diff options
Diffstat (limited to 'nerv/lib/matrix/generic/cukernel.cu')
| -rw-r--r-- | nerv/lib/matrix/generic/cukernel.cu | 19 |
1 files changed, 19 insertions, 0 deletions
diff --git a/nerv/lib/matrix/generic/cukernel.cu b/nerv/lib/matrix/generic/cukernel.cu index 82bea14..4447f7c 100644 --- a/nerv/lib/matrix/generic/cukernel.cu +++ b/nerv/lib/matrix/generic/cukernel.cu @@ -263,6 +263,14 @@ __global__ void cudak_(add_row)(const MATRIX_ELEM *a, MATRIX_ELEM *b, b[j + i * stride] += beta * a[j]; } +__global__ void cudak_(add_col)(const MATRIX_ELEM *a, MATRIX_ELEM *b, + int nrow, int ncol, int astride, int bstride, double beta) { + int j = blockIdx.x * blockDim.x + threadIdx.x; + int i = blockIdx.y * blockDim.y + threadIdx.y; + if (i >= nrow || j >= ncol) return; + b[j + i * bstride] += beta * a[i * astride]; +} + __global__ void cudak_(fill)(MATRIX_ELEM *a, int nrow, int ncol, int stride, double val) { int j = blockIdx.x * blockDim.x + threadIdx.x; @@ -735,6 +743,17 @@ extern "C" { cudaStreamSynchronize(0); } + void cudak_(cuda_add_col)(const Matrix *a, Matrix *b, double beta) { + dim3 threadsPerBlock(CUDA_THREADS_N, CUDA_THREADS_N); + dim3 numBlocks(CEIL_DIV(b->ncol, threadsPerBlock.x), + CEIL_DIV(b->nrow, threadsPerBlock.y)); + cudak_(add_col)<<<numBlocks, threadsPerBlock>>> \ + (MATRIX_ELEM_PTR(a), MATRIX_ELEM_PTR(b), b->nrow, b->ncol, + a->stride / sizeof(MATRIX_ELEM), + b->stride / sizeof(MATRIX_ELEM), beta); + cudaStreamSynchronize(0); + } + void cudak_(cuda_fill)(Matrix *a, double val) { dim3 threadsPerBlock(CUDA_THREADS_N, CUDA_THREADS_N); dim3 numBlocks(CEIL_DIV(a->ncol, threadsPerBlock.x), |