add comments for the tricky block matrix multiplication

apache · Sep 2, 2019 · 1a97e58 · 1a97e58
1 parent c9e9779
commit 1a97e58
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/src/operator/numpy/linalg/np_gesvd-inl.h b/src/operator/numpy/linalg/np_gesvd-inl.h
@@ -236,6 +236,7 @@ struct gesvd_backward {
 
     // G3:
     gemm::op(tempM, V, dA, DType(1.0), DType(1.0), false, false, s);
+    // dA <- dot(UT, dA). Loop over (k, m, m) blocks to avoid large temporary memory
     for (int i = 0; i < n; i += m) {
       int ncols = n - i < m ? n - i : m;
       Tensor<xpu, 3, DType> t = Tensor<xpu, 3, DType>(dA.dptr_ + i,