wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

sgemm_tcore: Fix casting error

Browse Source
This commit is contained in:
Hansung Kim
2024-08-12 17:57:50 -07:00
parent 95e3e96c6c
commit e69fbea83a
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/regression/sgemm_tcore/kernel.cpp
View File

@@ -84,10 +84,10 @@ inline void global_dmem_load(const uint32_t dim_n, const uint32_t dim_k,
// this is equivalent to threadblock_dim_y (assuming threadblock_dim_x == // this is equivalent to threadblock_dim_y (assuming threadblock_dim_x ==
// BK) // BK)
constexpr uint32_t row_stride_a = threads_in_threadblock / BK_adjusted; constexpr uint32_t row_stride_a = threads_in_threadblock / BK_adjusted;
const float *global_a = reinterpret_cast<float *>(A) + const float *global_a = reinterpret_cast<const float *>(A) +
dim_k_adjusted * global_a_row + dim_k_adjusted * global_a_row +
(k_adjusted + local_a_col); (k_adjusted + local_a_col);
volatile float *local_a_tmp = reinterpret_cast<float *>(local_a) + volatile float *local_a_tmp = reinterpret_cast<volatile float *>(local_a) +
BK_adjusted * local_a_row + local_a_col; BK_adjusted * local_a_row + local_a_col;
#pragma GCC unroll 1 #pragma GCC unroll 1