From f3e42f96c736d1416b309b4a023d5d3cb677c901 Mon Sep 17 00:00:00 2001 From: Przemek Tredak Date: Thu, 22 Aug 2019 15:11:14 -0700 Subject: [PATCH] Fix get_rows_per_block --- src/common/cuda_utils.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/common/cuda_utils.cc b/src/common/cuda_utils.cc index f38b2f8b5490..893b34e6ff29 100644 --- a/src/common/cuda_utils.cc +++ b/src/common/cuda_utils.cc @@ -79,7 +79,7 @@ int get_rows_per_block(size_t row_size, int num_threads_per_block) { int actual_num_warps_per_row = std::min(desired_num_warps_per_row, num_threads_per_block / warp_size); // actual number of warps needs to be power of 2 - actual_num_warps_per_row = RoundToPower2(desired_num_warps_per_row); + actual_num_warps_per_row = RoundToPower2(actual_num_warps_per_row); return num_threads_per_block / (warp_size * actual_num_warps_per_row); }