Add the full quantized matmul kernels for cuda. (#2057)

Add the full quantized matmul kernels for cuda. (#2057) #1352