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

added reduce_min/max_float, packed_store_active for CUDA, and now kerenls1.ispc just work :)

Browse Source
This commit is contained in:
Evghenii
2013-11-11 12:33:39 +01:00
parent 9c7a842163
commit a91c8e15e2
6 changed files with 193 additions and 106 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
examples_cuda/deferred/kernels.ispc
View File

@@ -116,7 +116,11 @@ ComputeZBounds(
float laneMinZ = cameraFar;
float laneMaxZ = cameraNear;
for (uniform int32 y = tileStartY; y < tileEndY; ++y) {
foreach (x = tileStartX ... tileEndX) {
// foreach (x = tileStartX ... tileEndX)
for (uniform int xb = tileStartX; xb < tileEndX; xb += programCount)
{
const int x = xb + programIndex;
if (x >= tileEndX) break;
// Unproject depth buffer Z value into view space
float z = zBuffer[y * gBufferWidth + x];
float viewSpaceZ = cameraProj_43 / (z - cameraProj_33);
@@ -178,7 +182,10 @@ IntersectLightsWithTileMinMax(
uniform int32 tileNumLights = 0;
foreach (lightIndex = 0 ... numLights) {
// foreach (lightIndex = 0 ... numLights)
for (uniform int lightIndexB = 0; lightIndexB < numLights; lightIndexB += programCount)
{
const int lightIndex = lightIndexB + programIndex;
float light_positionView_z = light_positionView_z_array[lightIndex];
float light_attenuationEnd = light_attenuationEnd_array[lightIndex];
float light_attenuationEndNeg = -light_attenuationEnd;
@@ -193,6 +200,7 @@ IntersectLightsWithTileMinMax(
// don't actually need to mask the rest of this function - this is
// just a greedy early-out. Could also structure all of this as
// nested if() statements, but this a bit easier to read
bool active = false;
if (any(inFrustum)) {
float light_positionView_x = light_positionView_x_array[lightIndex];
float light_positionView_y = light_positionView_y_array[lightIndex];
@@ -214,12 +222,14 @@ IntersectLightsWithTileMinMax(
inFrustum = inFrustum && (d >= light_attenuationEndNeg);
// Pack and store intersecting lights
cif (inFrustum) {
tileNumLights += packed_store_active(&tileLightIndices[tileNumLights],
lightIndex);
if (inFrustum)
active = true;
}
if (lightIndex >= numLights)
active = false;
tileNumLights += packed_store_active(active, &tileLightIndices[tileNumLights], lightIndex);
}
}
return tileNumLights;
}
@@ -285,7 +295,11 @@ ShadeTile(
if (tileNumLights == 0 || visualizeLightCount) {
uniform unsigned int8 c = (unsigned int8)(min(tileNumLights << 2, 255));
for (uniform int32 y = tileStartY; y < tileEndY; ++y) {
foreach (x = tileStartX ... tileEndX) {
// foreach (x = tileStartX ... tileEndX)
for (uniform int xb = tileStartX ; xb < tileEndX; xb += programCount)
{
const int x = xb + programIndex;
if (x >= tileEndX) continue;
int32 framebufferIndex = (y * gBufferWidth + x);
framebuffer_r[framebufferIndex] = c;
framebuffer_g[framebufferIndex] = c;
@@ -299,7 +313,10 @@ ShadeTile(
for (uniform int32 y = tileStartY; y < tileEndY; ++y) {
uniform float positionScreen_y = -(((0.5f + y) * twoOverGBufferHeight) - 1.f);
foreach (x = tileStartX ... tileEndX) {
// foreach (x = tileStartX ... tileEndX) {
for (uniform int xb = tileStartX ; xb < tileEndX; xb += programCount)
{
const int x = xb + programIndex;
int32 gBufferOffset = y * gBufferWidth + x;
// Reconstruct position and (negative) view vector from G-buffer