Initial commit for better brightness.

Needs more work, uses too many VGPRs. Also, maybe work on a single
shared variable?

3 files changed, 52 insertions, 5 deletions

diff --git a/BuddhaTest/Shaders/BuddhaCompute.glsl b/BuddhaTest/Shaders/BuddhaCompute.glsl
index caf10ee..f0b7d43 100644
--- a/BuddhaTest/Shaders/BuddhaCompute.glsl
+++ b/BuddhaTest/Shaders/BuddhaCompute.glsl
@@ -7,6 +7,11 @@ layout(std430, binding=2) restrict buffer renderedDataRed
     restrict uint counts_SSBO[];
 };
 
+layout(std430, binding=3) restrict buffer brightnessData
+{
+    restrict uvec3 brightness;
+};
+
 struct individualData
 {
     uint phase;
@@ -28,12 +33,20 @@ uniform uvec4 orbitLength;
 uniform uint iterationsPerDispatch;
 uniform uint totalIterations;
 
+shared uvec3[gl_WorkGroupSize.x*gl_WorkGroupSize.y*gl_WorkGroupSize.z] brightnesses;
+
 void addToColorOfCell(uvec2 cell, uvec3 toAdd)
 {
     uint firstIndex = 3*(cell.x + cell.y * width);
-    atomicAdd(counts_SSBO[firstIndex],toAdd.x);
-    atomicAdd(counts_SSBO[firstIndex+1],toAdd.y);
-    atomicAdd(counts_SSBO[firstIndex+2],toAdd.z);
+    uvec3 b;
+    b.x = atomicAdd(counts_SSBO[firstIndex],toAdd.x);
+    b.y = atomicAdd(counts_SSBO[firstIndex+1],toAdd.y);
+    b.z = atomicAdd(counts_SSBO[firstIndex+2],toAdd.z);
+    for(int i = 0; i < 3;++i)
+    {
+        if(brightnesses[gl_LocalInvocationIndex][i] < b[i])
+            brightnesses[gl_LocalInvocationIndex][i] = b[i];
+    }
 }
 
 uvec2 getCell(vec2 complex)
@@ -194,6 +207,11 @@ vec2 getCurrentOrbitOffset(const uint orbitNumber, const uint totalWorkers, cons
 }
 
 void main() {
+    for(int i = 0; i < 3;++i)
+    {
+        brightnesses[gl_LocalInvocationIndex][i] = 0;
+    }
+
     //we need to know how many total work groups are running this iteration
     const uvec3 totalWorkersPerDimension = gl_WorkGroupSize * gl_NumWorkGroups;
     const uint totalWorkers = totalWorkersPerDimension.x*totalWorkersPerDimension.y*totalWorkersPerDimension.z;
@@ -258,4 +276,24 @@ void main() {
         }
     }
     stateArray[uniqueWorkerID] = state;
+
+    barrier();
+    groupMemoryBarrier();
+    if(gl_LocalInvocationIndex == 0)
+    {
+        uvec3 maxbrightness = uvec3(0,0,0);
+        for(int i = 0; i < brightnesses.length(); ++i)
+        {
+            for(int j=0; j< 3; ++j)
+            {
+                if(brightnesses[i][j] > maxbrightness[j])
+                    maxbrightness[j] = brightnesses[i][j];
+            }
+
+        }
+        for(int i = 0; i < 3; ++i)
+        {
+            atomicMax(brightness[i],maxbrightness[i]);
+        }
+    }
 }
diff --git a/BuddhaTest/Shaders/BuddhaFragment.glsl b/BuddhaTest/Shaders/BuddhaFragment.glsl
index 449aa29..16d25f2 100644
--- a/BuddhaTest/Shaders/BuddhaFragment.glsl
+++ b/BuddhaTest/Shaders/BuddhaFragment.glsl
@@ -8,6 +8,10 @@ layout(std430, binding=2) restrict readonly buffer renderedDataRed
 {
         restrict readonly uint counts_SSBO[];
 };
+layout(std430, binding=3) restrict readonly buffer brightnessData
+{
+    restrict readonly uvec3 brightness;
+};
 
 uniform uint width;
 uniform uint height;
@@ -22,7 +26,6 @@ uvec3 getColorAt(vec2 fragCoord)
 
 void main(){
     uvec3 totalCount = getColorAt(uv);
-    uvec3 brightness = getColorAt(vec2(-0.2390625,0));
 
     vec3 scaled = vec3(totalCount)/max(length(vec3(brightness)),1.0);
     color = scaled;
diff --git a/BuddhaTest/src/BuddhaTest.cpp b/BuddhaTest/src/BuddhaTest.cpp
index 0ae94b0..4de252d 100644
--- a/BuddhaTest/src/BuddhaTest.cpp
+++ b/BuddhaTest/src/BuddhaTest.cpp
@@ -118,7 +118,13 @@ int main(int argc, char * argv[])
     glBufferData(GL_SHADER_STORAGE_BUFFER, 4 *3* pixelCount, nullptr, GL_DYNAMIC_COPY);
     glClearBufferData(GL_SHADER_STORAGE_BUFFER,GL_R8,GL_RED,GL_UNSIGNED_INT,nullptr);
     glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, drawBuffer);
-    glBindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
+
+    GLuint brightnessBuffer;
+    glGenBuffers(1,&brightnessBuffer);
+    glBindBuffer(GL_SHADER_STORAGE_BUFFER, brightnessBuffer);
+    glBufferData(GL_SHADER_STORAGE_BUFFER, 16,nullptr, GL_DYNAMIC_COPY);
+    glClearBufferData(GL_SHADER_STORAGE_BUFFER,GL_R8,GL_RED,GL_UNSIGNED_INT,nullptr);
+    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, brightnessBuffer);
 
     //the state buffer is special. While making each entry 8 bytes large and using std430 layout, the data is never read back,
     //so we can get away with being more lenient and allowing the compiler to choose layout without much extra work.