FastGPUConvolution

Part of UIUC ECE408/CS483 Final project.

An optimized convolution layer of CNN on CUDA

• GEMM implementation of convolution.
• Shared memory tiling of matrix multiplication.
• Fp16 vectorized matrix multiplication.
• Accelerating GEMM with the Tensor Cores (wmma instruction).