src/gallium/drivers/r300/r300_fs.c

   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_context.h"
  32 #include "r300_screen.h"
  33 #include "r300_fs.h"
  34 #include "r300_reg.h"
  35 #include "r300_tgsi_to_rc.h"
  36
  37 #include "radeon_code.h"
  38 #include "radeon_compiler.h"
  39
  40 /* Convert info about FS input semantics to r300_shader_semantics. */
  41 void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
  42                                 struct r300_shader_semantics* fs_inputs)
  43 {
  44     int i;
  45     unsigned index;
  46
  47     r300_shader_semantics_reset(fs_inputs);
  48
  49     for (i = 0; i < info->num_inputs; i++) {
  50         index = info->input_semantic_index[i];
  51
  52         switch (info->input_semantic_name[i]) {
  53             case TGSI_SEMANTIC_COLOR:
  54                 assert(index < ATTR_COLOR_COUNT);
  55                 fs_inputs->color[index] = i;
  56                 break;
  57
  58             case TGSI_SEMANTIC_GENERIC:
  59                 assert(index < ATTR_GENERIC_COUNT);
  60                 fs_inputs->generic[index] = i;
  61                 break;
  62
  63             case TGSI_SEMANTIC_FOG:
  64                 assert(index == 0);
  65                 fs_inputs->fog = i;
  66                 break;
  67
  68             case TGSI_SEMANTIC_POSITION:
  69                 assert(index == 0);
  70                 fs_inputs->wpos = i;
  71                 break;
  72
  73             default:
  74                 fprintf(stderr, "r300: FP: Unknown input semantic: %i\n",
  75                         info->input_semantic_name[i]);
  76         }
  77     }
  78 }
  79
  80 static void find_output_registers(struct r300_fragment_program_compiler * compiler,
  81                                   struct r300_fragment_shader_code *shader)
  82 {
  83     unsigned i, colorbuf_count = 0;
  84
  85     /* Mark the outputs as not present initially */
  86     compiler->OutputColor[0] = shader->info.num_outputs;
  87     compiler->OutputColor[1] = shader->info.num_outputs;
  88     compiler->OutputColor[2] = shader->info.num_outputs;
  89     compiler->OutputColor[3] = shader->info.num_outputs;
  90     compiler->OutputDepth = shader->info.num_outputs;
  91
  92     /* Now see where they really are. */
  93     for(i = 0; i < shader->info.num_outputs; ++i) {
  94         switch(shader->info.output_semantic_name[i]) {
  95             case TGSI_SEMANTIC_COLOR:
  96                 compiler->OutputColor[colorbuf_count] = i;
  97                 colorbuf_count++;
  98                 break;
  99             case TGSI_SEMANTIC_POSITION:
 100                 compiler->OutputDepth = i;
 101                 break;
 102         }
 103     }
 104 }
 105
 106 static void allocate_hardware_inputs(
 107     struct r300_fragment_program_compiler * c,
 108     void (*allocate)(void * data, unsigned input, unsigned hwreg),
 109     void * mydata)
 110 {
 111     struct r300_shader_semantics* inputs =
 112         (struct r300_shader_semantics*)c->UserData;
 113     int i, reg = 0;
 114
 115     /* Allocate input registers. */
 116     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 117         if (inputs->color[i] != ATTR_UNUSED) {
 118             allocate(mydata, inputs->color[i], reg++);
 119         }
 120     }
 121     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 122         if (inputs->generic[i] != ATTR_UNUSED) {
 123             allocate(mydata, inputs->generic[i], reg++);
 124         }
 125     }
 126     if (inputs->fog != ATTR_UNUSED) {
 127         allocate(mydata, inputs->fog, reg++);
 128     }
 129     if (inputs->wpos != ATTR_UNUSED) {
 130         allocate(mydata, inputs->wpos, reg++);
 131     }
 132 }
 133
 134 static void get_external_state(
 135     struct r300_context* r300,
 136     struct r300_fragment_program_external_state* state)
 137 {
 138     struct r300_textures_state *texstate = r300->textures_state.state;
 139     unsigned i;
 140
 141     for (i = 0; i < texstate->sampler_state_count; i++) {
 142         struct r300_sampler_state* s = texstate->sampler_states[i];
 143
 144         if (!s) {
 145             continue;
 146         }
 147
 148         if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
 149             /* XXX Gallium doesn't provide us with any information regarding
 150              * this mode, so we are screwed. I'm setting 0 = LUMINANCE. */
 151             state->unit[i].depth_texture_mode = 0;
 152
 153             /* Fortunately, no need to translate this. */
 154             state->unit[i].texture_compare_func = s->state.compare_func;
 155         }
 156
 157         state->unit[i].non_normalized_coords = !s->state.normalized_coords;
 158
 159         if (texstate->sampler_views[i]) {
 160             struct r300_texture *t;
 161             t = (struct r300_texture*)texstate->sampler_views[i]->base.texture;
 162
 163             /* XXX this should probably take into account STR, not just S. */
 164             if (t->uses_pitch) {
 165                 switch (s->state.wrap_s) {
 166                     case PIPE_TEX_WRAP_REPEAT:
 167                         state->unit[i].wrap_mode = RC_WRAP_REPEAT;
 168                         state->unit[i].fake_npot = TRUE;
 169                         break;
 170
 171                     case PIPE_TEX_WRAP_CLAMP:
 172                     case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
 173                     case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
 174                         state->unit[i].wrap_mode = RC_WRAP_CLAMP;
 175                         break;
 176
 177                     case PIPE_TEX_WRAP_MIRROR_REPEAT:
 178                     case PIPE_TEX_WRAP_MIRROR_CLAMP:
 179                     case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
 180                     case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
 181                         state->unit[i].wrap_mode = RC_WRAP_MIRROR;
 182                         state->unit[i].fake_npot = TRUE;
 183                         break;
 184
 185                     default:
 186                         state->unit[i].wrap_mode = RC_WRAP_NONE;
 187                         break;
 188                 }
 189             }
 190         }
 191     }
 192 }
 193
 194 static void r300_translate_fragment_shader(
 195     struct r300_context* r300,
 196     struct r300_fragment_shader_code* shader,
 197     const struct tgsi_token *tokens);
 198
 199 static void r300_dummy_fragment_shader(
 200     struct r300_context* r300,
 201     struct r300_fragment_shader_code* shader)
 202 {
 203     struct pipe_shader_state state;
 204     struct ureg_program *ureg;
 205     struct ureg_dst out;
 206     struct ureg_src imm;
 207
 208     /* Make a simple fragment shader which outputs (0, 0, 0, 1) */
 209     ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 210     out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
 211     imm = ureg_imm4f(ureg, 0, 0, 0, 1);
 212
 213     ureg_MOV(ureg, out, imm);
 214     ureg_END(ureg);
 215
 216     state.tokens = ureg_finalize(ureg);
 217
 218     shader->dummy = TRUE;
 219     r300_translate_fragment_shader(r300, shader, state.tokens);
 220
 221     ureg_destroy(ureg);
 222 }
 223
 224 static void r300_translate_fragment_shader(
 225     struct r300_context* r300,
 226     struct r300_fragment_shader_code* shader,
 227     const struct tgsi_token *tokens)
 228 {
 229     struct r300_fragment_program_compiler compiler;
 230     struct tgsi_to_rc ttr;
 231     int wpos;
 232     unsigned i;
 233
 234     tgsi_scan_shader(tokens, &shader->info);
 235     r300_shader_read_fs_inputs(&shader->info, &shader->inputs);
 236
 237     wpos = shader->inputs.wpos;
 238
 239     /* Setup the compiler. */
 240     memset(&compiler, 0, sizeof(compiler));
 241     rc_init(&compiler.Base);
 242     compiler.Base.Debug = DBG_ON(r300, DBG_FP);
 243
 244     compiler.code = &shader->code;
 245     compiler.state = shader->compare_state;
 246     compiler.is_r500 = r300->screen->caps.is_r500;
 247     compiler.max_temp_regs = compiler.is_r500 ? 128 : 32;
 248     compiler.AllocateHwInputs = &allocate_hardware_inputs;
 249     compiler.UserData = &shader->inputs;
 250
 251     find_output_registers(&compiler, shader);
 252
 253     if (compiler.Base.Debug) {
 254         debug_printf("r300: Initial fragment program\n");
 255         tgsi_dump(tokens, 0);
 256     }
 257
 258     /* Translate TGSI to our internal representation */
 259     ttr.compiler = &compiler.Base;
 260     ttr.info = &shader->info;
 261     ttr.use_half_swizzles = TRUE;
 262
 263     r300_tgsi_to_rc(&ttr, tokens);
 264
 265     /**
 266      * Transform the program to support WPOS.
 267      *
 268      * Introduce a small fragment at the start of the program that will be
 269      * the only code that directly reads the WPOS input.
 270      * All other code pieces that reference that input will be rewritten
 271      * to read from a newly allocated temporary. */
 272     if (wpos != ATTR_UNUSED) {
 273         /* Moving the input to some other reg is not really necessary. */
 274         rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE);
 275     }
 276
 277     /* Invoke the compiler */
 278     r3xx_compile_fragment_program(&compiler);
 279
 280     if (compiler.Base.Error) {
 281         fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader"
 282                 " instead.\n", compiler.Base.ErrorMsg);
 283
 284         if (shader->dummy) {
 285             fprintf(stderr, "r300 FP: Cannot compile the dummy shader! "
 286                     "Giving up...\n");
 287             abort();
 288         }
 289
 290         rc_destroy(&compiler.Base);
 291         r300_dummy_fragment_shader(r300, shader);
 292         return;
 293     }
 294
 295     /* Initialize numbers of constants for each type. */
 296     shader->externals_count = ttr.immediate_offset;
 297     shader->immediates_count = 0;
 298     shader->rc_state_count = 0;
 299
 300     for (i = shader->externals_count; i < shader->code.constants.Count; i++) {
 301         switch (shader->code.constants.Constants[i].Type) {
 302             case RC_CONSTANT_IMMEDIATE:
 303                 ++shader->immediates_count;
 304                 break;
 305             case RC_CONSTANT_STATE:
 306                 ++shader->rc_state_count;
 307                 break;
 308             default:
 309                 assert(0);
 310         }
 311     }
 312
 313     /* Setup shader depth output. */
 314     if (shader->code.writes_depth) {
 315         shader->fg_depth_src = R300_FG_DEPTH_SRC_SHADER;
 316         shader->us_out_w = R300_W_FMT_W24 | R300_W_SRC_US;
 317     } else {
 318         shader->fg_depth_src = R300_FG_DEPTH_SRC_SCAN;
 319         shader->us_out_w = R300_W_FMT_W0 | R300_W_SRC_US;
 320     }
 321
 322     /* And, finally... */
 323     rc_destroy(&compiler.Base);
 324 }
 325
 326 boolean r300_pick_fragment_shader(struct r300_context* r300)
 327 {
 328     struct r300_fragment_shader* fs = r300_fs(r300);
 329     struct r300_fragment_program_external_state state = {{{ 0 }}};
 330     struct r300_fragment_shader_code* ptr;
 331
 332     get_external_state(r300, &state);
 333
 334     if (!fs->first) {
 335         /* Build the fragment shader for the first time. */
 336         fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code);
 337
 338         memcpy(&fs->shader->compare_state, &state,
 339             sizeof(struct r300_fragment_program_external_state));
 340         r300_translate_fragment_shader(r300, fs->shader, fs->state.tokens);
 341         return TRUE;
 342
 343     } else {
 344         /* Check if the currently-bound shader has been compiled
 345          * with the texture-compare state we need. */
 346         if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) {
 347             /* Search for the right shader. */
 348             ptr = fs->first;
 349             while (ptr) {
 350                 if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) {
 351                     if (fs->shader != ptr) {
 352                         fs->shader = ptr;
 353                         return TRUE;
 354                     }
 355                     /* The currently-bound one is OK. */
 356                     return FALSE;
 357                 }
 358                 ptr = ptr->next;
 359             }
 360
 361             /* Not found, gotta compile a new one. */
 362             ptr = CALLOC_STRUCT(r300_fragment_shader_code);
 363             ptr->next = fs->first;
 364             fs->first = fs->shader = ptr;
 365
 366             ptr->compare_state = state;
 367             r300_translate_fragment_shader(r300, ptr, fs->state.tokens);
 368             return TRUE;
 369         }
 370     }
 371
 372     return FALSE;
 373 }