diff options
author | Tim Dettmers <tim.dettmers@gmail.com> | 2022-07-26 17:39:30 -0700 |
---|---|---|
committer | Tim Dettmers <tim.dettmers@gmail.com> | 2022-07-26 17:39:30 -0700 |
commit | bcab99ec877ba063543bd7c03ba1cdd1b06e8078 (patch) | |
tree | 3d1dcfec5d3361ad4e5c5ba552a432e045445851 /csrc/ops.cu | |
parent | cbb901ac51bd6c41e4243ffb936ef0e2f7ca8ada (diff) |
Working outlier extraction for Turing.
Diffstat (limited to 'csrc/ops.cu')
-rw-r--r-- | csrc/ops.cu | 5 |
1 files changed, 2 insertions, 3 deletions
diff --git a/csrc/ops.cu b/csrc/ops.cu index fe2d7fe..e6227ae 100644 --- a/csrc/ops.cu +++ b/csrc/ops.cu @@ -586,8 +586,7 @@ template <int FORMAT> void extractOutliers(char * A, int *idx, char *out, int id int tiledCols = tiledCols = fill_up_to_nearest_multiple(cols, 32); int tiledRows = 0; - int elements = idx_size*cols; // matrix A is transposed, so we extract columns - int num_blocks = (elements+threads-1)/threads; + int num_blocks = idx_size; if(FORMAT == COL_TURING) { @@ -598,7 +597,7 @@ template <int FORMAT> void extractOutliers(char * A, int *idx, char *out, int id tiledRows = fill_up_to_nearest_multiple(rows, 32); } - kExtractOutliers<FORMAT><<<num_blocks, threads>>>(A, idx, out, rows, cols, tiledRows, tiledCols); + kExtractOutliers<FORMAT><<<num_blocks, threads>>>(A, idx, out, idx_size, rows, cols, tiledRows, tiledCols); CUDA_CHECK_RETURN(cudaPeekAtLastError()); } |