aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

+fix

Browse Source
This commit is contained in:
Evghenii
2013-11-14 23:05:44 +01:00
parent 86652738c0
commit 53bf4573f0
1 changed files with 41 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

132
examples_cuda/rt/rt.ispc
View File

@@ -33,30 +33,11 @@
#define bool int #define bool int
#ifdef __NVPTX__
#warning "emitting DEVICE code"
#define programCount warpSize()
#define programIndex laneIndex()
#define taskIndex blockIndex0()
#define taskCount blockCount0()
#else
#warning "emitting HOST code"
#endif
typedef float<3> float3; typedef float<3> float3;
#if 0
#define DIRISNEG
#endif
struct Ray { struct Ray {
float3 origin, dir, invDir; float3 origin, dir, invDir;
#ifdef DIRISNEG /* this fails to compile with nvvm */
uniform unsigned int dirIsNeg[3]; uniform unsigned int dirIsNeg[3];
#else
unsigned int dirIsNeg0, dirIsNeg1, dirIsNeg2;
#endif
float mint, maxt; float mint, maxt;
int hitId; int hitId;
}; };
@@ -90,7 +71,7 @@ static inline float Dot(const float3 a, const float3 b) {
} }
static inline void generateRay(uniform const float raster2camera[4][4], static void generateRay(uniform const float raster2camera[4][4],
uniform const float camera2world[4][4], uniform const float camera2world[4][4],
float x, float y, Ray &ray) { float x, float y, Ray &ray) {
ray.mint = 0.f; ray.mint = 0.f;
@@ -120,19 +101,13 @@ static inline void generateRay(uniform const float raster2camera[4][4],
ray.invDir = 1.f / ray.dir; ray.invDir = 1.f / ray.dir;
#ifdef DIRISNEG
ray.dirIsNeg[0] = any(ray.invDir.x < 0) ? 1 : 0; ray.dirIsNeg[0] = any(ray.invDir.x < 0) ? 1 : 0;
ray.dirIsNeg[1] = any(ray.invDir.y < 0) ? 1 : 0; ray.dirIsNeg[1] = any(ray.invDir.y < 0) ? 1 : 0;
ray.dirIsNeg[2] = any(ray.invDir.z < 0) ? 1 : 0; ray.dirIsNeg[2] = any(ray.invDir.z < 0) ? 1 : 0;
#else
ray.dirIsNeg0 = any(ray.invDir.x < 0) ? 1 : 0;
ray.dirIsNeg1 = any(ray.invDir.y < 0) ? 1 : 0;
ray.dirIsNeg2 = any(ray.invDir.z < 0) ? 1 : 0;
#endif
} }
static inline bool BBoxIntersect(const uniform float bounds[2][3], static bool BBoxIntersect(const uniform float bounds[2][3],
const Ray &ray) { const Ray &ray) {
uniform float3 bounds0 = { bounds[0][0], bounds[0][1], bounds[0][2] }; uniform float3 bounds0 = { bounds[0][0], bounds[0][1], bounds[0][2] };
uniform float3 bounds1 = { bounds[1][0], bounds[1][1], bounds[1][2] }; uniform float3 bounds1 = { bounds[1][0], bounds[1][1], bounds[1][2] };
@@ -171,7 +146,7 @@ static inline bool BBoxIntersect(const uniform float bounds[2][3],
static inline bool TriIntersect(const uniform Triangle &tri, Ray &ray) { static bool TriIntersect(const uniform Triangle &tri, Ray &ray) {
uniform float3 p0 = { tri.p[0][0], tri.p[0][1], tri.p[0][2] }; uniform float3 p0 = { tri.p[0][0], tri.p[0][1], tri.p[0][2] };
uniform float3 p1 = { tri.p[1][0], tri.p[1][1], tri.p[1][2] }; uniform float3 p1 = { tri.p[1][0], tri.p[1][1], tri.p[1][2] };
uniform float3 p2 = { tri.p[2][0], tri.p[2][1], tri.p[2][2] }; uniform float3 p2 = { tri.p[2][0], tri.p[2][1], tri.p[2][2] };
@@ -211,7 +186,7 @@ static inline bool TriIntersect(const uniform Triangle &tri, Ray &ray) {
} }
inline inline bool BVHIntersect(const uniform LinearBVHNode nodes[], bool BVHIntersect(const uniform LinearBVHNode nodes[],
const uniform Triangle tris[], Ray &r) { const uniform Triangle tris[], Ray &r) {
Ray ray = r; Ray ray = r;
bool hit = false; bool hit = false;
@@ -219,62 +194,39 @@ inline inline bool BVHIntersect(const uniform LinearBVHNode nodes[],
uniform int todoOffset = 0, nodeNum = 0; uniform int todoOffset = 0, nodeNum = 0;
uniform int todo[64]; uniform int todo[64];
#define FAST
while (true) { while (true) {
// Check ray against BVH node // Check ray against BVH node
uniform LinearBVHNode node = nodes[nodeNum]; uniform LinearBVHNode node = nodes[nodeNum];
if (any(BBoxIntersect(node.bounds, ray))) if (any(BBoxIntersect(node.bounds, ray))) {
{
uniform unsigned int nPrimitives = node.nPrimitives; uniform unsigned int nPrimitives = node.nPrimitives;
if (nPrimitives > 0) if (nPrimitives > 0) {
{
// Intersect ray with primitives in leaf BVH node // Intersect ray with primitives in leaf BVH node
uniform unsigned int primitivesOffset = node.offset; uniform unsigned int primitivesOffset = node.offset;
for (uniform unsigned int i = 0; i < nPrimitives; ++i) { for (uniform unsigned int i = 0; i < nPrimitives; ++i) {
if (TriIntersect(tris[primitivesOffset+i], ray)) if (TriIntersect(tris[primitivesOffset+i], ray))
hit = true; hit = true;
} }
#ifndef FAST
if (todoOffset == 0) if (todoOffset == 0)
break; break;
#endif
nodeNum = todo[--todoOffset]; nodeNum = todo[--todoOffset];
} }
else else {
{
// Put far BVH node on _todo_ stack, advance to near node // Put far BVH node on _todo_ stack, advance to near node
#ifdef DIRISNEG if (r.dirIsNeg[node.splitAxis]) {
const int dirIsNeg = r.dirIsNeg[node.splitAxis];
#else
int dirIsNeg;
if (node.splitAxis == 0) dirIsNeg = r.dirIsNeg0;
if (node.splitAxis == 1) dirIsNeg = r.dirIsNeg1;
if (node.splitAxis == 2) dirIsNeg = r.dirIsNeg2;
#endif
if (dirIsNeg)
{
todo[todoOffset++] = nodeNum + 1; todo[todoOffset++] = nodeNum + 1;
nodeNum = node.offset; nodeNum = node.offset;
} }
else else {
{
todo[todoOffset++] = node.offset; todo[todoOffset++] = node.offset;
nodeNum = nodeNum + 1; nodeNum = nodeNum + 1;
} }
} }
} }
else { else {
#ifndef FAST
if (todoOffset == 0) if (todoOffset == 0)
break; break;
#endif
nodeNum = todo[--todoOffset]; nodeNum = todo[--todoOffset];
} }
#ifdef FAST
if (todoOffset == 0)
break;
#endif
} }
r.maxt = ray.maxt; r.maxt = ray.maxt;
r.hitId = ray.hitId; r.hitId = ray.hitId;
@@ -283,7 +235,7 @@ inline inline bool BVHIntersect(const uniform LinearBVHNode nodes[],
} }
static inline void raytrace_tile(uniform int x0, uniform int x1, static void raytrace_tile(uniform int x0, uniform int x1,
uniform int y0, uniform int y1, uniform int y0, uniform int y1,
uniform int width, uniform int height, uniform int width, uniform int height,
uniform int baseWidth, uniform int baseHeight, uniform int baseWidth, uniform int baseHeight,
@@ -295,15 +247,15 @@ static inline void raytrace_tile(uniform int x0, uniform int x1,
uniform float widthScale = (float)(baseWidth) / (float)(width); uniform float widthScale = (float)(baseWidth) / (float)(width);
uniform float heightScale = (float)(baseHeight) / (float)(height); uniform float heightScale = (float)(baseHeight) / (float)(height);
foreach_tiled (y = y0 ... y1, x = x0 ... x1) foreach_tiled (y = y0 ... y1, x = x0 ... x1) {
{ Ray ray;
Ray ray; generateRay(raster2camera, camera2world, x*widthScale,
generateRay(raster2camera, camera2world, x*widthScale, y*heightScale, ray); y*heightScale, ray);
BVHIntersect(nodes, triangles, ray); BVHIntersect(nodes, triangles, ray);
int offset = y * width + x; int offset = y * width + x;
image[offset] = ray.maxt; image[offset] = ray.maxt;
id[offset] = ray.hitId; id[offset] = ray.hitId;
} }
} }
@@ -328,34 +280,32 @@ task void raytrace_tile_task(uniform int width, uniform int height,
uniform float image[], uniform int id[], uniform float image[], uniform int id[],
const uniform LinearBVHNode nodes[], const uniform LinearBVHNode nodes[],
const uniform Triangle triangles[]) { const uniform Triangle triangles[]) {
if (taskIndex >= taskCount) return; uniform int dx = 16, dy = 16; // must match dx, dy below
uniform int xBuckets = (width + (dx-1)) / dx;
uniform int dx = 32, dy = 16; // must match dx, dy below uniform int x0 = (taskIndex % xBuckets) * dx;
uniform int xBuckets = (width + (dx-1)) / dx; uniform int x1 = min(x0 + dx, width);
uniform int x0 = (taskIndex % xBuckets) * dx; uniform int y0 = (taskIndex / xBuckets) * dy;
uniform int x1 = min(x0 + dx, width); uniform int y1 = min(y0 + dy, height);
uniform int y0 = (taskIndex / xBuckets) * dy;
uniform int y1 = min(y0 + dy, height); raytrace_tile(x0, x1, y0, y1, width, height, baseWidth, baseHeight,
raster2camera, camera2world, image,
raytrace_tile(x0, x1, y0, y1, width, height, baseWidth, baseHeight, id, nodes, triangles);
raster2camera, camera2world, image,
id, nodes, triangles);
} }
export void raytrace_ispc_tasks(uniform int width, uniform int height, export void raytrace_ispc_tasks(uniform int width, uniform int height,
uniform int baseWidth, uniform int baseHeight, uniform int baseWidth, uniform int baseHeight,
const uniform float raster2camera[4][4], const uniform float raster2camera[4][4],
const uniform float camera2world[4][4], const uniform float camera2world[4][4],
uniform float image[], uniform int id[], uniform float image[], uniform int id[],
const uniform LinearBVHNode nodes[], const uniform LinearBVHNode nodes[],
const uniform Triangle triangles[]) { const uniform Triangle triangles[]) {
uniform int dx = 32, dy = 16; uniform int dx = 16, dy = 16;
uniform int xBuckets = (width + (dx-1)) / dx; uniform int xBuckets = (width + (dx-1)) / dx;
uniform int yBuckets = (height + (dy-1)) / dy; uniform int yBuckets = (height + (dy-1)) / dy;
uniform int nTasks = xBuckets * yBuckets; uniform int nTasks = xBuckets * yBuckets;
launch[nTasks] raytrace_tile_task(width, height, baseWidth, baseHeight, launch[nTasks] raytrace_tile_task(width, height, baseWidth, baseHeight,
raster2camera, camera2world, raster2camera, camera2world,
image, id, nodes, triangles); image, id, nodes, triangles);
} }