clean up & script for shader compilation & readme update

Readme.md

@@ -14,13 +14,28 @@ $ glslc -o dense.comp.spv dense.comp
 
 To compile and run use:
 ```bash
+$ chmod u+x compile_shaders.sh
+$ ./compile_shaders.sh
 $ mkdir build
 $ cd build
 $ cmake ..
 $ make
 $ ./vulkan_perceptron
 ```
 
+## Quick reference
+
+to define perceptron you need vector with output sizes of each layer (input size will be inferred from previous layer)
+and vector of strings containing names of activations for each layer.
+also you need to specify input size and batch size (**Note:** that for sake of simplicity you can use forward only with matrix of size `batch_size x input_size`)
+
+Activation list:
+- relu - Rectified linear unit 
+- id - identity (not a layer actually, it's just omitted)
+- softmax - softmax function, not actually an activation function, but it's hear for sake of simplicity
+
+You can refer to `line 10,11,12 of main.cpp` for usage example.
+
 ## TODOs
 - Switch to using Tensor instead of VkBuffer
 - Implement trainers for different tasks

compile_shaders.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+SHADER_DIR=shaders
+
+shader_list=$(ls $SHADER_DIR | grep -E '.*\.comp$')
+
+for shader_name in $shader_list; do
+  glslc -o ${SHADER_DIR}/${shader_name}.spv ${SHADER_DIR}/${shader_name}
+done

main.cpp

@@ -7,8 +7,8 @@
 
 int main() {
 
-    std::vector<int> layers{10};
-    std::vector<std::string> activations{"softmax"};
+    std::vector<int> layers{512, 10};
+    std::vector<std::string> activations{"relu", "softmax"};
     MLP mlp = MLP(784, 32, layers, activations);
 
     mlp.forward_initialize();