1/*-------------------------------------------------------------------------
2 * drawElements Quality Program Reference Renderer
3 * -----------------------------------------------
4 *
5 * Copyright 2014 The Android Open Source Project
6 *
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
10 *
11 *      http://www.apache.org/licenses/LICENSE-2.0
12 *
13 * Unless required by applicable law or agreed to in writing, software
14 * distributed under the License is distributed on an "AS IS" BASIS,
15 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16 * See the License for the specific language governing permissions and
17 * limitations under the License.
18 *
19 *//*!
20 * \file
21 * \brief Reference rasterizer
22 *//*--------------------------------------------------------------------*/
23
24#include "rrRasterizer.hpp"
25#include "deMath.h"
26#include "tcuVectorUtil.hpp"
27
28namespace rr
29{
30
31inline deInt64 toSubpixelCoord (float v)
32{
33	return (deInt64)(v * (1<<RASTERIZER_SUBPIXEL_BITS) + (v < 0.f ? -0.5f : 0.5f));
34}
35
36inline deInt64 toSubpixelCoord (deInt32 v)
37{
38	return v << RASTERIZER_SUBPIXEL_BITS;
39}
40
41inline deInt32 ceilSubpixelToPixelCoord (deInt64 coord, bool fillEdge)
42{
43	if (coord >= 0)
44		return (deInt32)((coord + ((1ll<<RASTERIZER_SUBPIXEL_BITS) - (fillEdge ? 0 : 1))) >> RASTERIZER_SUBPIXEL_BITS);
45	else
46		return (deInt32)((coord + (fillEdge ? 1 : 0)) >> RASTERIZER_SUBPIXEL_BITS);
47}
48
49inline deInt32 floorSubpixelToPixelCoord (deInt64 coord, bool fillEdge)
50{
51	if (coord >= 0)
52		return (deInt32)((coord - (fillEdge ? 1 : 0)) >> RASTERIZER_SUBPIXEL_BITS);
53	else
54		return (deInt32)((coord - ((1ll<<RASTERIZER_SUBPIXEL_BITS) - (fillEdge ? 0 : 1))) >> RASTERIZER_SUBPIXEL_BITS);
55}
56
57static inline void initEdgeCCW (EdgeFunction& edge, const HorizontalFill horizontalFill, const VerticalFill verticalFill, const deInt64 x0, const deInt64 y0, const deInt64 x1, const deInt64 y1)
58{
59	// \note See EdgeFunction documentation for details.
60
61	const deInt64	xd			= x1-x0;
62	const deInt64	yd			= y1-y0;
63	bool			inclusive	= false;	//!< Inclusive in CCW orientation.
64
65	if (yd == 0)
66		inclusive = verticalFill == FILL_BOTTOM ? xd >= 0 : xd <= 0;
67	else
68		inclusive = horizontalFill == FILL_LEFT ? yd <= 0 : yd >= 0;
69
70	edge.a			= (y0 - y1);
71	edge.b			= (x1 - x0);
72	edge.c			= x0*y1 - y0*x1;
73	edge.inclusive	= inclusive; //!< \todo [pyry] Swap for CW triangles
74}
75
76static inline void reverseEdge (EdgeFunction& edge)
77{
78	edge.a			= -edge.a;
79	edge.b			= -edge.b;
80	edge.c			= -edge.c;
81	edge.inclusive	= !edge.inclusive;
82}
83
84static inline deInt64 evaluateEdge (const EdgeFunction& edge, const deInt64 x, const deInt64 y)
85{
86	return edge.a*x + edge.b*y + edge.c;
87}
88
89static inline bool isInsideCCW (const EdgeFunction& edge, const deInt64 edgeVal)
90{
91	return edge.inclusive ? (edgeVal >= 0) : (edgeVal > 0);
92}
93
94namespace LineRasterUtil
95{
96
97struct SubpixelLineSegment
98{
99	const tcu::Vector<deInt64,2>	m_v0;
100	const tcu::Vector<deInt64,2>	m_v1;
101
102	SubpixelLineSegment (const tcu::Vector<deInt64,2>& v0, const tcu::Vector<deInt64,2>& v1)
103		: m_v0(v0)
104		, m_v1(v1)
105	{
106	}
107
108	tcu::Vector<deInt64,2> direction (void) const
109	{
110		return m_v1 - m_v0;
111	}
112};
113
114enum LINE_SIDE
115{
116	LINE_SIDE_INTERSECT = 0,
117	LINE_SIDE_LEFT,
118	LINE_SIDE_RIGHT
119};
120
121static tcu::Vector<deInt64,2> toSubpixelVector (const tcu::Vec2& v)
122{
123	return tcu::Vector<deInt64,2>(toSubpixelCoord(v.x()), toSubpixelCoord(v.y()));
124}
125
126static tcu::Vector<deInt64,2> toSubpixelVector (const tcu::IVec2& v)
127{
128	return tcu::Vector<deInt64,2>(toSubpixelCoord(v.x()), toSubpixelCoord(v.y()));
129}
130
131#if defined(DE_DEBUG)
132static bool isTheCenterOfTheFragment (const tcu::Vector<deInt64,2>& a)
133{
134	const deUint64 pixelSize = 1ll << (RASTERIZER_SUBPIXEL_BITS);
135	const deUint64 halfPixel = 1ll << (RASTERIZER_SUBPIXEL_BITS-1);
136	return	((a.x() & (pixelSize-1)) == halfPixel &&
137				(a.y() & (pixelSize-1)) == halfPixel);
138}
139
140static bool inViewport (const tcu::IVec2& p, const tcu::IVec4& viewport)
141{
142	return	p.x() >= viewport.x() &&
143			p.y() >= viewport.y() &&
144			p.x() <  viewport.x() + viewport.z() &&
145			p.y() <  viewport.y() + viewport.w();
146}
147#endif // DE_DEBUG
148
149// returns true if vertex is on the left side of the line
150static bool vertexOnLeftSideOfLine (const tcu::Vector<deInt64,2>& p, const SubpixelLineSegment& l)
151{
152	const tcu::Vector<deInt64,2> u = l.direction();
153	const tcu::Vector<deInt64,2> v = ( p - l.m_v0);
154	const deInt64 crossProduct = (u.x() * v.y() - u.y() * v.x());
155	return crossProduct < 0;
156}
157
158// returns true if vertex is on the right side of the line
159static bool vertexOnRightSideOfLine (const tcu::Vector<deInt64,2>& p, const SubpixelLineSegment& l)
160{
161	const tcu::Vector<deInt64,2> u = l.direction();
162	const tcu::Vector<deInt64,2> v = ( p - l.m_v0);
163	const deInt64 crossProduct = (u.x() * v.y() - u.y() * v.x());
164	return crossProduct > 0;
165}
166
167// returns true if vertex is on the line
168static bool vertexOnLine (const tcu::Vector<deInt64,2>& p, const SubpixelLineSegment& l)
169{
170	const tcu::Vector<deInt64,2> u = l.direction();
171	const tcu::Vector<deInt64,2> v = ( p - l.m_v0);
172	const deInt64 crossProduct = (u.x() * v.y() - u.y() * v.x());
173	return crossProduct == 0; // cross product == 0
174}
175
176// returns true if vertex is on the line segment
177static bool vertexOnLineSegment (const tcu::Vector<deInt64,2>& p, const SubpixelLineSegment& l)
178{
179	if (!vertexOnLine(p, l))
180		return false;
181
182	const tcu::Vector<deInt64,2> v	= l.direction();
183	const tcu::Vector<deInt64,2> u1	= ( p - l.m_v0);
184	const tcu::Vector<deInt64,2> u2	= ( p - l.m_v1);
185
186	if (v.x() == 0 && v.y() == 0)
187		return false;
188
189	return	tcu::dot( v, u1) >= 0 &&
190			tcu::dot(-v, u2) >= 0; // dot (A->B, A->V) >= 0 and dot (B->A, B->V) >= 0
191}
192
193static LINE_SIDE getVertexSide (const tcu::Vector<deInt64,2>& v, const SubpixelLineSegment& l)
194{
195	if (vertexOnLeftSideOfLine(v, l))
196		return LINE_SIDE_LEFT;
197	else if (vertexOnRightSideOfLine(v, l))
198		return LINE_SIDE_RIGHT;
199	else if (vertexOnLine(v, l))
200		return LINE_SIDE_INTERSECT;
201	else
202	{
203		DE_ASSERT(false);
204		return LINE_SIDE_INTERSECT;
205	}
206}
207
208// returns true if angle between line and given cornerExitNormal is in range (-45, 45)
209bool lineInCornerAngleRange (const SubpixelLineSegment& line, const tcu::Vector<deInt64,2>& cornerExitNormal)
210{
211	// v0 -> v1 has angle difference to cornerExitNormal in range (-45, 45)
212	const tcu::Vector<deInt64,2> v = line.direction();
213	const deInt64 dotProduct = dot(v, cornerExitNormal);
214
215	// dotProduct > |v1-v0|*|cornerExitNormal|/sqrt(2)
216	if (dotProduct < 0)
217		return false;
218	return 2 * dotProduct * dotProduct > tcu::lengthSquared(v)*tcu::lengthSquared(cornerExitNormal);
219}
220
221// returns true if angle between line and given cornerExitNormal is in range (-135, 135)
222bool lineInCornerOutsideAngleRange (const SubpixelLineSegment& line, const tcu::Vector<deInt64,2>& cornerExitNormal)
223{
224	// v0 -> v1 has angle difference to cornerExitNormal in range (-135, 135)
225	const tcu::Vector<deInt64,2> v = line.direction();
226	const deInt64 dotProduct = dot(v, cornerExitNormal);
227
228	// dotProduct > -|v1-v0|*|cornerExitNormal|/sqrt(2)
229	if (dotProduct >= 0)
230		return true;
231	return 2 * (-dotProduct) * (-dotProduct) < tcu::lengthSquared(v)*tcu::lengthSquared(cornerExitNormal);
232}
233
234bool doesLineSegmentExitDiamond (const SubpixelLineSegment& line, const tcu::Vector<deInt64,2>& diamondCenter)
235{
236	DE_ASSERT(isTheCenterOfTheFragment(diamondCenter));
237
238	// Diamond Center is at diamondCenter in subpixel coords
239
240	const deInt64 halfPixel = 1ll << (RASTERIZER_SUBPIXEL_BITS-1);
241
242	// Reject distant diamonds early
243	{
244		const tcu::Vector<deInt64,2>	u				= line.direction();
245		const tcu::Vector<deInt64,2>	v				= (diamondCenter - line.m_v0);
246		const deInt64					crossProduct	= (u.x() * v.y() - u.y() * v.x());
247
248		// crossProduct = |p| |l| sin(theta)
249		// distanceFromLine = |p| sin(theta)
250		// => distanceFromLine = crossProduct / |l|
251		//
252		// |distanceFromLine| > C
253		// => distanceFromLine^2 > C^2
254		// => crossProduct^2 / |l|^2 > C^2
255		// => crossProduct^2 > |l|^2 * C^2
256
257		const deInt64	floorSqrtMaxInt64			= 3037000499LL; //!< floor(sqrt(MAX_INT64))
258
259		const deInt64	broadRejectDistance			= 2 * halfPixel;
260		const deInt64	broadRejectDistanceSquared	= broadRejectDistance * broadRejectDistance;
261		const bool		crossProductOverflows		= (crossProduct > floorSqrtMaxInt64 || crossProduct < -floorSqrtMaxInt64);
262		const deInt64	crossProductSquared			= (crossProductOverflows) ? (0) : (crossProduct * crossProduct); // avoid overflow
263		const deInt64	lineLengthSquared			= tcu::lengthSquared(u);
264		const bool		limitValueCouldOverflow		= ((64 - deClz64(lineLengthSquared)) + (64 - deClz64(broadRejectDistanceSquared))) > 63;
265		const deInt64	limitValue					= (limitValueCouldOverflow) ? (0) : (lineLengthSquared * broadRejectDistanceSquared); // avoid overflow
266
267		// only cross overflows
268		if (crossProductOverflows && !limitValueCouldOverflow)
269			return false;
270
271		// both representable
272		if (!crossProductOverflows && !limitValueCouldOverflow)
273		{
274			if (crossProductSquared > limitValue)
275				return false;
276		}
277	}
278
279	const struct DiamondBound
280	{
281		tcu::Vector<deInt64,2>	p0;
282		tcu::Vector<deInt64,2>	p1;
283		bool					edgeInclusive; // would a point on the bound be inside of the region
284	} bounds[] =
285	{
286		{ diamondCenter + tcu::Vector<deInt64,2>(0,				-halfPixel),	diamondCenter + tcu::Vector<deInt64,2>(-halfPixel,	0),				 false	},
287		{ diamondCenter + tcu::Vector<deInt64,2>(-halfPixel,	0),				diamondCenter + tcu::Vector<deInt64,2>(0,			halfPixel),		 false	},
288		{ diamondCenter + tcu::Vector<deInt64,2>(0,				halfPixel),		diamondCenter + tcu::Vector<deInt64,2>(halfPixel,	0),				 true	},
289		{ diamondCenter + tcu::Vector<deInt64,2>(halfPixel,		0),				diamondCenter + tcu::Vector<deInt64,2>(0,			-halfPixel),	 true	},
290	};
291
292	const struct DiamondCorners
293	{
294		enum CORNER_EDGE_CASE_BEHAVIOR
295		{
296			CORNER_EDGE_CASE_NONE,							// if the line intersects just a corner, no entering or exiting
297			CORNER_EDGE_CASE_HIT,							// if the line intersects just a corner, entering and exit
298			CORNER_EDGE_CASE_HIT_FIRST_QUARTER,				// if the line intersects just a corner and the line has either endpoint in (+X,-Y) direction (preturbing moves the line inside)
299			CORNER_EDGE_CASE_HIT_SECOND_QUARTER				// if the line intersects just a corner and the line has either endpoint in (+X,+Y) direction (preturbing moves the line inside)
300		};
301		enum CORNER_START_CASE_BEHAVIOR
302		{
303			CORNER_START_CASE_NONE,							// the line starting point is outside, no exiting
304			CORNER_START_CASE_OUTSIDE,						// exit, if line does not intersect the region (preturbing moves the start point inside)
305			CORNER_START_CASE_POSITIVE_Y_45,				// exit, if line the angle of line vector and X-axis is in range (0, 45] in positive Y side.
306			CORNER_START_CASE_NEGATIVE_Y_45					// exit, if line the angle of line vector and X-axis is in range [0, 45] in negative Y side.
307		};
308		enum CORNER_END_CASE_BEHAVIOR
309		{
310			CORNER_END_CASE_NONE,							// end is inside, no exiting (preturbing moves the line end inside)
311			CORNER_END_CASE_DIRECTION,						// exit, if line intersected the region (preturbing moves the line end outside)
312			CORNER_END_CASE_DIRECTION_AND_FIRST_QUARTER,	// exit, if line intersected the region, or line originates from (+X,-Y) direction (preturbing moves the line end outside)
313			CORNER_END_CASE_DIRECTION_AND_SECOND_QUARTER	// exit, if line intersected the region, or line originates from (+X,+Y) direction (preturbing moves the line end outside)
314		};
315
316		tcu::Vector<deInt64,2>		dp;
317		bool						pointInclusive;			// would a point in this corner intersect with the region
318		CORNER_EDGE_CASE_BEHAVIOR	lineBehavior;			// would a line segment going through this corner intersect with the region
319		CORNER_START_CASE_BEHAVIOR	startBehavior;			// how the corner behaves if the start point at the corner
320		CORNER_END_CASE_BEHAVIOR	endBehavior;			// how the corner behaves if the end point at the corner
321	} corners[] =
322	{
323		{ tcu::Vector<deInt64,2>(0,				-halfPixel),	false,	DiamondCorners::CORNER_EDGE_CASE_HIT_SECOND_QUARTER,	DiamondCorners::CORNER_START_CASE_POSITIVE_Y_45,	DiamondCorners::CORNER_END_CASE_DIRECTION_AND_SECOND_QUARTER},
324		{ tcu::Vector<deInt64,2>(-halfPixel,	0),				false,	DiamondCorners::CORNER_EDGE_CASE_NONE,					DiamondCorners::CORNER_START_CASE_NONE,				DiamondCorners::CORNER_END_CASE_DIRECTION					},
325		{ tcu::Vector<deInt64,2>(0,				halfPixel),		false,	DiamondCorners::CORNER_EDGE_CASE_HIT_FIRST_QUARTER,		DiamondCorners::CORNER_START_CASE_NEGATIVE_Y_45,	DiamondCorners::CORNER_END_CASE_DIRECTION_AND_FIRST_QUARTER	},
326		{ tcu::Vector<deInt64,2>(halfPixel,		0),				true,	DiamondCorners::CORNER_EDGE_CASE_HIT,					DiamondCorners::CORNER_START_CASE_OUTSIDE,			DiamondCorners::CORNER_END_CASE_NONE						},
327	};
328
329	// Corner cases at the corners
330	for (int ndx = 0; ndx < DE_LENGTH_OF_ARRAY(corners); ++ndx)
331	{
332		const tcu::Vector<deInt64,2> p	= diamondCenter + corners[ndx].dp;
333		const bool intersectsAtCorner	= LineRasterUtil::vertexOnLineSegment(p, line);
334
335		if (!intersectsAtCorner)
336			continue;
337
338		// line segment body intersects with the corner
339		if (p != line.m_v0 && p != line.m_v1)
340		{
341			if (corners[ndx].lineBehavior == DiamondCorners::CORNER_EDGE_CASE_HIT)
342				return true;
343
344			// endpoint in (+X, -Y) (X or Y may be 0) direction <==> x*y <= 0
345			if (corners[ndx].lineBehavior == DiamondCorners::CORNER_EDGE_CASE_HIT_FIRST_QUARTER &&
346				(line.direction().x() * line.direction().y()) <= 0)
347				return true;
348
349			// endpoint in (+X, +Y) (Y > 0) direction <==> x*y > 0
350			if (corners[ndx].lineBehavior == DiamondCorners::CORNER_EDGE_CASE_HIT_SECOND_QUARTER &&
351				(line.direction().x() * line.direction().y()) > 0)
352				return true;
353		}
354
355		// line exits the area at the corner
356		if (lineInCornerAngleRange(line, corners[ndx].dp))
357		{
358			const bool startIsInside = corners[ndx].pointInclusive || p != line.m_v0;
359			const bool endIsOutside = !corners[ndx].pointInclusive || p != line.m_v1;
360
361			// starting point is inside the region and end endpoint is outside
362			if (startIsInside && endIsOutside)
363				return true;
364		}
365
366		// line end is at the corner
367		if (p == line.m_v1)
368		{
369			if (corners[ndx].endBehavior == DiamondCorners::CORNER_END_CASE_DIRECTION ||
370				corners[ndx].endBehavior == DiamondCorners::CORNER_END_CASE_DIRECTION_AND_FIRST_QUARTER ||
371				corners[ndx].endBehavior == DiamondCorners::CORNER_END_CASE_DIRECTION_AND_SECOND_QUARTER)
372			{
373				// did the line intersect the region
374				if (lineInCornerAngleRange(line, corners[ndx].dp))
375					return true;
376			}
377
378			// due to the perturbed endpoint, lines at this the angle will cause and enter-exit pair
379			if (corners[ndx].endBehavior == DiamondCorners::CORNER_END_CASE_DIRECTION_AND_FIRST_QUARTER &&
380				line.direction().x() < 0 &&
381				line.direction().y() > 0)
382				return true;
383			if (corners[ndx].endBehavior == DiamondCorners::CORNER_END_CASE_DIRECTION_AND_SECOND_QUARTER &&
384				line.direction().x() > 0 &&
385				line.direction().y() > 0)
386				return true;
387		}
388
389		// line start is at the corner
390		if (p == line.m_v0)
391		{
392			if (corners[ndx].startBehavior == DiamondCorners::CORNER_START_CASE_OUTSIDE)
393			{
394				// if the line is not going inside, it will exit
395				if (lineInCornerOutsideAngleRange(line, corners[ndx].dp))
396					return true;
397			}
398
399			// exit, if line the angle between line vector and X-axis is in range (0, 45] in positive Y side.
400			if (corners[ndx].startBehavior == DiamondCorners::CORNER_START_CASE_POSITIVE_Y_45 &&
401				line.direction().x() > 0 &&
402				line.direction().y() > 0 &&
403				line.direction().y() <= line.direction().x())
404				return true;
405
406			// exit, if line the angle between line vector and X-axis is in range [0, 45] in negative Y side.
407			if (corners[ndx].startBehavior == DiamondCorners::CORNER_START_CASE_NEGATIVE_Y_45 &&
408				 line.direction().x() > 0 &&
409				 line.direction().y() <= 0 &&
410				-line.direction().y() <= line.direction().x())
411				return true;
412		}
413	}
414
415	// Does the line intersect boundary at the left == exits the diamond
416	for (int ndx = 0; ndx < DE_LENGTH_OF_ARRAY(bounds); ++ndx)
417	{
418		const bool startVertexInside =	LineRasterUtil::vertexOnLeftSideOfLine						(line.m_v0, LineRasterUtil::SubpixelLineSegment(bounds[ndx].p0, bounds[ndx].p1)) ||
419										(bounds[ndx].edgeInclusive && LineRasterUtil::vertexOnLine	(line.m_v0, LineRasterUtil::SubpixelLineSegment(bounds[ndx].p0, bounds[ndx].p1)));
420		const bool endVertexInside =	LineRasterUtil::vertexOnLeftSideOfLine						(line.m_v1, LineRasterUtil::SubpixelLineSegment(bounds[ndx].p0, bounds[ndx].p1)) ||
421										(bounds[ndx].edgeInclusive && LineRasterUtil::vertexOnLine	(line.m_v1, LineRasterUtil::SubpixelLineSegment(bounds[ndx].p0, bounds[ndx].p1)));
422
423		// start must be on inside this half space (left or at the inclusive boundary)
424		if (!startVertexInside)
425			continue;
426
427		// end must be outside of this half-space (right or at non-inclusive boundary)
428		if (endVertexInside)
429			continue;
430
431		// Does the line via v0 and v1 intersect the line segment p0-p1
432		// <==> p0 and p1 are the different sides (LEFT, RIGHT) of the v0-v1 line.
433		// Corners are not allowed, they are checked already
434		LineRasterUtil::LINE_SIDE sideP0 = LineRasterUtil::getVertexSide(bounds[ndx].p0, line);
435		LineRasterUtil::LINE_SIDE sideP1 = LineRasterUtil::getVertexSide(bounds[ndx].p1, line);
436
437		if (sideP0 != LineRasterUtil::LINE_SIDE_INTERSECT &&
438			sideP1 != LineRasterUtil::LINE_SIDE_INTERSECT &&
439			sideP0 != sideP1)
440			return true;
441	}
442
443	return false;
444}
445
446} // LineRasterUtil
447
448TriangleRasterizer::TriangleRasterizer (const tcu::IVec4& viewport, const int numSamples, const RasterizationState& state)
449	: m_viewport		(viewport)
450	, m_numSamples		(numSamples)
451	, m_winding			(state.winding)
452	, m_horizontalFill	(state.horizontalFill)
453	, m_verticalFill	(state.verticalFill)
454	, m_face			(FACETYPE_LAST)
455{
456}
457
458/*--------------------------------------------------------------------*//*!
459 * \brief Initialize triangle rasterization
460 * \param v0 Screen-space coordinates (x, y, z) and 1/w for vertex 0.
461 * \param v1 Screen-space coordinates (x, y, z) and 1/w for vertex 1.
462 * \param v2 Screen-space coordinates (x, y, z) and 1/w for vertex 2.
463 *//*--------------------------------------------------------------------*/
464void TriangleRasterizer::init (const tcu::Vec4& v0, const tcu::Vec4& v1, const tcu::Vec4& v2)
465{
466	m_v0 = v0;
467	m_v1 = v1;
468	m_v2 = v2;
469
470	// Positions in fixed-point coordinates.
471	const deInt64	x0		= toSubpixelCoord(v0.x());
472	const deInt64	y0		= toSubpixelCoord(v0.y());
473	const deInt64	x1		= toSubpixelCoord(v1.x());
474	const deInt64	y1		= toSubpixelCoord(v1.y());
475	const deInt64	x2		= toSubpixelCoord(v2.x());
476	const deInt64	y2		= toSubpixelCoord(v2.y());
477
478	// Initialize edge functions.
479	if (m_winding == WINDING_CCW)
480	{
481		initEdgeCCW(m_edge01, m_horizontalFill, m_verticalFill, x0, y0, x1, y1);
482		initEdgeCCW(m_edge12, m_horizontalFill, m_verticalFill, x1, y1, x2, y2);
483		initEdgeCCW(m_edge20, m_horizontalFill, m_verticalFill, x2, y2, x0, y0);
484	}
485	else
486	{
487		// Reverse edges
488		initEdgeCCW(m_edge01, m_horizontalFill, m_verticalFill, x1, y1, x0, y0);
489		initEdgeCCW(m_edge12, m_horizontalFill, m_verticalFill, x2, y2, x1, y1);
490		initEdgeCCW(m_edge20, m_horizontalFill, m_verticalFill, x0, y0, x2, y2);
491	}
492
493	// Determine face.
494	const deInt64	s				= evaluateEdge(m_edge01, x2, y2);
495	const bool		positiveArea	= (m_winding == WINDING_CCW) ? (s > 0) : (s < 0);
496	m_face = positiveArea ? FACETYPE_FRONT : FACETYPE_BACK;
497
498	if (!positiveArea)
499	{
500		// Reverse edges so that we can use CCW area tests & interpolation
501		reverseEdge(m_edge01);
502		reverseEdge(m_edge12);
503		reverseEdge(m_edge20);
504	}
505
506	// Bounding box
507	const deInt64	xMin	= de::min(de::min(x0, x1), x2);
508	const deInt64	xMax	= de::max(de::max(x0, x1), x2);
509	const deInt64	yMin	= de::min(de::min(y0, y1), y2);
510	const deInt64	yMax	= de::max(de::max(y0, y1), y2);
511
512	m_bboxMin.x() = floorSubpixelToPixelCoord	(xMin, m_horizontalFill	== FILL_LEFT);
513	m_bboxMin.y() = floorSubpixelToPixelCoord	(yMin, m_verticalFill	== FILL_BOTTOM);
514	m_bboxMax.x() = ceilSubpixelToPixelCoord	(xMax, m_horizontalFill	== FILL_RIGHT);
515	m_bboxMax.y() = ceilSubpixelToPixelCoord	(yMax, m_verticalFill	== FILL_TOP);
516
517	// Clamp to viewport
518	const int		wX0		= m_viewport.x();
519	const int		wY0		= m_viewport.y();
520	const int		wX1		= wX0 + m_viewport.z() - 1;
521	const int		wY1		= wY0 + m_viewport.w() -1;
522
523	m_bboxMin.x() = de::clamp(m_bboxMin.x(), wX0, wX1);
524	m_bboxMin.y() = de::clamp(m_bboxMin.y(), wY0, wY1);
525	m_bboxMax.x() = de::clamp(m_bboxMax.x(), wX0, wX1);
526	m_bboxMax.y() = de::clamp(m_bboxMax.y(), wY0, wY1);
527
528	m_curPos = m_bboxMin;
529}
530
531void TriangleRasterizer::rasterizeSingleSample (FragmentPacket* const fragmentPackets, float* const depthValues, const int maxFragmentPackets, int& numPacketsRasterized)
532{
533	DE_ASSERT(maxFragmentPackets > 0);
534
535	const deUint64	halfPixel	= 1ll << (RASTERIZER_SUBPIXEL_BITS-1);
536	int				packetNdx	= 0;
537
538	// For depth interpolation; given barycentrics A, B, C = (1 - A - B)
539	// we can reformulate the usual z = z0*A + z1*B + z2*C into more
540	// stable equation z = A*(z0 - z2) + B*(z1 - z2) + z2.
541	const float		za			= m_v0.z()-m_v2.z();
542	const float		zb			= m_v1.z()-m_v2.z();
543	const float		zc			= m_v2.z();
544
545	while (m_curPos.y() <= m_bboxMax.y() && packetNdx < maxFragmentPackets)
546	{
547		const int		x0		= m_curPos.x();
548		const int		y0		= m_curPos.y();
549
550		// Subpixel coords
551		const deInt64	sx0		= toSubpixelCoord(x0)	+ halfPixel;
552		const deInt64	sx1		= toSubpixelCoord(x0+1)	+ halfPixel;
553		const deInt64	sy0		= toSubpixelCoord(y0)	+ halfPixel;
554		const deInt64	sy1		= toSubpixelCoord(y0+1)	+ halfPixel;
555
556		const deInt64	sx[4]	= { sx0, sx1, sx0, sx1 };
557		const deInt64	sy[4]	= { sy0, sy0, sy1, sy1 };
558
559		// Viewport test
560		const bool		outX1	= x0+1 == m_viewport.x()+m_viewport.z();
561		const bool		outY1	= y0+1 == m_viewport.y()+m_viewport.w();
562
563		DE_ASSERT(x0 < m_viewport.x()+m_viewport.z());
564		DE_ASSERT(y0 < m_viewport.y()+m_viewport.w());
565
566		// Edge values
567		tcu::Vector<deInt64, 4>	e01;
568		tcu::Vector<deInt64, 4>	e12;
569		tcu::Vector<deInt64, 4>	e20;
570
571		// Coverage
572		deUint64		coverage	= 0;
573
574		// Evaluate edge values
575		for (int i = 0; i < 4; i++)
576		{
577			e01[i] = evaluateEdge(m_edge01, sx[i], sy[i]);
578			e12[i] = evaluateEdge(m_edge12, sx[i], sy[i]);
579			e20[i] = evaluateEdge(m_edge20, sx[i], sy[i]);
580		}
581
582		// Compute coverage mask
583		coverage = setCoverageValue(coverage, 1, 0, 0, 0,						isInsideCCW(m_edge01, e01[0]) && isInsideCCW(m_edge12, e12[0]) && isInsideCCW(m_edge20, e20[0]));
584		coverage = setCoverageValue(coverage, 1, 1, 0, 0, !outX1 &&				isInsideCCW(m_edge01, e01[1]) && isInsideCCW(m_edge12, e12[1]) && isInsideCCW(m_edge20, e20[1]));
585		coverage = setCoverageValue(coverage, 1, 0, 1, 0, !outY1 &&				isInsideCCW(m_edge01, e01[2]) && isInsideCCW(m_edge12, e12[2]) && isInsideCCW(m_edge20, e20[2]));
586		coverage = setCoverageValue(coverage, 1, 1, 1, 0, !outX1 && !outY1 &&	isInsideCCW(m_edge01, e01[3]) && isInsideCCW(m_edge12, e12[3]) && isInsideCCW(m_edge20, e20[3]));
587
588		// Advance to next location
589		m_curPos.x() += 2;
590		if (m_curPos.x() > m_bboxMax.x())
591		{
592			m_curPos.y() += 2;
593			m_curPos.x()  = m_bboxMin.x();
594		}
595
596		if (coverage == 0)
597			continue; // Discard.
598
599		// Floating-point edge values for barycentrics etc.
600		const tcu::Vec4		e01f	= e01.asFloat();
601		const tcu::Vec4		e12f	= e12.asFloat();
602		const tcu::Vec4		e20f	= e20.asFloat();
603
604		// Compute depth values.
605		if (depthValues)
606		{
607			const tcu::Vec4		edgeSum	= e01f + e12f + e20f;
608			const tcu::Vec4		z0		= e12f / edgeSum;
609			const tcu::Vec4		z1		= e20f / edgeSum;
610
611			depthValues[packetNdx*4+0] = z0[0]*za + z1[0]*zb + zc;
612			depthValues[packetNdx*4+1] = z0[1]*za + z1[1]*zb + zc;
613			depthValues[packetNdx*4+2] = z0[2]*za + z1[2]*zb + zc;
614			depthValues[packetNdx*4+3] = z0[3]*za + z1[3]*zb + zc;
615		}
616
617		// Compute barycentrics and write out fragment packet
618		{
619			FragmentPacket& packet = fragmentPackets[packetNdx];
620
621			const tcu::Vec4		b0		= e12f * m_v0.w();
622			const tcu::Vec4		b1		= e20f * m_v1.w();
623			const tcu::Vec4		b2		= e01f * m_v2.w();
624			const tcu::Vec4		bSum	= b0 + b1 + b2;
625
626			packet.position			= tcu::IVec2(x0, y0);
627			packet.coverage			= coverage;
628			packet.barycentric[0]	= b0 / bSum;
629			packet.barycentric[1]	= b1 / bSum;
630			packet.barycentric[2]	= 1.0f - packet.barycentric[0] - packet.barycentric[1];
631
632			packetNdx += 1;
633		}
634	}
635
636	DE_ASSERT(packetNdx <= maxFragmentPackets);
637	numPacketsRasterized = packetNdx;
638}
639
640// Sample positions - ordered as (x, y) list.
641
642// \note Macros are used to eliminate function calls even in debug builds.
643#define SAMPLE_POS_TO_SUBPIXEL_COORD(POS)	\
644	(deInt64)((POS) * (1<<RASTERIZER_SUBPIXEL_BITS) + 0.5f)
645
646#define SAMPLE_POS(X, Y)	\
647	SAMPLE_POS_TO_SUBPIXEL_COORD(X), SAMPLE_POS_TO_SUBPIXEL_COORD(Y)
648
649static const deInt64 s_samplePos2[] =
650{
651	SAMPLE_POS(0.3f, 0.3f),
652	SAMPLE_POS(0.7f, 0.7f)
653};
654
655static const deInt64 s_samplePos4[] =
656{
657	SAMPLE_POS(0.25f, 0.25f),
658	SAMPLE_POS(0.75f, 0.25f),
659	SAMPLE_POS(0.25f, 0.75f),
660	SAMPLE_POS(0.75f, 0.75f)
661};
662DE_STATIC_ASSERT(DE_LENGTH_OF_ARRAY(s_samplePos4) == 4*2);
663
664static const deInt64 s_samplePos8[] =
665{
666	SAMPLE_POS( 7.f/16.f,  9.f/16.f),
667	SAMPLE_POS( 9.f/16.f, 13.f/16.f),
668	SAMPLE_POS(11.f/16.f,  3.f/16.f),
669	SAMPLE_POS(13.f/16.f, 11.f/16.f),
670	SAMPLE_POS( 1.f/16.f,  7.f/16.f),
671	SAMPLE_POS( 5.f/16.f,  1.f/16.f),
672	SAMPLE_POS(15.f/16.f,  5.f/16.f),
673	SAMPLE_POS( 3.f/16.f, 15.f/16.f)
674};
675DE_STATIC_ASSERT(DE_LENGTH_OF_ARRAY(s_samplePos8) == 8*2);
676
677static const deInt64 s_samplePos16[] =
678{
679	SAMPLE_POS(1.f/8.f, 1.f/8.f),
680	SAMPLE_POS(3.f/8.f, 1.f/8.f),
681	SAMPLE_POS(5.f/8.f, 1.f/8.f),
682	SAMPLE_POS(7.f/8.f, 1.f/8.f),
683	SAMPLE_POS(1.f/8.f, 3.f/8.f),
684	SAMPLE_POS(3.f/8.f, 3.f/8.f),
685	SAMPLE_POS(5.f/8.f, 3.f/8.f),
686	SAMPLE_POS(7.f/8.f, 3.f/8.f),
687	SAMPLE_POS(1.f/8.f, 5.f/8.f),
688	SAMPLE_POS(3.f/8.f, 5.f/8.f),
689	SAMPLE_POS(5.f/8.f, 5.f/8.f),
690	SAMPLE_POS(7.f/8.f, 5.f/8.f),
691	SAMPLE_POS(1.f/8.f, 7.f/8.f),
692	SAMPLE_POS(3.f/8.f, 7.f/8.f),
693	SAMPLE_POS(5.f/8.f, 7.f/8.f),
694	SAMPLE_POS(7.f/8.f, 7.f/8.f)
695};
696DE_STATIC_ASSERT(DE_LENGTH_OF_ARRAY(s_samplePos16) == 16*2);
697
698#undef SAMPLE_POS
699#undef SAMPLE_POS_TO_SUBPIXEL_COORD
700
701template<int NumSamples>
702void TriangleRasterizer::rasterizeMultiSample (FragmentPacket* const fragmentPackets, float* const depthValues, const int maxFragmentPackets, int& numPacketsRasterized)
703{
704	DE_ASSERT(maxFragmentPackets > 0);
705
706	const deInt64*	samplePos	= DE_NULL;
707	const deUint64	halfPixel	= 1ll << (RASTERIZER_SUBPIXEL_BITS-1);
708	int				packetNdx	= 0;
709
710	// For depth interpolation, see rasterizeSingleSample
711	const float		za			= m_v0.z()-m_v2.z();
712	const float		zb			= m_v1.z()-m_v2.z();
713	const float		zc			= m_v2.z();
714
715	switch (NumSamples)
716	{
717		case 2:		samplePos = s_samplePos2;	break;
718		case 4:		samplePos = s_samplePos4;	break;
719		case 8:		samplePos = s_samplePos8;	break;
720		case 16:	samplePos = s_samplePos16;	break;
721		default:
722			DE_ASSERT(false);
723	}
724
725	while (m_curPos.y() <= m_bboxMax.y() && packetNdx < maxFragmentPackets)
726	{
727		const int		x0		= m_curPos.x();
728		const int		y0		= m_curPos.y();
729
730		// Base subpixel coords
731		const deInt64	sx0		= toSubpixelCoord(x0);
732		const deInt64	sx1		= toSubpixelCoord(x0+1);
733		const deInt64	sy0		= toSubpixelCoord(y0);
734		const deInt64	sy1		= toSubpixelCoord(y0+1);
735
736		const deInt64	sx[4]	= { sx0, sx1, sx0, sx1 };
737		const deInt64	sy[4]	= { sy0, sy0, sy1, sy1 };
738
739		// Viewport test
740		const bool		outX1	= x0+1 == m_viewport.x()+m_viewport.z();
741		const bool		outY1	= y0+1 == m_viewport.y()+m_viewport.w();
742
743		DE_ASSERT(x0 < m_viewport.x()+m_viewport.z());
744		DE_ASSERT(y0 < m_viewport.y()+m_viewport.w());
745
746		// Edge values
747		tcu::Vector<deInt64, 4>	e01[NumSamples];
748		tcu::Vector<deInt64, 4>	e12[NumSamples];
749		tcu::Vector<deInt64, 4>	e20[NumSamples];
750
751		// Coverage
752		deUint64		coverage	= 0;
753
754		// Evaluate edge values at sample positions
755		for (int sampleNdx = 0; sampleNdx < NumSamples; sampleNdx++)
756		{
757			const deInt64 ox = samplePos[sampleNdx*2 + 0];
758			const deInt64 oy = samplePos[sampleNdx*2 + 1];
759
760			for (int fragNdx = 0; fragNdx < 4; fragNdx++)
761			{
762				e01[sampleNdx][fragNdx] = evaluateEdge(m_edge01, sx[fragNdx] + ox, sy[fragNdx] + oy);
763				e12[sampleNdx][fragNdx] = evaluateEdge(m_edge12, sx[fragNdx] + ox, sy[fragNdx] + oy);
764				e20[sampleNdx][fragNdx] = evaluateEdge(m_edge20, sx[fragNdx] + ox, sy[fragNdx] + oy);
765			}
766		}
767
768		// Compute coverage mask
769		for (int sampleNdx = 0; sampleNdx < NumSamples; sampleNdx++)
770		{
771			coverage = setCoverageValue(coverage, NumSamples, 0, 0, sampleNdx,						isInsideCCW(m_edge01, e01[sampleNdx][0]) && isInsideCCW(m_edge12, e12[sampleNdx][0]) && isInsideCCW(m_edge20, e20[sampleNdx][0]));
772			coverage = setCoverageValue(coverage, NumSamples, 1, 0, sampleNdx, !outX1 &&			isInsideCCW(m_edge01, e01[sampleNdx][1]) && isInsideCCW(m_edge12, e12[sampleNdx][1]) && isInsideCCW(m_edge20, e20[sampleNdx][1]));
773			coverage = setCoverageValue(coverage, NumSamples, 0, 1, sampleNdx, !outY1 &&			isInsideCCW(m_edge01, e01[sampleNdx][2]) && isInsideCCW(m_edge12, e12[sampleNdx][2]) && isInsideCCW(m_edge20, e20[sampleNdx][2]));
774			coverage = setCoverageValue(coverage, NumSamples, 1, 1, sampleNdx, !outX1 && !outY1 &&	isInsideCCW(m_edge01, e01[sampleNdx][3]) && isInsideCCW(m_edge12, e12[sampleNdx][3]) && isInsideCCW(m_edge20, e20[sampleNdx][3]));
775		}
776
777		// Advance to next location
778		m_curPos.x() += 2;
779		if (m_curPos.x() > m_bboxMax.x())
780		{
781			m_curPos.y() += 2;
782			m_curPos.x()  = m_bboxMin.x();
783		}
784
785		if (coverage == 0)
786			continue; // Discard.
787
788		// Compute depth values.
789		if (depthValues)
790		{
791			for (int sampleNdx = 0; sampleNdx < NumSamples; sampleNdx++)
792			{
793				// Floating-point edge values at sample coordinates.
794				const tcu::Vec4&	e01f	= e01[sampleNdx].asFloat();
795				const tcu::Vec4&	e12f	= e12[sampleNdx].asFloat();
796				const tcu::Vec4&	e20f	= e20[sampleNdx].asFloat();
797
798				const tcu::Vec4		edgeSum	= e01f + e12f + e20f;
799				const tcu::Vec4		z0		= e12f / edgeSum;
800				const tcu::Vec4		z1		= e20f / edgeSum;
801
802				depthValues[(packetNdx*4+0)*NumSamples + sampleNdx] = z0[0]*za + z1[0]*zb + zc;
803				depthValues[(packetNdx*4+1)*NumSamples + sampleNdx] = z0[1]*za + z1[1]*zb + zc;
804				depthValues[(packetNdx*4+2)*NumSamples + sampleNdx] = z0[2]*za + z1[2]*zb + zc;
805				depthValues[(packetNdx*4+3)*NumSamples + sampleNdx] = z0[3]*za + z1[3]*zb + zc;
806			}
807		}
808
809		// Compute barycentrics and write out fragment packet
810		{
811			FragmentPacket& packet = fragmentPackets[packetNdx];
812
813			// Floating-point edge values at pixel center.
814			tcu::Vec4			e01f;
815			tcu::Vec4			e12f;
816			tcu::Vec4			e20f;
817
818			for (int i = 0; i < 4; i++)
819			{
820				e01f[i] = float(evaluateEdge(m_edge01, sx[i] + halfPixel, sy[i] + halfPixel));
821				e12f[i] = float(evaluateEdge(m_edge12, sx[i] + halfPixel, sy[i] + halfPixel));
822				e20f[i] = float(evaluateEdge(m_edge20, sx[i] + halfPixel, sy[i] + halfPixel));
823			}
824
825			// Barycentrics & scale.
826			const tcu::Vec4		b0		= e12f * m_v0.w();
827			const tcu::Vec4		b1		= e20f * m_v1.w();
828			const tcu::Vec4		b2		= e01f * m_v2.w();
829			const tcu::Vec4		bSum	= b0 + b1 + b2;
830
831			packet.position			= tcu::IVec2(x0, y0);
832			packet.coverage			= coverage;
833			packet.barycentric[0]	= b0 / bSum;
834			packet.barycentric[1]	= b1 / bSum;
835			packet.barycentric[2]	= 1.0f - packet.barycentric[0] - packet.barycentric[1];
836
837			packetNdx += 1;
838		}
839	}
840
841	DE_ASSERT(packetNdx <= maxFragmentPackets);
842	numPacketsRasterized = packetNdx;
843}
844
845void TriangleRasterizer::rasterize (FragmentPacket* const fragmentPackets, float* const depthValues, const int maxFragmentPackets, int& numPacketsRasterized)
846{
847	DE_ASSERT(maxFragmentPackets > 0);
848
849	switch (m_numSamples)
850	{
851		case 1:		rasterizeSingleSample		(fragmentPackets, depthValues, maxFragmentPackets, numPacketsRasterized);	break;
852		case 2:		rasterizeMultiSample<2>		(fragmentPackets, depthValues, maxFragmentPackets, numPacketsRasterized);	break;
853		case 4:		rasterizeMultiSample<4>		(fragmentPackets, depthValues, maxFragmentPackets, numPacketsRasterized);	break;
854		case 8:		rasterizeMultiSample<8>		(fragmentPackets, depthValues, maxFragmentPackets, numPacketsRasterized);	break;
855		case 16:	rasterizeMultiSample<16>	(fragmentPackets, depthValues, maxFragmentPackets, numPacketsRasterized);	break;
856		default:
857			DE_ASSERT(DE_FALSE);
858	}
859}
860
861SingleSampleLineRasterizer::SingleSampleLineRasterizer (const tcu::IVec4& viewport)
862	: m_viewport		(viewport)
863	, m_curRowFragment	(0)
864	, m_lineWidth		(0.0f)
865{
866}
867
868SingleSampleLineRasterizer::~SingleSampleLineRasterizer (void)
869{
870}
871
872void SingleSampleLineRasterizer::init (const tcu::Vec4& v0, const tcu::Vec4& v1, float lineWidth)
873{
874	const bool						isXMajor		= de::abs((v1 - v0).x()) >= de::abs((v1 - v0).y());
875
876	// Bounding box \note: with wide lines, the line is actually moved as in the spec
877	const deInt32					lineWidthPixels	= (lineWidth > 1.0f) ? (deInt32)floor(lineWidth + 0.5f) : 1;
878
879	const tcu::Vector<deInt64,2>	widthOffset		= (isXMajor ? tcu::Vector<deInt64,2>(0, -1) : tcu::Vector<deInt64,2>(-1, 0)) * (toSubpixelCoord(lineWidthPixels - 1) / 2);
880
881	const deInt64					x0				= toSubpixelCoord(v0.x()) + widthOffset.x();
882	const deInt64					y0				= toSubpixelCoord(v0.y()) + widthOffset.y();
883	const deInt64					x1				= toSubpixelCoord(v1.x()) + widthOffset.x();
884	const deInt64					y1				= toSubpixelCoord(v1.y()) + widthOffset.y();
885
886	// line endpoints might be perturbed, add some margin
887	const deInt64					xMin			= de::min(x0, x1) - toSubpixelCoord(1);
888	const deInt64					xMax			= de::max(x0, x1) + toSubpixelCoord(1);
889	const deInt64					yMin			= de::min(y0, y1) - toSubpixelCoord(1);
890	const deInt64					yMax			= de::max(y0, y1) + toSubpixelCoord(1);
891
892	// Remove invisible area
893
894	if (isXMajor)
895	{
896		// clamp to viewport in major direction
897		m_bboxMin.x() = de::clamp(floorSubpixelToPixelCoord(xMin, true), m_viewport.x(), m_viewport.x() + m_viewport.z() - 1);
898		m_bboxMax.x() = de::clamp(ceilSubpixelToPixelCoord (xMax, true), m_viewport.x(), m_viewport.x() + m_viewport.z() - 1);
899
900		// clamp to padded viewport in minor direction (wide lines might bleed over viewport in minor direction)
901		m_bboxMin.y() = de::clamp(floorSubpixelToPixelCoord(yMin, true), m_viewport.y() - lineWidthPixels, m_viewport.y() + m_viewport.w() - 1);
902		m_bboxMax.y() = de::clamp(ceilSubpixelToPixelCoord (yMax, true), m_viewport.y() - lineWidthPixels, m_viewport.y() + m_viewport.w() - 1);
903	}
904	else
905	{
906		// clamp to viewport in major direction
907		m_bboxMin.y() = de::clamp(floorSubpixelToPixelCoord(yMin, true), m_viewport.y(), m_viewport.y() + m_viewport.w() - 1);
908		m_bboxMax.y() = de::clamp(ceilSubpixelToPixelCoord (yMax, true), m_viewport.y(), m_viewport.y() + m_viewport.w() - 1);
909
910		// clamp to padded viewport in minor direction (wide lines might bleed over viewport in minor direction)
911		m_bboxMin.x() = de::clamp(floorSubpixelToPixelCoord(xMin, true), m_viewport.x() - lineWidthPixels, m_viewport.x() + m_viewport.z() - 1);
912		m_bboxMax.x() = de::clamp(ceilSubpixelToPixelCoord (xMax, true), m_viewport.x() - lineWidthPixels, m_viewport.x() + m_viewport.z() - 1);
913	}
914
915	m_lineWidth = lineWidth;
916
917	m_v0 = v0;
918	m_v1 = v1;
919
920	m_curPos = m_bboxMin;
921	m_curRowFragment = 0;
922}
923
924void SingleSampleLineRasterizer::rasterize (FragmentPacket* const fragmentPackets, float* const depthValues, const int maxFragmentPackets, int& numPacketsRasterized)
925{
926	DE_ASSERT(maxFragmentPackets > 0);
927
928	const deInt64								halfPixel			= 1ll << (RASTERIZER_SUBPIXEL_BITS-1);
929	const deInt32								lineWidth			= (m_lineWidth > 1.0f) ? deFloorFloatToInt32(m_lineWidth + 0.5f) : 1;
930	const bool									isXMajor			= de::abs((m_v1 - m_v0).x()) >= de::abs((m_v1 - m_v0).y());
931	const tcu::IVec2							minorDirection		= (isXMajor) ? (tcu::IVec2(0, 1)) : (tcu::IVec2(1, 0));
932	const int									minViewportLimit	= (isXMajor) ? (m_viewport.y()) : (m_viewport.x());
933	const int									maxViewportLimit	= (isXMajor) ? (m_viewport.y() + m_viewport.w()) : (m_viewport.x() + m_viewport.z());
934	const tcu::Vector<deInt64,2>				widthOffset			= -minorDirection.cast<deInt64>() * (toSubpixelCoord(lineWidth - 1) / 2);
935	const tcu::Vector<deInt64,2>				pa					= LineRasterUtil::toSubpixelVector(m_v0.xy()) + widthOffset;
936	const tcu::Vector<deInt64,2>				pb					= LineRasterUtil::toSubpixelVector(m_v1.xy()) + widthOffset;
937	const LineRasterUtil::SubpixelLineSegment	line				= LineRasterUtil::SubpixelLineSegment(pa, pb);
938
939	int											packetNdx			= 0;
940
941	while (m_curPos.y() <= m_bboxMax.y() && packetNdx < maxFragmentPackets)
942	{
943		const tcu::Vector<deInt64,2> diamondPosition = LineRasterUtil::toSubpixelVector(m_curPos) + tcu::Vector<deInt64,2>(halfPixel,halfPixel);
944
945		// Should current fragment be drawn? == does the segment exit this diamond?
946		if (LineRasterUtil::doesLineSegmentExitDiamond(line, diamondPosition))
947		{
948			const tcu::Vector<deInt64,2>	pr					= diamondPosition;
949			const float						t					= tcu::dot((pr - pa).asFloat(), (pb - pa).asFloat()) / tcu::lengthSquared(pb.asFloat() - pa.asFloat());
950
951			// Rasterize on only fragments that are would end up in the viewport (i.e. visible)
952			const int						fragmentLocation	= (isXMajor) ? (m_curPos.y()) : (m_curPos.x());
953			const int						rowFragBegin		= de::max(0, minViewportLimit - fragmentLocation);
954			const int						rowFragEnd			= de::min(maxViewportLimit - fragmentLocation, lineWidth);
955
956			// Wide lines require multiple fragments.
957			for (; rowFragBegin + m_curRowFragment < rowFragEnd; m_curRowFragment++)
958			{
959				const int			replicationId	= rowFragBegin + m_curRowFragment;
960				const tcu::IVec2	fragmentPos		= m_curPos + minorDirection * replicationId;
961
962				// We only rasterize visible area
963				DE_ASSERT(LineRasterUtil::inViewport(fragmentPos, m_viewport));
964
965				// Compute depth values.
966				if (depthValues)
967				{
968					const float za = m_v0.z();
969					const float zb = m_v1.z();
970
971					depthValues[packetNdx*4+0] = (1 - t) * za + t * zb;
972					depthValues[packetNdx*4+1] = 0;
973					depthValues[packetNdx*4+2] = 0;
974					depthValues[packetNdx*4+3] = 0;
975				}
976
977				{
978					// output this fragment
979					// \note In order to make consistent output with multisampled line rasterization, output "barycentric" coordinates
980					FragmentPacket& packet = fragmentPackets[packetNdx];
981
982					const tcu::Vec4		b0		= tcu::Vec4(1 - t);
983					const tcu::Vec4		b1		= tcu::Vec4(t);
984					const tcu::Vec4		ooSum	= 1.0f / (b0 + b1);
985
986					packet.position			= fragmentPos;
987					packet.coverage			= getCoverageBit(1, 0, 0, 0);
988					packet.barycentric[0]	= b0 * ooSum;
989					packet.barycentric[1]	= b1 * ooSum;
990					packet.barycentric[2]	= tcu::Vec4(0.0f);
991
992					packetNdx += 1;
993				}
994
995				if (packetNdx == maxFragmentPackets)
996				{
997					m_curRowFragment++; // don't redraw this fragment again next time
998					numPacketsRasterized = packetNdx;
999					return;
1000				}
1001			}
1002
1003			m_curRowFragment = 0;
1004		}
1005
1006		++m_curPos.x();
1007		if (m_curPos.x() > m_bboxMax.x())
1008		{
1009			++m_curPos.y();
1010			m_curPos.x() = m_bboxMin.x();
1011		}
1012	}
1013
1014	DE_ASSERT(packetNdx <= maxFragmentPackets);
1015	numPacketsRasterized = packetNdx;
1016}
1017
1018MultiSampleLineRasterizer::MultiSampleLineRasterizer (const int numSamples, const tcu::IVec4& viewport)
1019	: m_numSamples			(numSamples)
1020	, m_triangleRasterizer0 (viewport, m_numSamples, RasterizationState())
1021	, m_triangleRasterizer1 (viewport, m_numSamples, RasterizationState())
1022{
1023}
1024
1025MultiSampleLineRasterizer::~MultiSampleLineRasterizer ()
1026{
1027}
1028
1029void MultiSampleLineRasterizer::init (const tcu::Vec4& v0, const tcu::Vec4& v1, float lineWidth)
1030{
1031	// allow creation of single sampled rasterizer objects but do not allow using them
1032	DE_ASSERT(m_numSamples > 1);
1033
1034	const tcu::Vec2 lineVec		= tcu::Vec2(tcu::Vec4(v1).xy()) - tcu::Vec2(tcu::Vec4(v0).xy());
1035	const tcu::Vec2 normal2		= tcu::normalize(tcu::Vec2(-lineVec[1], lineVec[0]));
1036	const tcu::Vec4 normal4		= tcu::Vec4(normal2.x(), normal2.y(), 0, 0);
1037	const float offset			= lineWidth / 2.0f;
1038
1039	const tcu::Vec4 p0 = v0 + normal4 * offset;
1040	const tcu::Vec4 p1 = v0 - normal4 * offset;
1041	const tcu::Vec4 p2 = v1 - normal4 * offset;
1042	const tcu::Vec4 p3 = v1 + normal4 * offset;
1043
1044	// Edge 0 -> 1 is always along the line and edge 1 -> 2 is in 90 degree angle to the line
1045	m_triangleRasterizer0.init(p0, p3, p2);
1046	m_triangleRasterizer1.init(p2, p1, p0);
1047}
1048
1049void MultiSampleLineRasterizer::rasterize (FragmentPacket* const fragmentPackets, float* const depthValues, const int maxFragmentPackets, int& numPacketsRasterized)
1050{
1051	DE_ASSERT(maxFragmentPackets > 0);
1052
1053	m_triangleRasterizer0.rasterize(fragmentPackets, depthValues, maxFragmentPackets, numPacketsRasterized);
1054
1055	// Remove 3rd barycentric value and rebalance. Lines do not have non-zero barycentric at index 2
1056	for (int packNdx = 0; packNdx < numPacketsRasterized; ++packNdx)
1057	for (int fragNdx = 0; fragNdx < 4; fragNdx++)
1058	{
1059		float removedValue = fragmentPackets[packNdx].barycentric[2][fragNdx];
1060		fragmentPackets[packNdx].barycentric[2][fragNdx] = 0.0f;
1061		fragmentPackets[packNdx].barycentric[1][fragNdx] += removedValue;
1062	}
1063
1064	// rasterizer 0 filled the whole buffer?
1065	if (numPacketsRasterized == maxFragmentPackets)
1066		return;
1067
1068	{
1069		FragmentPacket* const nextFragmentPackets	= fragmentPackets + numPacketsRasterized;
1070		float* nextDepthValues						= (depthValues) ? (depthValues+4*numPacketsRasterized*m_numSamples) : (DE_NULL);
1071		int numPacketsRasterized2					= 0;
1072
1073		m_triangleRasterizer1.rasterize(nextFragmentPackets, nextDepthValues, maxFragmentPackets - numPacketsRasterized, numPacketsRasterized2);
1074
1075		numPacketsRasterized += numPacketsRasterized2;
1076
1077		// Fix swapped barycentrics in the second triangle
1078		for (int packNdx = 0; packNdx < numPacketsRasterized2; ++packNdx)
1079		for (int fragNdx = 0; fragNdx < 4; fragNdx++)
1080		{
1081			float removedValue = nextFragmentPackets[packNdx].barycentric[2][fragNdx];
1082			nextFragmentPackets[packNdx].barycentric[2][fragNdx] = 0.0f;
1083			nextFragmentPackets[packNdx].barycentric[1][fragNdx] += removedValue;
1084
1085			// edge has reversed direction
1086			std::swap(nextFragmentPackets[packNdx].barycentric[0][fragNdx], nextFragmentPackets[packNdx].barycentric[1][fragNdx]);
1087		}
1088	}
1089}
1090
1091LineExitDiamondGenerator::LineExitDiamondGenerator (void)
1092{
1093}
1094
1095LineExitDiamondGenerator::~LineExitDiamondGenerator (void)
1096{
1097}
1098
1099void LineExitDiamondGenerator::init (const tcu::Vec4& v0, const tcu::Vec4& v1)
1100{
1101	const deInt64					x0				= toSubpixelCoord(v0.x());
1102	const deInt64					y0				= toSubpixelCoord(v0.y());
1103	const deInt64					x1				= toSubpixelCoord(v1.x());
1104	const deInt64					y1				= toSubpixelCoord(v1.y());
1105
1106	// line endpoints might be perturbed, add some margin
1107	const deInt64					xMin			= de::min(x0, x1) - toSubpixelCoord(1);
1108	const deInt64					xMax			= de::max(x0, x1) + toSubpixelCoord(1);
1109	const deInt64					yMin			= de::min(y0, y1) - toSubpixelCoord(1);
1110	const deInt64					yMax			= de::max(y0, y1) + toSubpixelCoord(1);
1111
1112	m_bboxMin.x() = floorSubpixelToPixelCoord(xMin, true);
1113	m_bboxMin.y() = floorSubpixelToPixelCoord(yMin, true);
1114	m_bboxMax.x() = ceilSubpixelToPixelCoord (xMax, true);
1115	m_bboxMax.y() = ceilSubpixelToPixelCoord (yMax, true);
1116
1117	m_v0 = v0;
1118	m_v1 = v1;
1119
1120	m_curPos = m_bboxMin;
1121}
1122
1123void LineExitDiamondGenerator::rasterize (LineExitDiamond* const lineDiamonds, const int maxDiamonds, int& numWritten)
1124{
1125	DE_ASSERT(maxDiamonds > 0);
1126
1127	const deInt64								halfPixel			= 1ll << (RASTERIZER_SUBPIXEL_BITS-1);
1128	const tcu::Vector<deInt64,2>				pa					= LineRasterUtil::toSubpixelVector(m_v0.xy());
1129	const tcu::Vector<deInt64,2>				pb					= LineRasterUtil::toSubpixelVector(m_v1.xy());
1130	const LineRasterUtil::SubpixelLineSegment	line				= LineRasterUtil::SubpixelLineSegment(pa, pb);
1131
1132	int											diamondNdx			= 0;
1133
1134	while (m_curPos.y() <= m_bboxMax.y() && diamondNdx < maxDiamonds)
1135	{
1136		const tcu::Vector<deInt64,2> diamondPosition = LineRasterUtil::toSubpixelVector(m_curPos) + tcu::Vector<deInt64,2>(halfPixel,halfPixel);
1137
1138		if (LineRasterUtil::doesLineSegmentExitDiamond(line, diamondPosition))
1139		{
1140			LineExitDiamond& packet = lineDiamonds[diamondNdx];
1141			packet.position = m_curPos;
1142			++diamondNdx;
1143		}
1144
1145		++m_curPos.x();
1146		if (m_curPos.x() > m_bboxMax.x())
1147		{
1148			++m_curPos.y();
1149			m_curPos.x() = m_bboxMin.x();
1150		}
1151	}
1152
1153	DE_ASSERT(diamondNdx <= maxDiamonds);
1154	numWritten = diamondNdx;
1155}
1156
1157} // rr
1158