->r390_mode ? (code->alu.length) : 0) +
319            /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0 */
320            code->alu.length * 4 +
321            /* R300_US_TEX_INST_0, R300_US_TEX_INST_[0-31] */
322            (code->tex.length > 0 ? code->tex.length + tex_iterations : 0) +
323            imm_count * 5;
324
325        NEW_CB(shader->cb_code, shader->cb_code_size);
326
327        OUT_CB_REG(R300_US_CONFIG, code->config);
328        OUT_CB_REG(R300_US_PIXSIZE, code->pixsize);
329        OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset);
330
331        if (code->r390_mode) {
332            OUT_CB_REG(R400_US_CODE_EXT, code->r400_code_offset_ext);
333        } else if (r300->screen->caps.is_r400) {
334            /* This register appears to affect shaders even if r390_mode is
335             * disabled, so it needs to be set to 0 for shaders that
336             * don't use r390_mode. */
337            OUT_CB_REG(R400_US_CODE_EXT, 0);
338        }
339
340        OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4);
341        OUT_CB_TABLE(code->code_addr, 4);
342
343        do {
344            unsigned int bank_alu_length = (alu_length < 64 ? alu_length : 64);
345            unsigned int bank_alu_offset = bank * 64;
346            unsigned int bank_tex_length = (tex_length < 32 ? tex_length : 32);
347            unsigned int bank_tex_offset = bank * 32;
348
349            if (r300->screen->caps.is_r400) {
350                OUT_CB_REG(R400_US_CODE_BANK, code->r390_mode ?
351                                (bank << R400_BANK_SHIFT) | R400_R390_MODE_ENABLE : 0);//2
352            }
353
354            if (bank_alu_length > 0) {
355                OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, bank_alu_length);
356                for (i = 0; i < bank_alu_length; i++)
357                    OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_inst);
358
359                OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, bank_alu_length);
360                for (i = 0; i < bank_alu_length; i++)
361                    OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_addr);
362
363                OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, bank_alu_length);
364                for (i = 0; i < bank_alu_length; i++)
365                    OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_inst);
366
367                OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, bank_alu_length);
368                for (i = 0; i < bank_alu_length; i++)
369                    OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_addr);
370
371                if (code->r390_mode) {
372                    OUT_CB_REG_SEQ(R400_US_ALU_EXT_ADDR_0, bank_alu_length);
373                    for (i = 0; i < bank_alu_length; i++)
374                        OUT_CB(code->alu.inst[i + bank_alu_offset].r400_ext_addr);
375                }
376            }
377
378            if (bank_tex_length > 0) {
379                OUT_CB_REG_SEQ(R300_US_TEX_INST_0, bank_tex_length);
380                OUT_CB_TABLE(code->tex.inst + bank_tex_offset, bank_tex_length);
381            }
382
383            alu_length -= bank_alu_length;
384            tex_length -= bank_tex_length;
385            bank++;
386        } while(code->r390_mode && (alu_length > 0 || tex_length > 0));
387
388        /* R400_US_CODE_BANK needs to be reset to 0, otherwise some shaders
389         * will be rendered incorrectly. */
390        if (r300->screen->caps.is_r400) {
391            OUT_CB_REG(R400_US_CODE_BANK,
392                code->r390_mode ? R400_R390_MODE_ENABLE : 0);
393        }
394
395        /* Emit immediates. */
396        if (imm_count) {
397            for(i = imm_first; i < imm_end; ++i) {
398                if (constants[i].Type == RC_CONSTANT_IMMEDIATE) {
399                    const float *data = constants[i].u.Immediate;
400
401                    OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
402                    OUT_CB(pack_float24(data[0]));
403                    OUT_CB(pack_float24(data[1]));
404                    OUT_CB(pack_float24(data[2]));
405                    OUT_CB(pack_float24(data[3]));
406                }
407            }
408        }
409    }
410
411    OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src);
412    OUT_CB_REG(R300_US_W_FMT, shader->us_out_w);
413    END_CB;
414}
415
416static void r300_translate_fragment_shader(
417    struct r300_context* r300,
418    struct r300_fragment_shader_code* shader,
419    const struct tgsi_token *tokens)
420{
421    struct r300_fragment_program_compiler compiler;
422    struct tgsi_to_rc ttr;
423    int wpos, face;
424    unsigned i;
425
426    tgsi_scan_shader(tokens, &shader->info);
427    r300_shader_read_fs_inputs(&shader->info, &shader->inputs);
428
429    wpos = shader->inputs.wpos;
430    face = shader->inputs.face;
431
432    /* Setup the compiler. */
433    memset(&compiler, 0, sizeof(compiler));
434    rc_init(&compiler.Base);
435    DBG_ON(r300, DBG_FP) ? compiler.Base.Debug |= RC_DBG_LOG : 0;
436    DBG_ON(r300, DBG_P_STAT) ? compiler.Base.Debug |= RC_DBG_STATS : 0;
437
438    compiler.code = &shader->code;
439    compiler.state = shader->compare_state;
440    compiler.Base.is_r500 = r300->screen->caps.is_r500;
441    compiler.Base.is_r400 = r300->screen->caps.is_r400;
442    compiler.Base.disable_optimizations = DBG_ON(r300, DBG_NO_OPT);
443    compiler.Base.has_half_swizzles = TRUE;
444    compiler.Base.has_presub = TRUE;
445    compiler.Base.max_temp_regs =
446        compiler.Base.is_r500 ? 128 : (compiler.Base.is_r400 ? 64 : 32);
447    compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32;
448    compiler.Base.max_alu_insts =
449        (compiler.Base.is_r500 || compiler.Base.is_r400) ? 512 : 64;
450    compiler.Base.max_tex_insts =
451        (compiler.Base.is_r500 || compiler.Base.is_r400) ? 512 : 32;
452    compiler.AllocateHwInputs = &allocate_hardware_inputs;
453    compiler.UserData = &shader->inputs;
454
455    find_output_registers(&compiler, shader);
456
457    shader->write_all = FALSE;
458    for (i = 0; i < shader->info.num_properties; i++) {
459        if (shader->info.properties[i].name == TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS) {
460            shader->write_all = TRUE;
461        }
462    }
463
464    if (compiler.Base.Debug & RC_DBG_LOG) {
465        DBG(r300, DBG_FP, "r300: Initial fragment program\n");
466        tgsi_dump(tokens, 0);
467    }
468
469    /* Translate TGSI to our internal representation */
470    ttr.compiler = &compiler.Base;
471    ttr.info = &shader->info;
472    ttr.use_half_swizzles = TRUE;
473
474    r300_tgsi_to_rc(&ttr, tokens);
475
476    if (ttr.error) {
477        fprintf(stderr, "r300 FP: Cannot translate a shader. "
478                "Using a dummy shader instead.\n");
479        r300_dummy_fragment_shader(r300, shader);
480        return;
481    }
482
483    if (!r300->screen->caps.is_r500 ||
484        compiler.Base.Program.Constants.Count > 200) {
485        compiler.Base.remove_unused_constants = TRUE;
486    }
487
488    /**
489     * Transform the program to support WPOS.
490     *
491     * Introduce a small fragment at the start of the program that will be
492     * the only code that directly reads the WPOS input.
493     * All other code pieces that reference that input will be rewritten
494     * to read from a newly allocated temporary. */
495    if (wpos != ATTR_UNUSED) {
496        /* Moving the input to some other reg is not really necessary. */
497        rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE);
498    }
499
500    if (face != ATTR_UNUSED) {
501        rc_transform_fragment_face(&compiler.Base, face);
502    }
503
504    /* Invoke the compiler */
505    r3xx_compile_fragment_program(&compiler);
506
507    if (compiler.Base.Error) {
508        fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader"
509                " instead.\n", compiler.Base.ErrorMsg);
510
511        if (shader->dummy) {
512            fprintf(stderr, "r300 FP: Cannot compile the dummy shader! "
513                    "Giving up...\n");
514            abort();
515        }
516
517        rc_destroy(&compiler.Base);
518        r300_dummy_fragment_shader(r300, shader);
519        return;
520    }
521
522    /* Shaders with zero instructions are invalid,
523     * use the dummy shader instead. */
524    if (shader->code.code.r500.inst_end == -1) {
525        rc_destroy(&compiler.Base);
526        r300_dummy_fragment_shader(r300, shader);
527        return;
528    }
529
530    /* Initialize numbers of constants for each type. */
531    shader->externals_count = 0;
532    for (i = 0;
533         i < shader->code.constants.Count &&
534         shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) {
535        shader->externals_count = i+1;
536    }
537    shader->immediates_count = 0;
538    shader->rc_state_count = 0;
539
540    for (i = shader->externals_count; i < shader->code.constants.Count; i++) {
541        switch (shader->code.constants.Constants[i].Type) {
542            case RC_CONSTANT_IMMEDIATE:
543                ++shader->immediates_count;
544                break;
545            case RC_CONSTANT_STATE:
546                ++shader->rc_state_count;
547                break;
548            default:
549                assert(0);
550        }
551    }
552
553    /* Setup shader depth output. */
554    if (shader->code.writes_depth) {
555        shader->fg_depth_src = R300_FG_DEPTH_SRC_SHADER;
556        shader->us_out_w = R300_W_FMT_W24 | R300_W_SRC_US;
557    } else {
558        shader->fg_depth_src = R300_FG_DEPTH_SRC_SCAN;
559        shader->us_out_w = R300_W_FMT_W0 | R300_W_SRC_US;
560    }
561
562    /* And, finally... */
563    rc_destroy(&compiler.Base);
564
565    /* Build the command buffer. */
566    r300_emit_fs_code_to_buffer(r300, shader);
567}
568
569boolean r300_pick_fragment_shader(struct r300_context* r300)
570{
571    struct r300_fragment_shader* fs = r300_fs(r300);
572    struct r300_fragment_program_external_state state = {{{ 0 }}};
573    struct r300_fragment_shader_code* ptr;
574
575    get_external_state(r300, &state);
576
577    if (!fs->first) {
578        /* Build the fragment shader for the first time. */
579        fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code);
580
581        memcpy(&fs->shader->compare_state, &state,
582            sizeof(struct r300_fragment_program_external_state));
583        r300_translate_fragment_shader(r300, fs->shader, fs->state.tokens);
584        return TRUE;
585
586    } else {
587        /* Check if the currently-bound shader has been compiled
588         * with the texture-compare state we need. */
589        if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) {
590            /* Search for the right shader. */
591            ptr = fs->first;
592            while (ptr) {
593                if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) {
594                    if (fs->shader != ptr) {
595                        fs->shader = ptr;
596                        return TRUE;
597                    }
598                    /* The currently-bound one is OK. */
599                    return FALSE;
600                }
601                ptr = ptr->next;
602            }
603
604            /* Not found, gotta compile a new one. */
605            ptr = CALLOC_STRUCT(r300_fragment_shader_code);
606            ptr->next = fs->first;
607            fs->first = fs->shader = ptr;
608
609            ptr->compare_state = state;
610            r300_translate_fragment_shader(r300, ptr, fs->state.tokens);
611            return TRUE;
612        }
613    }
614
615    return FALSE;
616}
617