merge-squashed: propedeuthic changes towards gemm cost reduction

make rank1 work just on non-deflated (single-threaded)
eth-cscs · Dec 4, 2023 · 477b2e2 · 477b2e2
1 parent 51109c9
commit 477b2e2
Show file tree

Hide file tree

Showing 2 changed files with 269 additions and 282 deletions.
diff --git a/include/dlaf/eigensolver/tridiag_solver/impl.h b/include/dlaf/eigensolver/tridiag_solver/impl.h
@@ -219,7 +219,8 @@ void TridiagSolver<B, D, T>::call(Matrix<T, Device::CPU>& tridiag, Matrix<T, D>&
                      Matrix<T, D>(vec_size, vec_tile_size),          // z0
                      Matrix<T, D>(vec_size, vec_tile_size),          // z1
                      Matrix<SizeType, D>(vec_size, vec_tile_size),   // i2
-                     Matrix<SizeType, D>(vec_size, vec_tile_size)};  // i5
+                     Matrix<SizeType, D>(vec_size, vec_tile_size),   // i5
+                     Matrix<SizeType, D>(vec_size, vec_tile_size)};  // i6
 
   WorkSpaceHost<T> ws_h{Matrix<T, Device::CPU>(vec_size, vec_tile_size),          // d0
                         Matrix<ColType, Device::CPU>(vec_size, vec_tile_size),    // c
@@ -380,7 +381,8 @@ void TridiagSolver<B, D, T>::call(comm::CommunicatorGrid grid, Matrix<T, Device:
                      Matrix<T, D>(dist_evals),          // z0
                      Matrix<T, D>(dist_evals),          // z1
                      Matrix<SizeType, D>(dist_evals),   // i2
-                     Matrix<SizeType, D>(dist_evals)};  // i5
+                     Matrix<SizeType, D>(dist_evals),   // i5
+                     Matrix<SizeType, D>(dist_evals)};  // i6
 
   WorkSpaceHost<T> ws_h{Matrix<T, Device::CPU>(dist_evals),          // d0
                         Matrix<ColType, Device::CPU>(dist_evals),    // c
@@ -392,7 +394,7 @@ void TridiagSolver<B, D, T>::call(comm::CommunicatorGrid grid, Matrix<T, Device:
   DistWorkSpaceHostMirror<T, D> ws_hm{initMirrorMatrix(ws.e0), initMirrorMatrix(ws.e2),
                                       initMirrorMatrix(ws.d1), initMirrorMatrix(ws.z0),
                                       initMirrorMatrix(ws.z1), initMirrorMatrix(ws.i2),
-                                      initMirrorMatrix(ws.i5)};
+                                      initMirrorMatrix(ws.i5), initMirrorMatrix(ws.i6)};
 
   // Set `ws.e0` to `zero` (needed for Given's rotation to make sure no random values are picked up)
   matrix::util::set0<B, T, D>(thread_priority::normal, ws.e0);
@@ -426,12 +428,12 @@ void TridiagSolver<B, D, T>::call(comm::CommunicatorGrid grid, Matrix<T, Device:
   copy(ws.e0, ws_hm.e0);
 
   // Note: ws_hm.d1 is the mirror of ws.d1 which is evals
-  applyIndex(0, n, ws_hm.i2, ws_h.d0, ws_hm.d1);
+  applyIndex(0, n, ws_h.i1, ws_h.d0, ws_hm.d1);
   copy(ws_hm.d1, evals);
 
   // Note: ws_hm.e2 is the mirror of ws.e2 which is evecs
   dlaf::permutations::permute<Backend::MC, Device::CPU, T, Coord::Col>(grid, row_task_chain, 0, n,
-                                                                       ws_hm.i2, ws_hm.e0, ws_hm.e2);
+                                                                       ws_h.i1, ws_hm.e0, ws_hm.e2);
   copy(ws_hm.e2, evecs);
 }