local group size adjustment

Zhylkaaa · May 19, 2020 · 39616fc · 39616fc
1 parent 2239650
commit 39616fc
Show file tree

Hide file tree

Showing 4 changed files with 4 additions and 4 deletions.
diff --git a/shaders/d_dense_b.comp b/shaders/d_dense_b.comp
@@ -1,6 +1,6 @@
 #version 450
 
-layout (local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
+layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
 
 layout(std430, binding=5) buffer outdb { float d_bias[]; };
 layout(std430, binding=6) buffer readonly indO { float layer_doutput[]; };

diff --git a/shaders/softmax.comp b/shaders/softmax.comp
@@ -1,6 +1,6 @@
 #version 450
 
-layout (local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
+layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
 
 layout(std430, binding=0) buffer inI { float layer_input[]; };
 layout(std430, binding=1) buffer outO { float layer_output[]; };

diff --git a/src/dense.cpp b/src/dense.cpp
@@ -124,7 +124,7 @@ void DenseLayer::backward_initialize(VkBuffer &d_out) {
                           backwardWeightPipeline,backwardDescriptorSet, (dim.inp_dim+15)/16, (dim.output_dim+15)/16, 1);
 
     recordComputePipeline(backwardBiasCommandBuffer, backwardPipelineLayout, sizeof(dims), reinterpret_cast<void*>(&dim),
-                          backwardBiasPipeline,backwardDescriptorSet, (dim.output_dim+31)/32, 1, 1);
+                          backwardBiasPipeline,backwardDescriptorSet, (dim.output_dim+63)/64, 1, 1);
     backward_initialized = true;
 }
 

diff --git a/src/softmax.cpp b/src/softmax.cpp
@@ -37,7 +37,7 @@ void SoftmaxLayer::forward_initialize(VkQueue &queue) {
     createCommandPoolAndBuffer(device, queueFamilyIndex, forwardCommandPool, forwardCommandBuffer);
 
     recordComputePipeline(forwardCommandBuffer, forwardPipelineLayout, sizeof(dims), reinterpret_cast<void*>(&dim),
-                          forwardPipeline,forwardDescriptorSet, (dim.batch_size+31)/32, 1, 1);
+                          forwardPipeline,forwardDescriptorSet, (dim.batch_size+63)/64, 1, 1);
 
 }