Undo splitting of buffers in three. Makes buffer access much faster

Not kidding here. For the default parameters this increased the maximum pixel brightness in the image by a factor of 2.25.
This has the drawback that the maximum possible image size got reduced, but hey, there's a command line switch to ignore all size limits ;-)

4 files changed, 21 insertions, 51 deletions

diff --git a/BuddhaTest/Shaders/BuddhaCompute.glsl b/BuddhaTest/Shaders/BuddhaCompute.glsl
index 7cd4efe..caf10ee 100644
--- a/BuddhaTest/Shaders/BuddhaCompute.glsl
+++ b/BuddhaTest/Shaders/BuddhaCompute.glsl
@@ -4,15 +4,7 @@
 
 layout(std430, binding=2) restrict buffer renderedDataRed
 {
-    restrict uint counts_SSBORed[];
-};
-layout(std430, binding=3) restrict buffer renderedDataGreen
-{
-    restrict uint counts_SSBOGreen[];
-};
-layout(std430, binding=4) restrict buffer renderedDataBlue
-{
-    restrict uint counts_SSBOBlue[];
+    restrict uint counts_SSBO[];
 };
 
 struct individualData
@@ -38,10 +30,10 @@ uniform uint totalIterations;
 
 void addToColorOfCell(uvec2 cell, uvec3 toAdd)
 {
-    uint firstIndex = (cell.x + cell.y * width);
-    atomicAdd(counts_SSBORed[firstIndex],toAdd.x);
-    atomicAdd(counts_SSBOGreen[firstIndex],toAdd.y);
-    atomicAdd(counts_SSBOBlue[firstIndex],toAdd.z);
+    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);
 }
 
 uvec2 getCell(vec2 complex)
diff --git a/BuddhaTest/Shaders/BuddhaFragment.glsl b/BuddhaTest/Shaders/BuddhaFragment.glsl
index b33ee1d..449aa29 100644
--- a/BuddhaTest/Shaders/BuddhaFragment.glsl
+++ b/BuddhaTest/Shaders/BuddhaFragment.glsl
@@ -6,15 +6,7 @@ out vec3 color;
 
 layout(std430, binding=2) restrict readonly buffer renderedDataRed
 {
-        restrict readonly uint counts_SSBORed[];
-};
-layout(std430, binding=3) restrict readonly buffer renderedDataGreen
-{
-        restrict readonly uint counts_SSBOGreen[];
-};
-layout(std430, binding=4) restrict readonly buffer renderedDataBlue
-{
-        restrict readonly uint counts_SSBOBlue[];
+        restrict readonly uint counts_SSBO[];
 };
 
 uniform uint width;
@@ -24,8 +16,8 @@ uvec3 getColorAt(vec2 fragCoord)
 {
     uint xIndex = uint(max(0.0,(fragCoord.x+1.0)*0.5*width));
     uint yIndex = uint(max(0.0,abs(fragCoord.y)*height));
-    uint firstIndex = (xIndex + yIndex * width);
-    return uvec3(counts_SSBORed[firstIndex],counts_SSBOGreen[firstIndex],counts_SSBOBlue[firstIndex]);
+    uint firstIndex = 3*(xIndex + yIndex * width);
+    return uvec3(counts_SSBO[firstIndex],counts_SSBO[firstIndex+1],counts_SSBO[firstIndex+2]);
 }
 
 void main(){
diff --git a/BuddhaTest/src/BuddhaTest.cpp b/BuddhaTest/src/BuddhaTest.cpp
index 0a99fee..c1f054d 100644
--- a/BuddhaTest/src/BuddhaTest.cpp
+++ b/BuddhaTest/src/BuddhaTest.cpp
@@ -112,17 +112,12 @@ int main(int argc, char * argv[])
     glBufferData(GL_ARRAY_BUFFER, sizeof(g_vertex_buffer_data), g_vertex_buffer_data, GL_STATIC_DRAW);
 
 
-    GLuint drawBuffer[3];
-    glGenBuffers(3, drawBuffer);
-    for(int i=0; i < 3; ++i)
-    {
-        glBindBuffer(GL_SHADER_STORAGE_BUFFER, drawBuffer[i]);
-        {
-            glBufferData(GL_SHADER_STORAGE_BUFFER, 4 * pixelCount, nullptr, GL_DYNAMIC_COPY);
-            glClearBufferData(GL_SHADER_STORAGE_BUFFER,GL_R8,GL_RED,GL_UNSIGNED_INT,nullptr);
-        }
-        glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2+i, drawBuffer[i]);
-    }
+    GLuint drawBuffer;
+    glGenBuffers(1, &drawBuffer);
+    glBindBuffer(GL_SHADER_STORAGE_BUFFER, drawBuffer);
+    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);
 
     //the state buffer is special. While making each entry 8 bytes large and using std430 layout, the data is never read back,
@@ -235,25 +230,16 @@ int main(int argc, char * argv[])
     if(!settings.pngFilename.empty())
     {
         glMemoryBarrier(GL_ALL_BARRIER_BITS);
-        std::vector<std::vector<uint32_t>> readBackBuffers(3,std::vector<uint32_t>(pixelCount));
-        for(int i = 0; i < 3; ++i)
-        {
-            glBindBuffer(GL_SHADER_STORAGE_BUFFER, drawBuffer[i]);
-            glGetBufferSubData(GL_SHADER_STORAGE_BUFFER,0,4 * pixelCount,readBackBuffers[i].data());
-        }
+        std::vector<uint32_t> readBackBuffer(pixelCount*3);
+        glBindBuffer(GL_SHADER_STORAGE_BUFFER, drawBuffer);
+        glGetBufferSubData(GL_SHADER_STORAGE_BUFFER,0,4 *3* pixelCount,readBackBuffer.data());
 
-        //too lazy to change WriteOutputPng...
-        std::vector<uint32_t> combinedBuffer(3*pixelCount);
-        for(uint32_t i=0;i<3*pixelCount;++i)
-        {
-            combinedBuffer[i] = readBackBuffers[i%3][i/3];
-        }
-        Helpers::WriteOutputPNG(settings.pngFilename,combinedBuffer,settings.imageWidth,bufferHeight, settings.pngGamma, settings.pngColorScale);
+        Helpers::WriteOutputPNG(settings.pngFilename,readBackBuffer,settings.imageWidth,bufferHeight, settings.pngGamma, settings.pngColorScale);
     }
 
     //a bit of cleanup
     glDeleteBuffers(1,&vertexbuffer);
-    glDeleteBuffers(3,drawBuffer);
+    glDeleteBuffers(1,&drawBuffer);
     glDeleteBuffers(1,&stateBuffer);
 
     glfwTerminate();
diff --git a/BuddhaTest/src/Helpers.cpp b/BuddhaTest/src/Helpers.cpp
index e499f0c..2a687d2 100644
--- a/BuddhaTest/src/Helpers.cpp
+++ b/BuddhaTest/src/Helpers.cpp
@@ -264,9 +264,9 @@ namespace Helpers
         }
         int maxSSBOSize;
         glGetIntegerv(GL_MAX_SHADER_STORAGE_BLOCK_SIZE,&maxSSBOSize);
-        if((static_cast<unsigned long>(imageWidth) * static_cast<unsigned long>(imageHeight)) > static_cast<unsigned long>(maxSSBOSize)/2) //divided by 2, as we have 4 bytes per int, but only half the image height
+        if((static_cast<unsigned long>(imageWidth) * static_cast<unsigned long>(imageHeight)) > static_cast<unsigned long>(maxSSBOSize)/6) //divided by 3*2, as we have three colors per pixel, 4 bytes per int, but only half the image height
         {
-            std::cerr << "Requested buffer size is larger than maximum allowed by graphics driver. Max pixel number: " << maxSSBOSize/2 << std::endl;
+            std::cerr << "Requested buffer size is larger than maximum allowed by graphics driver. Max pixel number: " << maxSSBOSize/6 << std::endl;
             std::cerr << "You can override this limit check using the --ignoreMaxBufferSize 1 command line parameter, but doing so is your own risk." << std::endl;
             if(ignoreMaxBufferSize == 0)
                 return false;