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