minor speedup for the glare compositor node

- pre calculate the UV dot product
- use image width and height converted to floats in the inner loop.
This commit is contained in:
Campbell Barton
2012-06-20 08:11:59 +00:00
parent fc1164c223
commit 586f202eac
2 changed files with 16 additions and 14 deletions

View File

@@ -72,19 +72,20 @@ void ScreenLensDistortionOperation::executePixel(float *outputColor, int x, int
float tc[4] = {0, 0, 0, 0}; float tc[4] = {0, 0, 0, 0};
const float v = sc * ((y + 0.5f) - cy) / cy; const float v = sc * ((y + 0.5f) - cy) / cy;
const float u = sc * ((x + 0.5f) - cx) / cx; const float u = sc * ((x + 0.5f) - cx) / cx;
const float uv_dot = u * u + v * v;
int sta = 0, mid = 0, end = 0; int sta = 0, mid = 0, end = 0;
if ((t = 1.f - kr4 * (u * u + v * v)) >= 0.f) { if ((t = 1.f - kr4 * uv_dot) >= 0.f) {
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
ln[0] = (u * d + 0.5f) * width - 0.5f, ln[1] = (v * d + 0.5f) * height - 0.5f; ln[0] = (u * d + 0.5f) * width - 0.5f, ln[1] = (v * d + 0.5f) * height - 0.5f;
sta = 1; sta = 1;
} }
if ((t = 1.f - kg4 * (u * u + v * v)) >= 0.f) { if ((t = 1.f - kg4 * uv_dot) >= 0.f) {
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
ln[2] = (u * d + 0.5f) * width - 0.5f, ln[3] = (v * d + 0.5f) * height - 0.5f; ln[2] = (u * d + 0.5f) * width - 0.5f, ln[3] = (v * d + 0.5f) * height - 0.5f;
mid = 1; mid = 1;
} }
if ((t = 1.f - kb4 * (u * u + v * v)) >= 0.f) { if ((t = 1.f - kb4 * uv_dot) >= 0.f) {
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
ln[4] = (u * d + 0.5f) * width - 0.5f, ln[5] = (v * d + 0.5f) * height - 0.5f; ln[4] = (u * d + 0.5f) * width - 0.5f, ln[5] = (v * d + 0.5f) * height - 0.5f;
end = 1; end = 1;
@@ -103,10 +104,10 @@ void ScreenLensDistortionOperation::executePixel(float *outputColor, int x, int
for (z = 0; z < ds; ++z) { for (z = 0; z < ds; ++z) {
const float tz = ((float)z + (jit ? BLI_frand() : 0.5f)) * sd; const float tz = ((float)z + (jit ? BLI_frand() : 0.5f)) * sd;
t = 1.f - (kr4 + tz * drg) * (u * u + v * v); t = 1.f - (kr4 + tz * drg) * uv_dot;
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
const float nx = (u * d + 0.5f) * getWidth() - 0.5f; const float nx = (u * d + 0.5f) * width - 0.5f;
const float ny = (v * d + 0.5f) * getHeight() - 0.5f; const float ny = (v * d + 0.5f) * height - 0.5f;
buffer->readCubic(color, nx, ny); buffer->readCubic(color, nx, ny);
tc[0] += (1.f - tz) * color[0], tc[1] += tz * color[1]; tc[0] += (1.f - tz) * color[0], tc[1] += tz * color[1];
dr++, dg++; dr++, dg++;
@@ -121,10 +122,10 @@ void ScreenLensDistortionOperation::executePixel(float *outputColor, int x, int
for (z = 0; z < ds; ++z) { for (z = 0; z < ds; ++z) {
const float tz = ((float)z + (jit ? BLI_frand() : 0.5f)) * sd; const float tz = ((float)z + (jit ? BLI_frand() : 0.5f)) * sd;
t = 1.f - (kg4 + tz * dgb) * (u * u + v * v); t = 1.f - (kg4 + tz * dgb) * uv_dot;
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
const float nx = (u * d + 0.5f) * getWidth() - 0.5f; const float nx = (u * d + 0.5f) * width - 0.5f;
const float ny = (v * d + 0.5f) * getHeight() - 0.5f; const float ny = (v * d + 0.5f) * height - 0.5f;
buffer->readCubic(color, nx, ny); buffer->readCubic(color, nx, ny);
tc[1] += (1.f - tz) * color[1], tc[2] += tz * color[2]; tc[1] += (1.f - tz) * color[1], tc[2] += tz * color[2];
dg++, db++; dg++, db++;

View File

@@ -98,19 +98,20 @@ static void lensDistort(CompBuf *dst, CompBuf *src, float kr, float kg, float kb
float d, t, ln[6] = {0, 0, 0, 0, 0, 0}; float d, t, ln[6] = {0, 0, 0, 0, 0, 0};
fRGB c1, tc = {0, 0, 0, 0}; fRGB c1, tc = {0, 0, 0, 0};
const float u = sc*((x + 0.5f) - cx)/cx; const float u = sc*((x + 0.5f) - cx)/cx;
const float uv_dot = u * u + v * v;
int sta = 0, mid = 0, end = 0; int sta = 0, mid = 0, end = 0;
if ((t = 1.f - kr*(u*u + v*v)) >= 0.f) { if ((t = 1.f - kr*uv_dot) >= 0.f) {
d = 1.f/(1.f + sqrtf(t)); d = 1.f/(1.f + sqrtf(t));
ln[0] = (u*d + 0.5f)*dst->x - 0.5f, ln[1] = (v*d + 0.5f)*dst->y - 0.5f; ln[0] = (u*d + 0.5f)*dst->x - 0.5f, ln[1] = (v*d + 0.5f)*dst->y - 0.5f;
sta = 1; sta = 1;
} }
if ((t = 1.f - kg*(u*u + v*v)) >= 0.f) { if ((t = 1.f - kg*uv_dot) >= 0.f) {
d = 1.f/(1.f + sqrtf(t)); d = 1.f/(1.f + sqrtf(t));
ln[2] = (u*d + 0.5f)*dst->x - 0.5f, ln[3] = (v*d + 0.5f)*dst->y - 0.5f; ln[2] = (u*d + 0.5f)*dst->x - 0.5f, ln[3] = (v*d + 0.5f)*dst->y - 0.5f;
mid = 1; mid = 1;
} }
if ((t = 1.f - kb*(u*u + v*v)) >= 0.f) { if ((t = 1.f - kb*uv_dot) >= 0.f) {
d = 1.f/(1.f + sqrtf(t)); d = 1.f/(1.f + sqrtf(t));
ln[4] = (u*d + 0.5f)*dst->x - 0.5f, ln[5] = (v*d + 0.5f)*dst->y - 0.5f; ln[4] = (u*d + 0.5f)*dst->x - 0.5f, ln[5] = (v*d + 0.5f)*dst->y - 0.5f;
end = 1; end = 1;
@@ -125,7 +126,7 @@ static void lensDistort(CompBuf *dst, CompBuf *src, float kr, float kg, float kb
for (z=0; z<ds; ++z) { for (z=0; z<ds; ++z) {
const float tz = ((float)z + (jit ? BLI_frand() : 0.5f))*sd; const float tz = ((float)z + (jit ? BLI_frand() : 0.5f))*sd;
t = 1.f - (kr + tz*drg)*(u*u + v*v); t = 1.f - (kr + tz*drg)*uv_dot;
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
qd_getPixelLerp(src, (u*d + 0.5f)*dst->x - 0.5f, (v*d + 0.5f)*dst->y - 0.5f, c1); qd_getPixelLerp(src, (u*d + 0.5f)*dst->x - 0.5f, (v*d + 0.5f)*dst->y - 0.5f, c1);
if (src->type == CB_VAL) c1[1] = c1[2] = c1[0]; if (src->type == CB_VAL) c1[1] = c1[2] = c1[0];
@@ -141,7 +142,7 @@ static void lensDistort(CompBuf *dst, CompBuf *src, float kr, float kg, float kb
for (z=0; z<ds; ++z) { for (z=0; z<ds; ++z) {
const float tz = ((float)z + (jit ? BLI_frand() : 0.5f))*sd; const float tz = ((float)z + (jit ? BLI_frand() : 0.5f))*sd;
t = 1.f - (kg + tz*dgb)*(u*u + v*v); t = 1.f - (kg + tz*dgb)*uv_dot;
d = 1.f / (1.f + sqrtf(t)); d = 1.f / (1.f + sqrtf(t));
qd_getPixelLerp(src, (u*d + 0.5f)*dst->x - 0.5f, (v*d + 0.5f)*dst->y - 0.5f, c1); qd_getPixelLerp(src, (u*d + 0.5f)*dst->x - 0.5f, (v*d + 0.5f)*dst->y - 0.5f, c1);
if (src->type == CB_VAL) c1[1] = c1[2] = c1[0]; if (src->type == CB_VAL) c1[1] = c1[2] = c1[0];