s_tritemp.h revision fcd7c37fd3d0f61cf6ac81170bc0b3fca64ad9bb
1/* 2 * Mesa 3-D graphics library 3 * Version: 7.0 4 * 5 * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the "Software"), 9 * to deal in the Software without restriction, including without limitation 10 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 11 * and/or sell copies of the Software, and to permit persons to whom the 12 * Software is furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included 15 * in all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN 21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 23 */ 24 25/* 26 * Triangle Rasterizer Template 27 * 28 * This file is #include'd to generate custom triangle rasterizers. 29 * 30 * The following macros may be defined to indicate what auxillary information 31 * must be interpolated across the triangle: 32 * INTERP_Z - if defined, interpolate integer Z values 33 * INTERP_RGB - if defined, interpolate integer RGB values 34 * INTERP_ALPHA - if defined, interpolate integer Alpha values 35 * INTERP_INDEX - if defined, interpolate color index values 36 * INTERP_INT_TEX - if defined, interpolate integer ST texcoords 37 * (fast, simple 2-D texture mapping, without 38 * perspective correction) 39 * INTERP_ATTRIBS - if defined, interpolate arbitrary attribs (texcoords, 40 * varying vars, etc) This also causes W to be 41 * computed for perspective correction). 42 * 43 * When one can directly address pixels in the color buffer the following 44 * macros can be defined and used to compute pixel addresses during 45 * rasterization (see pRow): 46 * PIXEL_TYPE - the datatype of a pixel (GLubyte, GLushort, GLuint) 47 * BYTES_PER_ROW - number of bytes per row in the color buffer 48 * PIXEL_ADDRESS(X,Y) - returns the address of pixel at (X,Y) where 49 * Y==0 at bottom of screen and increases upward. 50 * 51 * Similarly, for direct depth buffer access, this type is used for depth 52 * buffer addressing (see zRow): 53 * DEPTH_TYPE - either GLushort or GLuint 54 * 55 * Optionally, one may provide one-time setup code per triangle: 56 * SETUP_CODE - code which is to be executed once per triangle 57 * 58 * The following macro MUST be defined: 59 * RENDER_SPAN(span) - code to write a span of pixels. 60 * 61 * This code was designed for the origin to be in the lower-left corner. 62 * 63 * Inspired by triangle rasterizer code written by Allen Akin. Thanks Allen! 64 * 65 * 66 * Some notes on rasterization accuracy: 67 * 68 * This code uses fixed point arithmetic (the GLfixed type) to iterate 69 * over the triangle edges and interpolate ancillary data (such as Z, 70 * color, secondary color, etc). The number of fractional bits in 71 * GLfixed and the value of SUB_PIXEL_BITS has a direct bearing on the 72 * accuracy of rasterization. 73 * 74 * If SUB_PIXEL_BITS=4 then we'll snap the vertices to the nearest 75 * 1/16 of a pixel. If we're walking up a long, nearly vertical edge 76 * (dx=1/16, dy=1024) we'll need 4 + 10 = 14 fractional bits in 77 * GLfixed to walk the edge without error. If the maximum viewport 78 * height is 4K pixels, then we'll need 4 + 12 = 16 fractional bits. 79 * 80 * Historically, Mesa has used 11 fractional bits in GLfixed, snaps 81 * vertices to 1/16 pixel and allowed a maximum viewport height of 2K 82 * pixels. 11 fractional bits is actually insufficient for accurately 83 * rasterizing some triangles. More recently, the maximum viewport 84 * height was increased to 4K pixels. Thus, Mesa should be using 16 85 * fractional bits in GLfixed. Unfortunately, there may be some issues 86 * with setting FIXED_FRAC_BITS=16, such as multiplication overflow. 87 * This will have to be examined in some detail... 88 * 89 * For now, if you find rasterization errors, particularly with tall, 90 * sliver triangles, try increasing FIXED_FRAC_BITS and/or decreasing 91 * SUB_PIXEL_BITS. 92 */ 93 94 95/* 96 * Some code we unfortunately need to prevent negative interpolated colors. 97 */ 98#ifndef CLAMP_INTERPOLANT 99#define CLAMP_INTERPOLANT(CHANNEL, CHANNELSTEP, LEN) \ 100do { \ 101 GLfixed endVal = span.CHANNEL + (LEN) * span.CHANNELSTEP; \ 102 if (endVal < 0) { \ 103 span.CHANNEL -= endVal; \ 104 } \ 105 if (span.CHANNEL < 0) { \ 106 span.CHANNEL = 0; \ 107 } \ 108} while (0) 109#endif 110 111 112static void NAME(GLcontext *ctx, const SWvertex *v0, 113 const SWvertex *v1, 114 const SWvertex *v2 ) 115{ 116 typedef struct { 117 const SWvertex *v0, *v1; /* Y(v0) < Y(v1) */ 118 GLfloat dx; /* X(v1) - X(v0) */ 119 GLfloat dy; /* Y(v1) - Y(v0) */ 120 GLfloat dxdy; /* dx/dy */ 121 GLfixed fdxdy; /* dx/dy in fixed-point */ 122 GLfloat adjy; /* adjust from v[0]->fy to fsy, scaled */ 123 GLfixed fsx; /* first sample point x coord */ 124 GLfixed fsy; 125 GLfixed fx0; /* fixed pt X of lower endpoint */ 126 GLint lines; /* number of lines to be sampled on this edge */ 127 } EdgeT; 128 129 const SWcontext *swrast = SWRAST_CONTEXT(ctx); 130#ifdef INTERP_Z 131 const GLint depthBits = ctx->DrawBuffer->Visual.depthBits; 132 const GLint fixedToDepthShift = depthBits <= 16 ? FIXED_SHIFT : 0; 133 const GLfloat maxDepth = ctx->DrawBuffer->_DepthMaxF; 134#define FixedToDepth(F) ((F) >> fixedToDepthShift) 135#endif 136 EdgeT eMaj, eTop, eBot; 137 GLfloat oneOverArea; 138 const SWvertex *vMin, *vMid, *vMax; /* Y(vMin)<=Y(vMid)<=Y(vMax) */ 139 GLfloat bf = SWRAST_CONTEXT(ctx)->_BackfaceCullSign; 140 const GLint snapMask = ~((FIXED_ONE / (1 << SUB_PIXEL_BITS)) - 1); /* for x/y coord snapping */ 141 GLfixed vMin_fx, vMin_fy, vMid_fx, vMid_fy, vMax_fx, vMax_fy; 142 143 SWspan span; 144 145 (void) swrast; 146 147 INIT_SPAN(span, GL_POLYGON); 148 span.y = 0; /* silence warnings */ 149 150#ifdef INTERP_Z 151 (void) fixedToDepthShift; 152#endif 153 154 /* 155 printf("%s()\n", __FUNCTION__); 156 printf(" %g, %g, %g\n", 157 v0->attrib[FRAG_ATTRIB_WPOS][0], 158 v0->attrib[FRAG_ATTRIB_WPOS][1], 159 v0->attrib[FRAG_ATTRIB_WPOS][2]); 160 printf(" %g, %g, %g\n", 161 v1->attrib[FRAG_ATTRIB_WPOS][0], 162 v1->attrib[FRAG_ATTRIB_WPOS][1], 163 v1->attrib[FRAG_ATTRIB_WPOS][2]); 164 printf(" %g, %g, %g\n", 165 v2->attrib[FRAG_ATTRIB_WPOS][0], 166 v2->attrib[FRAG_ATTRIB_WPOS][1], 167 v2->attrib[FRAG_ATTRIB_WPOS][2]); 168 */ 169 170 /* Compute fixed point x,y coords w/ half-pixel offsets and snapping. 171 * And find the order of the 3 vertices along the Y axis. 172 */ 173 { 174 const GLfixed fy0 = FloatToFixed(v0->attrib[FRAG_ATTRIB_WPOS][1] - 0.5F) & snapMask; 175 const GLfixed fy1 = FloatToFixed(v1->attrib[FRAG_ATTRIB_WPOS][1] - 0.5F) & snapMask; 176 const GLfixed fy2 = FloatToFixed(v2->attrib[FRAG_ATTRIB_WPOS][1] - 0.5F) & snapMask; 177 if (fy0 <= fy1) { 178 if (fy1 <= fy2) { 179 /* y0 <= y1 <= y2 */ 180 vMin = v0; vMid = v1; vMax = v2; 181 vMin_fy = fy0; vMid_fy = fy1; vMax_fy = fy2; 182 } 183 else if (fy2 <= fy0) { 184 /* y2 <= y0 <= y1 */ 185 vMin = v2; vMid = v0; vMax = v1; 186 vMin_fy = fy2; vMid_fy = fy0; vMax_fy = fy1; 187 } 188 else { 189 /* y0 <= y2 <= y1 */ 190 vMin = v0; vMid = v2; vMax = v1; 191 vMin_fy = fy0; vMid_fy = fy2; vMax_fy = fy1; 192 bf = -bf; 193 } 194 } 195 else { 196 if (fy0 <= fy2) { 197 /* y1 <= y0 <= y2 */ 198 vMin = v1; vMid = v0; vMax = v2; 199 vMin_fy = fy1; vMid_fy = fy0; vMax_fy = fy2; 200 bf = -bf; 201 } 202 else if (fy2 <= fy1) { 203 /* y2 <= y1 <= y0 */ 204 vMin = v2; vMid = v1; vMax = v0; 205 vMin_fy = fy2; vMid_fy = fy1; vMax_fy = fy0; 206 bf = -bf; 207 } 208 else { 209 /* y1 <= y2 <= y0 */ 210 vMin = v1; vMid = v2; vMax = v0; 211 vMin_fy = fy1; vMid_fy = fy2; vMax_fy = fy0; 212 } 213 } 214 215 /* fixed point X coords */ 216 vMin_fx = FloatToFixed(vMin->attrib[FRAG_ATTRIB_WPOS][0] + 0.5F) & snapMask; 217 vMid_fx = FloatToFixed(vMid->attrib[FRAG_ATTRIB_WPOS][0] + 0.5F) & snapMask; 218 vMax_fx = FloatToFixed(vMax->attrib[FRAG_ATTRIB_WPOS][0] + 0.5F) & snapMask; 219 } 220 221 /* vertex/edge relationship */ 222 eMaj.v0 = vMin; eMaj.v1 = vMax; /*TODO: .v1's not needed */ 223 eTop.v0 = vMid; eTop.v1 = vMax; 224 eBot.v0 = vMin; eBot.v1 = vMid; 225 226 /* compute deltas for each edge: vertex[upper] - vertex[lower] */ 227 eMaj.dx = FixedToFloat(vMax_fx - vMin_fx); 228 eMaj.dy = FixedToFloat(vMax_fy - vMin_fy); 229 eTop.dx = FixedToFloat(vMax_fx - vMid_fx); 230 eTop.dy = FixedToFloat(vMax_fy - vMid_fy); 231 eBot.dx = FixedToFloat(vMid_fx - vMin_fx); 232 eBot.dy = FixedToFloat(vMid_fy - vMin_fy); 233 234 /* compute area, oneOverArea and perform backface culling */ 235 { 236 const GLfloat area = eMaj.dx * eBot.dy - eBot.dx * eMaj.dy; 237 /* Do backface culling */ 238 239 if (area * bf < 0.0) 240 return; 241 242 if (IS_INF_OR_NAN(area) || area == 0.0F) 243 return; 244 245 oneOverArea = 1.0F / area; 246 247 /* 0 = front, 1 = back */ 248 span.facing = oneOverArea * swrast->_BackfaceSign > 0.0F; 249 } 250 251 /* Edge setup. For a triangle strip these could be reused... */ 252 { 253 eMaj.fsy = FixedCeil(vMin_fy); 254 eMaj.lines = FixedToInt(FixedCeil(vMax_fy - eMaj.fsy)); 255 if (eMaj.lines > 0) { 256 eMaj.dxdy = eMaj.dx / eMaj.dy; 257 eMaj.fdxdy = SignedFloatToFixed(eMaj.dxdy); 258 eMaj.adjy = (GLfloat) (eMaj.fsy - vMin_fy); /* SCALED! */ 259 eMaj.fx0 = vMin_fx; 260 eMaj.fsx = eMaj.fx0 + (GLfixed) (eMaj.adjy * eMaj.dxdy); 261 } 262 else { 263 return; /*CULLED*/ 264 } 265 266 eTop.fsy = FixedCeil(vMid_fy); 267 eTop.lines = FixedToInt(FixedCeil(vMax_fy - eTop.fsy)); 268 if (eTop.lines > 0) { 269 eTop.dxdy = eTop.dx / eTop.dy; 270 eTop.fdxdy = SignedFloatToFixed(eTop.dxdy); 271 eTop.adjy = (GLfloat) (eTop.fsy - vMid_fy); /* SCALED! */ 272 eTop.fx0 = vMid_fx; 273 eTop.fsx = eTop.fx0 + (GLfixed) (eTop.adjy * eTop.dxdy); 274 } 275 276 eBot.fsy = FixedCeil(vMin_fy); 277 eBot.lines = FixedToInt(FixedCeil(vMid_fy - eBot.fsy)); 278 if (eBot.lines > 0) { 279 eBot.dxdy = eBot.dx / eBot.dy; 280 eBot.fdxdy = SignedFloatToFixed(eBot.dxdy); 281 eBot.adjy = (GLfloat) (eBot.fsy - vMin_fy); /* SCALED! */ 282 eBot.fx0 = vMin_fx; 283 eBot.fsx = eBot.fx0 + (GLfixed) (eBot.adjy * eBot.dxdy); 284 } 285 } 286 287 /* 288 * Conceptually, we view a triangle as two subtriangles 289 * separated by a perfectly horizontal line. The edge that is 290 * intersected by this line is one with maximal absolute dy; we 291 * call it a ``major'' edge. The other two edges are the 292 * ``top'' edge (for the upper subtriangle) and the ``bottom'' 293 * edge (for the lower subtriangle). If either of these two 294 * edges is horizontal or very close to horizontal, the 295 * corresponding subtriangle might cover zero sample points; 296 * we take care to handle such cases, for performance as well 297 * as correctness. 298 * 299 * By stepping rasterization parameters along the major edge, 300 * we can avoid recomputing them at the discontinuity where 301 * the top and bottom edges meet. However, this forces us to 302 * be able to scan both left-to-right and right-to-left. 303 * Also, we must determine whether the major edge is at the 304 * left or right side of the triangle. We do this by 305 * computing the magnitude of the cross-product of the major 306 * and top edges. Since this magnitude depends on the sine of 307 * the angle between the two edges, its sign tells us whether 308 * we turn to the left or to the right when travelling along 309 * the major edge to the top edge, and from this we infer 310 * whether the major edge is on the left or the right. 311 * 312 * Serendipitously, this cross-product magnitude is also a 313 * value we need to compute the iteration parameter 314 * derivatives for the triangle, and it can be used to perform 315 * backface culling because its sign tells us whether the 316 * triangle is clockwise or counterclockwise. In this code we 317 * refer to it as ``area'' because it's also proportional to 318 * the pixel area of the triangle. 319 */ 320 321 { 322 GLint scan_from_left_to_right; /* true if scanning left-to-right */ 323#ifdef INTERP_INDEX 324 GLfloat didx, didy; 325#endif 326 327 /* 328 * Execute user-supplied setup code 329 */ 330#ifdef SETUP_CODE 331 SETUP_CODE 332#endif 333 334 scan_from_left_to_right = (oneOverArea < 0.0F); 335 336 337 /* compute d?/dx and d?/dy derivatives */ 338#ifdef INTERP_Z 339 span.interpMask |= SPAN_Z; 340 { 341 GLfloat eMaj_dz = vMax->attrib[FRAG_ATTRIB_WPOS][2] - vMin->attrib[FRAG_ATTRIB_WPOS][2]; 342 GLfloat eBot_dz = vMid->attrib[FRAG_ATTRIB_WPOS][2] - vMin->attrib[FRAG_ATTRIB_WPOS][2]; 343 span.attrStepX[FRAG_ATTRIB_WPOS][2] = oneOverArea * (eMaj_dz * eBot.dy - eMaj.dy * eBot_dz); 344 if (span.attrStepX[FRAG_ATTRIB_WPOS][2] > maxDepth || 345 span.attrStepX[FRAG_ATTRIB_WPOS][2] < -maxDepth) { 346 /* probably a sliver triangle */ 347 span.attrStepX[FRAG_ATTRIB_WPOS][2] = 0.0; 348 span.attrStepY[FRAG_ATTRIB_WPOS][2] = 0.0; 349 } 350 else { 351 span.attrStepY[FRAG_ATTRIB_WPOS][2] = oneOverArea * (eMaj.dx * eBot_dz - eMaj_dz * eBot.dx); 352 } 353 if (depthBits <= 16) 354 span.zStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_WPOS][2]); 355 else 356 span.zStep = (GLint) span.attrStepX[FRAG_ATTRIB_WPOS][2]; 357 } 358#endif 359#ifdef INTERP_RGB 360 span.interpMask |= SPAN_RGBA; 361 if (ctx->Light.ShadeModel == GL_SMOOTH) { 362 GLfloat eMaj_dr = (GLfloat) (vMax->color[RCOMP] - vMin->color[RCOMP]); 363 GLfloat eBot_dr = (GLfloat) (vMid->color[RCOMP] - vMin->color[RCOMP]); 364 GLfloat eMaj_dg = (GLfloat) (vMax->color[GCOMP] - vMin->color[GCOMP]); 365 GLfloat eBot_dg = (GLfloat) (vMid->color[GCOMP] - vMin->color[GCOMP]); 366 GLfloat eMaj_db = (GLfloat) (vMax->color[BCOMP] - vMin->color[BCOMP]); 367 GLfloat eBot_db = (GLfloat) (vMid->color[BCOMP] - vMin->color[BCOMP]); 368# ifdef INTERP_ALPHA 369 GLfloat eMaj_da = (GLfloat) (vMax->color[ACOMP] - vMin->color[ACOMP]); 370 GLfloat eBot_da = (GLfloat) (vMid->color[ACOMP] - vMin->color[ACOMP]); 371# endif 372 span.attrStepX[FRAG_ATTRIB_COL0][0] = oneOverArea * (eMaj_dr * eBot.dy - eMaj.dy * eBot_dr); 373 span.attrStepY[FRAG_ATTRIB_COL0][0] = oneOverArea * (eMaj.dx * eBot_dr - eMaj_dr * eBot.dx); 374 span.attrStepX[FRAG_ATTRIB_COL0][1] = oneOverArea * (eMaj_dg * eBot.dy - eMaj.dy * eBot_dg); 375 span.attrStepY[FRAG_ATTRIB_COL0][1] = oneOverArea * (eMaj.dx * eBot_dg - eMaj_dg * eBot.dx); 376 span.attrStepX[FRAG_ATTRIB_COL0][2] = oneOverArea * (eMaj_db * eBot.dy - eMaj.dy * eBot_db); 377 span.attrStepY[FRAG_ATTRIB_COL0][2] = oneOverArea * (eMaj.dx * eBot_db - eMaj_db * eBot.dx); 378 span.redStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][0]); 379 span.greenStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][1]); 380 span.blueStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][2]); 381# ifdef INTERP_ALPHA 382 span.attrStepX[FRAG_ATTRIB_COL0][3] = oneOverArea * (eMaj_da * eBot.dy - eMaj.dy * eBot_da); 383 span.attrStepY[FRAG_ATTRIB_COL0][3] = oneOverArea * (eMaj.dx * eBot_da - eMaj_da * eBot.dx); 384 span.alphaStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][3]); 385# endif /* INTERP_ALPHA */ 386 } 387 else { 388 ASSERT(ctx->Light.ShadeModel == GL_FLAT); 389 span.interpMask |= SPAN_FLAT; 390 span.attrStepX[FRAG_ATTRIB_COL0][0] = span.attrStepY[FRAG_ATTRIB_COL0][0] = 0.0F; 391 span.attrStepX[FRAG_ATTRIB_COL0][1] = span.attrStepY[FRAG_ATTRIB_COL0][1] = 0.0F; 392 span.attrStepX[FRAG_ATTRIB_COL0][2] = span.attrStepY[FRAG_ATTRIB_COL0][2] = 0.0F; 393 span.redStep = 0; 394 span.greenStep = 0; 395 span.blueStep = 0; 396# ifdef INTERP_ALPHA 397 span.attrStepX[FRAG_ATTRIB_COL0][3] = span.attrStepY[FRAG_ATTRIB_COL0][3] = 0.0F; 398 span.alphaStep = 0; 399# endif 400 } 401#endif /* INTERP_RGB */ 402#ifdef INTERP_INDEX 403 span.interpMask |= SPAN_INDEX; 404 if (ctx->Light.ShadeModel == GL_SMOOTH) { 405 GLfloat eMaj_di = vMax->attrib[FRAG_ATTRIB_CI][0] - vMin->attrib[FRAG_ATTRIB_CI][0]; 406 GLfloat eBot_di = vMid->attrib[FRAG_ATTRIB_CI][0] - vMin->attrib[FRAG_ATTRIB_CI][0]; 407 didx = oneOverArea * (eMaj_di * eBot.dy - eMaj.dy * eBot_di); 408 didy = oneOverArea * (eMaj.dx * eBot_di - eMaj_di * eBot.dx); 409 span.indexStep = SignedFloatToFixed(didx); 410 } 411 else { 412 span.interpMask |= SPAN_FLAT; 413 didx = didy = 0.0F; 414 span.indexStep = 0; 415 } 416#endif 417#ifdef INTERP_INT_TEX 418 { 419 GLfloat eMaj_ds = (vMax->attrib[FRAG_ATTRIB_TEX0][0] - vMin->attrib[FRAG_ATTRIB_TEX0][0]) * S_SCALE; 420 GLfloat eBot_ds = (vMid->attrib[FRAG_ATTRIB_TEX0][0] - vMin->attrib[FRAG_ATTRIB_TEX0][0]) * S_SCALE; 421 GLfloat eMaj_dt = (vMax->attrib[FRAG_ATTRIB_TEX0][1] - vMin->attrib[FRAG_ATTRIB_TEX0][1]) * T_SCALE; 422 GLfloat eBot_dt = (vMid->attrib[FRAG_ATTRIB_TEX0][1] - vMin->attrib[FRAG_ATTRIB_TEX0][1]) * T_SCALE; 423 span.attrStepX[FRAG_ATTRIB_TEX0][0] = oneOverArea * (eMaj_ds * eBot.dy - eMaj.dy * eBot_ds); 424 span.attrStepY[FRAG_ATTRIB_TEX0][0] = oneOverArea * (eMaj.dx * eBot_ds - eMaj_ds * eBot.dx); 425 span.attrStepX[FRAG_ATTRIB_TEX0][1] = oneOverArea * (eMaj_dt * eBot.dy - eMaj.dy * eBot_dt); 426 span.attrStepY[FRAG_ATTRIB_TEX0][1] = oneOverArea * (eMaj.dx * eBot_dt - eMaj_dt * eBot.dx); 427 span.intTexStep[0] = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_TEX0][0]); 428 span.intTexStep[1] = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_TEX0][1]); 429 } 430#endif 431#ifdef INTERP_ATTRIBS 432 { 433 /* attrib[FRAG_ATTRIB_WPOS][3] is 1/W */ 434 const GLfloat wMax = vMax->attrib[FRAG_ATTRIB_WPOS][3]; 435 const GLfloat wMin = vMin->attrib[FRAG_ATTRIB_WPOS][3]; 436 const GLfloat wMid = vMid->attrib[FRAG_ATTRIB_WPOS][3]; 437 { 438 const GLfloat eMaj_dw = wMax - wMin; 439 const GLfloat eBot_dw = wMid - wMin; 440 span.attrStepX[FRAG_ATTRIB_WPOS][3] = oneOverArea * (eMaj_dw * eBot.dy - eMaj.dy * eBot_dw); 441 span.attrStepY[FRAG_ATTRIB_WPOS][3] = oneOverArea * (eMaj.dx * eBot_dw - eMaj_dw * eBot.dx); 442 } 443 ATTRIB_LOOP_BEGIN 444 if (swrast->_InterpMode[attr] == GL_FLAT) { 445 ASSIGN_4V(span.attrStepX[attr], 0.0, 0.0, 0.0, 0.0); 446 ASSIGN_4V(span.attrStepY[attr], 0.0, 0.0, 0.0, 0.0); 447 } 448 else { 449 GLuint c; 450 for (c = 0; c < 4; c++) { 451 GLfloat eMaj_da = vMax->attrib[attr][c] * wMax - vMin->attrib[attr][c] * wMin; 452 GLfloat eBot_da = vMid->attrib[attr][c] * wMid - vMin->attrib[attr][c] * wMin; 453 span.attrStepX[attr][c] = oneOverArea * (eMaj_da * eBot.dy - eMaj.dy * eBot_da); 454 span.attrStepY[attr][c] = oneOverArea * (eMaj.dx * eBot_da - eMaj_da * eBot.dx); 455 } 456 } 457 ATTRIB_LOOP_END 458 } 459#endif 460 461 /* 462 * We always sample at pixel centers. However, we avoid 463 * explicit half-pixel offsets in this code by incorporating 464 * the proper offset in each of x and y during the 465 * transformation to window coordinates. 466 * 467 * We also apply the usual rasterization rules to prevent 468 * cracks and overlaps. A pixel is considered inside a 469 * subtriangle if it meets all of four conditions: it is on or 470 * to the right of the left edge, strictly to the left of the 471 * right edge, on or below the top edge, and strictly above 472 * the bottom edge. (Some edges may be degenerate.) 473 * 474 * The following discussion assumes left-to-right scanning 475 * (that is, the major edge is on the left); the right-to-left 476 * case is a straightforward variation. 477 * 478 * We start by finding the half-integral y coordinate that is 479 * at or below the top of the triangle. This gives us the 480 * first scan line that could possibly contain pixels that are 481 * inside the triangle. 482 * 483 * Next we creep down the major edge until we reach that y, 484 * and compute the corresponding x coordinate on the edge. 485 * Then we find the half-integral x that lies on or just 486 * inside the edge. This is the first pixel that might lie in 487 * the interior of the triangle. (We won't know for sure 488 * until we check the other edges.) 489 * 490 * As we rasterize the triangle, we'll step down the major 491 * edge. For each step in y, we'll move an integer number 492 * of steps in x. There are two possible x step sizes, which 493 * we'll call the ``inner'' step (guaranteed to land on the 494 * edge or inside it) and the ``outer'' step (guaranteed to 495 * land on the edge or outside it). The inner and outer steps 496 * differ by one. During rasterization we maintain an error 497 * term that indicates our distance from the true edge, and 498 * select either the inner step or the outer step, whichever 499 * gets us to the first pixel that falls inside the triangle. 500 * 501 * All parameters (z, red, etc.) as well as the buffer 502 * addresses for color and z have inner and outer step values, 503 * so that we can increment them appropriately. This method 504 * eliminates the need to adjust parameters by creeping a 505 * sub-pixel amount into the triangle at each scanline. 506 */ 507 508 { 509 GLint subTriangle; 510 GLfixed fxLeftEdge = 0, fxRightEdge = 0; 511 GLfixed fdxLeftEdge = 0, fdxRightEdge = 0; 512 GLfixed fError = 0, fdError = 0; 513#ifdef PIXEL_ADDRESS 514 PIXEL_TYPE *pRow = NULL; 515 GLint dPRowOuter = 0, dPRowInner; /* offset in bytes */ 516#endif 517#ifdef INTERP_Z 518# ifdef DEPTH_TYPE 519 struct gl_renderbuffer *zrb 520 = ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer; 521 DEPTH_TYPE *zRow = NULL; 522 GLint dZRowOuter = 0, dZRowInner; /* offset in bytes */ 523# endif 524 GLuint zLeft = 0; 525 GLfixed fdzOuter = 0, fdzInner; 526#endif 527#ifdef INTERP_RGB 528 GLint rLeft = 0, fdrOuter = 0, fdrInner; 529 GLint gLeft = 0, fdgOuter = 0, fdgInner; 530 GLint bLeft = 0, fdbOuter = 0, fdbInner; 531#endif 532#ifdef INTERP_ALPHA 533 GLint aLeft = 0, fdaOuter = 0, fdaInner; 534#endif 535#ifdef INTERP_INDEX 536 GLfixed iLeft=0, diOuter=0, diInner; 537#endif 538#ifdef INTERP_INT_TEX 539 GLfixed sLeft=0, dsOuter=0, dsInner; 540 GLfixed tLeft=0, dtOuter=0, dtInner; 541#endif 542#ifdef INTERP_ATTRIBS 543 GLfloat wLeft = 0, dwOuter = 0, dwInner; 544 GLfloat attrLeft[FRAG_ATTRIB_MAX][4]; 545 GLfloat daOuter[FRAG_ATTRIB_MAX][4], daInner[FRAG_ATTRIB_MAX][4]; 546#endif 547 548 for (subTriangle=0; subTriangle<=1; subTriangle++) { 549 EdgeT *eLeft, *eRight; 550 int setupLeft, setupRight; 551 int lines; 552 553 if (subTriangle==0) { 554 /* bottom half */ 555 if (scan_from_left_to_right) { 556 eLeft = &eMaj; 557 eRight = &eBot; 558 lines = eRight->lines; 559 setupLeft = 1; 560 setupRight = 1; 561 } 562 else { 563 eLeft = &eBot; 564 eRight = &eMaj; 565 lines = eLeft->lines; 566 setupLeft = 1; 567 setupRight = 1; 568 } 569 } 570 else { 571 /* top half */ 572 if (scan_from_left_to_right) { 573 eLeft = &eMaj; 574 eRight = &eTop; 575 lines = eRight->lines; 576 setupLeft = 0; 577 setupRight = 1; 578 } 579 else { 580 eLeft = &eTop; 581 eRight = &eMaj; 582 lines = eLeft->lines; 583 setupLeft = 1; 584 setupRight = 0; 585 } 586 if (lines == 0) 587 return; 588 } 589 590 if (setupLeft && eLeft->lines > 0) { 591 const SWvertex *vLower = eLeft->v0; 592 const GLfixed fsy = eLeft->fsy; 593 const GLfixed fsx = eLeft->fsx; /* no fractional part */ 594 const GLfixed fx = FixedCeil(fsx); /* no fractional part */ 595 const GLfixed adjx = (GLfixed) (fx - eLeft->fx0); /* SCALED! */ 596 const GLfixed adjy = (GLfixed) eLeft->adjy; /* SCALED! */ 597 GLint idxOuter; 598 GLfloat dxOuter; 599 GLfixed fdxOuter; 600 601 fError = fx - fsx - FIXED_ONE; 602 fxLeftEdge = fsx - FIXED_EPSILON; 603 fdxLeftEdge = eLeft->fdxdy; 604 fdxOuter = FixedFloor(fdxLeftEdge - FIXED_EPSILON); 605 fdError = fdxOuter - fdxLeftEdge + FIXED_ONE; 606 idxOuter = FixedToInt(fdxOuter); 607 dxOuter = (GLfloat) idxOuter; 608 span.y = FixedToInt(fsy); 609 610 /* silence warnings on some compilers */ 611 (void) dxOuter; 612 (void) adjx; 613 (void) adjy; 614 (void) vLower; 615 616#ifdef PIXEL_ADDRESS 617 { 618 pRow = (PIXEL_TYPE *) PIXEL_ADDRESS(FixedToInt(fxLeftEdge), span.y); 619 dPRowOuter = -((int)BYTES_PER_ROW) + idxOuter * sizeof(PIXEL_TYPE); 620 /* negative because Y=0 at bottom and increases upward */ 621 } 622#endif 623 /* 624 * Now we need the set of parameter (z, color, etc.) values at 625 * the point (fx, fsy). This gives us properly-sampled parameter 626 * values that we can step from pixel to pixel. Furthermore, 627 * although we might have intermediate results that overflow 628 * the normal parameter range when we step temporarily outside 629 * the triangle, we shouldn't overflow or underflow for any 630 * pixel that's actually inside the triangle. 631 */ 632 633#ifdef INTERP_Z 634 { 635 GLfloat z0 = vLower->attrib[FRAG_ATTRIB_WPOS][2]; 636 if (depthBits <= 16) { 637 /* interpolate fixed-pt values */ 638 GLfloat tmp = (z0 * FIXED_SCALE 639 + span.attrStepX[FRAG_ATTRIB_WPOS][2] * adjx 640 + span.attrStepY[FRAG_ATTRIB_WPOS][2] * adjy) + FIXED_HALF; 641 if (tmp < MAX_GLUINT / 2) 642 zLeft = (GLfixed) tmp; 643 else 644 zLeft = MAX_GLUINT / 2; 645 fdzOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_WPOS][2] + 646 dxOuter * span.attrStepX[FRAG_ATTRIB_WPOS][2]); 647 } 648 else { 649 /* interpolate depth values w/out scaling */ 650 zLeft = (GLuint) (z0 + span.attrStepX[FRAG_ATTRIB_WPOS][2] * FixedToFloat(adjx) 651 + span.attrStepY[FRAG_ATTRIB_WPOS][2] * FixedToFloat(adjy)); 652 fdzOuter = (GLint) (span.attrStepY[FRAG_ATTRIB_WPOS][2] + 653 dxOuter * span.attrStepX[FRAG_ATTRIB_WPOS][2]); 654 } 655# ifdef DEPTH_TYPE 656 zRow = (DEPTH_TYPE *) 657 zrb->GetPointer(ctx, zrb, FixedToInt(fxLeftEdge), span.y); 658 dZRowOuter = (ctx->DrawBuffer->Width + idxOuter) * sizeof(DEPTH_TYPE); 659# endif 660 } 661#endif 662#ifdef INTERP_RGB 663 if (ctx->Light.ShadeModel == GL_SMOOTH) { 664 rLeft = (GLint)(ChanToFixed(vLower->color[RCOMP]) 665 + span.attrStepX[FRAG_ATTRIB_COL0][0] * adjx 666 + span.attrStepY[FRAG_ATTRIB_COL0][0] * adjy) + FIXED_HALF; 667 gLeft = (GLint)(ChanToFixed(vLower->color[GCOMP]) 668 + span.attrStepX[FRAG_ATTRIB_COL0][1] * adjx 669 + span.attrStepY[FRAG_ATTRIB_COL0][1] * adjy) + FIXED_HALF; 670 bLeft = (GLint)(ChanToFixed(vLower->color[BCOMP]) 671 + span.attrStepX[FRAG_ATTRIB_COL0][2] * adjx 672 + span.attrStepY[FRAG_ATTRIB_COL0][2] * adjy) + FIXED_HALF; 673 fdrOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][0] 674 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][0]); 675 fdgOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][1] 676 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][1]); 677 fdbOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][2] 678 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][2]); 679# ifdef INTERP_ALPHA 680 aLeft = (GLint)(ChanToFixed(vLower->color[ACOMP]) 681 + span.attrStepX[FRAG_ATTRIB_COL0][3] * adjx 682 + span.attrStepY[FRAG_ATTRIB_COL0][3] * adjy) + FIXED_HALF; 683 fdaOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][3] 684 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][3]); 685# endif 686 } 687 else { 688 ASSERT(ctx->Light.ShadeModel == GL_FLAT); 689 rLeft = ChanToFixed(v2->color[RCOMP]); 690 gLeft = ChanToFixed(v2->color[GCOMP]); 691 bLeft = ChanToFixed(v2->color[BCOMP]); 692 fdrOuter = fdgOuter = fdbOuter = 0; 693# ifdef INTERP_ALPHA 694 aLeft = ChanToFixed(v2->color[ACOMP]); 695 fdaOuter = 0; 696# endif 697 } 698#endif /* INTERP_RGB */ 699 700 701#ifdef INTERP_INDEX 702 if (ctx->Light.ShadeModel == GL_SMOOTH) { 703 iLeft = (GLfixed)(vLower->attrib[FRAG_ATTRIB_CI][0] * FIXED_SCALE 704 + didx * adjx + didy * adjy) + FIXED_HALF; 705 diOuter = SignedFloatToFixed(didy + dxOuter * didx); 706 } 707 else { 708 ASSERT(ctx->Light.ShadeModel == GL_FLAT); 709 iLeft = FloatToFixed(v2->attrib[FRAG_ATTRIB_CI][0]); 710 diOuter = 0; 711 } 712#endif 713#ifdef INTERP_INT_TEX 714 { 715 GLfloat s0, t0; 716 s0 = vLower->attrib[FRAG_ATTRIB_TEX0][0] * S_SCALE; 717 sLeft = (GLfixed)(s0 * FIXED_SCALE + span.attrStepX[FRAG_ATTRIB_TEX0][0] * adjx 718 + span.attrStepY[FRAG_ATTRIB_TEX0][0] * adjy) + FIXED_HALF; 719 dsOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_TEX0][0] 720 + dxOuter * span.attrStepX[FRAG_ATTRIB_TEX0][0]); 721 722 t0 = vLower->attrib[FRAG_ATTRIB_TEX0][1] * T_SCALE; 723 tLeft = (GLfixed)(t0 * FIXED_SCALE + span.attrStepX[FRAG_ATTRIB_TEX0][1] * adjx 724 + span.attrStepY[FRAG_ATTRIB_TEX0][1] * adjy) + FIXED_HALF; 725 dtOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_TEX0][1] 726 + dxOuter * span.attrStepX[FRAG_ATTRIB_TEX0][1]); 727 } 728#endif 729#ifdef INTERP_ATTRIBS 730 { 731 const GLuint attr = FRAG_ATTRIB_WPOS; 732 wLeft = vLower->attrib[FRAG_ATTRIB_WPOS][3] 733 + (span.attrStepX[attr][3] * adjx 734 + span.attrStepY[attr][3] * adjy) * (1.0F/FIXED_SCALE); 735 dwOuter = span.attrStepY[attr][3] + dxOuter * span.attrStepX[attr][3]; 736 } 737 ATTRIB_LOOP_BEGIN 738 const GLfloat invW = vLower->attrib[FRAG_ATTRIB_WPOS][3]; 739 if (swrast->_InterpMode[attr] == GL_FLAT) { 740 GLuint c; 741 for (c = 0; c < 4; c++) { 742 attrLeft[attr][c] = v2->attrib[attr][c] * invW; 743 daOuter[attr][c] = 0.0; 744 } 745 } 746 else { 747 GLuint c; 748 for (c = 0; c < 4; c++) { 749 const GLfloat a = vLower->attrib[attr][c] * invW; 750 attrLeft[attr][c] = a + ( span.attrStepX[attr][c] * adjx 751 + span.attrStepY[attr][c] * adjy) * (1.0F/FIXED_SCALE); 752 daOuter[attr][c] = span.attrStepY[attr][c] + dxOuter * span.attrStepX[attr][c]; 753 } 754 } 755 ATTRIB_LOOP_END 756#endif 757 } /*if setupLeft*/ 758 759 760 if (setupRight && eRight->lines>0) { 761 fxRightEdge = eRight->fsx - FIXED_EPSILON; 762 fdxRightEdge = eRight->fdxdy; 763 } 764 765 if (lines==0) { 766 continue; 767 } 768 769 770 /* Rasterize setup */ 771#ifdef PIXEL_ADDRESS 772 dPRowInner = dPRowOuter + sizeof(PIXEL_TYPE); 773#endif 774#ifdef INTERP_Z 775# ifdef DEPTH_TYPE 776 dZRowInner = dZRowOuter + sizeof(DEPTH_TYPE); 777# endif 778 fdzInner = fdzOuter + span.zStep; 779#endif 780#ifdef INTERP_RGB 781 fdrInner = fdrOuter + span.redStep; 782 fdgInner = fdgOuter + span.greenStep; 783 fdbInner = fdbOuter + span.blueStep; 784#endif 785#ifdef INTERP_ALPHA 786 fdaInner = fdaOuter + span.alphaStep; 787#endif 788#ifdef INTERP_INDEX 789 diInner = diOuter + span.indexStep; 790#endif 791#ifdef INTERP_INT_TEX 792 dsInner = dsOuter + span.intTexStep[0]; 793 dtInner = dtOuter + span.intTexStep[1]; 794#endif 795#ifdef INTERP_ATTRIBS 796 dwInner = dwOuter + span.attrStepX[FRAG_ATTRIB_WPOS][3]; 797 ATTRIB_LOOP_BEGIN 798 GLuint c; 799 for (c = 0; c < 4; c++) { 800 daInner[attr][c] = daOuter[attr][c] + span.attrStepX[attr][c]; 801 } 802 ATTRIB_LOOP_END 803#endif 804 805 while (lines > 0) { 806 /* initialize the span interpolants to the leftmost value */ 807 /* ff = fixed-pt fragment */ 808 const GLint right = FixedToInt(fxRightEdge); 809 span.x = FixedToInt(fxLeftEdge); 810 if (right <= span.x) 811 span.end = 0; 812 else 813 span.end = right - span.x; 814 815#ifdef INTERP_Z 816 span.z = zLeft; 817#endif 818#ifdef INTERP_RGB 819 span.red = rLeft; 820 span.green = gLeft; 821 span.blue = bLeft; 822#endif 823#ifdef INTERP_ALPHA 824 span.alpha = aLeft; 825#endif 826#ifdef INTERP_INDEX 827 span.index = iLeft; 828#endif 829#ifdef INTERP_INT_TEX 830 span.intTex[0] = sLeft; 831 span.intTex[1] = tLeft; 832#endif 833 834#ifdef INTERP_ATTRIBS 835 span.attrStart[FRAG_ATTRIB_WPOS][3] = wLeft; 836 ATTRIB_LOOP_BEGIN 837 GLuint c; 838 for (c = 0; c < 4; c++) { 839 span.attrStart[attr][c] = attrLeft[attr][c]; 840 } 841 ATTRIB_LOOP_END 842#endif 843 844 /* This is where we actually generate fragments */ 845 /* XXX the test for span.y > 0 _shouldn't_ be needed but 846 * it fixes a problem on 64-bit Opterons (bug 4842). 847 */ 848 if (span.end > 0 && span.y >= 0) { 849 const GLint len = span.end - 1; 850 (void) len; 851#ifdef INTERP_RGB 852 CLAMP_INTERPOLANT(red, redStep, len); 853 CLAMP_INTERPOLANT(green, greenStep, len); 854 CLAMP_INTERPOLANT(blue, blueStep, len); 855#endif 856#ifdef INTERP_ALPHA 857 CLAMP_INTERPOLANT(alpha, alphaStep, len); 858#endif 859#ifdef INTERP_INDEX 860 CLAMP_INTERPOLANT(index, indexStep, len); 861#endif 862 { 863 RENDER_SPAN( span ); 864 } 865 } 866 867 /* 868 * Advance to the next scan line. Compute the 869 * new edge coordinates, and adjust the 870 * pixel-center x coordinate so that it stays 871 * on or inside the major edge. 872 */ 873 span.y++; 874 lines--; 875 876 fxLeftEdge += fdxLeftEdge; 877 fxRightEdge += fdxRightEdge; 878 879 fError += fdError; 880 if (fError >= 0) { 881 fError -= FIXED_ONE; 882 883#ifdef PIXEL_ADDRESS 884 pRow = (PIXEL_TYPE *) ((GLubyte *) pRow + dPRowOuter); 885#endif 886#ifdef INTERP_Z 887# ifdef DEPTH_TYPE 888 zRow = (DEPTH_TYPE *) ((GLubyte *) zRow + dZRowOuter); 889# endif 890 zLeft += fdzOuter; 891#endif 892#ifdef INTERP_RGB 893 rLeft += fdrOuter; 894 gLeft += fdgOuter; 895 bLeft += fdbOuter; 896#endif 897#ifdef INTERP_ALPHA 898 aLeft += fdaOuter; 899#endif 900#ifdef INTERP_INDEX 901 iLeft += diOuter; 902#endif 903#ifdef INTERP_INT_TEX 904 sLeft += dsOuter; 905 tLeft += dtOuter; 906#endif 907#ifdef INTERP_ATTRIBS 908 wLeft += dwOuter; 909 ATTRIB_LOOP_BEGIN 910 GLuint c; 911 for (c = 0; c < 4; c++) { 912 attrLeft[attr][c] += daOuter[attr][c]; 913 } 914 ATTRIB_LOOP_END 915#endif 916 } 917 else { 918#ifdef PIXEL_ADDRESS 919 pRow = (PIXEL_TYPE *) ((GLubyte *) pRow + dPRowInner); 920#endif 921#ifdef INTERP_Z 922# ifdef DEPTH_TYPE 923 zRow = (DEPTH_TYPE *) ((GLubyte *) zRow + dZRowInner); 924# endif 925 zLeft += fdzInner; 926#endif 927#ifdef INTERP_RGB 928 rLeft += fdrInner; 929 gLeft += fdgInner; 930 bLeft += fdbInner; 931#endif 932#ifdef INTERP_ALPHA 933 aLeft += fdaInner; 934#endif 935#ifdef INTERP_INDEX 936 iLeft += diInner; 937#endif 938#ifdef INTERP_INT_TEX 939 sLeft += dsInner; 940 tLeft += dtInner; 941#endif 942#ifdef INTERP_ATTRIBS 943 wLeft += dwInner; 944 ATTRIB_LOOP_BEGIN 945 GLuint c; 946 for (c = 0; c < 4; c++) { 947 attrLeft[attr][c] += daInner[attr][c]; 948 } 949 ATTRIB_LOOP_END 950#endif 951 } 952 } /*while lines>0*/ 953 954 } /* for subTriangle */ 955 956 } 957 } 958} 959 960#undef SETUP_CODE 961#undef RENDER_SPAN 962 963#undef PIXEL_TYPE 964#undef BYTES_PER_ROW 965#undef PIXEL_ADDRESS 966#undef DEPTH_TYPE 967 968#undef INTERP_Z 969#undef INTERP_RGB 970#undef INTERP_ALPHA 971#undef INTERP_INDEX 972#undef INTERP_INT_TEX 973#undef INTERP_ATTRIBS 974 975#undef S_SCALE 976#undef T_SCALE 977 978#undef FixedToDepth 979 980#undef NAME 981