[apps] Add fmatmul scalar kernel

pulp-platform · Jan 16, 2024 · 57f3228 · 57f3228
1 parent 594ce81
commit 57f3228
Show file tree

Hide file tree

Showing 3 changed files with 25 additions and 0 deletions.
diff --git a/apps/benchmarks/benchmark/fmatmul.bmark b/apps/benchmarks/benchmark/fmatmul.bmark
@@ -41,7 +41,11 @@ void warm_caches(uint64_t heat) {
   volatile double buf;
 
   for (uint64_t k = 0; k < heat; ++k)
+#ifndef SCALAR
     fmatmul(c, a, b, M, N, P);
+#else
+    fmatmul_scalar(c, a, b, M, N, P);
+#endif
 #ifdef AD_HOC_WARMING
     // Vector stores have invalidated the A mtx cache lines!
     // Fetch them again
@@ -62,7 +66,11 @@ int main() {
   // Measure runtime with a hot cache
   HW_CNT_READY;
   start_timer();
+#ifndef SCALAR
   fmatmul(c, a, b, M, N, P);
+#else
+  fmatmul_scalar(c, a, b, M, N, P);
+#endif
   stop_timer();
 
   int64_t runtime = get_timer();

diff --git a/apps/fmatmul/kernel/fmatmul.c b/apps/fmatmul/kernel/fmatmul.c
@@ -21,6 +21,19 @@
 
 #define MIN(a, b) ((a) < (b) ? (a) : (b))
 
+void fmatmul_scalar(double *c, const double *a, const double *b,
+             const unsigned long int M, const unsigned long int N,
+             const unsigned long int P) {
+    for (unsigned int i = 0; i < M; i++) {
+        for (unsigned int j = 0; j < P; j++) {
+            c[i][j] = 0;
+            for (unsigned int k = 0; k < N; k++) {
+                c[i][j] += a[i][k] * b[k][j];
+            }
+        }
+    }
+}
+
 void fmatmul(double *c, const double *a, const double *b,
              const unsigned long int M, const unsigned long int N,
              const unsigned long int P) {

diff --git a/apps/fmatmul/kernel/fmatmul.h b/apps/fmatmul/kernel/fmatmul.h
@@ -25,6 +25,10 @@
 void fmatmul(double *c, const double *a, const double *b, unsigned long int m,
              unsigned long int n, unsigned long int p);
 
+void fmatmul_scalar(double *c, const double *a, const double *b,
+             const unsigned long int M, const unsigned long int N,
+             const unsigned long int P);
+
 void fmatmul_4x4(double *c, const double *a, const double *b,
                  unsigned long int m, unsigned long int n, unsigned long int p);
 void fmatmul_vec_4x4_slice_init();