mul_strassen: avoid computing some entries twice

fredrik-johansson · Sep 26, 2024 · b4976c2 · b4976c2
1 parent 02d4e5d
commit b4976c2
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 4 deletions.
diff --git a/src/fmpz_mat/mul_strassen.c b/src/fmpz_mat/mul_strassen.c
@@ -122,10 +122,22 @@ void fmpz_mat_mul_strassen(fmpz_mat_t C, const fmpz_mat_t A, const fmpz_mat_t B)
 
     if (a > 2*anr)
     {
-        fmpz_mat_t Ar, Cr;
+        fmpz_mat_t Ar, Br, Cr;
         fmpz_mat_window_init(Ar, A, 2*anr, 0, a, b);
         fmpz_mat_window_init(Cr, C, 2*anr, 0, a, c);
-        fmpz_mat_mul(Cr, Ar, B);
+
+        /* don't compute the overlapping entries twice */
+        if (c > 2 * bnc)
+        {
+            fmpz_mat_window_init(Br, B, 0, 0, b, 2*bnc);
+            fmpz_mat_mul(Cr, Ar, Br);
+            fmpz_mat_window_clear(Br);
+        }
+        else
+        {
+            fmpz_mat_mul(Cr, Ar, B);
+        }
+
         fmpz_mat_window_clear(Ar);
         fmpz_mat_window_clear(Cr);
     }

diff --git a/src/gr_mat/mul_strassen.c b/src/gr_mat/mul_strassen.c
@@ -147,10 +147,22 @@ int gr_mat_mul_strassen(gr_mat_t C, const gr_mat_t A, const gr_mat_t B, gr_ctx_t
 
     if (ar > 2 * anr)
     {
-        gr_mat_t Ar, Cr;
+        gr_mat_t Ar, Br, Cr;
         gr_mat_window_init(Ar, A, 2 * anr, 0, ar, ac, ctx);
         gr_mat_window_init(Cr, C, 2 * anr, 0, ar, bc, ctx);
-        status |= gr_mat_mul(Cr, Ar, B, ctx);
+
+        /* don't compute the overlapping entries twice */
+        if (bc > 2 * bnc)
+        {
+            gr_mat_window_init(Br, B, 0, 0, ac, 2 * bnc, ctx);
+            status |= gr_mat_mul(Cr, Ar, Br, ctx);
+            gr_mat_window_clear(Br, ctx);
+        }
+        else
+        {
+            status |= gr_mat_mul(Cr, Ar, B, ctx);
+        }
+
         gr_mat_window_clear(Ar, ctx);
         gr_mat_window_clear(Cr, ctx);
     }