src/gallium/auxiliary/draw/draw_vs.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28  /*
  29   * Authors:
  30   *   Keith Whitwell <keith@tungstengraphics.com>
  31   *   Brian Paul
  32   */
  33
  34 #include "util/u_math.h"
  35 #include "util/u_memory.h"
  36
  37 #include "pipe/p_shader_tokens.h"
  38
  39 #include "draw_private.h"
  40 #include "draw_context.h"
  41 #include "draw_vs.h"
  42
  43 #include "translate/translate.h"
  44 #include "translate/translate_cache.h"
  45
  46 #include "tgsi/tgsi_dump.h"
  47 #include "tgsi/tgsi_exec.h"
  48
  49 DEBUG_GET_ONCE_BOOL_OPTION(gallium_dump_vs, "GALLIUM_DUMP_VS", FALSE)
  50
  51
  52 /**
  53  * Set a vertex shader constant buffer.
  54  * \param slot  which constant buffer in [0, PIPE_MAX_CONSTANT_BUFFERS-1]
  55  * \param constants  the mapped buffer
  56  * \param size  size of buffer in bytes
  57  */
  58 void
  59 draw_vs_set_constants(struct draw_context *draw,
  60                       unsigned slot,
  61                       const void *constants,
  62                       unsigned size)
  63 {
  64    const int alignment = 16;
  65
  66    /* check if buffer is 16-byte aligned */
  67    if (((uintptr_t)constants) & (alignment - 1)) {
  68       /* if not, copy the constants into a new, 16-byte aligned buffer */
  69       if (size > draw->vs.const_storage_size[slot]) {
  70          if (draw->vs.aligned_constant_storage[slot]) {
  71             align_free((void *)draw->vs.aligned_constant_storage[slot]);
  72          }
  73          draw->vs.aligned_constant_storage[slot] =
  74             align_malloc(size, alignment);
  75       }
  76       assert(constants);
  77       memcpy((void *)draw->vs.aligned_constant_storage[slot],
  78              constants,
  79              size);
  80       constants = draw->vs.aligned_constant_storage[slot];
  81    }
  82
  83    draw->vs.aligned_constants[slot] = constants;
  84    draw_vs_aos_machine_constants(draw->vs.aos_machine, slot, constants);
  85 }
  86
  87
  88 void draw_vs_set_viewport( struct draw_context *draw,
  89                            const struct pipe_viewport_state *viewport )
  90 {
  91    draw_vs_aos_machine_viewport( draw->vs.aos_machine, viewport );
  92 }
  93
  94
  95
  96 struct draw_vertex_shader *
  97 draw_create_vertex_shader(struct draw_context *draw,
  98                           const struct pipe_shader_state *shader)
  99 {
 100    struct draw_vertex_shader *vs = NULL;
 101
 102    if (draw->dump_vs) {
 103       tgsi_dump(shader->tokens, 0);
 104    }
 105
 106    if (!draw->pt.middle.llvm) {
 107 #if 0
 108 /* these paths don't support vertex clamping
 109  * TODO: either add it, or remove them completely
 110  * use LLVM instead if you want performance
 111  * use exec instead if you want debugging/more correctness
 112  */
 113 #if defined(PIPE_ARCH_X86)
 114       vs = draw_create_vs_sse( draw, shader );
 115 #elif defined(PIPE_ARCH_PPC)
 116       vs = draw_create_vs_ppc( draw, shader );
 117 #endif
 118 #endif
 119    }
 120 #if HAVE_LLVM
 121    else {
 122       vs = draw_create_vs_llvm(draw, shader);
 123    }
 124 #endif
 125
 126    if (!vs) {
 127       vs = draw_create_vs_exec( draw, shader );
 128    }
 129
 130    if (vs)
 131    {
 132       uint i;
 133       for (i = 0; i < vs->info.num_outputs; i++) {
 134          if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
 135              vs->info.output_semantic_index[i] == 0)
 136             vs->position_output = i;
 137          else if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_EDGEFLAG &&
 138              vs->info.output_semantic_index[i] == 0)
 139             vs->edgeflag_output = i;
 140       }
 141    }
 142
 143    assert(vs);
 144    return vs;
 145 }
 146
 147
 148 void
 149 draw_bind_vertex_shader(struct draw_context *draw,
 150                         struct draw_vertex_shader *dvs)
 151 {
 152    draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
 153
 154    if (dvs)
 155    {
 156       draw->vs.vertex_shader = dvs;
 157       draw->vs.num_vs_outputs = dvs->info.num_outputs;
 158       draw->vs.position_output = dvs->position_output;
 159       draw->vs.edgeflag_output = dvs->edgeflag_output;
 160       dvs->prepare( dvs, draw );
 161    }
 162    else {
 163       draw->vs.vertex_shader = NULL;
 164       draw->vs.num_vs_outputs = 0;
 165    }
 166 }
 167
 168
 169 void
 170 draw_delete_vertex_shader(struct draw_context *draw,
 171                           struct draw_vertex_shader *dvs)
 172 {
 173    unsigned i;
 174
 175    for (i = 0; i < dvs->nr_variants; i++)
 176       dvs->variant[i]->destroy( dvs->variant[i] );
 177
 178    dvs->nr_variants = 0;
 179
 180    dvs->delete( dvs );
 181 }
 182
 183
 184
 185 boolean
 186 draw_vs_init( struct draw_context *draw )
 187 {
 188    draw->dump_vs = debug_get_option_gallium_dump_vs();
 189
 190    draw->vs.machine = tgsi_exec_machine_create();
 191    if (!draw->vs.machine)
 192       return FALSE;
 193
 194    draw->vs.emit_cache = translate_cache_create();
 195    if (!draw->vs.emit_cache)
 196       return FALSE;
 197
 198    draw->vs.fetch_cache = translate_cache_create();
 199    if (!draw->vs.fetch_cache)
 200       return FALSE;
 201
 202    draw->vs.aos_machine = draw_vs_aos_machine();
 203 #ifdef PIPE_ARCH_X86
 204    if (!draw->vs.aos_machine)
 205       return FALSE;
 206 #endif
 207
 208    return TRUE;
 209 }
 210
 211 void
 212 draw_vs_destroy( struct draw_context *draw )
 213 {
 214    uint i;
 215
 216    if (draw->vs.fetch_cache)
 217       translate_cache_destroy(draw->vs.fetch_cache);
 218
 219    if (draw->vs.emit_cache)
 220       translate_cache_destroy(draw->vs.emit_cache);
 221
 222    if (draw->vs.aos_machine)
 223       draw_vs_aos_machine_destroy(draw->vs.aos_machine);
 224
 225    for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
 226       if (draw->vs.aligned_constant_storage[i]) {
 227          align_free((void *)draw->vs.aligned_constant_storage[i]);
 228       }
 229    }
 230
 231    tgsi_exec_machine_destroy(draw->vs.machine);
 232 }
 233
 234
 235 struct draw_vs_variant *
 236 draw_vs_lookup_variant( struct draw_vertex_shader *vs,
 237                         const struct draw_vs_variant_key *key )
 238 {
 239    struct draw_vs_variant *variant;
 240    unsigned i;
 241
 242    /* Lookup existing variant:
 243     */
 244    for (i = 0; i < vs->nr_variants; i++)
 245       if (draw_vs_variant_key_compare(key, &vs->variant[i]->key) == 0)
 246          return vs->variant[i];
 247
 248    /* Else have to create a new one:
 249     */
 250    variant = vs->create_variant( vs, key );
 251    if (variant == NULL)
 252       return NULL;
 253
 254    /* Add it to our list, could be smarter:
 255     */
 256    if (vs->nr_variants < Elements(vs->variant)) {
 257       vs->variant[vs->nr_variants++] = variant;
 258    }
 259    else {
 260       vs->last_variant++;
 261       vs->last_variant %= Elements(vs->variant);
 262       vs->variant[vs->last_variant]->destroy(vs->variant[vs->last_variant]);
 263       vs->variant[vs->last_variant] = variant;
 264    }
 265
 266    /* Done
 267     */
 268    return variant;
 269 }
 270
 271
 272 struct translate *
 273 draw_vs_get_fetch( struct draw_context *draw,
 274                    struct translate_key *key )
 275 {
 276    if (!draw->vs.fetch ||
 277        translate_key_compare(&draw->vs.fetch->key, key) != 0)
 278    {
 279       translate_key_sanitize(key);
 280       draw->vs.fetch = translate_cache_find(draw->vs.fetch_cache, key);
 281    }
 282
 283    return draw->vs.fetch;
 284 }
 285
 286 struct translate *
 287 draw_vs_get_emit( struct draw_context *draw,
 288                   struct translate_key *key )
 289 {
 290    if (!draw->vs.emit ||
 291        translate_key_compare(&draw->vs.emit->key, key) != 0)
 292    {
 293       translate_key_sanitize(key);
 294       draw->vs.emit = translate_cache_find(draw->vs.emit_cache, key);
 295    }
 296
 297    return draw->vs.emit;
 298 }