diff options
Diffstat (limited to 'matrix/generic/cukernel.cu')
| -rw-r--r-- | matrix/generic/cukernel.cu | 64 |
1 files changed, 53 insertions, 11 deletions
diff --git a/matrix/generic/cukernel.cu b/matrix/generic/cukernel.cu index a37ccf4..4b6af61 100644 --- a/matrix/generic/cukernel.cu +++ b/matrix/generic/cukernel.cu @@ -27,7 +27,7 @@ __global__ void cudak_(softmax_final)(const MATRIX_ELEM *a, MATRIX_ELEM *b, b[idx] = exp(a[idx] - max[0 + i * mstride]) / deno[0 + i * mstride]; } -__global__ void cudak_(block_reduce_sum)(const MATRIX_ELEM *input, +__global__ void cudak_(block_reduce_rowsum)(const MATRIX_ELEM *input, MATRIX_ELEM *output, const int istride, const int ostride, const int n) { @@ -45,7 +45,25 @@ __global__ void cudak_(block_reduce_sum)(const MATRIX_ELEM *input, output[blockIdx.x + ostride * blockIdx.y] = cudak_(arr)[0]; } -__global__ void cudak_(block_reduce_softmax_sum)(const MATRIX_ELEM *input, +__global__ void cudak_(block_reduce_colsum)(const MATRIX_ELEM *input, + MATRIX_ELEM *output, + const int istride, const int ostride, + const int n) { + extern __shared__ MATRIX_ELEM cudak_(arr)[]; + int i = blockIdx.y * blockDim.y + threadIdx.y; + cudak_(arr)[threadIdx.y] = i < n ? input[blockIdx.x + istride * i] : 0; + __syncthreads(); + for (int offset = blockDim.y >> 1; offset; offset >>= 1) + { + if (threadIdx.y < offset) + cudak_(arr)[threadIdx.y] += cudak_(arr)[threadIdx.y + offset]; + __syncthreads(); + } + if (threadIdx.y == 0) + output[blockIdx.x + ostride * blockIdx.y] = cudak_(arr)[0]; +} + +__global__ void cudak_(block_reduce_softmax_rowsum)(const MATRIX_ELEM *input, MATRIX_ELEM *output, const MATRIX_ELEM *max, const int istride, const int ostride, @@ -65,7 +83,7 @@ __global__ void cudak_(block_reduce_softmax_sum)(const MATRIX_ELEM *input, output[blockIdx.x + ostride * blockIdx.y] = cudak_(arr)[0]; } -__global__ void cudak_(block_reduce_max)(const MATRIX_ELEM *input, +__global__ void cudak_(block_reduce_rowmax)(const MATRIX_ELEM *input, MATRIX_ELEM *output, const int istride, const int ostride, const int n) { @@ -99,7 +117,7 @@ extern "C" { b->stride / sizeof(MATRIX_ELEM)); } - void cudak_(cuda_colsum)(const Matrix *a, Matrix *b) { + void cudak_(cuda_rowsum)(const Matrix *a, Matrix *b) { dim3 block(CUDA_THREADS_NN, 1); int ncol = a->ncol; int blocks_per_row = CEIL_DIV(ncol, block.x); @@ -107,20 +125,42 @@ extern "C" { MATRIX_ELEM *res; size_t stride; cudaMallocPitch(&res, &stride, blocks_per_row * sizeof(MATRIX_ELEM), a->nrow); - cudak_(block_reduce_sum)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ + cudak_(block_reduce_rowsum)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ (MATRIX_ELEM_PTR(a), res, a->stride / sizeof(MATRIX_ELEM), stride / sizeof(MATRIX_ELEM), ncol); ncol = blocks_per_row; assert((unsigned long)ncol <= block.x); grid.x = 1; - cudak_(block_reduce_sum)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ + cudak_(block_reduce_rowsum)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ (res, MATRIX_ELEM_PTR(b), stride / sizeof(MATRIX_ELEM), b->stride / sizeof(MATRIX_ELEM), ncol); cudaFree(res); } + void cudak_(cuda_colsum)(const Matrix *a, Matrix *b) { + dim3 block(1, CUDA_THREADS_NN); + int nrow = a->nrow; + int blocks_per_col = CEIL_DIV(nrow, block.x); + dim3 grid(a->ncol, blocks_per_col); + MATRIX_ELEM *res; + size_t stride; + cudaMallocPitch(&res, &stride, a->ncol * sizeof(MATRIX_ELEM), blocks_per_col); + cudak_(block_reduce_colsum)<<<grid, block, block.y * sizeof(MATRIX_ELEM)>>> \ + (MATRIX_ELEM_PTR(a), res, + a->stride / sizeof(MATRIX_ELEM), stride / sizeof(MATRIX_ELEM), + nrow); + nrow = blocks_per_col; + assert((unsigned long)nrow <= block.y); + grid.y = 1; + cudak_(block_reduce_colsum)<<<grid, block, block.y * sizeof(MATRIX_ELEM)>>> \ + (res, MATRIX_ELEM_PTR(b), + stride / sizeof(MATRIX_ELEM), b->stride / sizeof(MATRIX_ELEM), + nrow); + cudaFree(res); + } + void cudak_(cuda_softmax_final)(const Matrix *a, const Matrix *max, const Matrix *deno, Matrix *b) { dim3 threadsPerBlock(CUDA_THREADS_N, @@ -144,7 +184,8 @@ extern "C" { size_t stride; assert(max->ncol == 1); cudaMallocPitch(&res, &stride, blocks_per_row * sizeof(MATRIX_ELEM), a->nrow); - cudak_(block_reduce_softmax_sum)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ + cudak_(block_reduce_softmax_rowsum) \ + <<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ (MATRIX_ELEM_PTR(a), res, MATRIX_ELEM_PTR(max), a->stride / sizeof(MATRIX_ELEM), stride / sizeof(MATRIX_ELEM), max->stride / sizeof(MATRIX_ELEM), @@ -152,14 +193,15 @@ extern "C" { ncol = blocks_per_row; assert((unsigned long)ncol <= block.x); grid.x = 1; - cudak_(block_reduce_sum)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ + cudak_(block_reduce_rowsum) \ + <<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ (res, MATRIX_ELEM_PTR(b), stride / sizeof(MATRIX_ELEM), b->stride / sizeof(MATRIX_ELEM), ncol); cudaFree(res); } - void cudak_(cuda_colmax)(const Matrix *a, Matrix *b) { + void cudak_(cuda_rowmax)(const Matrix *a, Matrix *b) { dim3 block(CUDA_THREADS_NN, 1); int ncol = a->ncol; int blocks_per_row = CEIL_DIV(ncol, block.x); @@ -167,14 +209,14 @@ extern "C" { MATRIX_ELEM *res; size_t stride; cudaMallocPitch(&res, &stride, blocks_per_row * sizeof(MATRIX_ELEM), a->nrow); - cudak_(block_reduce_max)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ + cudak_(block_reduce_rowmax)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ (MATRIX_ELEM_PTR(a), res, a->stride / sizeof(MATRIX_ELEM), stride / sizeof(MATRIX_ELEM), ncol); ncol = blocks_per_row; assert((unsigned long)ncol <= block.x); grid.x = 1; - cudak_(block_reduce_max)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ + cudak_(block_reduce_rowmax)<<<grid, block, block.x * sizeof(MATRIX_ELEM)>>> \ (res, MATRIX_ELEM_PTR(b), stride / sizeof(MATRIX_ELEM), b->stride / sizeof(MATRIX_ELEM), ncol); |