24
|
1 |
/*------------------------------------------------------------------------
|
|
2 |
*
|
|
3 |
* OpenVG 1.1 Reference Implementation
|
|
4 |
* -----------------------------------
|
|
5 |
*
|
|
6 |
* Copyright (c) 2007 The Khronos Group Inc.
|
|
7 |
* Portions copyright (c) 2010 Nokia Corporation and/or its subsidiary(-ies).
|
|
8 |
*
|
|
9 |
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
10 |
* copy of this software and /or associated documentation files
|
|
11 |
* (the "Materials "), to deal in the Materials without restriction,
|
|
12 |
* including without limitation the rights to use, copy, modify, merge,
|
|
13 |
* publish, distribute, sublicense, and/or sell copies of the Materials,
|
|
14 |
* and to permit persons to whom the Materials are furnished to do so,
|
|
15 |
* subject to the following conditions:
|
|
16 |
*
|
|
17 |
* The above copyright notice and this permission notice shall be included
|
|
18 |
* in all copies or substantial portions of the Materials.
|
|
19 |
*
|
|
20 |
* THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|
21 |
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
|
22 |
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
|
|
23 |
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
|
|
24 |
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
|
|
25 |
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE MATERIALS OR
|
|
26 |
* THE USE OR OTHER DEALINGS IN THE MATERIALS.
|
|
27 |
*
|
|
28 |
*//**
|
|
29 |
* \file
|
|
30 |
* \brief Implementation of polygon rasterizer.
|
|
31 |
* \note
|
|
32 |
*//*-------------------------------------------------------------------*/
|
|
33 |
|
|
34 |
#include "riRasterizer.h"
|
|
35 |
|
|
36 |
// TEMP!
|
|
37 |
#ifndef __SFCOMPILER_H
|
|
38 |
# include "sfCompiler.h"
|
|
39 |
#endif
|
|
40 |
|
|
41 |
|
|
42 |
namespace OpenVGRI
|
|
43 |
{
|
|
44 |
|
|
45 |
/*-------------------------------------------------------------------*//*!
|
|
46 |
* \brief Rasterizer constructor.
|
|
47 |
* \param
|
|
48 |
* \return
|
|
49 |
* \note
|
|
50 |
*//*-------------------------------------------------------------------*/
|
|
51 |
|
|
52 |
Rasterizer::Rasterizer() :
|
|
53 |
m_covBuffer(NULL),
|
|
54 |
m_covBufferSz(0),
|
|
55 |
m_edges(),
|
|
56 |
m_scissorEdges(),
|
|
57 |
m_scissor(false),
|
|
58 |
m_aa(true),
|
|
59 |
m_vpx(0),
|
|
60 |
m_vpy(0),
|
|
61 |
m_vpwidth(0),
|
|
62 |
m_vpheight(0),
|
|
63 |
m_fillRule(VG_EVEN_ODD),
|
|
64 |
m_pixelPipe(NULL),
|
|
65 |
m_nSpans(0)
|
|
66 |
{}
|
|
67 |
|
|
68 |
/*-------------------------------------------------------------------*//*!
|
|
69 |
* \brief Rasterizer destructor.
|
|
70 |
* \param
|
|
71 |
* \return
|
|
72 |
* \note
|
|
73 |
*//*-------------------------------------------------------------------*/
|
|
74 |
|
|
75 |
Rasterizer::~Rasterizer()
|
|
76 |
{
|
|
77 |
if(m_covBuffer)
|
|
78 |
RI_DELETE_ARRAY(m_covBuffer);
|
|
79 |
}
|
|
80 |
|
|
81 |
/*-------------------------------------------------------------------*//*!
|
|
82 |
* \brief Removes all appended edges.
|
|
83 |
* \param
|
|
84 |
* \return
|
|
85 |
* \note
|
|
86 |
*//*-------------------------------------------------------------------*/
|
|
87 |
|
|
88 |
#define EDGE_TERMINATOR 0xFFFFFFFFu
|
|
89 |
|
|
90 |
void Rasterizer::clear()
|
|
91 |
{
|
|
92 |
//m_edges.clear();
|
|
93 |
for (int i = 0; i < m_edges.size(); i++)
|
|
94 |
m_edges[i] = EDGE_TERMINATOR;
|
|
95 |
|
|
96 |
m_edgePool.clear();
|
|
97 |
|
|
98 |
m_edgeMin.set(0x7fffffffu, 0x7fffffffu);
|
|
99 |
m_edgeMax.set(0x80000000, 0x80000000);
|
|
100 |
}
|
|
101 |
|
|
102 |
/*-------------------------------------------------------------------*//*!
|
|
103 |
* \brief Appends an edge to the rasterizer.
|
|
104 |
* \param
|
|
105 |
* \return
|
|
106 |
* \note
|
|
107 |
*//*-------------------------------------------------------------------*/
|
|
108 |
|
|
109 |
void Rasterizer::addBBox(const IVector2& v)
|
|
110 |
{
|
|
111 |
if(v.x < m_edgeMin.x) m_edgeMin.x = v.x;
|
|
112 |
if(v.y < m_edgeMin.y) m_edgeMin.y = v.y;
|
|
113 |
if(v.x > m_edgeMax.x) m_edgeMax.x = v.x;
|
|
114 |
if(v.y > m_edgeMax.y) m_edgeMax.y = v.y;
|
|
115 |
}
|
|
116 |
|
|
117 |
void Rasterizer::pushEdge(const Edge& edge)
|
|
118 |
{
|
|
119 |
addBBox(edge.v0);
|
|
120 |
addBBox(edge.v1);
|
|
121 |
|
|
122 |
// Only add processed edges.
|
|
123 |
|
|
124 |
RI_ASSERT(edge.v0.y >= 0);
|
|
125 |
RI_ASSERT(edge.v0.y < edge.v1.y); //horizontal edges should have been dropped already
|
|
126 |
|
|
127 |
ActiveEdge ae;
|
|
128 |
ae.direction = edge.direction;
|
|
129 |
|
|
130 |
// \todo Adjust for non-AA cases
|
|
131 |
// \todo verySteep is temporary. Either clip to right edge also, or validate that a proper slope can be
|
|
132 |
// calculated here.
|
|
133 |
const int slope = RI_SAT_SHL((edge.v1.x - edge.v0.x), RASTERIZER_BITS - X_BITS) / (edge.v1.y - edge.v0.y);
|
|
134 |
//const bool verySteep = RI_INT_ABS(edge.v1.x - edge.v0.x) > (1 << (30-RASTERIZER_BITS)) ? true : false;
|
|
135 |
//const int slope = verySteep ? 1 << 30 : RI_SHL((edge.v1.x - edge.v0.x), RASTERIZER_BITS - X_BITS) / (edge.v1.y - edge.v0.y);
|
|
136 |
// slope: SI.(RASTERIZER_BITS - Y_BITS)
|
|
137 |
const int yF = edge.v0.y & Y_MASK;
|
|
138 |
// \todo See verySteep note for this hack also. (Clip to right edge?)
|
|
139 |
const int xRef = RI_SAT_SHL(edge.v0.x, RASTERIZER_BITS - X_BITS) - (yF * slope);
|
|
140 |
//const int xRef = edge.v0.x > (1<<(30-RASTERIZER_BITS)) ? 1<<30 : RI_SHL(edge.v0.x, RASTERIZER_BITS - X_BITS) - (yF * slope);
|
|
141 |
|
|
142 |
RI_ASSERT(RI_INT_ABS(edge.v0.y <= 32767));
|
|
143 |
RI_ASSERT(RI_INT_ABS(edge.v1.y <= 32767));
|
|
144 |
|
|
145 |
ae.yStart = (RIint16)edge.v0.y;
|
|
146 |
ae.yEnd = (RIint16)edge.v1.y;
|
|
147 |
ae.xRef = xRef;
|
|
148 |
ae.slope = slope;
|
|
149 |
// Scanline range.
|
|
150 |
ae.minx = xRef >> RASTERIZER_BITS;
|
|
151 |
ae.maxx = (xRef + slope * (1<<Y_BITS)) >> RASTERIZER_BITS;
|
|
152 |
|
|
153 |
if (ae.minx > ae.maxx)
|
|
154 |
RI_ANY_SWAP(ActiveEdge::XCoord, ae.minx, ae.maxx);
|
|
155 |
|
|
156 |
if (ae.maxx < 0)
|
|
157 |
ae.minx = ae.maxx = LEFT_DISCARD_SHORT;
|
|
158 |
|
|
159 |
if (m_edges[ae.yStart>>Y_BITS] == EDGE_TERMINATOR)
|
|
160 |
ae.next = EDGE_TERMINATOR;
|
|
161 |
else
|
|
162 |
ae.next = m_edges[ae.yStart>>Y_BITS];
|
|
163 |
|
|
164 |
m_edgePool.push_back(ae); //throws bad_alloc
|
|
165 |
|
|
166 |
RI_ASSERT(m_edgePool.size() > 0);
|
|
167 |
m_edges[ae.yStart>>Y_BITS] = m_edgePool.size()-1;
|
|
168 |
}
|
|
169 |
|
|
170 |
/**
|
|
171 |
* \brief Clips an edge and if something remains, adds it to the list of edges.
|
|
172 |
* \todo Enhance precision: Currently this just uses doubles and gets away with
|
|
173 |
* it in most cases.
|
|
174 |
*/
|
|
175 |
void Rasterizer::clipAndAddEdge(Edge& edge)
|
|
176 |
{
|
|
177 |
//if (m_edges.size() > 48)
|
|
178 |
//return;
|
|
179 |
// Check y-clips
|
|
180 |
// \todo Reduce amount of clips.
|
|
181 |
bool outLeft[2] = {(edge.v0.x < m_vpMinx), (edge.v1.x < m_vpMinx)};
|
|
182 |
bool outRight[2] = {(edge.v0.x > m_vpMaxx), (edge.v1.x > m_vpMaxx)};
|
|
183 |
bool outTop[2] = {(edge.v0.y < m_vpMiny), (edge.v1.y < m_vpMiny)};
|
|
184 |
bool outBottom[2] = {(edge.v0.y > m_vpMaxy), (edge.v1.y > m_vpMaxy)};
|
|
185 |
|
|
186 |
if (!(outLeft[0] || outLeft[1] || outRight[0] || outRight[1] || outTop[0] || outTop[1] || outBottom[0] || outBottom[1]))
|
|
187 |
{
|
|
188 |
pushEdge(edge);
|
|
189 |
return;
|
|
190 |
}
|
|
191 |
|
|
192 |
// \todo Make sure that checking out-of-right works with the scanconverter.
|
|
193 |
if ((outBottom[0] && outBottom[1]) || (outTop[0] && outTop[1]))
|
|
194 |
return; // Out of bounds
|
|
195 |
|
|
196 |
// \todo Clip to right edge of screen.
|
|
197 |
// \todo Make slope-calculation and signs consistent.
|
|
198 |
//
|
|
199 |
if (outTop[0] || outBottom[1])
|
|
200 |
{
|
|
201 |
// Clip to top/bottom.
|
|
202 |
double slope = (double)(edge.v1.x - edge.v0.x)/(edge.v1.y - edge.v0.y);
|
|
203 |
|
|
204 |
if (outTop[0])
|
|
205 |
{
|
|
206 |
RI_ASSERT(-(RIint64)edge.v0.y >= 0);
|
|
207 |
RIint32 dx = RI_ROUND_TO_INT(-slope * edge.v0.y);
|
|
208 |
edge.v0.y = 0;
|
|
209 |
edge.v0.x += dx;
|
|
210 |
}
|
|
211 |
|
|
212 |
if (outBottom[1])
|
|
213 |
{
|
|
214 |
RIint32 dy = edge.v1.y - m_vpMaxy;
|
|
215 |
RI_ASSERT(dy >= 0);
|
|
216 |
RIint32 dx = -RI_ROUND_TO_INT(slope * dy);
|
|
217 |
edge.v1.y = m_vpMaxy;
|
|
218 |
edge.v1.x += dx;
|
|
219 |
}
|
|
220 |
|
|
221 |
}
|
|
222 |
|
|
223 |
if (edge.v0.y >= edge.v1.y)
|
|
224 |
return;
|
|
225 |
|
|
226 |
// \todo Recheck left/right.
|
|
227 |
outLeft[0] = (edge.v0.x < m_vpMinx); outLeft[1] = (edge.v1.x < m_vpMinx);
|
|
228 |
outRight[1] = (edge.v0.x > m_vpMaxx); outRight[1] = (edge.v1.x > m_vpMaxx);
|
|
229 |
|
|
230 |
if (outLeft[0] && outLeft[1])
|
|
231 |
{
|
|
232 |
edge.v0.x = m_vpMinx;
|
|
233 |
edge.v1.x = m_vpMinx;
|
|
234 |
pushEdge(edge);
|
|
235 |
return;
|
|
236 |
}
|
|
237 |
if (outRight[0] && outRight[1])
|
|
238 |
{
|
|
239 |
edge.v0.x = m_vpMaxx;
|
|
240 |
edge.v1.x = m_vpMaxx;
|
|
241 |
pushEdge(edge);
|
|
242 |
return;
|
|
243 |
}
|
|
244 |
|
|
245 |
// From outside -> screen
|
|
246 |
if (outLeft[0] || outRight[1])
|
|
247 |
{
|
|
248 |
// infinite slope?
|
|
249 |
double slope = (double)((RIint64)edge.v1.y - edge.v0.y)/((RIint64)edge.v1.x - edge.v0.x);
|
|
250 |
|
|
251 |
if (outLeft[0])
|
|
252 |
{
|
|
253 |
RIint32 dx = edge.v0.x;
|
|
254 |
//RI_ASSERT(dx >= 0);
|
|
255 |
// Note the sign.
|
|
256 |
RIint32 dy = RI_ROUND_TO_INT(-slope * dx);
|
|
257 |
|
|
258 |
Edge vpart = edge;
|
|
259 |
vpart.v1.y = edge.v0.y + dy;
|
|
260 |
//vpart.v1.x = edge.v0.x; // = 0?
|
|
261 |
// \note This should be flagged instead of setting the smallest possible
|
|
262 |
// value because of extremely gentle slopes may cause bugs:
|
|
263 |
vpart.v1.x = vpart.v0.x = -0x100000;
|
|
264 |
|
|
265 |
if (vpart.v1.y > vpart.v0.y)
|
|
266 |
pushEdge(vpart);
|
|
267 |
|
|
268 |
edge.v0.y += dy;
|
|
269 |
edge.v0.x = 0;
|
|
270 |
}
|
|
271 |
}
|
|
272 |
// From screen -> outside
|
|
273 |
if (outLeft[1] || outRight[0])
|
|
274 |
{
|
|
275 |
// infinite slope?
|
|
276 |
double slope = (double)((RIint64)edge.v1.y - edge.v0.y)/((RIint64)edge.v1.x - edge.v0.x);
|
|
277 |
|
|
278 |
if (outLeft[1])
|
|
279 |
{
|
|
280 |
RIint32 dx = edge.v0.x;
|
|
281 |
RI_ASSERT(dx >= 0);
|
|
282 |
RIint32 dy = RI_ROUND_TO_INT(-slope * dx);
|
|
283 |
|
|
284 |
Edge vpart = edge;
|
|
285 |
vpart.v0.y = edge.v0.y + dy;
|
|
286 |
vpart.v1.x = vpart.v0.x = LEFT_DISCARD;
|
|
287 |
|
|
288 |
if (vpart.v1.y > vpart.v0.y)
|
|
289 |
pushEdge(vpart);
|
|
290 |
|
|
291 |
edge.v1.y = edge.v0.y + dy;
|
|
292 |
edge.v1.x = 0;
|
|
293 |
}
|
|
294 |
}
|
|
295 |
|
|
296 |
if (edge.v0.y >= edge.v1.y)
|
|
297 |
return;
|
|
298 |
|
|
299 |
// Finally, add the edge:
|
|
300 |
pushEdge(edge);
|
|
301 |
}
|
|
302 |
|
|
303 |
void Rasterizer::addEdge(const Vector2& v0, const Vector2& v1)
|
|
304 |
{
|
|
305 |
if( m_edges.size() >= RI_MAX_EDGES )
|
|
306 |
throw std::bad_alloc(); //throw an out of memory error if there are too many edges
|
|
307 |
|
|
308 |
Edge e;
|
|
309 |
|
|
310 |
{
|
|
311 |
IVector2 i0(RI_ROUND_TO_INT(v0.x * (1<<X_BITS)), RI_ROUND_TO_INT(v0.y * (1<<Y_BITS)));
|
|
312 |
IVector2 i1(RI_ROUND_TO_INT(v1.x * (1<<X_BITS)), RI_ROUND_TO_INT(v1.y * (1<<Y_BITS)));
|
|
313 |
|
|
314 |
if(i0.y == i1.y)
|
|
315 |
return; //skip horizontal edges (they don't affect rasterization since we scan horizontally)
|
|
316 |
|
|
317 |
if (i0.y < i1.y)
|
|
318 |
{
|
|
319 |
// Edge is going upward
|
|
320 |
e.v0 = i0;
|
|
321 |
e.v1 = i1;
|
|
322 |
e.direction = 1;
|
|
323 |
}
|
|
324 |
else
|
|
325 |
{
|
|
326 |
// Edge is going downward
|
|
327 |
e.v0 = i1;
|
|
328 |
e.v1 = i0;
|
|
329 |
e.direction = -1;
|
|
330 |
}
|
|
331 |
}
|
|
332 |
|
|
333 |
// Clip and insert.
|
|
334 |
|
|
335 |
clipAndAddEdge(e);
|
|
336 |
}
|
|
337 |
|
|
338 |
/*-------------------------------------------------------------------*//*!
|
|
339 |
* \brief Set up rasterizer
|
|
340 |
* \param
|
|
341 |
* \return
|
|
342 |
* \note
|
|
343 |
*//*-------------------------------------------------------------------*/
|
|
344 |
|
|
345 |
void Rasterizer::setup(int vpx, int vpy, int vpwidth, int vpheight, VGFillRule fillRule, const PixelPipe* pixelPipe)
|
|
346 |
{
|
|
347 |
RI_ASSERT(vpwidth >= 0 && vpheight >= 0);
|
|
348 |
RI_ASSERT(vpx + vpwidth >= vpx && vpy + vpheight >= vpy);
|
|
349 |
RI_ASSERT(fillRule == VG_EVEN_ODD || fillRule == VG_NON_ZERO);
|
|
350 |
RI_ASSERT(pixelPipe);
|
|
351 |
|
|
352 |
clear();
|
|
353 |
|
|
354 |
m_vpx = vpx;
|
|
355 |
m_vpy = vpy;
|
|
356 |
m_vpwidth = vpwidth;
|
|
357 |
m_vpheight = vpheight;
|
|
358 |
|
|
359 |
if (m_vpheight > m_edges.size())
|
|
360 |
{
|
|
361 |
int os = m_edges.size();
|
|
362 |
m_edges.resize(m_vpheight);
|
|
363 |
for (int i = os; i < m_edges.size(); i++)
|
|
364 |
m_edges[i] = EDGE_TERMINATOR;
|
|
365 |
}
|
|
366 |
|
|
367 |
m_vpMinx = RI_SHL(vpx, X_BITS);
|
|
368 |
m_vpMiny = RI_SHL(vpy, Y_BITS);
|
|
369 |
m_vpMaxx = RI_SHL(vpx + vpwidth, X_BITS);
|
|
370 |
m_vpMaxy = RI_SHL(vpy + vpheight, Y_BITS);
|
|
371 |
|
|
372 |
m_fillRule = fillRule;
|
|
373 |
|
|
374 |
RIuint32 fillRuleMask = fillRule == VG_NON_ZERO ? 0xffffffffu : 1;
|
|
375 |
m_fillRuleMask = fillRuleMask;
|
|
376 |
|
|
377 |
m_pixelPipe = pixelPipe;
|
|
378 |
m_covMinx = vpx+vpwidth;
|
|
379 |
m_covMiny = vpy+vpheight;
|
|
380 |
m_covMaxx = vpx;
|
|
381 |
m_covMaxy = vpy;
|
|
382 |
}
|
|
383 |
|
|
384 |
/*-------------------------------------------------------------------*//*!
|
|
385 |
* \brief Sets scissor rectangles.
|
|
386 |
* \param
|
|
387 |
* \return
|
|
388 |
* \note
|
|
389 |
*//*-------------------------------------------------------------------*/
|
|
390 |
|
|
391 |
void Rasterizer::setScissor(const Array<Rectangle>& scissors)
|
|
392 |
{
|
|
393 |
try
|
|
394 |
{
|
|
395 |
m_scissorEdges.clear();
|
|
396 |
for(int i=0;i<scissors.size();i++)
|
|
397 |
{
|
|
398 |
if(scissors[i].width > 0 && scissors[i].height > 0)
|
|
399 |
{
|
|
400 |
ScissorEdge e;
|
|
401 |
e.miny = scissors[i].y;
|
|
402 |
e.maxy = RI_INT_ADDSATURATE(scissors[i].y, scissors[i].height);
|
|
403 |
|
|
404 |
e.x = scissors[i].x;
|
|
405 |
e.direction = 1;
|
|
406 |
m_scissorEdges.push_back(e); //throws bad_alloc
|
|
407 |
e.x = RI_INT_ADDSATURATE(scissors[i].x, scissors[i].width);
|
|
408 |
e.direction = -1;
|
|
409 |
m_scissorEdges.push_back(e); //throws bad_alloc
|
|
410 |
}
|
|
411 |
}
|
|
412 |
}
|
|
413 |
catch(std::bad_alloc)
|
|
414 |
{
|
|
415 |
m_scissorEdges.clear();
|
|
416 |
throw;
|
|
417 |
}
|
|
418 |
}
|
|
419 |
|
|
420 |
void Rasterizer::setScissoring(bool enabled)
|
|
421 |
{
|
|
422 |
m_scissor = enabled;
|
|
423 |
}
|
|
424 |
|
|
425 |
static RI_INLINE void small_memcpy32(void* dst, const void* src, size_t n)
|
|
426 |
{
|
|
427 |
RIuint32 *d = (RIuint32*)dst;
|
|
428 |
const RIuint32 *s = (const RIuint32*)src;
|
|
429 |
while(n)
|
|
430 |
{
|
|
431 |
*d++ = *s++;
|
|
432 |
n-=4;
|
|
433 |
}
|
|
434 |
}
|
|
435 |
|
|
436 |
// \todo Move this to some debug file or remove.
|
|
437 |
#if defined(USE_SSE2) && !defined(_WIN32)
|
|
438 |
RI_INLINE static void print128(__m128i ll)
|
|
439 |
{
|
|
440 |
#if defined(RI_DEBUG)
|
|
441 |
unsigned long long v[2];
|
|
442 |
_mm_storeu_pd((double*)v, (__m128d)ll);
|
|
443 |
RI_PRINTF("0x%016llx %016llx\n", v[0], v[1]);
|
|
444 |
#else
|
|
445 |
(void)ll;
|
|
446 |
#endif
|
|
447 |
}
|
|
448 |
#endif
|
|
449 |
|
|
450 |
#if defined(USE_SSE2)
|
|
451 |
RI_INLINE static __m128i mm_mul4x32(const __m128i a, const __m128i b) {
|
|
452 |
__m128i res;
|
|
453 |
#if (_MSC_VER > 1400 )
|
|
454 |
// \todo Simpler way to do this on intel?
|
|
455 |
__m128i m0 = _mm_mul_epu32(a, _mm_shuffle_epi32(b, _MM_SHUFFLE(1, 1, 0, 0)));
|
|
456 |
__m128i m1 = _mm_mul_epu32(a, _mm_shuffle_epi32(b, _MM_SHUFFLE(3, 3, 2, 2)));
|
|
457 |
|
|
458 |
res = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(m0), _mm_castsi128_ps(m1), _MM_SHUFFLE(2, 0, 2, 0)));
|
|
459 |
#else
|
|
460 |
__asm {
|
|
461 |
movdqa xmm1, a;
|
|
462 |
movdqa xmm2, b;
|
|
463 |
pshufd xmm3, xmm2, 80;
|
|
464 |
movdqa xmm0, xmm1;
|
|
465 |
|
|
466 |
pshufd xmm2, xmm2, 250;
|
|
467 |
pmuludq xmm0, xmm3;
|
|
468 |
pmuludq xmm1, xmm2;
|
|
469 |
|
|
470 |
shufps xmm0, xmm1, 136;
|
|
471 |
movdqa res, xmm0;
|
|
472 |
}
|
|
473 |
#endif
|
|
474 |
return res;
|
|
475 |
}
|
|
476 |
#endif
|
|
477 |
|
|
478 |
#if defined(USE_SSE2)
|
|
479 |
RI_INLINE static void mm_get_xmasks(const __m128i& coords, const __m128i& sampleCoords, __m128i& slWindMask, __m128i& pxWindMask)
|
|
480 |
{
|
|
481 |
const __m128i z = _mm_setzero_si128();
|
|
482 |
const __m128i xMask = _mm_cmpeq_epi16(_mm_srai_epi16(coords, Rasterizer::RASTERIZER_BITS), z);
|
|
483 |
const __m128i sCmp = _mm_or_si128(_mm_cmpgt_epi16(sampleCoords, coords), _mm_cmpeq_epi16(sampleCoords, coords));
|
|
484 |
//const __m128i sCmp = _mm_cmplt_epi16(coords, sampleCoords);
|
|
485 |
slWindMask = xMask;
|
|
486 |
pxWindMask = _mm_and_si128(xMask, sCmp);
|
|
487 |
}
|
|
488 |
#endif
|
|
489 |
|
|
490 |
RI_INLINE static void getVerticalSubpixels(int iY, int yStart, int yEnd, int& py0, int& py1)
|
|
491 |
{
|
|
492 |
const int cy = iY << Rasterizer::Y_BITS;
|
|
493 |
py0 = cy > yStart ? 0 : yStart & Rasterizer::Y_MASK;
|
|
494 |
py1 = (RI_INT_MIN(yEnd, cy + (1<<Rasterizer::Y_BITS)) - 1) & Rasterizer::Y_MASK;
|
|
495 |
}
|
|
496 |
|
|
497 |
RI_INLINE static void applyLeftEdge(const Rasterizer::ActiveEdge& currAe, Rasterizer::Windings& scanline, int intY)
|
|
498 |
{
|
|
499 |
// Applies the whole edge at a time. Make sure xRight < x for all y.
|
|
500 |
// \todo Remove duplicate code for determining the active samples
|
|
501 |
#if defined(USE_SSE2)
|
|
502 |
int py0, py1;
|
|
503 |
|
|
504 |
getVerticalSubpixels(intY, currAe.yStart, currAe.yEnd, py0, py1);
|
|
505 |
|
|
506 |
const __m128i csteps = _mm_set_epi16(7,6,5,4,3,2,1,0);
|
|
507 |
|
|
508 |
const __m128i ssePy0 = _mm_set1_epi16(py0-1);
|
|
509 |
const __m128i ssePy1 = _mm_set1_epi16(py1+1);
|
|
510 |
|
|
511 |
const __m128i yMask = _mm_and_si128(_mm_cmpgt_epi16(csteps, ssePy0), _mm_cmplt_epi16(csteps, ssePy1));
|
|
512 |
const __m128i dir = _mm_set1_epi16(currAe.direction);
|
|
513 |
|
|
514 |
scanline.sseWinding = _mm_add_epi16(scanline.sseWinding, _mm_and_si128(yMask, dir));
|
|
515 |
|
|
516 |
#else
|
|
517 |
RI_ASSERT(false); // Not implemented yet.
|
|
518 |
#endif
|
|
519 |
}
|
|
520 |
|
|
521 |
RI_INLINE static void applyLeftEdgeNoAA(const Rasterizer::ActiveEdge& currAe, Rasterizer::Windings& scanline, int intY)
|
|
522 |
{
|
|
523 |
// Applies the whole edge at a time. Make sure xRight < x for all y.
|
|
524 |
// \todo Remove duplicate code for determining the active samples?
|
|
525 |
#if defined(USE_SSE2)
|
|
526 |
int py0, py1;
|
|
527 |
|
|
528 |
getVerticalSubpixels(intY, currAe.yStart, currAe.yEnd, py0, py1);
|
|
529 |
|
|
530 |
//const __m128i csteps = _mm_set_epi16(4,4,4,4,4,4,4,4);
|
|
531 |
|
|
532 |
__m128i yMask;
|
|
533 |
|
|
534 |
if (py0 <= 4 && py1 >= 4)
|
|
535 |
yMask = _mm_set1_epi8(-1);
|
|
536 |
else
|
|
537 |
yMask = _mm_set1_epi8(0);
|
|
538 |
|
|
539 |
const __m128i dir = _mm_set1_epi16(currAe.direction);
|
|
540 |
|
|
541 |
scanline.sseWinding = _mm_add_epi16(scanline.sseWinding, _mm_and_si128(yMask, dir));
|
|
542 |
//scanline.sseWinding = _mm_add_epi32(scanline.sseWinding, dir);
|
|
543 |
|
|
544 |
#else
|
|
545 |
RI_ASSERT(false); // Not implemented yet.
|
|
546 |
#endif
|
|
547 |
}
|
|
548 |
|
|
549 |
RI_INLINE void calculateAEWinding(const Rasterizer::ActiveEdge& currAe, Rasterizer::Windings& pixel, Rasterizer::Windings& scanline, int intY, int pixelX)
|
|
550 |
{
|
|
551 |
#define QUEEN_COORD(Y) ((Y<<(Rasterizer::RASTERIZER_BITS - Rasterizer::SAMPLE_BITS)) + (1<<(Rasterizer::RASTERIZER_BITS-Rasterizer::SAMPLE_BITS-1)))
|
|
552 |
|
|
553 |
#if !defined(USE_SSE2)
|
|
554 |
static const int queenCoords[(1<<Rasterizer::SAMPLE_BITS)] = {
|
|
555 |
QUEEN_COORD(3), QUEEN_COORD(7), QUEEN_COORD(0), QUEEN_COORD(2),
|
|
556 |
QUEEN_COORD(5), QUEEN_COORD(1), QUEEN_COORD(6), QUEEN_COORD(4)
|
|
557 |
};
|
|
558 |
|
|
559 |
const int ix = pixelX >> Rasterizer::RASTERIZER_BITS;
|
|
560 |
const int cy = intY << Rasterizer::Y_BITS;
|
|
561 |
|
|
562 |
const int py0 = cy > currAe.yStart ? 0 : currAe.yStart & Rasterizer::Y_MASK;
|
|
563 |
const int py1 = (RI_INT_MIN(currAe.yEnd, cy + (1<<Rasterizer::Y_BITS)) - 1) & Rasterizer::Y_MASK;
|
|
564 |
|
|
565 |
int edgeX = currAe.xRef + (cy + py0 - (currAe.yStart & ~Rasterizer::Y_MASK)) * currAe.slope;
|
|
566 |
|
|
567 |
RI_ASSERT(py1 >= py0);
|
|
568 |
|
|
569 |
for (int s = py0; s <= py1; s++)
|
|
570 |
{
|
|
571 |
const int sampleX = pixelX + queenCoords[s];
|
|
572 |
|
|
573 |
//compute winding number by evaluating the edge functions of edges to the left of the sampling point
|
|
574 |
if(((edgeX >> Rasterizer::RASTERIZER_BITS) == ix))
|
|
575 |
{
|
|
576 |
if (sampleX >= edgeX)
|
|
577 |
{
|
|
578 |
pixel.winding[s] += currAe.direction;
|
|
579 |
}
|
|
580 |
scanline.winding[s] += currAe.direction;
|
|
581 |
}
|
|
582 |
|
|
583 |
edgeX += currAe.slope;
|
|
584 |
}
|
|
585 |
#else
|
|
586 |
|
|
587 |
__m128i qCoords = _mm_set_epi16(
|
|
588 |
QUEEN_COORD(4), QUEEN_COORD(6), QUEEN_COORD(1), QUEEN_COORD(5),
|
|
589 |
QUEEN_COORD(2), QUEEN_COORD(0), QUEEN_COORD(7), QUEEN_COORD(3));
|
|
590 |
|
|
591 |
RI_ASSERT(Rasterizer::RASTERIZER_BITS <= 14);
|
|
592 |
|
|
593 |
// TEROP: Optimize conditions.
|
|
594 |
int py0, py1;
|
|
595 |
getVerticalSubpixels(intY, currAe.yStart, currAe.yEnd, py0, py1);
|
|
596 |
|
|
597 |
const int cy = intY << Rasterizer::Y_BITS;
|
|
598 |
|
|
599 |
const __m128i csteps0 = _mm_set_epi32(3,2,1,0);
|
|
600 |
const __m128i csteps1 = _mm_set_epi32(7,6,5,4);
|
|
601 |
|
|
602 |
const __m128i ssePy0 = _mm_set1_epi32(py0-1);
|
|
603 |
const __m128i ssePy1 = _mm_set1_epi32(py1+1);
|
|
604 |
|
|
605 |
const __m128i yMask0 = _mm_and_si128(_mm_cmpgt_epi32(csteps0, ssePy0), _mm_cmplt_epi32(csteps0, ssePy1));
|
|
606 |
const __m128i yMask1 = _mm_and_si128(_mm_cmpgt_epi32(csteps1, ssePy0), _mm_cmplt_epi32(csteps1, ssePy1));
|
|
607 |
|
|
608 |
const int edgeX = currAe.xRef + (cy - (currAe.yStart & ~Rasterizer::Y_MASK)) * currAe.slope;
|
|
609 |
const __m128i xStart = _mm_set1_epi32(edgeX - pixelX);
|
|
610 |
|
|
611 |
const __m128i xs0 = _mm_set1_epi32(currAe.slope);
|
|
612 |
|
|
613 |
__m128i xAdd0 = mm_mul4x32(xs0, csteps0);
|
|
614 |
__m128i xAdd1 = mm_mul4x32(xs0, csteps1);
|
|
615 |
__m128i coords0 = _mm_add_epi32(xStart, xAdd0);
|
|
616 |
__m128i coords1 = _mm_add_epi32(xStart, xAdd1);
|
|
617 |
__m128i coords = _mm_packs_epi32(coords0, coords1);
|
|
618 |
|
|
619 |
__m128i dir = _mm_set1_epi16(currAe.direction);
|
|
620 |
__m128i yMask = _mm_packs_epi32(yMask0, yMask1);
|
|
621 |
__m128i mDir = _mm_and_si128(dir, yMask);
|
|
622 |
|
|
623 |
__m128i sampleCoords = qCoords;
|
|
624 |
|
|
625 |
__m128i sw, pw;
|
|
626 |
mm_get_xmasks(coords, sampleCoords, sw, pw);
|
|
627 |
|
|
628 |
pixel.sseWinding = _mm_add_epi16(pixel.sseWinding, _mm_and_si128(pw, mDir));
|
|
629 |
scanline.sseWinding = _mm_add_epi16(scanline.sseWinding, _mm_and_si128(sw, mDir));
|
|
630 |
#endif
|
|
631 |
|
|
632 |
#undef QUEEN_COORD
|
|
633 |
|
|
634 |
}
|
|
635 |
|
|
636 |
/**
|
|
637 |
* \brief Calculate winding using one sample only.
|
|
638 |
* \note This uses most of the same code as the AA-case even though it is not
|
|
639 |
* necessary (one sample would be enough).
|
|
640 |
*/
|
|
641 |
RI_INLINE void calculateAEWindingNoAA(const Rasterizer::ActiveEdge& currAe, Rasterizer::Windings& pixel, Rasterizer::Windings& scanline, int intY, int pixelX)
|
|
642 |
{
|
|
643 |
#if defined(USE_SSE2)
|
|
644 |
|
|
645 |
#define QUEEN_COORD(Y) ((Y<<(Rasterizer::RASTERIZER_BITS - Rasterizer::SAMPLE_BITS)) + (1<<(Rasterizer::RASTERIZER_BITS-Rasterizer::SAMPLE_BITS-1)))
|
|
646 |
const int half = 1<<(Rasterizer::RASTERIZER_BITS-1);
|
|
647 |
|
|
648 |
__m128i sampleCoords = _mm_set1_epi16(half);
|
|
649 |
|
|
650 |
RI_ASSERT(Rasterizer::RASTERIZER_BITS <= 14);
|
|
651 |
|
|
652 |
const int cy = intY << Rasterizer::Y_BITS;
|
|
653 |
|
|
654 |
int py0, py1;
|
|
655 |
getVerticalSubpixels(intY, currAe.yStart, currAe.yEnd, py0, py1);
|
|
656 |
|
|
657 |
__m128i yMask;
|
|
658 |
|
|
659 |
if (py0 <= 4 && py1 >= 4)
|
|
660 |
yMask = _mm_set1_epi8(-1);
|
|
661 |
else
|
|
662 |
yMask = _mm_set1_epi8(0);
|
|
663 |
|
|
664 |
const __m128i csteps0 = _mm_set_epi32(4,4,4,4);
|
|
665 |
const __m128i csteps1 = _mm_set_epi32(4,4,4,4);
|
|
666 |
|
|
667 |
const int edgeX = currAe.xRef + (cy - (currAe.yStart & ~Rasterizer::Y_MASK)) * currAe.slope;
|
|
668 |
const __m128i xStart = _mm_set1_epi32(edgeX - pixelX);
|
|
669 |
|
|
670 |
const __m128i xs0 = _mm_set1_epi32(currAe.slope);
|
|
671 |
|
|
672 |
__m128i xAdd0 = mm_mul4x32(xs0, csteps0);
|
|
673 |
__m128i xAdd1 = mm_mul4x32(xs0, csteps1);
|
|
674 |
__m128i coords0 = _mm_add_epi32(xStart, xAdd0);
|
|
675 |
__m128i coords1 = _mm_add_epi32(xStart, xAdd1);
|
|
676 |
__m128i coords = _mm_packs_epi32(coords0, coords1);
|
|
677 |
|
|
678 |
__m128i dir = _mm_set1_epi16(currAe.direction);
|
|
679 |
__m128i mDir = _mm_and_si128(dir, yMask);
|
|
680 |
//__m128i mDir = dir;
|
|
681 |
|
|
682 |
__m128i sw, pw;
|
|
683 |
mm_get_xmasks(coords, sampleCoords, sw, pw);
|
|
684 |
|
|
685 |
pixel.sseWinding = _mm_add_epi16(pixel.sseWinding, _mm_and_si128(pw, mDir));
|
|
686 |
scanline.sseWinding = _mm_add_epi16(scanline.sseWinding, _mm_and_si128(sw, mDir));
|
|
687 |
|
|
688 |
#undef QUEEN_COORD
|
|
689 |
|
|
690 |
#else
|
|
691 |
RI_ASSERT(false); // Not implemented.
|
|
692 |
#endif
|
|
693 |
}
|
|
694 |
|
|
695 |
#if defined(USE_SSE2)
|
|
696 |
RI_INLINE static int mm_winding_to_coverage(const Rasterizer::Windings& pixel, int fillRuleMask)
|
|
697 |
{
|
|
698 |
// This version uses SSE2 counters.
|
|
699 |
__m128i mask = _mm_set1_epi16(fillRuleMask);
|
|
700 |
__m128i t = _mm_and_si128(mask, pixel.sseWinding);
|
|
701 |
__m128i z = _mm_setzero_si128();
|
|
702 |
__m128i isz = _mm_cmpeq_epi16(t, z);
|
|
703 |
__m128i ones = _mm_set1_epi16(1);
|
|
704 |
__m128i res = _mm_add_epi16(ones, isz);
|
|
705 |
__m128i add0 = _mm_add_epi16(res, _mm_shuffle_epi32(res, _MM_SHUFFLE(2, 3, 2, 3)));
|
|
706 |
__m128i add1 = _mm_add_epi16(add0, _mm_shuffle_epi32(add0, _MM_SHUFFLE(1, 1, 1, 1)));
|
|
707 |
__m128i add2 = _mm_add_epi16(add1, _mm_shufflelo_epi16(add1, _MM_SHUFFLE(1, 1, 1, 1)));
|
|
708 |
|
|
709 |
int nSamples = _mm_cvtsi128_si32(add2) & 0xff;
|
|
710 |
return nSamples;
|
|
711 |
}
|
|
712 |
#endif
|
|
713 |
|
|
714 |
#define RI_DEBUG
|
|
715 |
#if defined(RI_DEBUG)
|
|
716 |
void maybeDumpEdges(Array<Rasterizer::ActiveEdge> &edgePool)
|
|
717 |
{
|
|
718 |
return;
|
|
719 |
// \note This gives an idea about the edges at the rasterization stage.
|
|
720 |
// Input edges must be output at a different stage.
|
|
721 |
RI_PRINTF("lines = []\n");
|
|
722 |
for (int i = 0 ; i < edgePool.size(); i++)
|
|
723 |
{
|
|
724 |
const int slope = edgePool[i].slope;
|
|
725 |
int x0, x1, y0, y1;
|
|
726 |
y0 = edgePool[i].yStart;
|
|
727 |
y1 = edgePool[i].yEnd;
|
|
728 |
x0 = edgePool[i].xRef + (slope * (y0 & Rasterizer::Y_MASK));
|
|
729 |
x1 = (edgePool[i].xRef + (slope * (y1 - (y0 & ~Rasterizer::Y_MASK))))>>(Rasterizer::RASTERIZER_BITS-Rasterizer::X_BITS);
|
|
730 |
RI_PRINTF("lines += [[%d, %d], [%d, %d]]\n",x0>>(Rasterizer::RASTERIZER_BITS-Rasterizer::X_BITS),y0,x1,y1);
|
|
731 |
}
|
|
732 |
}
|
|
733 |
#endif
|
|
734 |
|
|
735 |
/*-------------------------------------------------------------------*//*!
|
|
736 |
* \brief Calls PixelPipe::pixelPipe for each pixel with coverage greater
|
|
737 |
* than zero.
|
|
738 |
* \param
|
|
739 |
* \return
|
|
740 |
* \note
|
|
741 |
*//*-------------------------------------------------------------------*/
|
|
742 |
void Rasterizer::fill()
|
|
743 |
{
|
|
744 |
if(m_scissor && !m_scissorEdges.size())
|
|
745 |
return; //scissoring is on, but there are no scissor rectangles => nothing is visible
|
|
746 |
|
|
747 |
int firstAe = 0;
|
|
748 |
|
|
749 |
//proceed scanline by scanline
|
|
750 |
//keep track of edges that can intersect the pixel filters of the current scanline (Active Edge Table)
|
|
751 |
//until all pixels of the scanline have been processed
|
|
752 |
// for all sampling points of the current pixel
|
|
753 |
// determine the winding number using edge functions
|
|
754 |
// add filter weight to coverage
|
|
755 |
// divide coverage by the number of samples
|
|
756 |
// determine a run of pixels with constant coverage
|
|
757 |
// call fill callback for each pixel of the run
|
|
758 |
|
|
759 |
const int fillRuleMask = m_fillRuleMask;
|
|
760 |
|
|
761 |
int bbminx = (m_edgeMin.x >> X_BITS);
|
|
762 |
int bbminy = (m_edgeMin.y >> Y_BITS);
|
|
763 |
int bbmaxx = (m_edgeMax.x >> X_BITS)+1;
|
|
764 |
int bbmaxy = (m_edgeMax.y >> Y_BITS)+1;
|
|
765 |
int sx = RI_INT_MAX(m_vpx, bbminx);
|
|
766 |
int ex = RI_INT_MIN(m_vpx+m_vpwidth, bbmaxx);
|
|
767 |
int sy = RI_INT_MAX(m_vpy, bbminy);
|
|
768 |
int ey = RI_INT_MIN(m_vpy+m_vpheight, bbmaxy);
|
|
769 |
if(sx < m_covMinx) m_covMinx = sx;
|
|
770 |
if(sy < m_covMiny) m_covMiny = sy;
|
|
771 |
if(ex > m_covMaxx) m_covMaxx = ex;
|
|
772 |
if(ey > m_covMaxy) m_covMaxy = ey;
|
|
773 |
|
|
774 |
#if 0
|
|
775 |
// Dump edges:
|
|
776 |
static bool dump = true;
|
|
777 |
if (dump)
|
|
778 |
{
|
|
779 |
RI_PRINTF("lines = []\n");
|
|
780 |
for (int ie = 0; dump && ie < m_edgePool.size(); ie++)
|
|
781 |
{
|
|
782 |
RI_PRINTF("lines += [[%d, %d], [%d, %d]]\n",m_edgePool[ie].v0.x, m_edgePool[ie].v0.y, m_edgePool[ie].v1.x, m_edgePool[ie].v1.y);
|
|
783 |
}
|
|
784 |
dump = false;
|
|
785 |
}
|
|
786 |
|
|
787 |
#endif
|
|
788 |
int debugMagic = 0;
|
|
789 |
|
|
790 |
m_aet.clear();
|
|
791 |
|
|
792 |
#if defined(RI_DEBUG)
|
|
793 |
maybeDumpEdges(m_edgePool);
|
|
794 |
#endif
|
|
795 |
|
|
796 |
//fill the screen
|
|
797 |
for(int j = sy; j < ey; j++)
|
|
798 |
{
|
|
799 |
Windings scanlineWinding;
|
|
800 |
const int cminy = j << Y_BITS;
|
|
801 |
|
|
802 |
if (m_scissor)
|
|
803 |
{
|
|
804 |
// Gather scissor edges intersecting this scanline
|
|
805 |
// \todo Don't clear, remove unused instead!
|
|
806 |
m_scissorAet.clear();
|
|
807 |
|
|
808 |
for(int e = 0; e < m_scissorEdges.size(); e++)
|
|
809 |
{
|
|
810 |
const ScissorEdge& se = m_scissorEdges[e];
|
|
811 |
|
|
812 |
if(j >= se.miny && j < se.maxy)
|
|
813 |
m_scissorAet.push_back(m_scissorEdges[e]); //throws bad_alloc
|
|
814 |
}
|
|
815 |
|
|
816 |
//sort scissor AET by edge x
|
|
817 |
if (m_scissor)
|
|
818 |
m_scissorAet.sort();
|
|
819 |
}
|
|
820 |
|
|
821 |
// Drop unused edges, update remaining.
|
|
822 |
// \todo Combine with full sweep. Use a sort-friendly edge-discard.
|
|
823 |
for (int iae = firstAe; iae < m_aet.size(); iae++)
|
|
824 |
{
|
|
825 |
ActiveEdge& ae = m_aet[iae];
|
|
826 |
|
|
827 |
if (cminy >= ae.yEnd)
|
|
828 |
{
|
|
829 |
m_aet[iae] = m_aet[firstAe];
|
|
830 |
firstAe++;
|
|
831 |
continue;
|
|
832 |
}
|
|
833 |
|
|
834 |
/* Update existing coordinates */
|
|
835 |
// \todo AND instead of shift. See other places also.
|
|
836 |
const int y0 = (ae.yStart & ~Y_MASK);
|
|
837 |
const int x = ae.xRef + ((j << Y_BITS) - y0) * ae.slope;
|
|
838 |
ae.minx = x >> RASTERIZER_BITS;
|
|
839 |
ae.maxx = (x + ae.slope * (1<<Y_BITS)) >> RASTERIZER_BITS;
|
|
840 |
|
|
841 |
if (ae.minx > ae.maxx)
|
|
842 |
RI_ANY_SWAP(ActiveEdge::XCoord, ae.minx, ae.maxx);
|
|
843 |
|
|
844 |
// If the edge is not visible, "mark" it as immediately applicable
|
|
845 |
// \todo Verify that this is the correct procedure.
|
|
846 |
|
|
847 |
if (ae.maxx < 0)
|
|
848 |
ae.minx = ae.maxx = LEFT_DISCARD_SHORT;
|
|
849 |
}
|
|
850 |
|
|
851 |
/* Add new edges */
|
|
852 |
|
|
853 |
RIuint32 aeIndex = m_edges[j];
|
|
854 |
while (aeIndex != EDGE_TERMINATOR)
|
|
855 |
{
|
|
856 |
const ActiveEdge& ae = m_edgePool[aeIndex];
|
|
857 |
m_aet.push_back(ae); // \todo Just copy pointers?
|
|
858 |
aeIndex = ae.next;
|
|
859 |
}
|
|
860 |
|
|
861 |
if (firstAe >= m_aet.size())
|
|
862 |
{
|
|
863 |
RI_ASSERT(firstAe == m_aet.size());
|
|
864 |
continue; //no edges on the whole scanline, skip it
|
|
865 |
}
|
|
866 |
|
|
867 |
//sort AET by edge minx
|
|
868 |
m_aet.sort(firstAe, m_aet.size() - 1);
|
|
869 |
|
|
870 |
// \todo Optimize adding and updating the edges?
|
|
871 |
if (m_scissor && !m_scissorAet.size())
|
|
872 |
continue; // Scissoring is on, but there are no scissor rectangles on this scanline.
|
|
873 |
|
|
874 |
//fill the scanline
|
|
875 |
int scissorWinding = m_scissor ? 0 : 1; //if scissoring is off, winding is always 1
|
|
876 |
int scissorIndex = 0;
|
|
877 |
int aes = firstAe;
|
|
878 |
int aen = firstAe;
|
|
879 |
|
|
880 |
RI_ASSERT(sx >= 0);
|
|
881 |
|
|
882 |
#if 1
|
|
883 |
if (m_aa)
|
|
884 |
{
|
|
885 |
while ((aen < m_aet.size()) && (m_aet[aen].maxx < 0))
|
|
886 |
{
|
|
887 |
applyLeftEdge(m_aet[aen], scanlineWinding, j);
|
|
888 |
aen++;
|
|
889 |
}
|
|
890 |
}
|
|
891 |
else
|
|
892 |
{
|
|
893 |
while ((aen < m_aet.size()) && (m_aet[aen].maxx < 0))
|
|
894 |
{
|
|
895 |
applyLeftEdgeNoAA(m_aet[aen], scanlineWinding, j);
|
|
896 |
aen++;
|
|
897 |
}
|
|
898 |
}
|
|
899 |
|
|
900 |
#if defined(RI_DEBUG)
|
|
901 |
for (int a = aen; a < m_aet.size(); a++)
|
|
902 |
{
|
|
903 |
RI_ASSERT(m_aet[a].maxx >= 0);
|
|
904 |
}
|
|
905 |
#endif
|
|
906 |
#endif
|
|
907 |
|
|
908 |
// \todo Combine this with the first check or reorganize the "clipping".
|
|
909 |
if (aen >= m_aet.size())
|
|
910 |
continue; // No edges within viewport. Can happen atm. when all edges are "left".
|
|
911 |
|
|
912 |
for(int i = sx; i < ex;)
|
|
913 |
{
|
|
914 |
//find edges that intersect or are to the left of the pixel antialiasing filter
|
|
915 |
while(aes < m_aet.size() && (i + 1) >= m_aet[aes].minx)
|
|
916 |
aes++;
|
|
917 |
//edges [0,aes[ may have an effect on winding, and need to be evaluated while sampling
|
|
918 |
|
|
919 |
// RIint8 winding[SF_SAMPLES];
|
|
920 |
Windings pixelWinding;
|
|
921 |
|
|
922 |
pixelWinding = scanlineWinding;
|
|
923 |
|
|
924 |
if (m_aa)
|
|
925 |
{
|
|
926 |
for(int e = aen; e < aes; e++)
|
|
927 |
{
|
|
928 |
const ActiveEdge& currAe = m_aet[e];
|
|
929 |
calculateAEWinding(currAe, pixelWinding, scanlineWinding, j, i << RASTERIZER_BITS);
|
|
930 |
}
|
|
931 |
}
|
|
932 |
else
|
|
933 |
{
|
|
934 |
for(int e = aen; e < aes; e++)
|
|
935 |
{
|
|
936 |
const ActiveEdge& currAe = m_aet[e];
|
|
937 |
calculateAEWindingNoAA(currAe, pixelWinding, scanlineWinding, j, i << RASTERIZER_BITS);
|
|
938 |
}
|
|
939 |
}
|
|
940 |
|
|
941 |
//compute coverage
|
|
942 |
int coverageSamples = 0;
|
|
943 |
#if !defined(USE_SSE2)
|
|
944 |
|
|
945 |
for (int s = 0; s < SF_SAMPLES; s++)
|
|
946 |
{
|
|
947 |
if(pixelWinding.winding[s])
|
|
948 |
{
|
|
949 |
coverageSamples++;
|
|
950 |
}
|
|
951 |
}
|
|
952 |
#else
|
|
953 |
coverageSamples = mm_winding_to_coverage(pixelWinding, fillRuleMask);
|
|
954 |
_mm_empty();
|
|
955 |
#endif
|
|
956 |
|
|
957 |
//constant coverage optimization:
|
|
958 |
//scan AET from left to right and skip all the edges that are completely to the left of the pixel filter.
|
|
959 |
//since AET is sorted by minx, the edge we stop at is the leftmost of the edges we haven't passed yet.
|
|
960 |
//if that edge is to the right of this pixel, coverage is constant between this pixel and the start of the edge.
|
|
961 |
while(aen < m_aet.size() && m_aet[aen].maxx < i)
|
|
962 |
aen++;
|
|
963 |
|
|
964 |
int endSpan = m_vpx + m_vpwidth; // endSpan is the first pixel NOT part of the span
|
|
965 |
|
|
966 |
if(aen < m_aet.size())
|
|
967 |
{
|
|
968 |
endSpan = RI_INT_MAX(i+1, RI_INT_MIN(endSpan, m_aet[aen].minx));
|
|
969 |
}
|
|
970 |
|
|
971 |
//fill a run of pixels with constant coverage
|
|
972 |
if(coverageSamples)
|
|
973 |
{
|
|
974 |
|
|
975 |
if (!m_scissor)
|
|
976 |
{
|
|
977 |
int fillStartX = i; /* Inclusive */
|
|
978 |
pushSpan(fillStartX, j, (endSpan - fillStartX), coverageSamples);
|
|
979 |
}
|
|
980 |
else // (scissor)
|
|
981 |
{
|
|
982 |
int fillStartX = i;
|
|
983 |
//update scissor winding number
|
|
984 |
|
|
985 |
/* \todo Sort the scissor edges and skip unnecessary checks when scissors are used */
|
|
986 |
while (scissorIndex < m_scissorAet.size() && m_scissorAet[scissorIndex].x <= fillStartX)
|
|
987 |
{
|
|
988 |
scissorWinding += m_scissorAet[scissorIndex++].direction;
|
|
989 |
}
|
|
990 |
|
|
991 |
while (!scissorWinding && scissorIndex < m_scissorAet.size() && m_scissorAet[scissorIndex].x < endSpan)
|
|
992 |
{
|
|
993 |
fillStartX = m_scissorAet[scissorIndex].x;
|
|
994 |
scissorWinding += m_scissorAet[scissorIndex++].direction;
|
|
995 |
RI_ASSERT(fillStartX >= i);
|
|
996 |
}
|
|
997 |
|
|
998 |
RI_ASSERT(scissorWinding >= 0);
|
|
999 |
|
|
1000 |
int endScissorSpan = endSpan;
|
|
1001 |
|
|
1002 |
while (scissorWinding && fillStartX < endSpan && (scissorIndex < m_scissorAet.size()))
|
|
1003 |
{
|
|
1004 |
|
|
1005 |
// Determine the end of renderable area:
|
|
1006 |
while (scissorWinding && scissorIndex < m_scissorAet.size() && m_scissorAet[scissorIndex].x <= endSpan)
|
|
1007 |
{
|
|
1008 |
endScissorSpan = m_scissorAet[scissorIndex].x;
|
|
1009 |
scissorWinding += m_scissorAet[scissorIndex++].direction;
|
|
1010 |
}
|
|
1011 |
|
|
1012 |
RI_ASSERT(fillStartX >= i);
|
|
1013 |
RI_ASSERT(endScissorSpan <= endSpan);
|
|
1014 |
|
|
1015 |
pushSpan(fillStartX, j, (endScissorSpan - fillStartX), coverageSamples);
|
|
1016 |
fillStartX = endScissorSpan;
|
|
1017 |
endScissorSpan = endSpan;
|
|
1018 |
|
|
1019 |
// Skip until within drawable area
|
|
1020 |
while (!scissorWinding && scissorIndex < m_scissorAet.size() && m_scissorAet[scissorIndex].x < endSpan)
|
|
1021 |
{
|
|
1022 |
fillStartX = m_scissorAet[scissorIndex].x;
|
|
1023 |
scissorWinding += m_scissorAet[scissorIndex++].direction;
|
|
1024 |
}
|
|
1025 |
|
|
1026 |
}
|
|
1027 |
}
|
|
1028 |
}
|
|
1029 |
i = endSpan;
|
|
1030 |
}
|
|
1031 |
}
|
|
1032 |
commitSpans();
|
|
1033 |
#if defined(USE_SSE2)
|
|
1034 |
_mm_empty();
|
|
1035 |
#endif
|
|
1036 |
clear();
|
|
1037 |
}
|
|
1038 |
|
|
1039 |
RI_INLINE void Rasterizer::commitSpans()
|
|
1040 |
{
|
|
1041 |
if (!m_nSpans)
|
|
1042 |
return;
|
|
1043 |
|
|
1044 |
m_pixelPipe->fillSpans(m_ppVariants, m_spanCache, m_nSpans);
|
|
1045 |
m_nSpans = 0;
|
|
1046 |
|
|
1047 |
}
|
|
1048 |
|
|
1049 |
RI_INLINE void Rasterizer::pushSpan(int x, int y, int len, int coverage)
|
|
1050 |
{
|
|
1051 |
//printf("x: %d, y: %d, len: %d, coverage: %d\n", x, y, len, coverage);
|
|
1052 |
// \todo Check what causes this with scissors
|
|
1053 |
if (len <= 0) return;
|
|
1054 |
//RI_ASSERT(len > 0);
|
|
1055 |
|
|
1056 |
Span& span = m_spanCache[m_nSpans];
|
|
1057 |
|
|
1058 |
span.x0 = x;
|
|
1059 |
span.y = y;
|
|
1060 |
span.len = (RIuint16)len;
|
|
1061 |
span.coverage = coverage;
|
|
1062 |
|
|
1063 |
m_nSpans++;
|
|
1064 |
|
|
1065 |
if (m_nSpans == N_CACHED_SPANS)
|
|
1066 |
{
|
|
1067 |
commitSpans();
|
|
1068 |
}
|
|
1069 |
}
|
|
1070 |
|
|
1071 |
//=======================================================================
|
|
1072 |
|
|
1073 |
} //namespace OpenVGRI
|