sgemm_impl: Fix wrong smem address for fp16

Verified results for fp16 256x256.
This commit is contained in:
Hansung Kim
2024-09-08 02:23:51 -07:00
parent c51dc4902d
commit adcd0a9d49
2 changed files with 10 additions and 9 deletions

View File

@@ -7,7 +7,7 @@
#include "include/gemmini.h"
#include "gemmini_mmio.h"
constexpr bool DEBUG = true;
constexpr bool DEBUG = false;
template <uint32_t tile_dim_row, uint32_t tile_dim_col>
inline void thread_block_copy_tile(const float *src, float *dest,
@@ -91,8 +91,8 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
/*write_to_gmem=*/true,
/*smem_a_offset=*/0,
/*smem_a_dbuf_offset=*/0,
/*smem_b_offset=*/2 * BM * BK * sizeof(float),
/*smem_b_dbuf_offset=*/2 * BM * BK * sizeof(float)>(
/*smem_b_offset=*/2 * BM * BK * sizeof(float_type),
/*smem_b_dbuf_offset=*/2 * BM * BK * sizeof(float_type)>(
(const float_type *)arg->addr_a, (const float_type *)arg->addr_b,
(float *)arg->addr_c, arg->dim_m, arg->dim_n, arg->dim_k,
tid_in_threadblock, threadblocks_per_cluster, threadblock_id_in_cluster,
@@ -102,7 +102,8 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
float *gmem_tmp_d1 = reinterpret_cast<float *>(0xd1000000UL);
const float *smem_A = reinterpret_cast<float *>(sharedmem_per_threadblock);
const float *smem_B = smem_A + 2 * BM * BK;
const float *smem_B = reinterpret_cast<float *>(
sharedmem_per_threadblock + 2 * BM * BK * sizeof(float_type));
if constexpr (DEBUG) {
threadblock_barrier(threadblock_id_in_cluster,