mirror of
https://github.com/ZDoom/raze-gles.git
synced 2024-12-25 03:00:46 +00:00
engine.cpp: lintersect(): fix integer overflow and negative case issue
git-svn-id: https://svn.eduke32.com/eduke32@7472 1a8010ca-5511-0410-912e-c29ae57300e0
This commit is contained in:
parent
a82eb1d6fe
commit
f7f62c4466
1 changed files with 24 additions and 10 deletions
|
@ -7459,7 +7459,7 @@ int32_t lintersect(const int32_t originX, const int32_t originY, const int32_t o
|
|||
const int32_t rayLengthSquared = ray.x*ray.x + ray.y*ray.y;
|
||||
const int32_t rayDotOriginDiff = ray.x*originDiff.x + ray.y*originDiff.y;
|
||||
const int32_t rayDotLineEndDiff = rayDotOriginDiff + ray.x*lineVec.x + ray.y*lineVec.y;
|
||||
int32_t t = min(rayDotOriginDiff, rayDotLineEndDiff);
|
||||
int64_t t = min(rayDotOriginDiff, rayDotLineEndDiff);
|
||||
if (rayDotOriginDiff < 0)
|
||||
{
|
||||
if (rayDotLineEndDiff < 0)
|
||||
|
@ -7474,7 +7474,8 @@ int32_t lintersect(const int32_t originX, const int32_t originY, const int32_t o
|
|||
|
||||
t = rayDotLineEndDiff;
|
||||
}
|
||||
t = divscale24(t, rayLengthSquared);
|
||||
t = tabledivide64(t << 24L, rayLengthSquared);
|
||||
|
||||
*intersectionX = originX + mulscale24(ray.x, t);
|
||||
*intersectionY = originY + mulscale24(ray.y, t);
|
||||
*intersectionZ = originZ + mulscale24(destZ-originZ, t);
|
||||
|
@ -7484,20 +7485,33 @@ int32_t lintersect(const int32_t originX, const int32_t originY, const int32_t o
|
|||
|
||||
const int32_t originDiffCrossLineVec = originDiff.x*lineVec.y - originDiff.y*lineVec.x;
|
||||
static const int32_t signBit = 1u<<31u;
|
||||
// Any point on either line can be expressed as p+t*r and q+u*s
|
||||
// The two line segments intersect when we can find a t & u such that p+t*r = q+u*s
|
||||
// If the point is outside of the bounds of the line segment, we know we don't have an intersection.
|
||||
// t is < 0 if (originDiffCrossLineVec^rayCrossLineVec) & signBit)
|
||||
// t is > 1 if originDiffCrossLineVec > rayCrossLineVec
|
||||
// u is < 0 if (originDiffCrossRay^rayCrossLineVec) & signBit
|
||||
// u is > 1 if originDiffCrossRay > rayCrossLineVec
|
||||
// where int32_t u = divscale24(originDiffCrossRay, rayCrossLineVec);
|
||||
if (originDiffCrossLineVec > rayCrossLineVec ||
|
||||
originDiffCrossRay > rayCrossLineVec ||
|
||||
((originDiffCrossLineVec^rayCrossLineVec) & signBit) ||
|
||||
((originDiffCrossRay^rayCrossLineVec) & signBit))
|
||||
// t is > 1 if abs(originDiffCrossLineVec) > abs(rayCrossLineVec)
|
||||
// u is > 1 if abs(originDiffCrossRay) > abs(rayCrossLineVec)
|
||||
// where int32_t u = tabledivide64(((int64_t) originDiffCrossRay) << 24L, rayCrossLineVec);
|
||||
if (((originDiffCrossLineVec^rayCrossLineVec) & signBit) ||
|
||||
((originDiffCrossRay^rayCrossLineVec) & signBit) ||
|
||||
abs(originDiffCrossLineVec) > abs(rayCrossLineVec) ||
|
||||
abs(originDiffCrossRay) > abs(rayCrossLineVec))
|
||||
{
|
||||
// line segments do not overlap
|
||||
return 0;
|
||||
}
|
||||
int32_t t = divscale24(originDiffCrossLineVec, rayCrossLineVec);
|
||||
|
||||
int64_t t = tabledivide64(((int64_t) originDiffCrossLineVec) << 24L, rayCrossLineVec);
|
||||
// For sake of completeness/readability, alternative to the above approach for an early out & avoidance of an extra division:
|
||||
#if 0
|
||||
int64_t u = tabledivide64(((int64_t) originDiffCrossRay) << 24L, rayCrossLineVec);
|
||||
if (u < 0 || u > 1 << 24 || t < 0 || t > 1 << 24)
|
||||
{
|
||||
return 0;
|
||||
}
|
||||
#endif
|
||||
|
||||
*intersectionX = originX + mulscale24(ray.x, t);
|
||||
*intersectionY = originY + mulscale24(ray.y, t);
|
||||
*intersectionZ = originZ + mulscale24(destZ-originZ, t);
|
||||
|
|
Loading…
Reference in a new issue