Add convertfp16tofloatCuda memory free

This commit is contained in:
kradchen
2023-12-19 13:10:12 +08:00
parent cd10ebb5e8
commit ea68e6c5af

View File

@@ -1631,5 +1631,6 @@ CudaMatrix Aurora::convertfp16tofloatCuda(short* aData, int aRows, int aColumns)
cudaMalloc((void**)&output,size*sizeof(float)); cudaMalloc((void**)&output,size*sizeof(float));
int blocksPerGrid = (size + THREADS_PER_BLOCK - 1) / THREADS_PER_BLOCK; int blocksPerGrid = (size + THREADS_PER_BLOCK - 1) / THREADS_PER_BLOCK;
convertValueKernel<<<blocksPerGrid, THREADS_PER_BLOCK>>>(input,output, size); convertValueKernel<<<blocksPerGrid, THREADS_PER_BLOCK>>>(input,output, size);
cudaFree(input);
return CudaMatrix::fromRawData(output, aRows, aColumns); return CudaMatrix::fromRawData(output, aRows, aColumns);
} }