s_tritemp.h revision c1f859d4
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)->_BackfaceSign; 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; 237 238 if (IS_INF_OR_NAN(area) || area == 0.0F) 239 return; 240 241 if (area * bf * swrast->_BackfaceCullSign < 0.0) 242 return; 243 244 oneOverArea = 1.0F / area; 245 246 /* 0 = front, 1 = back */ 247 span.facing = oneOverArea * bf > 0.0F; 248 } 249 250 /* Edge setup. For a triangle strip these could be reused... */ 251 { 252 eMaj.fsy = FixedCeil(vMin_fy); 253 eMaj.lines = FixedToInt(FixedCeil(vMax_fy - eMaj.fsy)); 254 if (eMaj.lines > 0) { 255 eMaj.dxdy = eMaj.dx / eMaj.dy; 256 eMaj.fdxdy = SignedFloatToFixed(eMaj.dxdy); 257 eMaj.adjy = (GLfloat) (eMaj.fsy - vMin_fy); /* SCALED! */ 258 eMaj.fx0 = vMin_fx; 259 eMaj.fsx = eMaj.fx0 + (GLfixed) (eMaj.adjy * eMaj.dxdy); 260 } 261 else { 262 return; /*CULLED*/ 263 } 264 265 eTop.fsy = FixedCeil(vMid_fy); 266 eTop.lines = FixedToInt(FixedCeil(vMax_fy - eTop.fsy)); 267 if (eTop.lines > 0) { 268 eTop.dxdy = eTop.dx / eTop.dy; 269 eTop.fdxdy = SignedFloatToFixed(eTop.dxdy); 270 eTop.adjy = (GLfloat) (eTop.fsy - vMid_fy); /* SCALED! */ 271 eTop.fx0 = vMid_fx; 272 eTop.fsx = eTop.fx0 + (GLfixed) (eTop.adjy * eTop.dxdy); 273 } 274 275 eBot.fsy = FixedCeil(vMin_fy); 276 eBot.lines = FixedToInt(FixedCeil(vMid_fy - eBot.fsy)); 277 if (eBot.lines > 0) { 278 eBot.dxdy = eBot.dx / eBot.dy; 279 eBot.fdxdy = SignedFloatToFixed(eBot.dxdy); 280 eBot.adjy = (GLfloat) (eBot.fsy - vMin_fy); /* SCALED! */ 281 eBot.fx0 = vMin_fx; 282 eBot.fsx = eBot.fx0 + (GLfixed) (eBot.adjy * eBot.dxdy); 283 } 284 } 285 286 /* 287 * Conceptually, we view a triangle as two subtriangles 288 * separated by a perfectly horizontal line. The edge that is 289 * intersected by this line is one with maximal absolute dy; we 290 * call it a ``major'' edge. The other two edges are the 291 * ``top'' edge (for the upper subtriangle) and the ``bottom'' 292 * edge (for the lower subtriangle). If either of these two 293 * edges is horizontal or very close to horizontal, the 294 * corresponding subtriangle might cover zero sample points; 295 * we take care to handle such cases, for performance as well 296 * as correctness. 297 * 298 * By stepping rasterization parameters along the major edge, 299 * we can avoid recomputing them at the discontinuity where 300 * the top and bottom edges meet. However, this forces us to 301 * be able to scan both left-to-right and right-to-left. 302 * Also, we must determine whether the major edge is at the 303 * left or right side of the triangle. We do this by 304 * computing the magnitude of the cross-product of the major 305 * and top edges. Since this magnitude depends on the sine of 306 * the angle between the two edges, its sign tells us whether 307 * we turn to the left or to the right when travelling along 308 * the major edge to the top edge, and from this we infer 309 * whether the major edge is on the left or the right. 310 * 311 * Serendipitously, this cross-product magnitude is also a 312 * value we need to compute the iteration parameter 313 * derivatives for the triangle, and it can be used to perform 314 * backface culling because its sign tells us whether the 315 * triangle is clockwise or counterclockwise. In this code we 316 * refer to it as ``area'' because it's also proportional to 317 * the pixel area of the triangle. 318 */ 319 320 { 321 GLint scan_from_left_to_right; /* true if scanning left-to-right */ 322#ifdef INTERP_INDEX 323 GLfloat didx, didy; 324#endif 325 326 /* 327 * Execute user-supplied setup code 328 */ 329#ifdef SETUP_CODE 330 SETUP_CODE 331#endif 332 333 scan_from_left_to_right = (oneOverArea < 0.0F); 334 335 336 /* compute d?/dx and d?/dy derivatives */ 337#ifdef INTERP_Z 338 span.interpMask |= SPAN_Z; 339 { 340 GLfloat eMaj_dz = vMax->attrib[FRAG_ATTRIB_WPOS][2] - vMin->attrib[FRAG_ATTRIB_WPOS][2]; 341 GLfloat eBot_dz = vMid->attrib[FRAG_ATTRIB_WPOS][2] - vMin->attrib[FRAG_ATTRIB_WPOS][2]; 342 span.attrStepX[FRAG_ATTRIB_WPOS][2] = oneOverArea * (eMaj_dz * eBot.dy - eMaj.dy * eBot_dz); 343 if (span.attrStepX[FRAG_ATTRIB_WPOS][2] > maxDepth || 344 span.attrStepX[FRAG_ATTRIB_WPOS][2] < -maxDepth) { 345 /* probably a sliver triangle */ 346 span.attrStepX[FRAG_ATTRIB_WPOS][2] = 0.0; 347 span.attrStepY[FRAG_ATTRIB_WPOS][2] = 0.0; 348 } 349 else { 350 span.attrStepY[FRAG_ATTRIB_WPOS][2] = oneOverArea * (eMaj.dx * eBot_dz - eMaj_dz * eBot.dx); 351 } 352 if (depthBits <= 16) 353 span.zStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_WPOS][2]); 354 else 355 span.zStep = (GLint) span.attrStepX[FRAG_ATTRIB_WPOS][2]; 356 } 357#endif 358#ifdef INTERP_RGB 359 span.interpMask |= SPAN_RGBA; 360 if (ctx->Light.ShadeModel == GL_SMOOTH) { 361 GLfloat eMaj_dr = (GLfloat) (vMax->color[RCOMP] - vMin->color[RCOMP]); 362 GLfloat eBot_dr = (GLfloat) (vMid->color[RCOMP] - vMin->color[RCOMP]); 363 GLfloat eMaj_dg = (GLfloat) (vMax->color[GCOMP] - vMin->color[GCOMP]); 364 GLfloat eBot_dg = (GLfloat) (vMid->color[GCOMP] - vMin->color[GCOMP]); 365 GLfloat eMaj_db = (GLfloat) (vMax->color[BCOMP] - vMin->color[BCOMP]); 366 GLfloat eBot_db = (GLfloat) (vMid->color[BCOMP] - vMin->color[BCOMP]); 367# ifdef INTERP_ALPHA 368 GLfloat eMaj_da = (GLfloat) (vMax->color[ACOMP] - vMin->color[ACOMP]); 369 GLfloat eBot_da = (GLfloat) (vMid->color[ACOMP] - vMin->color[ACOMP]); 370# endif 371 span.attrStepX[FRAG_ATTRIB_COL0][0] = oneOverArea * (eMaj_dr * eBot.dy - eMaj.dy * eBot_dr); 372 span.attrStepY[FRAG_ATTRIB_COL0][0] = oneOverArea * (eMaj.dx * eBot_dr - eMaj_dr * eBot.dx); 373 span.attrStepX[FRAG_ATTRIB_COL0][1] = oneOverArea * (eMaj_dg * eBot.dy - eMaj.dy * eBot_dg); 374 span.attrStepY[FRAG_ATTRIB_COL0][1] = oneOverArea * (eMaj.dx * eBot_dg - eMaj_dg * eBot.dx); 375 span.attrStepX[FRAG_ATTRIB_COL0][2] = oneOverArea * (eMaj_db * eBot.dy - eMaj.dy * eBot_db); 376 span.attrStepY[FRAG_ATTRIB_COL0][2] = oneOverArea * (eMaj.dx * eBot_db - eMaj_db * eBot.dx); 377 span.redStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][0]); 378 span.greenStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][1]); 379 span.blueStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][2]); 380# ifdef INTERP_ALPHA 381 span.attrStepX[FRAG_ATTRIB_COL0][3] = oneOverArea * (eMaj_da * eBot.dy - eMaj.dy * eBot_da); 382 span.attrStepY[FRAG_ATTRIB_COL0][3] = oneOverArea * (eMaj.dx * eBot_da - eMaj_da * eBot.dx); 383 span.alphaStep = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_COL0][3]); 384# endif /* INTERP_ALPHA */ 385 } 386 else { 387 ASSERT(ctx->Light.ShadeModel == GL_FLAT); 388 span.interpMask |= SPAN_FLAT; 389 span.attrStepX[FRAG_ATTRIB_COL0][0] = span.attrStepY[FRAG_ATTRIB_COL0][0] = 0.0F; 390 span.attrStepX[FRAG_ATTRIB_COL0][1] = span.attrStepY[FRAG_ATTRIB_COL0][1] = 0.0F; 391 span.attrStepX[FRAG_ATTRIB_COL0][2] = span.attrStepY[FRAG_ATTRIB_COL0][2] = 0.0F; 392 span.redStep = 0; 393 span.greenStep = 0; 394 span.blueStep = 0; 395# ifdef INTERP_ALPHA 396 span.attrStepX[FRAG_ATTRIB_COL0][3] = span.attrStepY[FRAG_ATTRIB_COL0][3] = 0.0F; 397 span.alphaStep = 0; 398# endif 399 } 400#endif /* INTERP_RGB */ 401#ifdef INTERP_INDEX 402 span.interpMask |= SPAN_INDEX; 403 if (ctx->Light.ShadeModel == GL_SMOOTH) { 404 GLfloat eMaj_di = vMax->attrib[FRAG_ATTRIB_CI][0] - vMin->attrib[FRAG_ATTRIB_CI][0]; 405 GLfloat eBot_di = vMid->attrib[FRAG_ATTRIB_CI][0] - vMin->attrib[FRAG_ATTRIB_CI][0]; 406 didx = oneOverArea * (eMaj_di * eBot.dy - eMaj.dy * eBot_di); 407 didy = oneOverArea * (eMaj.dx * eBot_di - eMaj_di * eBot.dx); 408 span.indexStep = SignedFloatToFixed(didx); 409 } 410 else { 411 span.interpMask |= SPAN_FLAT; 412 didx = didy = 0.0F; 413 span.indexStep = 0; 414 } 415#endif 416#ifdef INTERP_INT_TEX 417 { 418 GLfloat eMaj_ds = (vMax->attrib[FRAG_ATTRIB_TEX0][0] - vMin->attrib[FRAG_ATTRIB_TEX0][0]) * S_SCALE; 419 GLfloat eBot_ds = (vMid->attrib[FRAG_ATTRIB_TEX0][0] - vMin->attrib[FRAG_ATTRIB_TEX0][0]) * S_SCALE; 420 GLfloat eMaj_dt = (vMax->attrib[FRAG_ATTRIB_TEX0][1] - vMin->attrib[FRAG_ATTRIB_TEX0][1]) * T_SCALE; 421 GLfloat eBot_dt = (vMid->attrib[FRAG_ATTRIB_TEX0][1] - vMin->attrib[FRAG_ATTRIB_TEX0][1]) * T_SCALE; 422 span.attrStepX[FRAG_ATTRIB_TEX0][0] = oneOverArea * (eMaj_ds * eBot.dy - eMaj.dy * eBot_ds); 423 span.attrStepY[FRAG_ATTRIB_TEX0][0] = oneOverArea * (eMaj.dx * eBot_ds - eMaj_ds * eBot.dx); 424 span.attrStepX[FRAG_ATTRIB_TEX0][1] = oneOverArea * (eMaj_dt * eBot.dy - eMaj.dy * eBot_dt); 425 span.attrStepY[FRAG_ATTRIB_TEX0][1] = oneOverArea * (eMaj.dx * eBot_dt - eMaj_dt * eBot.dx); 426 span.intTexStep[0] = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_TEX0][0]); 427 span.intTexStep[1] = SignedFloatToFixed(span.attrStepX[FRAG_ATTRIB_TEX0][1]); 428 } 429#endif 430#ifdef INTERP_ATTRIBS 431 { 432 /* attrib[FRAG_ATTRIB_WPOS][3] is 1/W */ 433 const GLfloat wMax = vMax->attrib[FRAG_ATTRIB_WPOS][3]; 434 const GLfloat wMin = vMin->attrib[FRAG_ATTRIB_WPOS][3]; 435 const GLfloat wMid = vMid->attrib[FRAG_ATTRIB_WPOS][3]; 436 { 437 const GLfloat eMaj_dw = wMax - wMin; 438 const GLfloat eBot_dw = wMid - wMin; 439 span.attrStepX[FRAG_ATTRIB_WPOS][3] = oneOverArea * (eMaj_dw * eBot.dy - eMaj.dy * eBot_dw); 440 span.attrStepY[FRAG_ATTRIB_WPOS][3] = oneOverArea * (eMaj.dx * eBot_dw - eMaj_dw * eBot.dx); 441 } 442 ATTRIB_LOOP_BEGIN 443 if (swrast->_InterpMode[attr] == GL_FLAT) { 444 ASSIGN_4V(span.attrStepX[attr], 0.0, 0.0, 0.0, 0.0); 445 ASSIGN_4V(span.attrStepY[attr], 0.0, 0.0, 0.0, 0.0); 446 } 447 else { 448 GLuint c; 449 for (c = 0; c < 4; c++) { 450 GLfloat eMaj_da = vMax->attrib[attr][c] * wMax - vMin->attrib[attr][c] * wMin; 451 GLfloat eBot_da = vMid->attrib[attr][c] * wMid - vMin->attrib[attr][c] * wMin; 452 span.attrStepX[attr][c] = oneOverArea * (eMaj_da * eBot.dy - eMaj.dy * eBot_da); 453 span.attrStepY[attr][c] = oneOverArea * (eMaj.dx * eBot_da - eMaj_da * eBot.dx); 454 } 455 } 456 ATTRIB_LOOP_END 457 } 458#endif 459 460 /* 461 * We always sample at pixel centers. However, we avoid 462 * explicit half-pixel offsets in this code by incorporating 463 * the proper offset in each of x and y during the 464 * transformation to window coordinates. 465 * 466 * We also apply the usual rasterization rules to prevent 467 * cracks and overlaps. A pixel is considered inside a 468 * subtriangle if it meets all of four conditions: it is on or 469 * to the right of the left edge, strictly to the left of the 470 * right edge, on or below the top edge, and strictly above 471 * the bottom edge. (Some edges may be degenerate.) 472 * 473 * The following discussion assumes left-to-right scanning 474 * (that is, the major edge is on the left); the right-to-left 475 * case is a straightforward variation. 476 * 477 * We start by finding the half-integral y coordinate that is 478 * at or below the top of the triangle. This gives us the 479 * first scan line that could possibly contain pixels that are 480 * inside the triangle. 481 * 482 * Next we creep down the major edge until we reach that y, 483 * and compute the corresponding x coordinate on the edge. 484 * Then we find the half-integral x that lies on or just 485 * inside the edge. This is the first pixel that might lie in 486 * the interior of the triangle. (We won't know for sure 487 * until we check the other edges.) 488 * 489 * As we rasterize the triangle, we'll step down the major 490 * edge. For each step in y, we'll move an integer number 491 * of steps in x. There are two possible x step sizes, which 492 * we'll call the ``inner'' step (guaranteed to land on the 493 * edge or inside it) and the ``outer'' step (guaranteed to 494 * land on the edge or outside it). The inner and outer steps 495 * differ by one. During rasterization we maintain an error 496 * term that indicates our distance from the true edge, and 497 * select either the inner step or the outer step, whichever 498 * gets us to the first pixel that falls inside the triangle. 499 * 500 * All parameters (z, red, etc.) as well as the buffer 501 * addresses for color and z have inner and outer step values, 502 * so that we can increment them appropriately. This method 503 * eliminates the need to adjust parameters by creeping a 504 * sub-pixel amount into the triangle at each scanline. 505 */ 506 507 { 508 GLint subTriangle; 509 GLfixed fxLeftEdge = 0, fxRightEdge = 0; 510 GLfixed fdxLeftEdge = 0, fdxRightEdge = 0; 511 GLfixed fError = 0, fdError = 0; 512#ifdef PIXEL_ADDRESS 513 PIXEL_TYPE *pRow = NULL; 514 GLint dPRowOuter = 0, dPRowInner; /* offset in bytes */ 515#endif 516#ifdef INTERP_Z 517# ifdef DEPTH_TYPE 518 struct gl_renderbuffer *zrb 519 = ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer; 520 DEPTH_TYPE *zRow = NULL; 521 GLint dZRowOuter = 0, dZRowInner; /* offset in bytes */ 522# endif 523 GLuint zLeft = 0; 524 GLfixed fdzOuter = 0, fdzInner; 525#endif 526#ifdef INTERP_RGB 527 GLint rLeft = 0, fdrOuter = 0, fdrInner; 528 GLint gLeft = 0, fdgOuter = 0, fdgInner; 529 GLint bLeft = 0, fdbOuter = 0, fdbInner; 530#endif 531#ifdef INTERP_ALPHA 532 GLint aLeft = 0, fdaOuter = 0, fdaInner; 533#endif 534#ifdef INTERP_INDEX 535 GLfixed iLeft=0, diOuter=0, diInner; 536#endif 537#ifdef INTERP_INT_TEX 538 GLfixed sLeft=0, dsOuter=0, dsInner; 539 GLfixed tLeft=0, dtOuter=0, dtInner; 540#endif 541#ifdef INTERP_ATTRIBS 542 GLfloat wLeft = 0, dwOuter = 0, dwInner; 543 GLfloat attrLeft[FRAG_ATTRIB_MAX][4]; 544 GLfloat daOuter[FRAG_ATTRIB_MAX][4], daInner[FRAG_ATTRIB_MAX][4]; 545#endif 546 547 for (subTriangle=0; subTriangle<=1; subTriangle++) { 548 EdgeT *eLeft, *eRight; 549 int setupLeft, setupRight; 550 int lines; 551 552 if (subTriangle==0) { 553 /* bottom half */ 554 if (scan_from_left_to_right) { 555 eLeft = &eMaj; 556 eRight = &eBot; 557 lines = eRight->lines; 558 setupLeft = 1; 559 setupRight = 1; 560 } 561 else { 562 eLeft = &eBot; 563 eRight = &eMaj; 564 lines = eLeft->lines; 565 setupLeft = 1; 566 setupRight = 1; 567 } 568 } 569 else { 570 /* top half */ 571 if (scan_from_left_to_right) { 572 eLeft = &eMaj; 573 eRight = &eTop; 574 lines = eRight->lines; 575 setupLeft = 0; 576 setupRight = 1; 577 } 578 else { 579 eLeft = &eTop; 580 eRight = &eMaj; 581 lines = eLeft->lines; 582 setupLeft = 1; 583 setupRight = 0; 584 } 585 if (lines == 0) 586 return; 587 } 588 589 if (setupLeft && eLeft->lines > 0) { 590 const SWvertex *vLower = eLeft->v0; 591 const GLfixed fsy = eLeft->fsy; 592 const GLfixed fsx = eLeft->fsx; /* no fractional part */ 593 const GLfixed fx = FixedCeil(fsx); /* no fractional part */ 594 const GLfixed adjx = (GLfixed) (fx - eLeft->fx0); /* SCALED! */ 595 const GLfixed adjy = (GLfixed) eLeft->adjy; /* SCALED! */ 596 GLint idxOuter; 597 GLfloat dxOuter; 598 GLfixed fdxOuter; 599 600 fError = fx - fsx - FIXED_ONE; 601 fxLeftEdge = fsx - FIXED_EPSILON; 602 fdxLeftEdge = eLeft->fdxdy; 603 fdxOuter = FixedFloor(fdxLeftEdge - FIXED_EPSILON); 604 fdError = fdxOuter - fdxLeftEdge + FIXED_ONE; 605 idxOuter = FixedToInt(fdxOuter); 606 dxOuter = (GLfloat) idxOuter; 607 span.y = FixedToInt(fsy); 608 609 /* silence warnings on some compilers */ 610 (void) dxOuter; 611 (void) adjx; 612 (void) adjy; 613 (void) vLower; 614 615#ifdef PIXEL_ADDRESS 616 { 617 pRow = (PIXEL_TYPE *) PIXEL_ADDRESS(FixedToInt(fxLeftEdge), span.y); 618 dPRowOuter = -((int)BYTES_PER_ROW) + idxOuter * sizeof(PIXEL_TYPE); 619 /* negative because Y=0 at bottom and increases upward */ 620 } 621#endif 622 /* 623 * Now we need the set of parameter (z, color, etc.) values at 624 * the point (fx, fsy). This gives us properly-sampled parameter 625 * values that we can step from pixel to pixel. Furthermore, 626 * although we might have intermediate results that overflow 627 * the normal parameter range when we step temporarily outside 628 * the triangle, we shouldn't overflow or underflow for any 629 * pixel that's actually inside the triangle. 630 */ 631 632#ifdef INTERP_Z 633 { 634 GLfloat z0 = vLower->attrib[FRAG_ATTRIB_WPOS][2]; 635 if (depthBits <= 16) { 636 /* interpolate fixed-pt values */ 637 GLfloat tmp = (z0 * FIXED_SCALE 638 + span.attrStepX[FRAG_ATTRIB_WPOS][2] * adjx 639 + span.attrStepY[FRAG_ATTRIB_WPOS][2] * adjy) + FIXED_HALF; 640 if (tmp < MAX_GLUINT / 2) 641 zLeft = (GLfixed) tmp; 642 else 643 zLeft = MAX_GLUINT / 2; 644 fdzOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_WPOS][2] + 645 dxOuter * span.attrStepX[FRAG_ATTRIB_WPOS][2]); 646 } 647 else { 648 /* interpolate depth values w/out scaling */ 649 zLeft = (GLuint) (z0 + span.attrStepX[FRAG_ATTRIB_WPOS][2] * FixedToFloat(adjx) 650 + span.attrStepY[FRAG_ATTRIB_WPOS][2] * FixedToFloat(adjy)); 651 fdzOuter = (GLint) (span.attrStepY[FRAG_ATTRIB_WPOS][2] + 652 dxOuter * span.attrStepX[FRAG_ATTRIB_WPOS][2]); 653 } 654# ifdef DEPTH_TYPE 655 zRow = (DEPTH_TYPE *) 656 zrb->GetPointer(ctx, zrb, FixedToInt(fxLeftEdge), span.y); 657 dZRowOuter = (ctx->DrawBuffer->Width + idxOuter) * sizeof(DEPTH_TYPE); 658# endif 659 } 660#endif 661#ifdef INTERP_RGB 662 if (ctx->Light.ShadeModel == GL_SMOOTH) { 663 rLeft = (GLint)(ChanToFixed(vLower->color[RCOMP]) 664 + span.attrStepX[FRAG_ATTRIB_COL0][0] * adjx 665 + span.attrStepY[FRAG_ATTRIB_COL0][0] * adjy) + FIXED_HALF; 666 gLeft = (GLint)(ChanToFixed(vLower->color[GCOMP]) 667 + span.attrStepX[FRAG_ATTRIB_COL0][1] * adjx 668 + span.attrStepY[FRAG_ATTRIB_COL0][1] * adjy) + FIXED_HALF; 669 bLeft = (GLint)(ChanToFixed(vLower->color[BCOMP]) 670 + span.attrStepX[FRAG_ATTRIB_COL0][2] * adjx 671 + span.attrStepY[FRAG_ATTRIB_COL0][2] * adjy) + FIXED_HALF; 672 fdrOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][0] 673 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][0]); 674 fdgOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][1] 675 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][1]); 676 fdbOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][2] 677 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][2]); 678# ifdef INTERP_ALPHA 679 aLeft = (GLint)(ChanToFixed(vLower->color[ACOMP]) 680 + span.attrStepX[FRAG_ATTRIB_COL0][3] * adjx 681 + span.attrStepY[FRAG_ATTRIB_COL0][3] * adjy) + FIXED_HALF; 682 fdaOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_COL0][3] 683 + dxOuter * span.attrStepX[FRAG_ATTRIB_COL0][3]); 684# endif 685 } 686 else { 687 ASSERT(ctx->Light.ShadeModel == GL_FLAT); 688 rLeft = ChanToFixed(v2->color[RCOMP]); 689 gLeft = ChanToFixed(v2->color[GCOMP]); 690 bLeft = ChanToFixed(v2->color[BCOMP]); 691 fdrOuter = fdgOuter = fdbOuter = 0; 692# ifdef INTERP_ALPHA 693 aLeft = ChanToFixed(v2->color[ACOMP]); 694 fdaOuter = 0; 695# endif 696 } 697#endif /* INTERP_RGB */ 698 699 700#ifdef INTERP_INDEX 701 if (ctx->Light.ShadeModel == GL_SMOOTH) { 702 iLeft = (GLfixed)(vLower->attrib[FRAG_ATTRIB_CI][0] * FIXED_SCALE 703 + didx * adjx + didy * adjy) + FIXED_HALF; 704 diOuter = SignedFloatToFixed(didy + dxOuter * didx); 705 } 706 else { 707 ASSERT(ctx->Light.ShadeModel == GL_FLAT); 708 iLeft = FloatToFixed(v2->attrib[FRAG_ATTRIB_CI][0]); 709 diOuter = 0; 710 } 711#endif 712#ifdef INTERP_INT_TEX 713 { 714 GLfloat s0, t0; 715 s0 = vLower->attrib[FRAG_ATTRIB_TEX0][0] * S_SCALE; 716 sLeft = (GLfixed)(s0 * FIXED_SCALE + span.attrStepX[FRAG_ATTRIB_TEX0][0] * adjx 717 + span.attrStepY[FRAG_ATTRIB_TEX0][0] * adjy) + FIXED_HALF; 718 dsOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_TEX0][0] 719 + dxOuter * span.attrStepX[FRAG_ATTRIB_TEX0][0]); 720 721 t0 = vLower->attrib[FRAG_ATTRIB_TEX0][1] * T_SCALE; 722 tLeft = (GLfixed)(t0 * FIXED_SCALE + span.attrStepX[FRAG_ATTRIB_TEX0][1] * adjx 723 + span.attrStepY[FRAG_ATTRIB_TEX0][1] * adjy) + FIXED_HALF; 724 dtOuter = SignedFloatToFixed(span.attrStepY[FRAG_ATTRIB_TEX0][1] 725 + dxOuter * span.attrStepX[FRAG_ATTRIB_TEX0][1]); 726 } 727#endif 728#ifdef INTERP_ATTRIBS 729 { 730 const GLuint attr = FRAG_ATTRIB_WPOS; 731 wLeft = vLower->attrib[FRAG_ATTRIB_WPOS][3] 732 + (span.attrStepX[attr][3] * adjx 733 + span.attrStepY[attr][3] * adjy) * (1.0F/FIXED_SCALE); 734 dwOuter = span.attrStepY[attr][3] + dxOuter * span.attrStepX[attr][3]; 735 } 736 ATTRIB_LOOP_BEGIN 737 const GLfloat invW = vLower->attrib[FRAG_ATTRIB_WPOS][3]; 738 if (swrast->_InterpMode[attr] == GL_FLAT) { 739 GLuint c; 740 for (c = 0; c < 4; c++) { 741 attrLeft[attr][c] = v2->attrib[attr][c] * invW; 742 daOuter[attr][c] = 0.0; 743 } 744 } 745 else { 746 GLuint c; 747 for (c = 0; c < 4; c++) { 748 const GLfloat a = vLower->attrib[attr][c] * invW; 749 attrLeft[attr][c] = a + ( span.attrStepX[attr][c] * adjx 750 + span.attrStepY[attr][c] * adjy) * (1.0F/FIXED_SCALE); 751 daOuter[attr][c] = span.attrStepY[attr][c] + dxOuter * span.attrStepX[attr][c]; 752 } 753 } 754 ATTRIB_LOOP_END 755#endif 756 } /*if setupLeft*/ 757 758 759 if (setupRight && eRight->lines>0) { 760 fxRightEdge = eRight->fsx - FIXED_EPSILON; 761 fdxRightEdge = eRight->fdxdy; 762 } 763 764 if (lines==0) { 765 continue; 766 } 767 768 769 /* Rasterize setup */ 770#ifdef PIXEL_ADDRESS 771 dPRowInner = dPRowOuter + sizeof(PIXEL_TYPE); 772#endif 773#ifdef INTERP_Z 774# ifdef DEPTH_TYPE 775 dZRowInner = dZRowOuter + sizeof(DEPTH_TYPE); 776# endif 777 fdzInner = fdzOuter + span.zStep; 778#endif 779#ifdef INTERP_RGB 780 fdrInner = fdrOuter + span.redStep; 781 fdgInner = fdgOuter + span.greenStep; 782 fdbInner = fdbOuter + span.blueStep; 783#endif 784#ifdef INTERP_ALPHA 785 fdaInner = fdaOuter + span.alphaStep; 786#endif 787#ifdef INTERP_INDEX 788 diInner = diOuter + span.indexStep; 789#endif 790#ifdef INTERP_INT_TEX 791 dsInner = dsOuter + span.intTexStep[0]; 792 dtInner = dtOuter + span.intTexStep[1]; 793#endif 794#ifdef INTERP_ATTRIBS 795 dwInner = dwOuter + span.attrStepX[FRAG_ATTRIB_WPOS][3]; 796 ATTRIB_LOOP_BEGIN 797 GLuint c; 798 for (c = 0; c < 4; c++) { 799 daInner[attr][c] = daOuter[attr][c] + span.attrStepX[attr][c]; 800 } 801 ATTRIB_LOOP_END 802#endif 803 804 while (lines > 0) { 805 /* initialize the span interpolants to the leftmost value */ 806 /* ff = fixed-pt fragment */ 807 const GLint right = FixedToInt(fxRightEdge); 808 span.x = FixedToInt(fxLeftEdge); 809 if (right <= span.x) 810 span.end = 0; 811 else 812 span.end = right - span.x; 813 814#ifdef INTERP_Z 815 span.z = zLeft; 816#endif 817#ifdef INTERP_RGB 818 span.red = rLeft; 819 span.green = gLeft; 820 span.blue = bLeft; 821#endif 822#ifdef INTERP_ALPHA 823 span.alpha = aLeft; 824#endif 825#ifdef INTERP_INDEX 826 span.index = iLeft; 827#endif 828#ifdef INTERP_INT_TEX 829 span.intTex[0] = sLeft; 830 span.intTex[1] = tLeft; 831#endif 832 833#ifdef INTERP_ATTRIBS 834 span.attrStart[FRAG_ATTRIB_WPOS][3] = wLeft; 835 ATTRIB_LOOP_BEGIN 836 GLuint c; 837 for (c = 0; c < 4; c++) { 838 span.attrStart[attr][c] = attrLeft[attr][c]; 839 } 840 ATTRIB_LOOP_END 841#endif 842 843 /* This is where we actually generate fragments */ 844 /* XXX the test for span.y > 0 _shouldn't_ be needed but 845 * it fixes a problem on 64-bit Opterons (bug 4842). 846 */ 847 if (span.end > 0 && span.y >= 0) { 848 const GLint len = span.end - 1; 849 (void) len; 850#ifdef INTERP_RGB 851 CLAMP_INTERPOLANT(red, redStep, len); 852 CLAMP_INTERPOLANT(green, greenStep, len); 853 CLAMP_INTERPOLANT(blue, blueStep, len); 854#endif 855#ifdef INTERP_ALPHA 856 CLAMP_INTERPOLANT(alpha, alphaStep, len); 857#endif 858#ifdef INTERP_INDEX 859 CLAMP_INTERPOLANT(index, indexStep, len); 860#endif 861 { 862 RENDER_SPAN( span ); 863 } 864 } 865 866 /* 867 * Advance to the next scan line. Compute the 868 * new edge coordinates, and adjust the 869 * pixel-center x coordinate so that it stays 870 * on or inside the major edge. 871 */ 872 span.y++; 873 lines--; 874 875 fxLeftEdge += fdxLeftEdge; 876 fxRightEdge += fdxRightEdge; 877 878 fError += fdError; 879 if (fError >= 0) { 880 fError -= FIXED_ONE; 881 882#ifdef PIXEL_ADDRESS 883 pRow = (PIXEL_TYPE *) ((GLubyte *) pRow + dPRowOuter); 884#endif 885#ifdef INTERP_Z 886# ifdef DEPTH_TYPE 887 zRow = (DEPTH_TYPE *) ((GLubyte *) zRow + dZRowOuter); 888# endif 889 zLeft += fdzOuter; 890#endif 891#ifdef INTERP_RGB 892 rLeft += fdrOuter; 893 gLeft += fdgOuter; 894 bLeft += fdbOuter; 895#endif 896#ifdef INTERP_ALPHA 897 aLeft += fdaOuter; 898#endif 899#ifdef INTERP_INDEX 900 iLeft += diOuter; 901#endif 902#ifdef INTERP_INT_TEX 903 sLeft += dsOuter; 904 tLeft += dtOuter; 905#endif 906#ifdef INTERP_ATTRIBS 907 wLeft += dwOuter; 908 ATTRIB_LOOP_BEGIN 909 GLuint c; 910 for (c = 0; c < 4; c++) { 911 attrLeft[attr][c] += daOuter[attr][c]; 912 } 913 ATTRIB_LOOP_END 914#endif 915 } 916 else { 917#ifdef PIXEL_ADDRESS 918 pRow = (PIXEL_TYPE *) ((GLubyte *) pRow + dPRowInner); 919#endif 920#ifdef INTERP_Z 921# ifdef DEPTH_TYPE 922 zRow = (DEPTH_TYPE *) ((GLubyte *) zRow + dZRowInner); 923# endif 924 zLeft += fdzInner; 925#endif 926#ifdef INTERP_RGB 927 rLeft += fdrInner; 928 gLeft += fdgInner; 929 bLeft += fdbInner; 930#endif 931#ifdef INTERP_ALPHA 932 aLeft += fdaInner; 933#endif 934#ifdef INTERP_INDEX 935 iLeft += diInner; 936#endif 937#ifdef INTERP_INT_TEX 938 sLeft += dsInner; 939 tLeft += dtInner; 940#endif 941#ifdef INTERP_ATTRIBS 942 wLeft += dwInner; 943 ATTRIB_LOOP_BEGIN 944 GLuint c; 945 for (c = 0; c < 4; c++) { 946 attrLeft[attr][c] += daInner[attr][c]; 947 } 948 ATTRIB_LOOP_END 949#endif 950 } 951 } /*while lines>0*/ 952 953 } /* for subTriangle */ 954 955 } 956 } 957} 958 959#undef SETUP_CODE 960#undef RENDER_SPAN 961 962#undef PIXEL_TYPE 963#undef BYTES_PER_ROW 964#undef PIXEL_ADDRESS 965#undef DEPTH_TYPE 966 967#undef INTERP_Z 968#undef INTERP_RGB 969#undef INTERP_ALPHA 970#undef INTERP_INDEX 971#undef INTERP_INT_TEX 972#undef INTERP_ATTRIBS 973 974#undef S_SCALE 975#undef T_SCALE 976 977#undef FixedToDepth 978 979#undef NAME 980