fix(gpu): guard qq synchronization without projectors (deepmodeling#7816)

* fix(gpu): guard qq synchronization without projectors

* test(gpu): re-enable cases fixed by deepmodeling#7681

---------

Co-authored-by: Gengze Lin <gengze@DESKTOP-RUU2CTC.localdomain>

Author: GenZeal-Lin

source/source_pw/module_pwdft/vnl_pw.cpp

@@ -768,10 +768,13 @@ void pseudopot_cell_vnl::init_vnl(UnitCell& cell, const ModulePW::PW_Basis* rho_
             castmem_d2s_h2d_op()(this->s_nhtol, this->nhtol.c, this->nhtol.nr * this->nhtol.nc);
             castmem_d2s_h2d_op()(this->s_nhtolm, this->nhtolm.c, this->nhtolm.nr * this->nhtolm.nc);
             castmem_d2s_h2d_op()(this->s_tab, this->tab.ptr, this->tab.getSize());
-            castmem_d2s_h2d_op()(this->s_qq_nt, this->qq_nt.ptr, this->qq_nt.getSize());
-            castmem_z2c_h2d_op()(this->c_qq_so, this->qq_so.ptr, this->qq_so.getSize());
+            if (this->nhm > 0)
+            {
+                castmem_d2s_h2d_op()(this->s_qq_nt, this->qq_nt.ptr, this->qq_nt.getSize());
+                castmem_z2c_h2d_op()(this->c_qq_so, this->qq_so.ptr, this->qq_so.getSize());
+            }
         }
-        if (PARAM.globalv.has_double_data)
+        if (PARAM.globalv.has_double_data && this->nhm > 0)
         {
             syncmem_z2z_h2d_op()(this->z_qq_so, this->qq_so.ptr, this->qq_so.getSize());
         }
@@ -782,7 +785,10 @@ void pseudopot_cell_vnl::init_vnl(UnitCell& cell, const ModulePW::PW_Basis* rho_
         syncmem_d2d_h2d_op()(this->d_nhtol, this->nhtol.c, this->nhtol.nr * this->nhtol.nc);
         syncmem_d2d_h2d_op()(this->d_nhtolm, this->nhtolm.c, this->nhtolm.nr * this->nhtolm.nc);
         syncmem_d2d_h2d_op()(this->d_tab, this->tab.ptr, this->tab.getSize());
-        syncmem_d2d_h2d_op()(this->d_qq_nt, this->qq_nt.ptr, this->qq_nt.getSize());
+        if (this->nhm > 0)
+        {
+            syncmem_d2d_h2d_op()(this->d_qq_nt, this->qq_nt.ptr, this->qq_nt.getSize());
+        }
     }
     else
     {
@@ -792,8 +798,11 @@ void pseudopot_cell_vnl::init_vnl(UnitCell& cell, const ModulePW::PW_Basis* rho_
             castmem_d2s_h2h_op()(this->s_nhtol, this->nhtol.c, this->nhtol.nr * this->nhtol.nc);
             castmem_d2s_h2h_op()(this->s_nhtolm, this->nhtolm.c, this->nhtolm.nr * this->nhtolm.nc);
             castmem_d2s_h2h_op()(this->s_tab, this->tab.ptr, this->tab.getSize());
-            castmem_d2s_h2h_op()(this->s_qq_nt, this->qq_nt.ptr, this->qq_nt.getSize());
-            castmem_z2c_h2h_op()(this->c_qq_so, this->qq_so.ptr, this->qq_so.getSize());
+            if (this->nhm > 0)
+            {
+                castmem_d2s_h2h_op()(this->s_qq_nt, this->qq_nt.ptr, this->qq_nt.getSize());
+                castmem_z2c_h2h_op()(this->c_qq_so, this->qq_so.ptr, this->qq_so.getSize());
+            }
         }
         // There's no need to synchronize double precision pointers while in a CPU environment.
     }

tests/01_PW/CASES_GPU.txt

@@ -17,10 +17,10 @@ scf_out_elf
 013_PW_ONCV_LDA
 014_PW_UPF201_BLPS
 015_PW_GTH
-#016_PW_BLPS
-#017_PW_LPS6
-#018_PW_LPS8
-#019_PW_Coulomb
+016_PW_BLPS
+017_PW_LPS6
+018_PW_LPS8
+019_PW_Coulomb
 020_PW_kspace
 021_PW_kspace3
 022_PW_CG
@@ -113,17 +113,17 @@ scf_out_elf
 #208_PW_CG_float
 209_PW_DFTHALF
 210_PW_kspace_shift
-#801_PW_LT_sc
-#802_PW_LT_fcc
+801_PW_LT_sc
+802_PW_LT_fcc
 #803_PW_LT_bcc
-#804_PW_LT_hex
-#805_PW_LT_trigonal
-#806_PW_LT_st
-#807_PW_LT_bct
-#808_PW_LT_so
-#809_PW_LT_baco
-#810_PW_LT_fco
-#811_PW_LT_bco
-#812_PW_LT_sm
-#813_PW_LT_bacm
-#814_PW_LT_triclinic
+804_PW_LT_hex
+805_PW_LT_trigonal
+806_PW_LT_st
+807_PW_LT_bct
+808_PW_LT_so
+809_PW_LT_baco
+810_PW_LT_fco
+811_PW_LT_bco
+812_PW_LT_sm
+813_PW_LT_bacm
+814_PW_LT_triclinic