1/* 2Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. 3 4The Weather Channel (TM) funded Tungsten Graphics to develop the 5initial release of the Radeon 8500 driver under the XFree86 license. 6This notice must be preserved. 7 8Permission is hereby granted, free of charge, to any person obtaining 9a copy of this software and associated documentation files (the 10"Software"), to deal in the Software without restriction, including 11without limitation the rights to use, copy, modify, merge, publish, 12distribute, sublicense, and/or sell copies of the Software, and to 13permit persons to whom the Software is furnished to do so, subject to 14the following conditions: 15 16The above copyright notice and this permission notice (including the 17next paragraph) shall be included in all copies or substantial 18portions of the Software. 19 20THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 21EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 22MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 23IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 24LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 25OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 26WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 27 28**************************************************************************/ 29 30/* 31 * Authors: 32 * Keith Whitwell <keithw@vmware.com> 33 */ 34 35#include "main/glheader.h" 36#include "main/mtypes.h" 37#include "main/enums.h" 38#include "main/image.h" 39 40#include "main/macros.h" 41#include "main/state.h" 42 43#include "swrast/s_context.h" 44#include "swrast/s_fog.h" 45#include "swrast_setup/swrast_setup.h" 46#include "tnl/tnl.h" 47#include "tnl/t_context.h" 48#include "tnl/t_pipeline.h" 49 50#include "r200_context.h" 51#include "r200_ioctl.h" 52#include "r200_state.h" 53#include "r200_swtcl.h" 54#include "r200_tcl.h" 55 56 57/*********************************************************************** 58 * Initialization 59 ***********************************************************************/ 60 61#define EMIT_ATTR( ATTR, STYLE, F0 ) \ 62do { \ 63 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR); \ 64 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE); \ 65 rmesa->radeon.swtcl.vertex_attr_count++; \ 66 fmt_0 |= F0; \ 67} while (0) 68 69#define EMIT_PAD( N ) \ 70do { \ 71 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0; \ 72 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD; \ 73 rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N); \ 74 rmesa->radeon.swtcl.vertex_attr_count++; \ 75} while (0) 76 77static void r200SetVertexFormat( struct gl_context *ctx ) 78{ 79 r200ContextPtr rmesa = R200_CONTEXT( ctx ); 80 TNLcontext *tnl = TNL_CONTEXT(ctx); 81 struct vertex_buffer *VB = &tnl->vb; 82 GLbitfield64 index_bitset = tnl->render_inputs_bitset; 83 int fmt_0 = 0; 84 int fmt_1 = 0; 85 int offset = 0; 86 87 /* Important: 88 */ 89 if ( VB->NdcPtr != NULL ) { 90 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr; 91 } 92 else { 93 VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr; 94 } 95 96 assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL ); 97 rmesa->radeon.swtcl.vertex_attr_count = 0; 98 99 /* EMIT_ATTR's must be in order as they tell t_vertex.c how to 100 * build up a hardware vertex. 101 */ 102 if ( !rmesa->swtcl.needproj || 103 (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) ) { 104 /* need w coord for projected textures */ 105 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F, R200_VTX_XY | R200_VTX_Z0 | R200_VTX_W0 ); 106 offset = 4; 107 } 108 else { 109 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F, R200_VTX_XY | R200_VTX_Z0 ); 110 offset = 3; 111 } 112 113 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_POINTSIZE)) { 114 EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F, R200_VTX_POINT_SIZE ); 115 offset += 1; 116 } 117 118 rmesa->swtcl.coloroffset = offset; 119#if MESA_LITTLE_ENDIAN 120 EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA, (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT) ); 121#else 122 EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR, (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT) ); 123#endif 124 offset += 1; 125 126 rmesa->swtcl.specoffset = 0; 127 if (index_bitset & 128 (BITFIELD64_BIT(_TNL_ATTRIB_COLOR1) | BITFIELD64_BIT(_TNL_ATTRIB_FOG))) { 129 130#if MESA_LITTLE_ENDIAN 131 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) { 132 rmesa->swtcl.specoffset = offset; 133 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_RGB, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) ); 134 } 135 else { 136 EMIT_PAD( 3 ); 137 } 138 139 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) { 140 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) ); 141 } 142 else { 143 EMIT_PAD( 1 ); 144 } 145#else 146 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_FOG)) { 147 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) ); 148 } 149 else { 150 EMIT_PAD( 1 ); 151 } 152 153 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_COLOR1)) { 154 rmesa->swtcl.specoffset = offset; 155 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, (R200_VTX_PK_RGBA << R200_VTX_COLOR_1_SHIFT) ); 156 } 157 else { 158 EMIT_PAD( 3 ); 159 } 160#endif 161 } 162 163 if (index_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) { 164 int i; 165 166 for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { 167 if (index_bitset & BITFIELD64_BIT(_TNL_ATTRIB_TEX(i))) { 168 GLuint sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size; 169 170 fmt_1 |= sz << (3 * i); 171 EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_1F + sz - 1, 0 ); 172 } 173 } 174 } 175 176 if ( (rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] & R200_FOG_USE_MASK) 177 != R200_FOG_USE_SPEC_ALPHA ) { 178 R200_STATECHANGE( rmesa, ctx ); 179 rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] &= ~R200_FOG_USE_MASK; 180 rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] |= R200_FOG_USE_SPEC_ALPHA; 181 } 182 183 if (rmesa->radeon.tnl_index_bitset != index_bitset || 184 (rmesa->hw.vtx.cmd[VTX_VTXFMT_0] != fmt_0) || 185 (rmesa->hw.vtx.cmd[VTX_VTXFMT_1] != fmt_1) ) { 186 R200_NEWPRIM(rmesa); 187 R200_STATECHANGE( rmesa, vtx ); 188 rmesa->hw.vtx.cmd[VTX_VTXFMT_0] = fmt_0; 189 rmesa->hw.vtx.cmd[VTX_VTXFMT_1] = fmt_1; 190 191 rmesa->radeon.swtcl.vertex_size = 192 _tnl_install_attrs( ctx, 193 rmesa->radeon.swtcl.vertex_attrs, 194 rmesa->radeon.swtcl.vertex_attr_count, 195 NULL, 0 ); 196 rmesa->radeon.swtcl.vertex_size /= 4; 197 rmesa->radeon.tnl_index_bitset = index_bitset; 198 } 199} 200 201static void r200_predict_emit_size( r200ContextPtr rmesa ) 202{ 203 if (RADEON_DEBUG & RADEON_VERTS) 204 fprintf(stderr, "%s\n", __func__); 205 const int vertex_array_size = 7; 206 const int prim_size = 3; 207 if (!rmesa->radeon.swtcl.emit_prediction) { 208 const int state_size = radeonCountStateEmitSize(&rmesa->radeon); 209 if (rcommonEnsureCmdBufSpace(&rmesa->radeon, 210 state_size + 211 vertex_array_size + prim_size, 212 __func__)) 213 rmesa->radeon.swtcl.emit_prediction = radeonCountStateEmitSize(&rmesa->radeon); 214 else 215 rmesa->radeon.swtcl.emit_prediction = state_size; 216 rmesa->radeon.swtcl.emit_prediction += vertex_array_size + prim_size 217 + rmesa->radeon.cmdbuf.cs->cdw; 218 } 219} 220 221 222static void r200RenderStart( struct gl_context *ctx ) 223{ 224 r200SetVertexFormat( ctx ); 225 if (RADEON_DEBUG & RADEON_VERTS) 226 fprintf(stderr, "%s\n", __func__); 227} 228 229 230/** 231 * Set vertex state for SW TCL. The primary purpose of this function is to 232 * determine in advance whether or not the hardware can / should do the 233 * projection divide or Mesa should do it. 234 */ 235void r200ChooseVertexState( struct gl_context *ctx ) 236{ 237 r200ContextPtr rmesa = R200_CONTEXT( ctx ); 238 TNLcontext *tnl = TNL_CONTEXT(ctx); 239 GLuint vte; 240 GLuint vap; 241 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL || 242 ctx->Polygon.BackMode != GL_FILL); 243 GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide; 244 245 /* We must ensure that we don't do _tnl_need_projected_coords while in a 246 * rasterization fallback. As this function will be called again when we 247 * leave a rasterization fallback, we can just skip it for now. 248 */ 249 if (rmesa->radeon.Fallback != 0) 250 return; 251 252 vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]; 253 vap = rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL]; 254 255 /* HW perspective divide is a win, but tiny vertex formats are a 256 * bigger one. 257 */ 258 if ((0 == (tnl->render_inputs_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX))) 259 || twosided 260 || unfilled) { 261 rmesa->swtcl.needproj = GL_TRUE; 262 vte |= R200_VTX_XY_FMT | R200_VTX_Z_FMT; 263 vte &= ~R200_VTX_W0_FMT; 264 if (tnl->render_inputs_bitset & BITFIELD64_RANGE(_TNL_ATTRIB_TEX0, _TNL_NUM_TEX)) { 265 vap &= ~R200_VAP_FORCE_W_TO_ONE; 266 } 267 else { 268 vap |= R200_VAP_FORCE_W_TO_ONE; 269 } 270 } 271 else { 272 rmesa->swtcl.needproj = GL_FALSE; 273 vte &= ~(R200_VTX_XY_FMT | R200_VTX_Z_FMT); 274 vte |= R200_VTX_W0_FMT; 275 vap &= ~R200_VAP_FORCE_W_TO_ONE; 276 } 277 278 _tnl_need_projected_coords( ctx, rmesa->swtcl.needproj ); 279 280 if (vte != rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]) { 281 R200_STATECHANGE( rmesa, vte ); 282 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] = vte; 283 } 284 285 if (vap != rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL]) { 286 R200_STATECHANGE( rmesa, vap ); 287 rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] = vap; 288 } 289} 290 291void r200_swtcl_flush(struct gl_context *ctx, uint32_t current_offset) 292{ 293 r200ContextPtr rmesa = R200_CONTEXT(ctx); 294 if (RADEON_DEBUG & RADEON_VERTS) 295 fprintf(stderr, "%s\n", __func__); 296 297 298 radeonEmitState(&rmesa->radeon); 299 r200EmitVertexAOS( rmesa, 300 rmesa->radeon.swtcl.vertex_size, 301 rmesa->radeon.swtcl.bo, 302 current_offset); 303 304 305 r200EmitVbufPrim( rmesa, 306 rmesa->radeon.swtcl.hw_primitive, 307 rmesa->radeon.swtcl.numverts); 308 if ( rmesa->radeon.swtcl.emit_prediction < rmesa->radeon.cmdbuf.cs->cdw ) 309 WARN_ONCE("Rendering was %d commands larger than predicted size." 310 " We might overflow command buffer.\n", 311 rmesa->radeon.cmdbuf.cs->cdw - rmesa->radeon.swtcl.emit_prediction ); 312 313 rmesa->radeon.swtcl.emit_prediction = 0; 314 315} 316 317/**************************************************************************/ 318 319 320static inline GLuint reduced_hw_prim( struct gl_context *ctx, GLuint prim) 321{ 322 switch (prim) { 323 case GL_POINTS: 324 return ((!ctx->Point.SmoothFlag) ? 325 R200_VF_PRIM_POINT_SPRITES : R200_VF_PRIM_POINTS); 326 case GL_LINES: 327 FALLTHROUGH; 328 case GL_LINE_LOOP: 329 FALLTHROUGH; 330 case GL_LINE_STRIP: 331 return R200_VF_PRIM_LINES; 332 default: 333 /* all others reduced to triangles */ 334 return R200_VF_PRIM_TRIANGLES; 335 } 336} 337 338 339static void r200RasterPrimitive( struct gl_context *ctx, GLuint hwprim ); 340static void r200RenderPrimitive( struct gl_context *ctx, GLenum prim ); 341static void r200ResetLineStipple( struct gl_context *ctx ); 342 343/*********************************************************************** 344 * Emit primitives as inline vertices * 345 ***********************************************************************/ 346 347#define HAVE_POINTS 1 348#define HAVE_LINES 1 349#define HAVE_LINE_STRIPS 1 350#define HAVE_TRIANGLES 1 351#define HAVE_TRI_STRIPS 1 352#define HAVE_TRI_FANS 1 353#define HAVE_QUADS 0 354#define HAVE_QUAD_STRIPS 0 355#define HAVE_POLYGONS 1 356#define HAVE_ELTS 0 357 358static void* r200_alloc_verts( r200ContextPtr rmesa, GLuint n, GLuint size) 359{ 360 void *rv; 361 do { 362 r200_predict_emit_size( rmesa ); 363 rv = rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 ); 364 } while(!rv); 365 return rv; 366} 367 368#undef LOCAL_VARS 369#undef ALLOC_VERTS 370#define CTX_ARG r200ContextPtr rmesa 371#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size 372#define ALLOC_VERTS( n, size ) r200_alloc_verts(rmesa, n, size) 373#define LOCAL_VARS \ 374 r200ContextPtr rmesa = R200_CONTEXT(ctx); \ 375 const char *r200verts = (char *)rmesa->radeon.swtcl.verts; 376#define VERT(x) (radeonVertex *)(r200verts + ((x) * vertsize * sizeof(int))) 377#define VERTEX radeonVertex 378#define DO_DEBUG_VERTS (1 && (R200_DEBUG & RADEON_VERTS)) 379 380#undef TAG 381#define TAG(x) r200_##x 382#include "tnl_dd/t_dd_triemit.h" 383 384 385/*********************************************************************** 386 * Macros for t_dd_tritmp.h to draw basic primitives * 387 ***********************************************************************/ 388 389#define QUAD( a, b, c, d ) r200_quad( rmesa, a, b, c, d ) 390#define TRI( a, b, c ) r200_triangle( rmesa, a, b, c ) 391#define LINE( a, b ) r200_line( rmesa, a, b ) 392#define POINT( a ) r200_point( rmesa, a ) 393 394/*********************************************************************** 395 * Build render functions from dd templates * 396 ***********************************************************************/ 397 398#define R200_TWOSIDE_BIT 0x01 399#define R200_UNFILLED_BIT 0x02 400#define R200_MAX_TRIFUNC 0x04 401 402 403static struct { 404 tnl_points_func points; 405 tnl_line_func line; 406 tnl_triangle_func triangle; 407 tnl_quad_func quad; 408} rast_tab[R200_MAX_TRIFUNC]; 409 410 411#define DO_FALLBACK 0 412#define DO_UNFILLED ((IND & R200_UNFILLED_BIT) != 0) 413#define DO_TWOSIDE ((IND & R200_TWOSIDE_BIT) != 0) 414#define DO_FLAT 0 415#define DO_OFFSET 0 416#define DO_TRI 1 417#define DO_QUAD 1 418#define DO_LINE 1 419#define DO_POINTS 1 420#define DO_FULL_QUAD 1 421 422#define HAVE_SPEC 1 423#define HAVE_BACK_COLORS 0 424#define HAVE_HW_FLATSHADE 1 425#define TAB rast_tab 426 427#define DEPTH_SCALE 1.0 428#define UNFILLED_TRI unfilled_tri 429#define UNFILLED_QUAD unfilled_quad 430#define VERT_X(_v) _v->v.x 431#define VERT_Y(_v) _v->v.y 432#define VERT_Z(_v) _v->v.z 433#define AREA_IS_CCW( a ) (a < 0) 434#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + (e*rmesa->radeon.swtcl.vertex_size*sizeof(int))) 435 436#define VERT_SET_RGBA( v, c ) \ 437do { \ 438 radeon_color_t *color = (radeon_color_t *)&((v)->ui[coloroffset]); \ 439 UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]); \ 440 UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]); \ 441 UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]); \ 442 UNCLAMPED_FLOAT_TO_UBYTE(color->alpha, (c)[3]); \ 443} while (0) 444 445#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset] 446 447#define VERT_SET_SPEC( v, c ) \ 448do { \ 449 if (specoffset) { \ 450 radeon_color_t *spec = (radeon_color_t *)&((v)->ui[specoffset]); \ 451 UNCLAMPED_FLOAT_TO_UBYTE(spec->red, (c)[0]); \ 452 UNCLAMPED_FLOAT_TO_UBYTE(spec->green, (c)[1]); \ 453 UNCLAMPED_FLOAT_TO_UBYTE(spec->blue, (c)[2]); \ 454 } \ 455} while (0) 456#define VERT_COPY_SPEC( v0, v1 ) \ 457do { \ 458 if (specoffset) { \ 459 radeon_color_t *spec0 = (radeon_color_t *)&((v0)->ui[specoffset]); \ 460 radeon_color_t *spec1 = (radeon_color_t *)&((v1)->ui[specoffset]); \ 461 spec0->red = spec1->red; \ 462 spec0->green = spec1->green; \ 463 spec0->blue = spec1->blue; \ 464 } \ 465} while (0) 466 467/* These don't need LE32_TO_CPU() as they used to save and restore 468 * colors which are already in the correct format. 469 */ 470#define VERT_SAVE_RGBA( idx ) color[idx] = v[idx]->ui[coloroffset] 471#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx] 472#define VERT_SAVE_SPEC( idx ) if (specoffset) spec[idx] = v[idx]->ui[specoffset] 473#define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx] 474 475#undef LOCAL_VARS 476#undef TAG 477#undef INIT 478 479#define LOCAL_VARS(n) \ 480 r200ContextPtr rmesa = R200_CONTEXT(ctx); \ 481 GLuint color[n] = {0}, spec[n] = {0}; \ 482 GLuint coloroffset = rmesa->swtcl.coloroffset; \ 483 GLuint specoffset = rmesa->swtcl.specoffset; \ 484 (void) color; (void) spec; (void) coloroffset; (void) specoffset; 485 486/*********************************************************************** 487 * Helpers for rendering unfilled primitives * 488 ***********************************************************************/ 489 490#define RASTERIZE(x) r200RasterPrimitive( ctx, reduced_hw_prim(ctx, x) ) 491#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive 492#undef TAG 493#define TAG(x) x 494#include "tnl_dd/t_dd_unfilled.h" 495#undef IND 496 497 498/*********************************************************************** 499 * Generate GL render functions * 500 ***********************************************************************/ 501 502 503#define IND (0) 504#define TAG(x) x 505#include "tnl_dd/t_dd_tritmp.h" 506 507#define IND (R200_TWOSIDE_BIT) 508#define TAG(x) x##_twoside 509#include "tnl_dd/t_dd_tritmp.h" 510 511#define IND (R200_UNFILLED_BIT) 512#define TAG(x) x##_unfilled 513#include "tnl_dd/t_dd_tritmp.h" 514 515#define IND (R200_TWOSIDE_BIT|R200_UNFILLED_BIT) 516#define TAG(x) x##_twoside_unfilled 517#include "tnl_dd/t_dd_tritmp.h" 518 519 520static void init_rast_tab( void ) 521{ 522 init(); 523 init_twoside(); 524 init_unfilled(); 525 init_twoside_unfilled(); 526} 527 528/**********************************************************************/ 529/* Render unclipped begin/end objects */ 530/**********************************************************************/ 531 532#define RENDER_POINTS( start, count ) \ 533 for ( ; start < count ; start++) \ 534 r200_point( rmesa, VERT(start) ) 535#define RENDER_LINE( v0, v1 ) \ 536 r200_line( rmesa, VERT(v0), VERT(v1) ) 537#define RENDER_TRI( v0, v1, v2 ) \ 538 r200_triangle( rmesa, VERT(v0), VERT(v1), VERT(v2) ) 539#define RENDER_QUAD( v0, v1, v2, v3 ) \ 540 r200_quad( rmesa, VERT(v0), VERT(v1), VERT(v2), VERT(v3) ) 541#define INIT(x) do { \ 542 r200RenderPrimitive( ctx, x ); \ 543} while (0) 544#undef LOCAL_VARS 545#define LOCAL_VARS \ 546 r200ContextPtr rmesa = R200_CONTEXT(ctx); \ 547 const GLuint vertsize = rmesa->radeon.swtcl.vertex_size; \ 548 const char *r200verts = (char *)rmesa->radeon.swtcl.verts; \ 549 const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts; \ 550 const GLboolean stipple = ctx->Line.StippleFlag; \ 551 (void) elt; (void) stipple; 552#define RESET_STIPPLE if ( stipple ) r200ResetLineStipple( ctx ); 553#define RESET_OCCLUSION 554#define PRESERVE_VB_DEFS 555#define ELT(x) (x) 556#define TAG(x) r200_##x##_verts 557#include "tnl/t_vb_rendertmp.h" 558#undef ELT 559#undef TAG 560#define TAG(x) r200_##x##_elts 561#define ELT(x) elt[x] 562#include "tnl/t_vb_rendertmp.h" 563 564 565 566/**********************************************************************/ 567/* Choose render functions */ 568/**********************************************************************/ 569 570void r200ChooseRenderState( struct gl_context *ctx ) 571{ 572 TNLcontext *tnl = TNL_CONTEXT(ctx); 573 r200ContextPtr rmesa = R200_CONTEXT(ctx); 574 GLuint index = 0; 575 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL || 576 ctx->Polygon.BackMode != GL_FILL); 577 GLboolean twosided = ctx->Light.Enabled && ctx->Light.Model.TwoSide; 578 579 if (!rmesa->radeon.TclFallback || rmesa->radeon.Fallback) 580 return; 581 582 if (twosided) 583 index |= R200_TWOSIDE_BIT; 584 if (unfilled) 585 index |= R200_UNFILLED_BIT; 586 587 if (index != rmesa->radeon.swtcl.RenderIndex) { 588 tnl->Driver.Render.Points = rast_tab[index].points; 589 tnl->Driver.Render.Line = rast_tab[index].line; 590 tnl->Driver.Render.ClippedLine = rast_tab[index].line; 591 tnl->Driver.Render.Triangle = rast_tab[index].triangle; 592 tnl->Driver.Render.Quad = rast_tab[index].quad; 593 594 if (index == 0) { 595 tnl->Driver.Render.PrimTabVerts = r200_render_tab_verts; 596 tnl->Driver.Render.PrimTabElts = r200_render_tab_elts; 597 tnl->Driver.Render.ClippedPolygon = r200_fast_clipped_poly; 598 } else { 599 tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts; 600 tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts; 601 tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon; 602 } 603 604 rmesa->radeon.swtcl.RenderIndex = index; 605 } 606} 607 608 609/**********************************************************************/ 610/* High level hooks for t_vb_render.c */ 611/**********************************************************************/ 612 613 614static void r200RasterPrimitive( struct gl_context *ctx, GLuint hwprim ) 615{ 616 r200ContextPtr rmesa = R200_CONTEXT(ctx); 617 618 radeon_prepare_render(&rmesa->radeon); 619 if (rmesa->radeon.NewGLState) 620 r200ValidateState( ctx ); 621 622 623 if (rmesa->radeon.swtcl.hw_primitive != hwprim) { 624 /* need to disable perspective-correct texturing for point sprites */ 625 if ((hwprim & 0xf) == R200_VF_PRIM_POINT_SPRITES && ctx->Point.PointSprite) { 626 if (rmesa->hw.set.cmd[SET_RE_CNTL] & R200_PERSPECTIVE_ENABLE) { 627 R200_STATECHANGE( rmesa, set ); 628 rmesa->hw.set.cmd[SET_RE_CNTL] &= ~R200_PERSPECTIVE_ENABLE; 629 } 630 } 631 else if (!(rmesa->hw.set.cmd[SET_RE_CNTL] & R200_PERSPECTIVE_ENABLE)) { 632 R200_STATECHANGE( rmesa, set ); 633 rmesa->hw.set.cmd[SET_RE_CNTL] |= R200_PERSPECTIVE_ENABLE; 634 } 635 R200_NEWPRIM( rmesa ); 636 rmesa->radeon.swtcl.hw_primitive = hwprim; 637 } 638} 639 640static void r200RenderPrimitive( struct gl_context *ctx, GLenum prim ) 641{ 642 r200ContextPtr rmesa = R200_CONTEXT(ctx); 643 GLboolean unfilled = (ctx->Polygon.FrontMode != GL_FILL || 644 ctx->Polygon.BackMode != GL_FILL); 645 646 rmesa->radeon.swtcl.render_primitive = prim; 647 if (prim < GL_TRIANGLES || !unfilled) 648 r200RasterPrimitive( ctx, reduced_hw_prim(ctx, prim) ); 649} 650 651static void r200RenderFinish( struct gl_context *ctx ) 652{ 653} 654 655static void r200ResetLineStipple( struct gl_context *ctx ) 656{ 657 r200ContextPtr rmesa = R200_CONTEXT(ctx); 658 R200_STATECHANGE( rmesa, lin ); 659} 660 661 662/**********************************************************************/ 663/* Transition to/from hardware rasterization. */ 664/**********************************************************************/ 665 666static const char * const fallbackStrings[] = { 667 "Texture mode", 668 "glDrawBuffer(GL_FRONT_AND_BACK)", 669 "glEnable(GL_STENCIL) without hw stencil buffer", 670 "glRenderMode(selection or feedback)", 671 "R200_NO_RAST", 672 "Mixing GL_CLAMP_TO_BORDER and GL_CLAMP (or GL_MIRROR_CLAMP_ATI)" 673}; 674 675 676static const char *getFallbackString(GLuint bit) 677{ 678 int i = 0; 679 while (bit > 1) { 680 i++; 681 bit >>= 1; 682 } 683 return fallbackStrings[i]; 684} 685 686 687void r200Fallback( struct gl_context *ctx, GLuint bit, GLboolean mode ) 688{ 689 r200ContextPtr rmesa = R200_CONTEXT(ctx); 690 TNLcontext *tnl = TNL_CONTEXT(ctx); 691 GLuint oldfallback = rmesa->radeon.Fallback; 692 693 if (mode) { 694 rmesa->radeon.Fallback |= bit; 695 if (oldfallback == 0) { 696 radeon_firevertices(&rmesa->radeon); 697 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_RASTER, GL_TRUE ); 698 _swsetup_Wakeup( ctx ); 699 rmesa->radeon.swtcl.RenderIndex = ~0; 700 if (R200_DEBUG & RADEON_FALLBACKS) { 701 fprintf(stderr, "R200 begin rasterization fallback: 0x%x %s\n", 702 bit, getFallbackString(bit)); 703 } 704 } 705 } 706 else { 707 rmesa->radeon.Fallback &= ~bit; 708 if (oldfallback == bit) { 709 710 _swrast_flush( ctx ); 711 tnl->Driver.Render.Start = r200RenderStart; 712 tnl->Driver.Render.PrimitiveNotify = r200RenderPrimitive; 713 tnl->Driver.Render.Finish = r200RenderFinish; 714 715 tnl->Driver.Render.BuildVertices = _tnl_build_vertices; 716 tnl->Driver.Render.CopyPV = _tnl_copy_pv; 717 tnl->Driver.Render.Interp = _tnl_interp; 718 719 tnl->Driver.Render.ResetLineStipple = r200ResetLineStipple; 720 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_RASTER, GL_FALSE ); 721 if (rmesa->radeon.TclFallback) { 722 /* These are already done if rmesa->radeon.TclFallback goes to 723 * zero above. But not if it doesn't (R200_NO_TCL for 724 * example?) 725 */ 726 _tnl_invalidate_vertex_state( ctx, ~0 ); 727 _tnl_invalidate_vertices( ctx, ~0 ); 728 rmesa->radeon.tnl_index_bitset = 0; 729 r200ChooseVertexState( ctx ); 730 r200ChooseRenderState( ctx ); 731 } 732 if (R200_DEBUG & RADEON_FALLBACKS) { 733 fprintf(stderr, "R200 end rasterization fallback: 0x%x %s\n", 734 bit, getFallbackString(bit)); 735 } 736 } 737 } 738} 739 740 741 742 743/** 744 * Cope with depth operations by drawing individual pixels as points. 745 * 746 * \todo 747 * The way the vertex state is set in this routine is hokey. It seems to 748 * work, but it's very hackish. This whole routine is pretty hackish. If 749 * the bitmap is small enough, it seems like it would be faster to copy it 750 * to AGP memory and use it as a non-power-of-two texture (i.e., 751 * NV_texture_rectangle). 752 */ 753void 754r200PointsBitmap( struct gl_context *ctx, GLint px, GLint py, 755 GLsizei width, GLsizei height, 756 const struct gl_pixelstore_attrib *unpack, 757 const GLubyte *bitmap ) 758{ 759 r200ContextPtr rmesa = R200_CONTEXT(ctx); 760 const GLfloat *rc = ctx->Current.RasterColor; 761 GLint row, col; 762 radeonVertex vert; 763 GLuint orig_vte; 764 GLuint h; 765 766 767 /* Turn off tcl. 768 */ 769 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_BITMAP, 1 ); 770 771 /* Choose tiny vertex format 772 */ 773 { 774 const GLuint fmt_0 = R200_VTX_XY | R200_VTX_Z0 | R200_VTX_W0 775 | (R200_VTX_PK_RGBA << R200_VTX_COLOR_0_SHIFT); 776 const GLuint fmt_1 = 0; 777 GLuint vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]; 778 GLuint vap = rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL]; 779 780 vte &= ~(R200_VTX_XY_FMT | R200_VTX_Z_FMT); 781 vte |= R200_VTX_W0_FMT; 782 vap &= ~R200_VAP_FORCE_W_TO_ONE; 783 784 rmesa->radeon.swtcl.vertex_size = 5; 785 786 if ( (rmesa->hw.vtx.cmd[VTX_VTXFMT_0] != fmt_0) 787 || (rmesa->hw.vtx.cmd[VTX_VTXFMT_1] != fmt_1) ) { 788 R200_NEWPRIM(rmesa); 789 R200_STATECHANGE( rmesa, vtx ); 790 rmesa->hw.vtx.cmd[VTX_VTXFMT_0] = fmt_0; 791 rmesa->hw.vtx.cmd[VTX_VTXFMT_1] = fmt_1; 792 } 793 794 if (vte != rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]) { 795 R200_STATECHANGE( rmesa, vte ); 796 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] = vte; 797 } 798 799 if (vap != rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL]) { 800 R200_STATECHANGE( rmesa, vap ); 801 rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] = vap; 802 } 803 } 804 805 /* Ready for point primitives: 806 */ 807 r200RenderPrimitive( ctx, GL_POINTS ); 808 809 /* Turn off the hw viewport transformation: 810 */ 811 R200_STATECHANGE( rmesa, vte ); 812 orig_vte = rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL]; 813 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] &= ~(R200_VPORT_X_SCALE_ENA | 814 R200_VPORT_Y_SCALE_ENA | 815 R200_VPORT_Z_SCALE_ENA | 816 R200_VPORT_X_OFFSET_ENA | 817 R200_VPORT_Y_OFFSET_ENA | 818 R200_VPORT_Z_OFFSET_ENA); 819 820 /* Turn off other stuff: Stipple?, texture?, blending?, etc. 821 */ 822 823 824 /* Populate the vertex 825 * 826 * Incorporate FOG into RGBA 827 */ 828 if (ctx->Fog.Enabled) { 829 const GLfloat *fc = ctx->Fog.Color; 830 GLfloat color[4]; 831 GLfloat f; 832 833 if (ctx->Fog.FogCoordinateSource == GL_FOG_COORDINATE_EXT) 834 f = _swrast_z_to_fogfactor(ctx, ctx->Current.Attrib[VERT_ATTRIB_FOG][0]); 835 else 836 f = _swrast_z_to_fogfactor(ctx, ctx->Current.RasterDistance); 837 838 color[0] = f * rc[0] + (1.F - f) * fc[0]; 839 color[1] = f * rc[1] + (1.F - f) * fc[1]; 840 color[2] = f * rc[2] + (1.F - f) * fc[2]; 841 color[3] = rc[3]; 842 843 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.red, color[0]); 844 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.green, color[1]); 845 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.blue, color[2]); 846 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.alpha, color[3]); 847 } 848 else { 849 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.red, rc[0]); 850 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.green, rc[1]); 851 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.blue, rc[2]); 852 UNCLAMPED_FLOAT_TO_CHAN(vert.tv.color.alpha, rc[3]); 853 } 854 855 856 vert.tv.z = ctx->Current.RasterPos[2]; 857 858 859 /* Update window height 860 */ 861 h = radeon_get_drawable(&rmesa->radeon)->h; 862 863 /* Clipping handled by existing mechansims in r200_ioctl.c? 864 */ 865 for (row=0; row<height; row++) { 866 const GLubyte *src = (const GLubyte *) 867 _mesa_image_address2d(unpack, bitmap, width, height, 868 GL_COLOR_INDEX, GL_BITMAP, row, 0 ); 869 870 if (unpack->LsbFirst) { 871 /* Lsb first */ 872 GLubyte mask = 1U << (unpack->SkipPixels & 0x7); 873 for (col=0; col<width; col++) { 874 if (*src & mask) { 875 vert.tv.x = px+col; 876 vert.tv.y = h - (py+row) - 1; 877 r200_point( rmesa, &vert ); 878 } 879 src += (mask >> 7); 880 mask = ((mask << 1) & 0xff) | (mask >> 7); 881 } 882 883 /* get ready for next row */ 884 if (mask != 1) 885 src++; 886 } 887 else { 888 /* Msb first */ 889 GLubyte mask = 128U >> (unpack->SkipPixels & 0x7); 890 for (col=0; col<width; col++) { 891 if (*src & mask) { 892 vert.tv.x = px+col; 893 vert.tv.y = h - (py+row) - 1; 894 r200_point( rmesa, &vert ); 895 } 896 src += mask & 1; 897 mask = ((mask << 7) & 0xff) | (mask >> 1); 898 } 899 /* get ready for next row */ 900 if (mask != 128) 901 src++; 902 } 903 } 904 905 /* Fire outstanding vertices, restore state 906 */ 907 R200_STATECHANGE( rmesa, vte ); 908 rmesa->hw.vte.cmd[VTE_SE_VTE_CNTL] = orig_vte; 909 910 /* Unfallback 911 */ 912 TCL_FALLBACK( ctx, R200_TCL_FALLBACK_BITMAP, 0 ); 913 914 /* Need to restore vertexformat? 915 */ 916 if (rmesa->radeon.TclFallback) 917 r200ChooseVertexState( ctx ); 918} 919 920 921 922/**********************************************************************/ 923/* Initialization. */ 924/**********************************************************************/ 925 926void r200InitSwtcl( struct gl_context *ctx ) 927{ 928 TNLcontext *tnl = TNL_CONTEXT(ctx); 929 r200ContextPtr rmesa = R200_CONTEXT(ctx); 930 static int firsttime = 1; 931 932 if (firsttime) { 933 init_rast_tab(); 934 firsttime = 0; 935 } 936 rmesa->radeon.swtcl.emit_prediction = 0; 937 938 tnl->Driver.Render.Start = r200RenderStart; 939 tnl->Driver.Render.Finish = r200RenderFinish; 940 tnl->Driver.Render.PrimitiveNotify = r200RenderPrimitive; 941 tnl->Driver.Render.ResetLineStipple = r200ResetLineStipple; 942 tnl->Driver.Render.BuildVertices = _tnl_build_vertices; 943 tnl->Driver.Render.CopyPV = _tnl_copy_pv; 944 tnl->Driver.Render.Interp = _tnl_interp; 945 946 /* FIXME: what are these numbers? */ 947 _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 948 36 * sizeof(GLfloat) ); 949 950 rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf; 951 rmesa->radeon.swtcl.RenderIndex = ~0; 952 rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES; 953 rmesa->radeon.swtcl.hw_primitive = 0; 954} 955 956