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

sgemm_impl: Remove #if 0, FP_SIZE 16

Browse Source
This commit is contained in:
Hansung Kim
2024-09-05 19:55:36 -07:00
parent a832fa7b84
commit 9f067acdb9
1 changed files with 1 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/regression/sgemm_tcore/sgemm_impl.hpp
View File

@@ -6,7 +6,7 @@
#include "include/gemmini.h" #include "include/gemmini.h"
#include "gemmini_mmio.h" #include "gemmini_mmio.h"
#define FP_SIZE 32 #define FP_SIZE 16
// "fake" fp16 type that only has the correct data width. // "fake" fp16 type that only has the correct data width.
using float16_t = uint16_t; using float16_t = uint16_t;
@@ -1038,7 +1038,6 @@ inline void thread_block_gemm(const T *A, const T *B, float *C,
warps_per_threadblock_per_core); warps_per_threadblock_per_core);
#endif #endif
#if 0
// consumer code: SMEM->RF and compute // consumer code: SMEM->RF and compute
// ---------------------------------------------------------------------- // ----------------------------------------------------------------------
// @perf: this loop spills to stack a lot because of all the flws in // @perf: this loop spills to stack a lot because of all the flws in
@@ -1087,7 +1086,6 @@ inline void thread_block_gemm(const T *A, const T *B, float *C,
threadblock_barrier(threadblock_id_in_cluster, threadblock_barrier(threadblock_id_in_cluster,
warps_per_threadblock_per_core); warps_per_threadblock_per_core);
#endif
} }
if constexpr (write_to_gmem) { if constexpr (write_to_gmem) {