one malloc for all scratch buffer

elinx · Oct 5, 2021 · 9416d59 · 9416d59
1 parent c7b7973
commit 9416d59
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 8 deletions.
diff --git a/README.md b/README.md
@@ -96,8 +96,8 @@ for my hardware in single thread
 
 |               |  Shape   | Memory  | Memory(Readable) |
 | :-----------: | :------: | :-----: | :--------------: |
-| L1 Cache Size |          | 196608  |      192 KB      |
-| L2 Cache Size |          | 1572864 |      1.5 MB      |
+| L1 Cache Size |          |  32768  |      32 KB       |
+| L2 Cache Size |          | 262144  |      256 KB      |
 | L3 Cache Size |          | 9437184 |       9 MB       |
 |       M       |   640    |         |                  |
 |       N       |   640    |         |                  |

diff --git a/manual_optimize_dgemm.cpp b/manual_optimize_dgemm.cpp
@@ -809,9 +809,10 @@ void manual_dgemm(const double *A, const double *B, double *C, const uint32_t M,
   static_assert(!(MC % MR));
   static_assert(!(KC % 4));
 
-  double *Bc = (double *)aligned_alloc(32, sizeof(double) * KC * NC);
-  double *Ac = (double *)aligned_alloc(32, sizeof(double) * MC * KC);
-  double *Cc = (double *)aligned_alloc(32, sizeof(double) * MR * NR);
+  double *buf = (double *)aligned_alloc(32, sizeof(double) * ((KC * NC) + (MC * KC) + (MR * NR)));
+  double *Bc = buf;
+  double *Ac = buf + KC * NC;
+  double *Cc = Ac + MC * KC;
 
   for (uint32_t n_outer = 0; n_outer < n_outer_bound; n_outer += NC) {
     for (uint32_t k_outer = 0; k_outer < k_outer_bound; k_outer += KC) {
@@ -829,9 +830,7 @@ void manual_dgemm(const double *A, const double *B, double *C, const uint32_t M,
       }
     }
   }
-  free(Cc);
-  free(Ac);
-  free(Bc);
+  free(buf);
 }
 
 int main() {