Skip to content

GitLab

Commit 6ed9a3e4 authored by wenjh's avatar wenjh
Browse files

[DCU] Add width to __shfl 


Signed-off-by: wenjh's avatarwenjh <wenjh@sugon.com>
parent b27e513d
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 2 deletions
transformer_engine/common/transpose/quantize_transpose_square_blockwise.cu
View file @ 6ed9a3e4
......@@ -135,7 +135,8 @@ __global__ void __launch_bounds__(THREADS_PER_BLOCK)
// broadcast the amax to all threads in a warp from the lane 0
constexpr int lane_zero = 0;
#ifdef __HIP_PLATFORM_AMD__
warp_tile_amax = __shfl(warp_tile_amax, lane_zero);
warp_tile_amax = __shfl(warp_tile_amax, lane_zero, THREADS_PER_WARP);
__syncthreads();
#else
warp_tile_amax = __shfl_sync(0xFFFFFFFF, warp_tile_amax, lane_zero);
#endif
......@@ -362,7 +363,8 @@ __global__ void __launch_bounds__(THREADS_PER_BLOCK) block_scaled_cast_transpose
// broadcast the amax to all threads in a warp from the lane 0
constexpr int lane_zero = 0;
#ifdef __HIP_PLATFORM_AMD__
warp_tile_amax = __shfl(warp_tile_amax, lane_zero);
warp_tile_amax = __shfl(warp_tile_amax, lane_zero, THREADS_PER_WARP);
__syncthreads();
#else
warp_tile_amax = __shfl_sync(0xFFFFFFFF, warp_tile_amax, lane_zero);
#endif
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment