Update matrix_opencl.cpp

matrix_opencl.cpp

@@ -76,6 +76,7 @@ const std::string kernel_source_transpose = R"(
         B[output_idx] = A[input_idx];
     }
 )";
+// NAIVE
 /*const std::string kernel_source_matrix_mul = R"(
     __kernel void matrix_mul(__global const float* A, __global const float* B, __global float* C, int A_rows, int A_cols, int B_cols) {
         int row = get_global_id(0);
@@ -85,6 +86,8 @@ const std::string kernel_source_transpose = R"(
         }
     }
 )";*/
+
+// FASTER
 const std::string kernel_source_matrix_mul = R"(
     __kernel void matrix_mul(__global const float* A,
                          __global const float* B,
@@ -98,7 +101,7 @@ const std::string kernel_source_matrix_mul = R"(
     int nloc = get_local_size(0);
 
     float tmp;
-    float Awrk[10000];
+    float Awrk[4096];
 
     for (k = 0; k < K; k++)
         Awrk[k] = A[i * K + k];
@@ -344,6 +347,7 @@ MatrixCL MatrixCL::operator+(const MatrixCL& other) const {
     return result;
 }
 
+// NAIVE VERSION
 /*MatrixCL MatrixCL::operator*(const MatrixCL& other) const {
     if (cols_ != other.rows_)
         throw std::runtime_error("Matrix dimension error.");
@@ -366,6 +370,7 @@ MatrixCL MatrixCL::operator+(const MatrixCL& other) const {
     return result;
 }*/
 
+// FASTER VERSION
 MatrixCL MatrixCL::operator*(const MatrixCL& other) const {
     if (cols_ != other.rows_)
         throw std::runtime_error("Matrix dimension error.");