src/mesa/drivers/dri/radeon/radeon_tcl.c

   1 /* $XFree86$ */
   2 /**************************************************************************
   3
   4 Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
   5                      Tungsten Graphics Inc., Austin, Texas.
   6
   7 All Rights Reserved.
   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 "light.h"
  39 #include "mtypes.h"
  40 #include "enums.h"
  41
  42 #include "array_cache/acache.h"
  43 #include "tnl/tnl.h"
  44 #include "tnl/t_pipeline.h"
  45
  46 #include "radeon_context.h"
  47 #include "radeon_state.h"
  48 #include "radeon_ioctl.h"
  49 #include "radeon_tex.h"
  50 #include "radeon_tcl.h"
  51 #include "radeon_swtcl.h"
  52 #include "radeon_maos.h"
  53
  54
  55
  56 /*
  57  * Render unclipped vertex buffers by emitting vertices directly to
  58  * dma buffers.  Use strip/fan hardware primitives where possible.
  59  * Try to simulate missing primitives with indexed vertices.
  60  */
  61 #define HAVE_POINTS      1
  62 #define HAVE_LINES       1
  63 #define HAVE_LINE_LOOP   0
  64 #define HAVE_LINE_STRIPS 1
  65 #define HAVE_TRIANGLES   1
  66 #define HAVE_TRI_STRIPS  1
  67 #define HAVE_TRI_STRIP_1 0
  68 #define HAVE_TRI_FANS    1
  69 #define HAVE_QUADS       0
  70 #define HAVE_QUAD_STRIPS 0
  71 #define HAVE_POLYGONS    1
  72 #define HAVE_ELTS        1
  73
  74
  75 #define HW_POINTS           RADEON_CP_VC_CNTL_PRIM_TYPE_POINT
  76 #define HW_LINES            RADEON_CP_VC_CNTL_PRIM_TYPE_LINE
  77 #define HW_LINE_LOOP        0
  78 #define HW_LINE_STRIP       RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP
  79 #define HW_TRIANGLES        RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST
  80 #define HW_TRIANGLE_STRIP_0 RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP
  81 #define HW_TRIANGLE_STRIP_1 0
  82 #define HW_TRIANGLE_FAN     RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN
  83 #define HW_QUADS            0
  84 #define HW_QUAD_STRIP       0
  85 #define HW_POLYGON          RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN
  86
  87
  88 static GLboolean discrete_prim[0x10] = {
  89    0,                           /* 0 none */
  90    1,                           /* 1 points */
  91    1,                           /* 2 lines */
  92    0,                           /* 3 line_strip */
  93    1,                           /* 4 tri_list */
  94    0,                           /* 5 tri_fan */
  95    0,                           /* 6 tri_type2 */
  96    1,                           /* 7 rect list (unused) */
  97    1,                           /* 8 3vert point */
  98    1,                           /* 9 3vert line */
  99    0,
 100    0,
 101    0,
 102    0,
 103    0,
 104    0,
 105 };
 106
 107
 108 #define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx); (void)rmesa
 109 #define ELT_TYPE  GLushort
 110
 111 #define ELT_INIT(prim, hw_prim) \
 112    radeonTclPrimitive( ctx, prim, hw_prim | RADEON_CP_VC_CNTL_PRIM_WALK_IND )
 113
 114 #define GET_MESA_ELTS() rmesa->tcl.Elts
 115
 116
 117 /* Don't really know how many elts will fit in what's left of cmdbuf,
 118  * as there is state to emit, etc:
 119  */
 120
 121 /* Testing on isosurf shows a maximum around here.  Don't know if it's
 122  * the card or driver or kernel module that is causing the behaviour.
 123  */
 124 #define GET_MAX_HW_ELTS() 300
 125
 126
 127 #define RESET_STIPPLE() do {                    \
 128    RADEON_STATECHANGE( rmesa, lin );            \
 129    radeonEmitState( rmesa );                    \
 130 } while (0)
 131
 132 #define AUTO_STIPPLE( mode )  do {              \
 133    RADEON_STATECHANGE( rmesa, lin );            \
 134    if (mode)                                    \
 135       rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] |= \
 136          RADEON_LINE_PATTERN_AUTO_RESET;        \
 137    else                                         \
 138       rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &= \
 139          ~RADEON_LINE_PATTERN_AUTO_RESET;       \
 140    radeonEmitState( rmesa );                    \
 141 } while (0)
 142
 143
 144
 145 #define ALLOC_ELTS(nr)  radeonAllocElts( rmesa, nr )
 146
 147 static GLushort *radeonAllocElts( radeonContextPtr rmesa, GLuint nr )
 148 {
 149    if (rmesa->dma.flush)
 150       rmesa->dma.flush( rmesa );
 151
 152    radeonEnsureCmdBufSpace(rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
 153                            rmesa->hw.max_state_size + ELTS_BUFSZ(nr));
 154
 155    radeonEmitAOS( rmesa,
 156                 rmesa->tcl.aos_components,
 157                 rmesa->tcl.nr_aos_components, 0 );
 158
 159    return radeonAllocEltsOpenEnded( rmesa,
 160                                     rmesa->tcl.vertex_format,
 161                                     rmesa->tcl.hw_primitive, nr );
 162 }
 163
 164 #define CLOSE_ELTS()  RADEON_NEWPRIM( rmesa )
 165
 166
 167
 168 /* TODO: Try to extend existing primitive if both are identical,
 169  * discrete and there are no intervening state changes.  (Somewhat
 170  * duplicates changes to DrawArrays code)
 171  */
 172 static void radeonEmitPrim( GLcontext *ctx,
 173                        GLenum prim,
 174                        GLuint hwprim,
 175                        GLuint start,
 176                        GLuint count)
 177 {
 178    radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
 179    radeonTclPrimitive( ctx, prim, hwprim );
 180
 181    radeonEnsureCmdBufSpace( rmesa, AOS_BUFSZ(rmesa->tcl.nr_aos_components) +
 182                             rmesa->hw.max_state_size + VBUF_BUFSZ );
 183
 184    radeonEmitAOS( rmesa,
 185                   rmesa->tcl.aos_components,
 186                   rmesa->tcl.nr_aos_components,
 187                   start );
 188
 189    /* Why couldn't this packet have taken an offset param?
 190     */
 191    radeonEmitVbufPrim( rmesa,
 192                        rmesa->tcl.vertex_format,
 193                        rmesa->tcl.hw_primitive,
 194                        count - start );
 195 }
 196
 197 #define EMIT_PRIM( ctx, prim, hwprim, start, count ) do {       \
 198    radeonEmitPrim( ctx, prim, hwprim, start, count );           \
 199    (void) rmesa; } while (0)
 200
 201 /* Try & join small primitives
 202  */
 203 #if 0
 204 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) 0
 205 #else
 206 #define PREFER_DISCRETE_ELT_PRIM( NR, PRIM )                    \
 207   ((NR) < 20 ||                                                 \
 208    ((NR) < 40 &&                                                \
 209     rmesa->tcl.hw_primitive == (PRIM|                           \
 210                             RADEON_CP_VC_CNTL_PRIM_WALK_IND|    \
 211                             RADEON_CP_VC_CNTL_TCL_ENABLE)))
 212 #endif
 213
 214 #ifdef MESA_BIG_ENDIAN
 215 /* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
 216 #define EMIT_ELT(dest, offset, x) do {                          \
 217         int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 );     \
 218         GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 );    \
 219         (des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x);     \
 220         (void)rmesa; } while (0)
 221 #else
 222 #define EMIT_ELT(dest, offset, x) do {                          \
 223         (dest)[offset] = (GLushort) (x);                        \
 224         (void)rmesa; } while (0)
 225 #endif
 226
 227 #define EMIT_TWO_ELTS(dest, offset, x, y)  *(GLuint *)(dest+offset) = ((y)<<16)|(x);
 228
 229
 230
 231 #define TAG(x) tcl_##x
 232 #include "tnl_dd/t_dd_dmatmp2.h"
 233
 234 /**********************************************************************/
 235 /*                          External entrypoints                     */
 236 /**********************************************************************/
 237
 238 void radeonEmitPrimitive( GLcontext *ctx,
 239                           GLuint first,
 240                           GLuint last,
 241                           GLuint flags )
 242 {
 243    tcl_render_tab_verts[flags&PRIM_MODE_MASK]( ctx, first, last, flags );
 244 }
 245
 246 void radeonEmitEltPrimitive( GLcontext *ctx,
 247                              GLuint first,
 248                              GLuint last,
 249                              GLuint flags )
 250 {
 251    tcl_render_tab_elts[flags&PRIM_MODE_MASK]( ctx, first, last, flags );
 252 }
 253
 254 void radeonTclPrimitive( GLcontext *ctx,
 255                          GLenum prim,
 256                          int hw_prim )
 257 {
 258    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 259    GLuint se_cntl;
 260    GLuint newprim = hw_prim | RADEON_CP_VC_CNTL_TCL_ENABLE;
 261
 262    if (newprim != rmesa->tcl.hw_primitive ||
 263        !discrete_prim[hw_prim&0xf]) {
 264       RADEON_NEWPRIM( rmesa );
 265       rmesa->tcl.hw_primitive = newprim;
 266    }
 267
 268    se_cntl = rmesa->hw.set.cmd[SET_SE_CNTL];
 269    se_cntl &= ~RADEON_FLAT_SHADE_VTX_LAST;
 270
 271    if (prim == GL_POLYGON && (ctx->_TriangleCaps & DD_FLATSHADE))
 272       se_cntl |= RADEON_FLAT_SHADE_VTX_0;
 273    else
 274       se_cntl |= RADEON_FLAT_SHADE_VTX_LAST;
 275
 276    if (se_cntl != rmesa->hw.set.cmd[SET_SE_CNTL]) {
 277       RADEON_STATECHANGE( rmesa, set );
 278       rmesa->hw.set.cmd[SET_SE_CNTL] = se_cntl;
 279    }
 280 }
 281
 282
 283 /**********************************************************************/
 284 /*                          Render pipeline stage                     */
 285 /**********************************************************************/
 286
 287
 288 /* TCL render.
 289  */
 290 static GLboolean radeon_run_tcl_render( GLcontext *ctx,
 291                                         struct tnl_pipeline_stage *stage )
 292 {
 293    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 294    TNLcontext *tnl = TNL_CONTEXT(ctx);
 295    struct vertex_buffer *VB = &tnl->vb;
 296    GLuint i;
 297
 298    /* TODO: separate this from the swtnl pipeline
 299     */
 300    if (rmesa->TclFallback)
 301       return GL_TRUE;   /* fallback to software t&l */
 302
 303    if (VB->Count == 0)
 304       return GL_FALSE;
 305
 306    radeonReleaseArrays( ctx, stage->changed_inputs );
 307    radeonEmitArrays( ctx, stage->inputs );
 308
 309    rmesa->tcl.Elts = VB->Elts;
 310
 311    for (i = 0 ; i < VB->PrimitiveCount ; i++)
 312    {
 313       GLuint prim = VB->Primitive[i].mode;
 314       GLuint start = VB->Primitive[i].start;
 315       GLuint length = VB->Primitive[i].count;
 316
 317       if (!length)
 318          continue;
 319
 320       if (rmesa->tcl.Elts)
 321          radeonEmitEltPrimitive( ctx, start, start+length, prim );
 322       else
 323          radeonEmitPrimitive( ctx, start, start+length, prim );
 324    }
 325
 326    return GL_FALSE;             /* finished the pipe */
 327 }
 328
 329
 330
 331 static void radeon_check_tcl_render( GLcontext *ctx,
 332                                      struct tnl_pipeline_stage *stage )
 333 {
 334    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 335    GLuint inputs = VERT_BIT_POS;
 336
 337    if (ctx->RenderMode == GL_RENDER) {
 338       /* Make all this event-driven:
 339        */
 340       if (ctx->Light.Enabled) {
 341          inputs |= VERT_BIT_NORMAL;
 342
 343          if (1 || ctx->Light.ColorMaterialEnabled) {
 344             inputs |= VERT_BIT_COLOR0;
 345          }
 346       }
 347       else {
 348          inputs |= VERT_BIT_COLOR0;
 349
 350          if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
 351             inputs |= VERT_BIT_COLOR1;
 352          }
 353       }
 354
 355       if (ctx->Texture.Unit[0]._ReallyEnabled) {
 356          if (ctx->Texture.Unit[0].TexGenEnabled) {
 357             if (rmesa->TexGenNeedNormals[0]) {
 358                inputs |= VERT_BIT_NORMAL;
 359             }
 360          } else {
 361             inputs |= VERT_BIT_TEX0;
 362          }
 363       }
 364
 365       if (ctx->Texture.Unit[1]._ReallyEnabled) {
 366          if (ctx->Texture.Unit[1].TexGenEnabled) {
 367             if (rmesa->TexGenNeedNormals[1]) {
 368                inputs |= VERT_BIT_NORMAL;
 369             }
 370          } else {
 371             inputs |= VERT_BIT_TEX1;
 372          }
 373       }
 374
 375       stage->inputs = inputs;
 376       stage->active = 1;
 377    }
 378    else
 379       stage->active = 0;
 380 }
 381
 382 static void radeon_init_tcl_render( GLcontext *ctx,
 383                                     struct tnl_pipeline_stage *stage )
 384 {
 385    stage->check = radeon_check_tcl_render;
 386    stage->check( ctx, stage );
 387 }
 388
 389 static void dtr( struct tnl_pipeline_stage *stage )
 390 {
 391    (void)stage;
 392 }
 393
 394
 395 /* Initial state for tcl stage.
 396  */
 397 const struct tnl_pipeline_stage _radeon_tcl_stage =
 398 {
 399    "radeon render",
 400    (_DD_NEW_SEPARATE_SPECULAR |
 401     _NEW_LIGHT|
 402     _NEW_TEXTURE|
 403     _NEW_FOG|
 404     _NEW_RENDERMODE),           /* re-check (new inputs) */
 405    0,                           /* re-run (always runs) */
 406    GL_TRUE,                     /* active */
 407    0, 0,                        /* inputs (set in check_render), outputs */
 408    0, 0,                        /* changed_inputs, private */
 409    dtr,                         /* destructor */
 410    radeon_init_tcl_render,      /* check - initially set to alloc data */
 411    radeon_run_tcl_render        /* run */
 412 };
 413
 414
 415
 416 /**********************************************************************/
 417 /*                 Validate state at pipeline start                   */
 418 /**********************************************************************/
 419
 420
 421 /*-----------------------------------------------------------------------
 422  * Manage TCL fallbacks
 423  */
 424
 425
 426 static void transition_to_swtnl( GLcontext *ctx )
 427 {
 428    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 429    TNLcontext *tnl = TNL_CONTEXT(ctx);
 430    GLuint se_cntl;
 431
 432    RADEON_NEWPRIM( rmesa );
 433    rmesa->swtcl.vertex_format = 0;
 434
 435    radeonChooseVertexState( ctx );
 436    radeonChooseRenderState( ctx );
 437
 438    _mesa_validate_all_lighting_tables( ctx );
 439
 440    tnl->Driver.NotifyMaterialChange =
 441       _mesa_validate_all_lighting_tables;
 442
 443    radeonReleaseArrays( ctx, ~0 );
 444
 445    se_cntl = rmesa->hw.set.cmd[SET_SE_CNTL];
 446    se_cntl |= RADEON_FLAT_SHADE_VTX_LAST;
 447
 448    if (se_cntl != rmesa->hw.set.cmd[SET_SE_CNTL]) {
 449       RADEON_STATECHANGE( rmesa, set );
 450       rmesa->hw.set.cmd[SET_SE_CNTL] = se_cntl;
 451    }
 452 }
 453
 454
 455 static void transition_to_hwtnl( GLcontext *ctx )
 456 {
 457    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 458    TNLcontext *tnl = TNL_CONTEXT(ctx);
 459    GLuint se_coord_fmt = (RADEON_VTX_W0_IS_NOT_1_OVER_W0 |
 460                           RADEON_TEX1_W_ROUTING_USE_Q1);
 461
 462    if ( se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT] ) {
 463       RADEON_STATECHANGE( rmesa, set );
 464       rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
 465       _tnl_need_projected_coords( ctx, GL_FALSE );
 466    }
 467
 468    radeonUpdateMaterial( ctx );
 469
 470    tnl->Driver.NotifyMaterialChange = radeonUpdateMaterial;
 471
 472    if ( rmesa->dma.flush )
 473       rmesa->dma.flush( rmesa );
 474
 475    rmesa->dma.flush = 0;
 476    rmesa->swtcl.vertex_format = 0;
 477
 478    if (rmesa->swtcl.indexed_verts.buf)
 479       radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts,
 480                               __FUNCTION__ );
 481
 482    if (RADEON_DEBUG & DEBUG_FALLBACKS)
 483       fprintf(stderr, "Radeon end tcl fallback\n");
 484 }
 485
 486 static char *fallbackStrings[] = {
 487    "Rasterization fallback",
 488    "Unfilled triangles",
 489    "Twosided lighting, differing materials",
 490    "Materials in VB (maybe between begin/end)",
 491    "Texgen unit 0",
 492    "Texgen unit 1",
 493    "Texgen unit 2",
 494    "User disable"
 495 };
 496
 497
 498 static char *getFallbackString(GLuint bit)
 499 {
 500    int i = 0;
 501    while (bit > 1) {
 502       i++;
 503       bit >>= 1;
 504    }
 505    return fallbackStrings[i];
 506 }
 507
 508
 509
 510 void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 511 {
 512    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 513    GLuint oldfallback = rmesa->TclFallback;
 514
 515    if (mode) {
 516       rmesa->TclFallback |= bit;
 517       if (oldfallback == 0) {
 518          if (RADEON_DEBUG & DEBUG_FALLBACKS)
 519             fprintf(stderr, "Radeon begin tcl fallback %s\n",
 520                     getFallbackString( bit ));
 521          transition_to_swtnl( ctx );
 522       }
 523    }
 524    else {
 525       rmesa->TclFallback &= ~bit;
 526       if (oldfallback == bit) {
 527          if (RADEON_DEBUG & DEBUG_FALLBACKS)
 528             fprintf(stderr, "Radeon end tcl fallback %s\n",
 529                     getFallbackString( bit ));
 530          transition_to_hwtnl( ctx );
 531       }
 532    }
 533 }