src/mesa/drivers/dri/i965/brw_wm.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a 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, sublicense, 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
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32 #include "main/texformat.h"
  33 #include "brw_context.h"
  34 #include "brw_util.h"
  35 #include "brw_wm.h"
  36 #include "brw_state.h"
  37
  38
  39 /** Return number of src args for given instruction */
  40 GLuint brw_wm_nr_args( GLuint opcode )
  41 {
  42    switch (opcode) {
  43    case WM_FRONTFACING:
  44       return 0;
  45    case WM_PIXELXY:
  46    case WM_CINTERP:
  47    case WM_WPOSXY:
  48       return 1;
  49    case WM_LINTERP:
  50    case WM_DELTAXY:
  51    case WM_PIXELW:
  52       return 2;
  53    case WM_FB_WRITE:
  54    case WM_PINTERP:
  55       return 3;
  56    default:
  57       assert(opcode < MAX_OPCODE);
  58       return _mesa_num_inst_src_regs(opcode);
  59    }
  60 }
  61
  62
  63 GLuint brw_wm_is_scalar_result( GLuint opcode )
  64 {
  65    switch (opcode) {
  66    case OPCODE_COS:
  67    case OPCODE_EX2:
  68    case OPCODE_LG2:
  69    case OPCODE_POW:
  70    case OPCODE_RCP:
  71    case OPCODE_RSQ:
  72    case OPCODE_SIN:
  73    case OPCODE_DP3:
  74    case OPCODE_DP4:
  75    case OPCODE_DPH:
  76    case OPCODE_DST:
  77       return 1;
  78
  79    default:
  80       return 0;
  81    }
  82 }
  83
  84
  85 /**
  86  * Do GPU code generation for non-GLSL shader.  non-GLSL shaders have
  87  * no flow control instructions so we can more readily do SSA-style
  88  * optimizations.
  89  */
  90 static void
  91 brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
  92 {
  93    /* Augment fragment program.  Add instructions for pre- and
  94     * post-fragment-program tasks such as interpolation and fogging.
  95     */
  96    brw_wm_pass_fp(c);
  97
  98    /* Translate to intermediate representation.  Build register usage
  99     * chains.
 100     */
 101    brw_wm_pass0(c);
 102
 103    /* Dead code removal.
 104     */
 105    brw_wm_pass1(c);
 106
 107    /* Register allocation.
 108     */
 109    c->grf_limit = BRW_WM_MAX_GRF / 2;
 110
 111    brw_wm_pass2(c);
 112
 113    c->prog_data.total_grf = c->max_wm_grf;
 114    if (c->last_scratch) {
 115       c->prog_data.total_scratch = c->last_scratch + 0x40;
 116    }
 117    else {
 118       c->prog_data.total_scratch = 0;
 119    }
 120
 121    /* Emit GEN4 code.
 122     */
 123    brw_wm_emit(c);
 124 }
 125
 126
 127 /**
 128  * All Mesa program -> GPU code generation goes through this function.
 129  * Depending on the instructions used (i.e. flow control instructions)
 130  * we'll use one of two code generators.
 131  */
 132 static void do_wm_prog( struct brw_context *brw,
 133                         struct brw_fragment_program *fp,
 134                         struct brw_wm_prog_key *key)
 135 {
 136    struct brw_wm_compile *c;
 137    const GLuint *program;
 138    GLuint program_size;
 139
 140    c = brw->wm.compile_data;
 141    if (c == NULL) {
 142       brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
 143       c = brw->wm.compile_data;
 144    } else {
 145       memset(c, 0, sizeof(*brw->wm.compile_data));
 146    }
 147    memcpy(&c->key, key, sizeof(*key));
 148
 149    c->fp = fp;
 150    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 151
 152    brw_init_compile(brw, &c->func);
 153
 154    /* temporary sanity check assertion */
 155    ASSERT(fp->isGLSL == brw_wm_is_glsl(&c->fp->program));
 156
 157    /*
 158     * Shader which use GLSL features such as flow control are handled
 159     * differently from "simple" shaders.
 160     */
 161    if (fp->isGLSL) {
 162       brw_wm_glsl_emit(brw, c);
 163    }
 164    else {
 165       brw_wm_non_glsl_emit(brw, c);
 166    }
 167
 168    if (INTEL_DEBUG & DEBUG_WM)
 169       fprintf(stderr, "\n");
 170
 171    /* get the program
 172     */
 173    program = brw_get_program(&c->func, &program_size);
 174
 175    dri_bo_unreference(brw->wm.prog_bo);
 176    brw->wm.prog_bo = brw_upload_cache( &brw->cache, BRW_WM_PROG,
 177                                        &c->key, sizeof(c->key),
 178                                        NULL, 0,
 179                                        program, program_size,
 180                                        &c->prog_data,
 181                                        &brw->wm.prog_data );
 182 }
 183
 184
 185
 186 static void brw_wm_populate_key( struct brw_context *brw,
 187                                  struct brw_wm_prog_key *key )
 188 {
 189    GLcontext *ctx = &brw->intel.ctx;
 190    /* BRW_NEW_FRAGMENT_PROGRAM */
 191    const struct brw_fragment_program *fp =
 192       (struct brw_fragment_program *)brw->fragment_program;
 193    GLuint lookup = 0;
 194    GLuint line_aa;
 195    GLuint i;
 196
 197    memset(key, 0, sizeof(*key));
 198
 199    /* Build the index for table lookup
 200     */
 201    /* _NEW_COLOR */
 202    if (fp->program.UsesKill ||
 203        ctx->Color.AlphaEnabled)
 204       lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 205
 206    if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPTH))
 207       lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 208
 209    /* _NEW_DEPTH */
 210    if (ctx->Depth.Test)
 211       lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 212
 213    if (ctx->Depth.Test &&
 214        ctx->Depth.Mask) /* ?? */
 215       lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 216
 217    /* _NEW_STENCIL */
 218    if (ctx->Stencil._Enabled) {
 219       lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 220
 221       if (ctx->Stencil.WriteMask[0] ||
 222           ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 223          lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 224    }
 225
 226    line_aa = AA_NEVER;
 227
 228    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 229    if (ctx->Line.SmoothFlag) {
 230       if (brw->intel.reduced_primitive == GL_LINES) {
 231          line_aa = AA_ALWAYS;
 232       }
 233       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 234          if (ctx->Polygon.FrontMode == GL_LINE) {
 235             line_aa = AA_SOMETIMES;
 236
 237             if (ctx->Polygon.BackMode == GL_LINE ||
 238                 (ctx->Polygon.CullFlag &&
 239                  ctx->Polygon.CullFaceMode == GL_BACK))
 240                line_aa = AA_ALWAYS;
 241          }
 242          else if (ctx->Polygon.BackMode == GL_LINE) {
 243             line_aa = AA_SOMETIMES;
 244
 245             if ((ctx->Polygon.CullFlag &&
 246                  ctx->Polygon.CullFaceMode == GL_FRONT))
 247                line_aa = AA_ALWAYS;
 248          }
 249       }
 250    }
 251
 252    brw_wm_lookup_iz(line_aa,
 253                     lookup,
 254                     key);
 255
 256
 257    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 258    key->projtex_mask = brw->wm.input_size_masks[4-1] >> (FRAG_ATTRIB_TEX0 - FRAG_ATTRIB_WPOS);
 259
 260    /* _NEW_LIGHT */
 261    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 262
 263    /* _NEW_TEXTURE */
 264    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 265       const struct gl_texture_unit *unit = &ctx->Texture.Unit[i];
 266
 267       if (unit->_ReallyEnabled) {
 268          const struct gl_texture_object *t = unit->_Current;
 269          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 270          if (img->InternalFormat == GL_YCBCR_MESA) {
 271             key->yuvtex_mask |= 1 << i;
 272             if (img->TexFormat->MesaFormat == MESA_FORMAT_YCBCR)
 273                 key->yuvtex_swap_mask |= 1 << i;
 274          }
 275
 276          key->tex_swizzles[i] = t->_Swizzle;
 277       }
 278       else {
 279          key->tex_swizzles[i] = SWIZZLE_NOOP;
 280       }
 281    }
 282
 283    /* Shadow */
 284    key->shadowtex_mask = fp->program.Base.ShadowSamplers;
 285
 286    /* _NEW_BUFFERS */
 287    /*
 288     * Include the draw buffer origin and height so that we can calculate
 289     * fragment position values relative to the bottom left of the drawable,
 290     * from the incoming screen origin relative position we get as part of our
 291     * payload.
 292     *
 293     * We could avoid recompiling by including this as a constant referenced by
 294     * our program, but if we were to do that it would also be nice to handle
 295     * getting that constant updated at batchbuffer submit time (when we
 296     * hold the lock and know where the buffer really is) rather than at emit
 297     * time when we don't hold the lock and are just guessing.  We could also
 298     * just avoid using this as key data if the program doesn't use
 299     * fragment.position.
 300     *
 301     * This pretty much becomes moot with DRI2 and redirected buffers anyway,
 302     * as our origins will always be zero then.
 303     */
 304    if (brw->intel.driDrawable != NULL) {
 305       key->origin_x = brw->intel.driDrawable->x;
 306       key->origin_y = brw->intel.driDrawable->y;
 307       key->drawable_height = brw->intel.driDrawable->h;
 308    }
 309
 310    /* The unique fragment program ID */
 311    key->program_string_id = fp->id;
 312 }
 313
 314
 315 static void brw_prepare_wm_prog(struct brw_context *brw)
 316 {
 317    struct brw_wm_prog_key key;
 318    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 319       brw->fragment_program;
 320
 321    brw_wm_populate_key(brw, &key);
 322
 323    /* Make an early check for the key.
 324     */
 325    dri_bo_unreference(brw->wm.prog_bo);
 326    brw->wm.prog_bo = brw_search_cache(&brw->cache, BRW_WM_PROG,
 327                                       &key, sizeof(key),
 328                                       NULL, 0,
 329                                       &brw->wm.prog_data);
 330    if (brw->wm.prog_bo == NULL)
 331       do_wm_prog(brw, fp, &key);
 332 }
 333
 334
 335 const struct brw_tracked_state brw_wm_prog = {
 336    .dirty = {
 337       .mesa  = (_NEW_COLOR |
 338                 _NEW_DEPTH |
 339                 _NEW_STENCIL |
 340                 _NEW_POLYGON |
 341                 _NEW_LINE |
 342                 _NEW_LIGHT |
 343                 _NEW_BUFFERS |
 344                 _NEW_TEXTURE),
 345       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 346                 BRW_NEW_WM_INPUT_DIMENSIONS |
 347                 BRW_NEW_REDUCED_PRIMITIVE),
 348       .cache = 0
 349    },
 350    .prepare = brw_prepare_wm_prog
 351 };
 352