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    case WM_PIXELXY:
  45       return 0;
  46    case WM_CINTERP:
  47    case WM_WPOSXY:
  48    case WM_DELTAXY:
  49       return 1;
  50    case WM_LINTERP:
  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     * Divide by two because we operate on 16 pixels at a time and require
 109     * two GRF entries for each logical shader register.
 110     */
 111    c->grf_limit = BRW_WM_MAX_GRF / 2;
 112
 113    brw_wm_pass2(c);
 114
 115    /* how many general-purpose registers are used */
 116    c->prog_data.total_grf = c->max_wm_grf;
 117
 118    /* Scratch space is used for register spilling */
 119    if (c->last_scratch) {
 120       c->prog_data.total_scratch = c->last_scratch + 0x40;
 121    }
 122    else {
 123       c->prog_data.total_scratch = 0;
 124    }
 125
 126    /* Emit GEN4 code.
 127     */
 128    brw_wm_emit(c);
 129 }
 130
 131
 132 /**
 133  * All Mesa program -> GPU code generation goes through this function.
 134  * Depending on the instructions used (i.e. flow control instructions)
 135  * we'll use one of two code generators.
 136  */
 137 static void do_wm_prog( struct brw_context *brw,
 138                         struct brw_fragment_program *fp,
 139                         struct brw_wm_prog_key *key)
 140 {
 141    struct brw_wm_compile *c;
 142    const GLuint *program;
 143    GLuint program_size;
 144
 145    c = brw->wm.compile_data;
 146    if (c == NULL) {
 147       brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
 148       c = brw->wm.compile_data;
 149       if (c == NULL) {
 150          /* Ouch - big out of memory problem.  Can't continue
 151           * without triggering a segfault, no way to signal,
 152           * so just return.
 153           */
 154          return;
 155       }
 156    } else {
 157       memset(c, 0, sizeof(*brw->wm.compile_data));
 158    }
 159    memcpy(&c->key, key, sizeof(*key));
 160
 161    c->fp = fp;
 162    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 163
 164    brw_init_compile(brw, &c->func);
 165
 166    /* temporary sanity check assertion */
 167    ASSERT(fp->isGLSL == brw_wm_is_glsl(&c->fp->program));
 168
 169    /*
 170     * Shader which use GLSL features such as flow control are handled
 171     * differently from "simple" shaders.
 172     */
 173    if (fp->isGLSL) {
 174       c->dispatch_width = 8;
 175       brw_wm_glsl_emit(brw, c);
 176    }
 177    else {
 178       c->dispatch_width = 16;
 179       brw_wm_non_glsl_emit(brw, c);
 180    }
 181
 182    if (INTEL_DEBUG & DEBUG_WM)
 183       fprintf(stderr, "\n");
 184
 185    /* get the program
 186     */
 187    program = brw_get_program(&c->func, &program_size);
 188
 189    dri_bo_unreference(brw->wm.prog_bo);
 190    brw->wm.prog_bo = brw_upload_cache( &brw->cache, BRW_WM_PROG,
 191                                        &c->key, sizeof(c->key),
 192                                        NULL, 0,
 193                                        program, program_size,
 194                                        &c->prog_data,
 195                                        &brw->wm.prog_data );
 196 }
 197
 198
 199
 200 static void brw_wm_populate_key( struct brw_context *brw,
 201                                  struct brw_wm_prog_key *key )
 202 {
 203    GLcontext *ctx = &brw->intel.ctx;
 204    /* BRW_NEW_FRAGMENT_PROGRAM */
 205    const struct brw_fragment_program *fp =
 206       (struct brw_fragment_program *)brw->fragment_program;
 207    GLboolean uses_depth = (fp->program.Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) != 0;
 208    GLuint lookup = 0;
 209    GLuint line_aa;
 210    GLuint i;
 211
 212    memset(key, 0, sizeof(*key));
 213
 214    /* Build the index for table lookup
 215     */
 216    /* _NEW_COLOR */
 217    if (fp->program.UsesKill ||
 218        ctx->Color.AlphaEnabled)
 219       lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 220
 221    if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPTH))
 222       lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 223
 224    /* _NEW_DEPTH */
 225    if (ctx->Depth.Test)
 226       lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 227
 228    if (ctx->Depth.Test &&
 229        ctx->Depth.Mask) /* ?? */
 230       lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 231
 232    /* _NEW_STENCIL */
 233    if (ctx->Stencil._Enabled) {
 234       lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 235
 236       if (ctx->Stencil.WriteMask[0] ||
 237           ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 238          lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 239    }
 240
 241    line_aa = AA_NEVER;
 242
 243    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 244    if (ctx->Line.SmoothFlag) {
 245       if (brw->intel.reduced_primitive == GL_LINES) {
 246          line_aa = AA_ALWAYS;
 247       }
 248       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 249          if (ctx->Polygon.FrontMode == GL_LINE) {
 250             line_aa = AA_SOMETIMES;
 251
 252             if (ctx->Polygon.BackMode == GL_LINE ||
 253                 (ctx->Polygon.CullFlag &&
 254                  ctx->Polygon.CullFaceMode == GL_BACK))
 255                line_aa = AA_ALWAYS;
 256          }
 257          else if (ctx->Polygon.BackMode == GL_LINE) {
 258             line_aa = AA_SOMETIMES;
 259
 260             if ((ctx->Polygon.CullFlag &&
 261                  ctx->Polygon.CullFaceMode == GL_FRONT))
 262                line_aa = AA_ALWAYS;
 263          }
 264       }
 265    }
 266
 267    brw_wm_lookup_iz(line_aa,
 268                     lookup,
 269                     uses_depth,
 270                     key);
 271
 272
 273    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 274    key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
 275
 276    /* _NEW_LIGHT */
 277    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 278
 279    /* _NEW_HINT */
 280    key->linear_color = (ctx->Hint.PerspectiveCorrection == GL_FASTEST);
 281
 282    /* _NEW_TEXTURE */
 283    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 284       const struct gl_texture_unit *unit = &ctx->Texture.Unit[i];
 285
 286       if (unit->_ReallyEnabled) {
 287          const struct gl_texture_object *t = unit->_Current;
 288          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 289          if (img->InternalFormat == GL_YCBCR_MESA) {
 290             key->yuvtex_mask |= 1 << i;
 291             if (img->TexFormat == MESA_FORMAT_YCBCR)
 292                 key->yuvtex_swap_mask |= 1 << i;
 293          }
 294
 295          key->tex_swizzles[i] = t->_Swizzle;
 296       }
 297       else {
 298          key->tex_swizzles[i] = SWIZZLE_NOOP;
 299       }
 300    }
 301
 302    /* Shadow */
 303    key->shadowtex_mask = fp->program.Base.ShadowSamplers;
 304
 305    /* _NEW_BUFFERS */
 306    /*
 307     * Include the draw buffer origin and height so that we can calculate
 308     * fragment position values relative to the bottom left of the drawable,
 309     * from the incoming screen origin relative position we get as part of our
 310     * payload.
 311     *
 312     * We could avoid recompiling by including this as a constant referenced by
 313     * our program, but if we were to do that it would also be nice to handle
 314     * getting that constant updated at batchbuffer submit time (when we
 315     * hold the lock and know where the buffer really is) rather than at emit
 316     * time when we don't hold the lock and are just guessing.  We could also
 317     * just avoid using this as key data if the program doesn't use
 318     * fragment.position.
 319     *
 320     * This pretty much becomes moot with DRI2 and redirected buffers anyway,
 321     * as our origins will always be zero then.
 322     */
 323    if (brw->intel.driDrawable != NULL) {
 324       key->origin_x = brw->intel.driDrawable->x;
 325       key->origin_y = brw->intel.driDrawable->y;
 326       key->drawable_height = brw->intel.driDrawable->h;
 327    }
 328
 329    /* CACHE_NEW_VS_PROG */
 330    key->vp_outputs_written = brw->vs.prog_data->outputs_written & DO_SETUP_BITS;
 331
 332    /* The unique fragment program ID */
 333    key->program_string_id = fp->id;
 334 }
 335
 336
 337 static void brw_prepare_wm_prog(struct brw_context *brw)
 338 {
 339    struct brw_wm_prog_key key;
 340    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 341       brw->fragment_program;
 342
 343    brw_wm_populate_key(brw, &key);
 344
 345    /* Make an early check for the key.
 346     */
 347    dri_bo_unreference(brw->wm.prog_bo);
 348    brw->wm.prog_bo = brw_search_cache(&brw->cache, BRW_WM_PROG,
 349                                       &key, sizeof(key),
 350                                       NULL, 0,
 351                                       &brw->wm.prog_data);
 352    if (brw->wm.prog_bo == NULL)
 353       do_wm_prog(brw, fp, &key);
 354 }
 355
 356
 357 const struct brw_tracked_state brw_wm_prog = {
 358    .dirty = {
 359       .mesa  = (_NEW_COLOR |
 360                 _NEW_DEPTH |
 361                 _NEW_HINT |
 362                 _NEW_STENCIL |
 363                 _NEW_POLYGON |
 364                 _NEW_LINE |
 365                 _NEW_LIGHT |
 366                 _NEW_BUFFERS |
 367                 _NEW_TEXTURE),
 368       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 369                 BRW_NEW_WM_INPUT_DIMENSIONS |
 370                 BRW_NEW_REDUCED_PRIMITIVE),
 371       .cache = CACHE_NEW_VS_PROG,
 372    },
 373    .prepare = brw_prepare_wm_prog
 374 };
 375