micro-optimization to mju_mulMatTVecSparse

PiperOrigin-RevId: 691431137 Change-Id: I95f94ff38ae911194b21aa220af35c7374af9739
2024-10-30 08:38:12 -07:00
parent 32df8a515a
commit 70d944b458
1 changed files with 6 additions and 3 deletions
@@ -224,13 +224,16 @@ void mju_mulMatTVecSparse(mjtNum* res, const mjtNum* mat, const mjtNum* vec, int
  mju_zero(res, nc);

  for (int i=0; i < nr; i++) {
+    mjtNum scl = vec[i];
+
+    // skip if 0
+    if (!scl) continue;
+
+    // add row scaled by the corresponding vector element
    int nnz = rownnz[i];
    int adr = rowadr[i];
    const int* ind = colind + adr;
    const mjtNum* row = mat + adr;
-    mjtNum scl = vec[i];
-
-    // add row scaled by the corresponding vector element
    for (int j=0; j < nnz; j++) {
      res[ind[j]] += row[j] * scl;
    }