src/mesa/drivers/dri/r200/r200_tcl.c

   1 /* $XFree86: xc/lib/GL/mesa/src/drv/r200/r200_tcl.c,v 1.2 2002/12/16 16:18:55 dawes Exp $ */
   2 /*
   3 Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
   4
   5 The Weather Channel (TM) funded Tungsten Graphics to develop the
   6 initial release of the Radeon 8500 driver under the XFree86 license.
   7 This notice must be preserved.
   8
   9 Permission is hereby granted, free of charge, to any person obtaining
  10 a copy of this software and associated documentation files (the
  11 "Software"), to deal in the Software without restriction, including
  12 without limitation the rights to use, copy, modify, merge, publish,
  13 distribute, sublicense, and/or sell copies of the Software, and to
  14 permit persons to whom the Software is furnished to do so, subject to
  15 the following conditions:
  16
  17 The above copyright notice and this permission notice (including the
  18 next paragraph) shall be included in all copies or substantial
  19 portions of the Software.
  20
  21 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  22 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  23 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  24 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  25 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  26 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  27 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  28
  29 **************************************************************************/
  30
  31 /*
  32  * Authors:
  33  *   Keith Whitwell <keith@tungstengraphics.com>
  34  */
  35
  36 #include "glheader.h"
  37 #include "imports.h"
  38 #include "mtypes.h"
  39 #include "enums.h"
  40 #include "colormac.h"
  41 #include "light.h"
  42
  43 #include "array_cache/acache.h"
  44 #include "tnl/tnl.h"
  45 #include "tnl/t_pipeline.h"
  46
  47 #include "r200_context.h"
  48 #include "r200_state.h"
  49 #include "r200_ioctl.h"
  50 #include "r200_tex.h"
  51 #include "r200_tcl.h"
  52 #include "r200_swtcl.h"
  53 #include "r200_maos.h"
  54
  55
  56
  57 #define HAVE_POINTS      1
  58 #define HAVE_LINES       1
  59 #define HAVE_LINE_LOOP   0
  60 #define HAVE_LINE_STRIPS 1
  61 #define HAVE_TRIANGLES   1
  62 #define HAVE_TRI_STRIPS  1
  63 #define HAVE_TRI_STRIP_1 0
  64 #define HAVE_TRI_FANS    1
  65 #define HAVE_QUADS       1
  66 #define HAVE_QUAD_STRIPS 1
  67 #define HAVE_POLYGONS    1
  68 #define HAVE_ELTS        1
  69
  70
  71 #define HW_POINTS           ((ctx->_TriangleCaps & DD_POINT_SIZE) ? \
  72                                 R200_VF_PRIM_POINT_SPRITES : R200_VF_PRIM_POINTS)
  73 #define HW_LINES            R200_VF_PRIM_LINES
  74 #define HW_LINE_LOOP        0
  75 #define HW_LINE_STRIP       R200_VF_PRIM_LINE_STRIP
  76 #define HW_TRIANGLES        R200_VF_PRIM_TRIANGLES
  77 #define HW_TRIANGLE_STRIP_0 R200_VF_PRIM_TRIANGLE_STRIP
  78 #define HW_TRIANGLE_STRIP_1 0
  79 #define HW_TRIANGLE_FAN     R200_VF_PRIM_TRIANGLE_FAN
  80 #define HW_QUADS            R200_VF_PRIM_QUADS
  81 #define HW_QUAD_STRIP       R200_VF_PRIM_QUAD_STRIP
  82 #define HW_POLYGON          R200_VF_PRIM_POLYGON
  83
  84
  85 static GLboolean discrete_prim[0x10] = {
  86    0,                           /* 0 none */
  87    1,                           /* 1 points */
  88    1,                           /* 2 lines */
  89    0,                           /* 3 line_strip */
  90    1,                           /* 4 tri_list */
  91    0,                           /* 5 tri_fan */
  92    0,                           /* 6 tri_strip */
  93    0,                           /* 7 tri_w_flags */
  94    1,                           /* 8 rect list (unused) */
  95    1,                           /* 9 3vert point */
  96    1,                           /* a 3vert line */
  97    0,                           /* b point sprite */
  98    0,                           /* c line loop */
  99    1,                           /* d quads */
 100    0,                           /* e quad strip */
 101    0,                           /* f polygon */
 102 };
 103
 104
 105 #define LOCAL_VARS r200ContextPtr rmesa = R200_CONTEXT(ctx)
 106 #define ELT_TYPE  GLushort
 107
 108 #define ELT_INIT(prim, hw_prim) \
 109    r200TclPrimitive( ctx, prim, hw_prim | R200_VF_PRIM_WALK_IND )
 110
 111 #define GET_MESA_ELTS() rmesa->tcl.Elts
 112
 113
 114 /* Don't really know how many elts will fit in what's left of cmdbuf,
 115  * as there is state to emit, etc:
 116  */
 117
 118 /* Testing on isosurf shows a maximum around here.  Don't know if it's
 119  * the card or driver or kernel module that is causing the behaviour.
 120  */
 121 #define GET_MAX_HW_ELTS() 300
 122
 123 #define RESET_STIPPLE() do {                    \
 124    R200_STATECHANGE( rmesa, lin );              \
 125    r200EmitState( rmesa );                      \
 126 } while (0)
 127
 128 #define AUTO_STIPPLE( mode )  do {              \
 129    R200_STATECHANGE( rmesa, lin );              \
 130    if (mode)                                    \
 131       rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] |= \
 132          R200_LINE_PATTERN_AUTO_RESET;  \
 133    else                                         \
 134       rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &= \
 135          ~R200_LINE_PATTERN_AUTO_RESET; \
 136    r200EmitState( rmesa );                      \
 137 } while (0)
 138
 139
 140 #define ALLOC_ELTS(nr)  r200AllocElts( rmesa, nr )
 141
 142 static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr )
 143 {
 144    if (rmesa->dma.flush == r200FlushElts &&
 145        rmesa->store.cmd_used + nr*2 < R200_CMD_BUF_SZ) {
 146
 147       GLushort *dest = (GLushort *)(rmesa->store.cmd_buf +
 148                                     rmesa->store.cmd_used);
 149
 150       rmesa->store.cmd_used += nr*2;
 151
 152       return dest;
 153    }
 154    else {
 155       if (rmesa->dma.flush)
 156          rmesa->dma.flush( rmesa );
 157
 158       r200EnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
 159                              rmesa->hw.max_state_size + ELTS_BUFSZ(nr) );
 160
 161       r200EmitAOS( rmesa,
 162                    rmesa->tcl.aos_components,
 163                    rmesa->tcl.nr_aos_components, 0 );
 164
 165       return r200AllocEltsOpenEnded( rmesa, rmesa->tcl.hw_primitive, nr );
 166    }
 167 }
 168
 169
 170 #define CLOSE_ELTS()                            \
 171 do {                                            \
 172    if (0) R200_NEWPRIM( rmesa );                \
 173 }                                               \
 174 while (0)
 175
 176
 177 /* TODO: Try to extend existing primitive if both are identical,
 178  * discrete and there are no intervening state changes.  (Somewhat
 179  * duplicates changes to DrawArrays code)
 180  */
 181 static void r200EmitPrim( GLcontext *ctx,
 182                           GLenum prim,
 183                           GLuint hwprim,
 184                           GLuint start,
 185                           GLuint count)
 186 {
 187    r200ContextPtr rmesa = R200_CONTEXT( ctx );
 188    r200TclPrimitive( ctx, prim, hwprim );
 189
 190    r200EnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
 191                           rmesa->hw.max_state_size + VBUF_BUFSZ );
 192
 193    r200EmitAOS( rmesa,
 194                   rmesa->tcl.aos_components,
 195                   rmesa->tcl.nr_aos_components,
 196                   start );
 197
 198    /* Why couldn't this packet have taken an offset param?
 199     */
 200    r200EmitVbufPrim( rmesa,
 201                      rmesa->tcl.hw_primitive,
 202                      count - start );
 203 }
 204
 205 #define EMIT_PRIM(ctx, prim, hwprim, start, count) do {         \
 206    r200EmitPrim( ctx, prim, hwprim, start, count );             \
 207    (void) rmesa; } while (0)
 208
 209 /* Try & join small primitives
 210  */
 211 #if 0
 212 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) 0
 213 #else
 214 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM )                    \
 215   ((NR) < 20 ||                                                 \
 216    ((NR) < 40 &&                                                \
 217     rmesa->tcl.hw_primitive == (PRIM|                           \
 218                             R200_VF_TCL_OUTPUT_VTX_ENABLE|      \
 219                                 R200_VF_PRIM_WALK_IND)))
 220 #endif
 221
 222 #ifdef MESA_BIG_ENDIAN
 223 /* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
 224 #define EMIT_ELT(dest, offset, x) do {                          \
 225         int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 );     \
 226         GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 );    \
 227         (des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x);     \
 228         (void)rmesa; } while (0)
 229 #else
 230 #define EMIT_ELT(dest, offset, x) do {                          \
 231         (dest)[offset] = (GLushort) (x);                        \
 232         (void)rmesa; } while (0)
 233 #endif
 234
 235 #define EMIT_TWO_ELTS(dest, offset, x, y)  *(GLuint *)((dest)+offset) = ((y)<<16)|(x);
 236
 237
 238
 239 #define TAG(x) tcl_##x
 240 #include "tnl_dd/t_dd_dmatmp2.h"
 241
 242 /**********************************************************************/
 243 /*                          External entrypoints                     */
 244 /**********************************************************************/
 245
 246 void r200EmitPrimitive( GLcontext *ctx,
 247                           GLuint first,
 248                           GLuint last,
 249                           GLuint flags )
 250 {
 251    tcl_render_tab_verts[flags&PRIM_MODE_MASK]( ctx, first, last, flags );
 252 }
 253
 254 void r200EmitEltPrimitive( GLcontext *ctx,
 255                              GLuint first,
 256                              GLuint last,
 257                              GLuint flags )
 258 {
 259    tcl_render_tab_elts[flags&PRIM_MODE_MASK]( ctx, first, last, flags );
 260 }
 261
 262 void r200TclPrimitive( GLcontext *ctx,
 263                          GLenum prim,
 264                          int hw_prim )
 265 {
 266    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 267    GLuint newprim = hw_prim | R200_VF_TCL_OUTPUT_VTX_ENABLE;
 268
 269    if (newprim != rmesa->tcl.hw_primitive ||
 270        !discrete_prim[hw_prim&0xf]) {
 271       R200_NEWPRIM( rmesa );
 272       rmesa->tcl.hw_primitive = newprim;
 273    }
 274 }
 275
 276
 277 /**********************************************************************/
 278 /*             Fog blend factor computation for hw tcl                */
 279 /*             same calculation used as in t_vb_fog.c                 */
 280 /**********************************************************************/
 281
 282 #define FOG_EXP_TABLE_SIZE 256
 283 #define FOG_MAX (10.0)
 284 #define EXP_FOG_MAX .0006595
 285 #define FOG_INCR (FOG_MAX/FOG_EXP_TABLE_SIZE)
 286 static GLfloat exp_table[FOG_EXP_TABLE_SIZE];
 287
 288 #if 1
 289 #define NEG_EXP( result, narg )                                         \
 290 do {                                                                    \
 291    GLfloat f = (GLfloat) (narg * (1.0/FOG_INCR));                       \
 292    GLint k = (GLint) f;                                                 \
 293    if (k > FOG_EXP_TABLE_SIZE-2)                                        \
 294       result = (GLfloat) EXP_FOG_MAX;                                   \
 295    else                                                                 \
 296       result = exp_table[k] + (f-k)*(exp_table[k+1]-exp_table[k]);      \
 297 } while (0)
 298 #else
 299 #define NEG_EXP( result, narg )                                 \
 300 do {                                                            \
 301    result = exp(-narg);                                         \
 302 } while (0)
 303 #endif
 304
 305
 306 /**
 307  * Initialize the exp_table[] lookup table for approximating exp().
 308  */
 309 void
 310 r200InitStaticFogData( void )
 311 {
 312    GLfloat f = 0.0F;
 313    GLint i = 0;
 314    for ( ; i < FOG_EXP_TABLE_SIZE ; i++, f += FOG_INCR) {
 315       exp_table[i] = (GLfloat) exp(-f);
 316    }
 317 }
 318
 319
 320 /**
 321  * Compute per-vertex fog blend factors from fog coordinates by
 322  * evaluating the GL_LINEAR, GL_EXP or GL_EXP2 fog function.
 323  * Fog coordinates are distances from the eye (typically between the
 324  * near and far clip plane distances).
 325  * Note the fog (eye Z) coords may be negative so we use ABS(z) below.
 326  * Fog blend factors are in the range [0,1].
 327  */
 328 float
 329 r200ComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord )
 330 {
 331    GLfloat end  = ctx->Fog.End;
 332    GLfloat d, temp;
 333    const GLfloat z = FABSF(fogcoord);
 334
 335    switch (ctx->Fog.Mode) {
 336    case GL_LINEAR:
 337       if (ctx->Fog.Start == ctx->Fog.End)
 338          d = 1.0F;
 339       else
 340          d = 1.0F / (ctx->Fog.End - ctx->Fog.Start);
 341       temp = (end - z) * d;
 342       return CLAMP(temp, 0.0F, 1.0F);
 343       break;
 344    case GL_EXP:
 345       d = ctx->Fog.Density;
 346       NEG_EXP( temp, d * z );
 347       return temp;
 348       break;
 349    case GL_EXP2:
 350       d = ctx->Fog.Density*ctx->Fog.Density;
 351       NEG_EXP( temp, d * z * z );
 352       return temp;
 353       break;
 354    default:
 355       _mesa_problem(ctx, "Bad fog mode in make_fog_coord");
 356       return 0;
 357    }
 358 }
 359
 360
 361 /**********************************************************************/
 362 /*                          Render pipeline stage                     */
 363 /**********************************************************************/
 364
 365
 366 /* TCL render.
 367  */
 368 static GLboolean r200_run_tcl_render( GLcontext *ctx,
 369                                       struct tnl_pipeline_stage *stage )
 370 {
 371    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 372    TNLcontext *tnl = TNL_CONTEXT(ctx);
 373    struct vertex_buffer *VB = &tnl->vb;
 374    GLuint inputs = VERT_BIT_POS | VERT_BIT_COLOR0;
 375    GLuint i;
 376
 377    /* TODO: separate this from the swtnl pipeline
 378     */
 379    if (rmesa->TclFallback)
 380       return GL_TRUE;   /* fallback to software t&l */
 381
 382    if (R200_DEBUG & DEBUG_PRIMS)
 383       fprintf(stderr, "%s\n", __FUNCTION__);
 384
 385    if (VB->Count == 0)
 386       return GL_FALSE;
 387
 388    /* Validate state:
 389     */
 390    if (rmesa->NewGLState)
 391       r200ValidateState( ctx );
 392
 393    /* NOTE: inputs != tnl->render_inputs - these are the untransformed
 394     * inputs.
 395     */
 396    if (ctx->Light.Enabled) {
 397       inputs |= VERT_BIT_NORMAL;
 398       if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
 399          inputs |= VERT_BIT_COLOR1;
 400       }
 401    }
 402
 403    if ( (ctx->Fog.FogCoordinateSource == GL_FOG_COORD) && ctx->Fog.Enabled ) {
 404       inputs |= VERT_BIT_FOG;
 405    }
 406
 407    for (i = 0 ; i < ctx->Const.MaxTextureUnits; i++) {
 408       if (ctx->Texture.Unit[i]._ReallyEnabled) {
 409          if (rmesa->TexGenNeedNormals[i]) {
 410             inputs |= VERT_BIT_NORMAL;
 411          }
 412          inputs |= VERT_BIT_TEX(i);
 413       }
 414    }
 415
 416    /* Do the actual work:
 417     */
 418    r200ReleaseArrays( ctx, ~0 /* stage->changed_inputs */ );
 419    r200EmitArrays( ctx, inputs );
 420
 421    rmesa->tcl.Elts = VB->Elts;
 422
 423    for (i = 0 ; i < VB->PrimitiveCount ; i++)
 424    {
 425       GLuint prim = VB->Primitive[i].mode;
 426       GLuint start = VB->Primitive[i].start;
 427       GLuint length = VB->Primitive[i].count;
 428
 429       if (!length)
 430          continue;
 431
 432       if (rmesa->tcl.Elts)
 433          r200EmitEltPrimitive( ctx, start, start+length, prim );
 434       else
 435          r200EmitPrimitive( ctx, start, start+length, prim );
 436    }
 437
 438    return GL_FALSE;             /* finished the pipe */
 439 }
 440
 441
 442
 443 /* Initial state for tcl stage.
 444  */
 445 const struct tnl_pipeline_stage _r200_tcl_stage =
 446 {
 447    "r200 render",
 448    NULL,                        /*  private */
 449    NULL,
 450    NULL,
 451    NULL,
 452    r200_run_tcl_render  /* run */
 453 };
 454
 455
 456
 457 /**********************************************************************/
 458 /*                 Validate state at pipeline start                   */
 459 /**********************************************************************/
 460
 461
 462 /*-----------------------------------------------------------------------
 463  * Manage TCL fallbacks
 464  */
 465
 466
 467 static void transition_to_swtnl( GLcontext *ctx )
 468 {
 469    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 470    TNLcontext *tnl = TNL_CONTEXT(ctx);
 471
 472    R200_NEWPRIM( rmesa );
 473
 474    r200ChooseVertexState( ctx );
 475    r200ChooseRenderState( ctx );
 476
 477    _mesa_validate_all_lighting_tables( ctx );
 478
 479    tnl->Driver.NotifyMaterialChange =
 480       _mesa_validate_all_lighting_tables;
 481
 482    r200ReleaseArrays( ctx, ~0 );
 483
 484    /* Still using the D3D based hardware-rasterizer from the radeon;
 485     * need to put the card into D3D mode to make it work:
 486     */
 487    R200_STATECHANGE( rmesa, vap );
 488    rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] &= ~R200_VAP_TCL_ENABLE;
 489 }
 490
 491 static void transition_to_hwtnl( GLcontext *ctx )
 492 {
 493    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 494    TNLcontext *tnl = TNL_CONTEXT(ctx);
 495
 496    _tnl_need_projected_coords( ctx, GL_FALSE );
 497
 498    r200UpdateMaterial( ctx );
 499
 500    tnl->Driver.NotifyMaterialChange = r200UpdateMaterial;
 501
 502    if ( rmesa->dma.flush )
 503       rmesa->dma.flush( rmesa );
 504
 505    rmesa->dma.flush = NULL;
 506
 507    if (rmesa->swtcl.indexed_verts.buf)
 508       r200ReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts,
 509                               __FUNCTION__ );
 510
 511    R200_STATECHANGE( rmesa, vap );
 512    rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] |= R200_VAP_TCL_ENABLE;
 513    rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] &= ~R200_VAP_FORCE_W_TO_ONE;
 514
 515    if ( ((rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] & R200_FOG_USE_MASK)
 516       == R200_FOG_USE_SPEC_ALPHA) &&
 517       (ctx->Fog.FogCoordinateSource == GL_FOG_COORD )) {
 518       R200_STATECHANGE( rmesa, ctx );
 519       rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] &= ~R200_FOG_USE_MASK;
 520       rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] |= R200_FOG_USE_VTX_FOG;
 521    }
 522
 523    R200_STATECHANGE( rmesa, vte );
 524    rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] &= ~(R200_VTX_XY_FMT|R200_VTX_Z_FMT);
 525    rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] |= R200_VTX_W0_FMT;
 526
 527    if (R200_DEBUG & DEBUG_FALLBACKS)
 528       fprintf(stderr, "R200 end tcl fallback\n");
 529 }
 530
 531
 532 static char *fallbackStrings[] = {
 533    "Rasterization fallback",
 534    "Unfilled triangles",
 535    "Twosided lighting, differing materials",
 536    "Materials in VB (maybe between begin/end)",
 537    "Texgen unit 0",
 538    "Texgen unit 1",
 539    "Texgen unit 2",
 540    "Texgen unit 3",
 541    "Texgen unit 4",
 542    "Texgen unit 5",
 543    "User disable",
 544    "Bitmap as points",
 545    "Vertex program"
 546 };
 547
 548
 549 static char *getFallbackString(GLuint bit)
 550 {
 551    int i = 0;
 552    while (bit > 1) {
 553       i++;
 554       bit >>= 1;
 555    }
 556    return fallbackStrings[i];
 557 }
 558
 559
 560
 561 void r200TclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 562 {
 563    r200ContextPtr rmesa = R200_CONTEXT(ctx);
 564    GLuint oldfallback = rmesa->TclFallback;
 565
 566    if (mode) {
 567       rmesa->TclFallback |= bit;
 568       if (oldfallback == 0) {
 569          if (R200_DEBUG & DEBUG_FALLBACKS)
 570             fprintf(stderr, "R200 begin tcl fallback %s\n",
 571                     getFallbackString( bit ));
 572          transition_to_swtnl( ctx );
 573       }
 574    }
 575    else {
 576       rmesa->TclFallback &= ~bit;
 577       if (oldfallback == bit) {
 578          if (R200_DEBUG & DEBUG_FALLBACKS)
 579             fprintf(stderr, "R200 end tcl fallback %s\n",
 580                     getFallbackString( bit ));
 581          transition_to_hwtnl( ctx );
 582       }
 583    }
 584 }