r300_fs.c revision d2ac3d5e79bdf5a32a2dca135403d963ac6c83e4 
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
30#include "r300_context.h"
31#include "r300_screen.h"
32#include "r300_fs.h"
33#include "r300_tgsi_to_rc.h"
34
35#include "radeon_code.h"
36#include "radeon_compiler.h"
37
38/* Convert info about FS input semantics to r300_shader_semantics. */
39void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
40                                struct r300_shader_semantics* fs_inputs)
41{
42    int i;
43    unsigned index;
44
45    r300_shader_semantics_reset(fs_inputs);
46
47    for (i = 0; i < info->num_inputs; i++) {
48        index = info->input_semantic_index[i];
49
50        switch (info->input_semantic_name[i]) {
51            case TGSI_SEMANTIC_COLOR:
52                assert(index < ATTR_COLOR_COUNT);
53                fs_inputs->color[index] = i;
54                break;
55
56            case TGSI_SEMANTIC_GENERIC:
57                assert(index < ATTR_GENERIC_COUNT);
58                fs_inputs->generic[index] = i;
59                break;
60
61            case TGSI_SEMANTIC_FOG:
62                assert(index == 0);
63                fs_inputs->fog = i;
64                break;
65
66            case TGSI_SEMANTIC_POSITION:
67                assert(index == 0);
68                fs_inputs->wpos = i;
69                break;
70
71            default:
72                assert(0);
73        }
74    }
75}
76
77static void find_output_registers(struct r300_fragment_program_compiler * compiler,
78                                  struct r300_fragment_shader * fs)
79{
80    unsigned i, colorbuf_count = 0;
81
82    /* Mark the outputs as not present initially */
83    compiler->OutputColor[0] = fs->info.num_outputs;
84    compiler->OutputColor[1] = fs->info.num_outputs;
85    compiler->OutputColor[2] = fs->info.num_outputs;
86    compiler->OutputColor[3] = fs->info.num_outputs;
87    compiler->OutputDepth = fs->info.num_outputs;
88
89    /* Now see where they really are. */
90    for(i = 0; i < fs->info.num_outputs; ++i) {
91        switch(fs->info.output_semantic_name[i]) {
92            case TGSI_SEMANTIC_COLOR:
93                compiler->OutputColor[colorbuf_count] = i;
94                colorbuf_count++;
95                break;
96            case TGSI_SEMANTIC_POSITION:
97                compiler->OutputDepth = i;
98                break;
99        }
100    }
101}
102
103static void allocate_hardware_inputs(
104    struct r300_fragment_program_compiler * c,
105    void (*allocate)(void * data, unsigned input, unsigned hwreg),
106    void * mydata)
107{
108    struct r300_shader_semantics* inputs =
109        (struct r300_shader_semantics*)c->UserData;
110    int i, reg = 0;
111
112    /* Allocate input registers. */
113    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
114        if (inputs->color[i] != ATTR_UNUSED) {
115            allocate(mydata, inputs->color[i], reg++);
116        }
117    }
118    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
119        if (inputs->generic[i] != ATTR_UNUSED) {
120            allocate(mydata, inputs->generic[i], reg++);
121        }
122    }
123    if (inputs->fog != ATTR_UNUSED) {
124        allocate(mydata, inputs->fog, reg++);
125    }
126    if (inputs->wpos != ATTR_UNUSED) {
127        allocate(mydata, inputs->wpos, reg++);
128    }
129}
130
131static void get_compare_state(
132    struct r300_context* r300,
133    struct r300_fragment_program_external_state* state,
134    unsigned shadow_samplers)
135{
136    struct r300_textures_state *texstate =
137        (struct r300_textures_state*)r300->textures_state.state;
138
139    memset(state, 0, sizeof(*state));
140
141    for (int i = 0; i < texstate->sampler_count; i++) {
142        struct r300_sampler_state* s = texstate->sampler_states[i];
143
144        if (s && s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
145            /* XXX Gallium doesn't provide us with any information regarding
146             * this mode, so we are screwed. I'm setting 0 = LUMINANCE. */
147            state->unit[i].depth_texture_mode = 0;
148
149            /* Fortunately, no need to translate this. */
150            state->unit[i].texture_compare_func = s->state.compare_func;
151        }
152    }
153}
154
155static void r300_translate_fragment_shader(
156    struct r300_context* r300,
157    struct r300_fragment_shader_code* shader)
158{
159    struct r300_fragment_shader* fs = r300->fs;
160    struct r300_fragment_program_compiler compiler;
161    struct tgsi_to_rc ttr;
162    int wpos = fs->inputs.wpos;
163
164    /* Setup the compiler. */
165    memset(&compiler, 0, sizeof(compiler));
166    rc_init(&compiler.Base);
167    compiler.Base.Debug = DBG_ON(r300, DBG_FP);
168
169    compiler.code = &shader->code;
170    compiler.state = shader->compare_state;
171    compiler.is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
172    compiler.AllocateHwInputs = &allocate_hardware_inputs;
173    compiler.UserData = &fs->inputs;
174
175    find_output_registers(&compiler, fs);
176
177    if (compiler.Base.Debug) {
178        debug_printf("r300: Initial fragment program\n");
179        tgsi_dump(fs->state.tokens, 0);
180    }
181
182    /* Translate TGSI to our internal representation */
183    ttr.compiler = &compiler.Base;
184    ttr.info = &fs->info;
185    ttr.use_half_swizzles = TRUE;
186
187    r300_tgsi_to_rc(&ttr, fs->state.tokens);
188
189    fs->shadow_samplers = compiler.Base.Program.ShadowSamplers;
190
191    /**
192     * Transform the program to support WPOS.
193     *
194     * Introduce a small fragment at the start of the program that will be
195     * the only code that directly reads the WPOS input.
196     * All other code pieces that reference that input will be rewritten
197     * to read from a newly allocated temporary. */
198    if (wpos != ATTR_UNUSED) {
199        /* Moving the input to some other reg is not really necessary. */
200        rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE);
201    }
202
203    /* Invoke the compiler */
204    r3xx_compile_fragment_program(&compiler);
205    if (compiler.Base.Error) {
206        /* XXX failover maybe? */
207        DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n",
208            compiler.Base.ErrorMsg);
209        assert(0);
210    }
211
212    /* And, finally... */
213    rc_destroy(&compiler.Base);
214}
215
216boolean r300_pick_fragment_shader(struct r300_context* r300)
217{
218    struct r300_fragment_shader* fs = r300->fs;
219    struct r300_fragment_program_external_state state;
220    struct r300_fragment_shader_code* ptr;
221
222    if (!fs->first) {
223        /* Build the fragment shader for the first time. */
224        fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code);
225
226        /* BTW shadow samplers will be known after the first translation,
227         * therefore we set ~0, which means it should look at all sampler
228         * states. This choice doesn't have any impact on the correctness. */
229        get_compare_state(r300, &fs->shader->compare_state, ~0);
230        r300_translate_fragment_shader(r300, fs->shader);
231        return TRUE;
232
233    } else if (fs->shadow_samplers) {
234        get_compare_state(r300, &state, fs->shadow_samplers);
235
236        /* Check if the currently-bound shader has been compiled
237         * with the texture-compare state we need. */
238        if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) {
239            /* Search for the right shader. */
240            ptr = fs->first;
241            while (ptr) {
242                if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) {
243                    fs->shader = ptr;
244                    return TRUE;
245                }
246                ptr = ptr->next;
247            }
248
249            /* Not found, gotta compile a new one. */
250            ptr = CALLOC_STRUCT(r300_fragment_shader_code);
251            ptr->next = fs->first;
252            fs->first = fs->shader = ptr;
253
254            ptr->compare_state = state;
255            r300_translate_fragment_shader(r300, ptr);
256            return TRUE;
257        }
258    }
259
260    return FALSE;
261}
262