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