Reduce H/S re-application frequency in PPCG block subspace

The Rayleigh-Ritz rotation already keeps hpsi_/spsi_ consistent with the
rotated psi up to rounding, so re-applying H/S exactly every iteration is
redundant.  Re-apply every rr_step_ iterations to reset the accumulated
rounding drift instead, removing one full-block H/S application on most
iterations (~1.5x wall-time speedup).

Author: cheerly-pku

source/source_hsolver/diago_ppcg.cpp

@@ -1636,8 +1636,15 @@ double DiagoPPCG<T, Device>::diag(const HPsiFunc& hpsi_func,
             // can otherwise drift into an ill-conditioned basis before the next
             // Ritz rotation.
             rayleigh_ritz(psi_in, eigenvalue_in, active_cols, ethr_band);
-            apply_h(hpsi_func, psi_in, hpsi_.data(), ncol);
-            apply_s_current(psi_in, spsi_.data(), ncol);
+            // The Rayleigh-Ritz rotation already keeps hpsi_/spsi_ consistent
+            // with the rotated psi up to rounding; re-applying H/S exactly is
+            // only needed every rr_step_ iterations to reset the accumulated
+            // rounding drift.
+            if ((iter % rr_step_) == 0)
+            {
+                apply_h(hpsi_func, psi_in, hpsi_.data(), ncol);
+                apply_s_current(psi_in, spsi_.data(), ncol);
+            }
             record_residual(iter, "rayleigh_ritz");
 
             ++iter;

tests/01_PW/817_PW_PPCG/result.ref

@@ -1,8 +1,8 @@
-etotref -4862.3309719757144194
+etotref -4862.3309719757116909
 etotperatomref -2431.1654859879
 totalforceref 9.131552
 totalstressref 37222.701329
 pointgroupref C_1
 spacegroupref C_1
 nksibzref 2
-totaltimeref 3.99
+totaltimeref 2.57