Add clearing kernel

include/kernels/matmul.cuh

@@ -62,6 +62,11 @@ __global__ void vec_scalar_sub(
     const unsigned int w
 );
 
+__global__ void clear(
+    float* __restrict__ d_vector,
+    const unsigned int w
+);
+
 }  // namespace CUDANet::Kernels
 
 #endif  // CUDANET_MATMUL_H

src/kernels/matmul.cu

@@ -83,4 +83,15 @@ __global__ void Kernels::vec_scalar_sub(
         return;
     }
     d_output[tid] = d_vector[tid] - d_scalar[0];
+}
+
+__global__ void Kernels::clear(
+    float* __restrict__ d_vector,
+    const unsigned int w
+) {
+    int tid = blockDim.x * blockIdx.x + threadIdx.x;
+    if (tid >= w) {
+        return;
+    }
+    d_vector[tid] = 0.0f;
 }