fix GPU bug

Qianruipku · Qianruipku · commit 6ffb2a36c4aa · 2025-03-23T13:29:08.000+08:00
diff --git a/source/module_base/parallel_device.h b/source/module_base/parallel_device.h
@@ -165,7 +165,7 @@ void gatherv_dev(const T* sendbuf,
     T* sendbuf_cpu = o1.get(sendbuf, sendcount, tmp_sspace);
     T* recvbuf_cpu = o2.get(recvbuf, gather_space, tmp_rspace);
     o1.sync_d2h(sendbuf_cpu, sendbuf, sendcount);
-    gatherv_data(sendbuf_cpu, sendcount, recvbuf, recvcounts, displs, comm);
+    gatherv_data(sendbuf_cpu, sendcount, recvbuf_cpu, recvcounts, displs, comm);
     o2.sync_h2d(recvbuf, recvbuf_cpu, gather_space);
     o1.del(sendbuf_cpu);
     o2.del(recvbuf_cpu);
diff --git a/source/module_hamilt_pw/hamilt_pwdft/VNL_in_pw.cpp b/source/module_hamilt_pw/hamilt_pwdft/VNL_in_pw.cpp
@@ -17,7 +17,7 @@
 
 pseudopot_cell_vnl::pseudopot_cell_vnl()
 {
-    this->use_gpu_ = (this->use_gpu_);
+    this->use_gpu_ = (PARAM.inp.device == "gpu");
 }
 
 pseudopot_cell_vnl::~pseudopot_cell_vnl()
diff --git a/source/module_hamilt_pw/hamilt_pwdft/elecond.cpp b/source/module_hamilt_pw/hamilt_pwdft/elecond.cpp
@@ -155,15 +155,16 @@ void EleCond<FPTYPE, Device>::jjresponse_ks(const int ik,
                                                             nbands);
 
         std::complex<FPTYPE>* pij_c = nullptr;
+        std::vector<std::complex<FPTYPE>> pij_h_;
         if(std::is_same<Device, base_device::DEVICE_CPU>::value)
         {
             pij_c = pij_d;
         }
         else
         {
-            std::vector<std::complex<FPTYPE>> pij_h(nbands * nbands);
-            syncmem_complex_d2h_op()(pij_h.data(), pij_d, nbands * nbands);
-            pij_c = pij_h.data();
+            pij_h_.resize(nbands * nbands);
+            syncmem_complex_d2h_op()(pij_h_.data(), pij_d, nbands * nbands);
+            pij_c = pij_h_.data();
         }
         
 #ifdef __MPI
diff --git a/source/module_hamilt_pw/hamilt_stodft/sto_elecond.cpp b/source/module_hamilt_pw/hamilt_stodft/sto_elecond.cpp
@@ -200,12 +200,7 @@ void Sto_EleCond<FPTYPE, Device>::cal_jmatrix(hamilt::HamiltSdftPW<std::complex<
     {
         vec_rightf_all.resize(allbands_ks);
         rightf_all = vec_rightf_all.data();
-        Parallel_Common::gatherv_dev<std::complex<FPTYPE>, Device>(rightfact,
-                                                                   perbands_ks,
-                                                                   rightf_all,
-                                                                   ks_fact->nrecv,
-                                                                   ks_fact->displs,
-                                                                   BP_WORLD);
+        Parallel_Common::gatherv_data(rightfact, perbands_ks, rightf_all, ks_fact->nrecv, ks_fact->displs, BP_WORLD);
     }
 #endif
 
@@ -747,10 +742,7 @@ void Sto_EleCond<FPTYPE, Device>::sKG(const int& smear_type,
         {
             for (int ib = 0; ib < perbands_ks; ++ib)
             {
-                for (int ig = 0; ig < npw; ++ig)
-                {
-                    kspsi(0, ib, ig) = this->p_psi[0](ib0_ks + ib, ig);
-                }
+                cpymem_complex_op()(&kspsi(0, ib, 0), &this->p_psi[0](ib0_ks + ib, 0), npw);
                 FPTYPE fi = this->stofunc.fd(FPTYPE(en[ib]));
                 expmtmf_fact[ib] = 1 - fi;
                 expmtf_fact[ib] = fi;
diff --git a/source/module_hamilt_pw/hamilt_stodft/sto_elecond.h b/source/module_hamilt_pw/hamilt_stodft/sto_elecond.h
@@ -20,6 +20,7 @@ class Sto_EleCond : protected EleCond<FPTYPE, Device>
     using delmem_lcomplex_op = base_device::memory::delete_memory_op<std::complex<lowTYPE>, Device>;
     using cpymem_lcomplex_op = base_device::memory::synchronize_memory_op<std::complex<lowTYPE>, Device, Device>;
     using castmem_lcomplex_op = base_device::memory::cast_memory_op<std::complex<lowTYPE>, std::complex<FPTYPE>, Device, Device>;
+    using cpymem_complex_op = base_device::memory::synchronize_memory_op<std::complex<FPTYPE>, Device, Device>;
   public:
     Sto_EleCond(UnitCell* p_ucell_in,
                 K_Vectors* p_kv_in,
diff --git a/tests/integrate/186_PW_SKG_MALL_GPU/INPUT b/tests/integrate/186_PW_SKG_MALL_GPU/INPUT
@@ -0,0 +1,43 @@
+INPUT_PARAMETERS
+#Parameters (1.General)
+suffix			autotest
+calculation     scf
+esolver_type    sdft
+method_sto      1
+
+nbands			5
+nbands_sto		all
+nche_sto		120
+emax_sto		0
+emin_sto		0
+seed_sto        20000
+pseudo_dir	../../PP_ORB
+symmetry	    1
+kpar            1
+bndpar          2
+device          gpu
+
+#Parameters (2.Iteration)
+ecutwfc			20
+scf_thr			1e-6
+scf_nmax	    20
+
+
+#Parameters (3.Basis)
+basis_type		pw
+
+#Parameters (4.Smearing)
+smearing_method		fd
+smearing_sigma			0.6
+
+#Parameters (5.Mixing)
+mixing_type		broyden
+mixing_beta		0.4
+mixing_gg0      0.0
+
+cal_cond      1
+cond_fwhm   16
+cond_wcut        10
+cond_dw          1
+cond_dt          0.237464
+cond_nonlocal    1
diff --git a/tests/integrate/186_PW_SKG_MALL_GPU/KPT b/tests/integrate/186_PW_SKG_MALL_GPU/KPT
@@ -0,0 +1,4 @@
+K_POINTS
+0
+Gamma
+1 1 1 0 0 0
diff --git a/tests/integrate/186_PW_SKG_MALL_GPU/README b/tests/integrate/186_PW_SKG_MALL_GPU/README
@@ -0,0 +1,8 @@
+This test for:
+*MDFT
+*Al
+*kpoints 2*1*1
+*complete basis
+*kpar 1
+*bndpar 2
+*cal_cond 1
diff --git a/tests/integrate/186_PW_SKG_MALL_GPU/STRU b/tests/integrate/186_PW_SKG_MALL_GPU/STRU
@@ -0,0 +1,18 @@
+ATOMIC_SPECIES
+Si 14 Si.pz-vbc.UPF
+
+LATTICE_CONSTANT
+5  // add lattice constant
+
+LATTICE_VECTORS
+1 0 0
+0 1 0
+0 0 1
+
+ATOMIC_POSITIONS
+Direct 
+
+Si	// Element type	
+0.0	// magnetism
+1	
+0.00 0.00 0.00 1 1 1
diff --git a/tests/integrate/186_PW_SKG_MALL_GPU/refOnsager.txt b/tests/integrate/186_PW_SKG_MALL_GPU/refOnsager.txt
@@ -0,0 +1,11 @@
+## w(eV)         sigma(Sm^-1)     kappa(W(mK)^-1)        L12/e(Am^-1)      L22/e^2(Wm^-1)
+     0.5              145953             177.614        -3.70776e+06         1.11017e+08
+     1.5              144420             175.681        -3.67128e+06          1.0997e+08
+     2.5              141416             171.896        -3.59972e+06         1.07914e+08
+     3.5              137057             166.415        -3.49581e+06          1.0493e+08
+     4.5              131513             159.459        -3.36351e+06         1.01129e+08
+     5.5              124997             151.302        -3.20782e+06         9.66558e+07
+     6.5              117754             142.252        -3.03448e+06         9.16732e+07
+     7.5              110048             132.637        -2.84972e+06         8.63592e+07
+     8.5              102146             122.779        -2.65987e+06         8.08944e+07
+     9.5             94302.5             112.982        -2.47103e+06         7.54522e+07
diff --git a/tests/integrate/186_PW_SKG_MALL_GPU/result.ref b/tests/integrate/186_PW_SKG_MALL_GPU/result.ref
@@ -0,0 +1,7 @@
+etotref -150.093373857575
+etotperatomref -150.0933738576
+CompareH_Failed 0
+pointgroupref O_h
+spacegroupref O_h
+nksibzref 1
+totaltimeref 26.09
diff --git a/tests/integrate/CASES_GPU.txt b/tests/integrate/CASES_GPU.txt
@@ -3,6 +3,7 @@
 102_PW_BPCG_GPU
 107_PW_outWfcPw_GPU
 186_PW_KG_100_GPU
+186_PW_SKG_MALL_GPU
 187_PW_SDFT_ALL_GPU
 187_PW_SDFT_MALL_GPU
 187_PW_SDFT_MALL_BPCG_GPU

Original file line number	Diff line number	Diff line change
`@@ -17,7 +17,7 @@`
`17`	`17`
`18`	`18`	`pseudopot_cell_vnl::pseudopot_cell_vnl()`
`19`	`19`	`{`
`20`		`- this->use_gpu_ = (this->use_gpu_);`
	`20`	`+ this->use_gpu_ = (PARAM.inp.device == "gpu");`
`21`	`21`	`}`
`22`	`22`
`23`	`23`	`pseudopot_cell_vnl::~pseudopot_cell_vnl()`
Original file line number	Diff line number	Diff line change
`@@ -155,15 +155,16 @@ void EleCond<FPTYPE, Device>::jjresponse_ks(const int ik,`
`155`	`155`	`nbands);`
`156`	`156`
`157`	`157`	`std::complex<FPTYPE>* pij_c = nullptr;`
	`158`	`+ std::vector<std::complex<FPTYPE>> pij_h_;`
`158`	`159`	`if(std::is_same<Device, base_device::DEVICE_CPU>::value)`
`159`	`160`	`{`
`160`	`161`	`pij_c = pij_d;`
`161`	`162`	`}`
`162`	`163`	`else`
`163`	`164`	`{`
`164`		`- std::vector<std::complex<FPTYPE>> pij_h(nbands * nbands);`
`165`		`- syncmem_complex_d2h_op()(pij_h.data(), pij_d, nbands * nbands);`
`166`		`- pij_c = pij_h.data();`
	`165`	`+ pij_h_.resize(nbands * nbands);`
	`166`	`+ syncmem_complex_d2h_op()(pij_h_.data(), pij_d, nbands * nbands);`
	`167`	`+ pij_c = pij_h_.data();`
`167`	`168`	`}`
`168`	`169`
`169`	`170`	`#ifdef __MPI`
Original file line number	Diff line number	Diff line change
`@@ -200,12 +200,7 @@ void Sto_EleCond<FPTYPE, Device>::cal_jmatrix(hamilt::HamiltSdftPW<std::complex<`
`200`	`200`	`{`
`201`	`201`	`vec_rightf_all.resize(allbands_ks);`
`202`	`202`	`rightf_all = vec_rightf_all.data();`
`203`		`- Parallel_Common::gatherv_dev<std::complex<FPTYPE>, Device>(rightfact,`
`204`		`- perbands_ks,`
`205`		`- rightf_all,`
`206`		`- ks_fact->nrecv,`
`207`		`- ks_fact->displs,`
`208`		`- BP_WORLD);`
	`203`	`+ Parallel_Common::gatherv_data(rightfact, perbands_ks, rightf_all, ks_fact->nrecv, ks_fact->displs, BP_WORLD);`
`209`	`204`	`}`
`210`	`205`	`#endif`
`211`	`206`
`@@ -747,10 +742,7 @@ void Sto_EleCond<FPTYPE, Device>::sKG(const int& smear_type,`
`747`	`742`	`{`
`748`	`743`	`for (int ib = 0; ib < perbands_ks; ++ib)`
`749`	`744`	`{`
`750`		`- for (int ig = 0; ig < npw; ++ig)`
`751`		`- {`
`752`		`- kspsi(0, ib, ig) = this->p_psi[0](ib0_ks + ib, ig);`
`753`		`- }`
	`745`	`+ cpymem_complex_op()(&kspsi(0, ib, 0), &this->p_psi[0](ib0_ks + ib, 0), npw);`
`754`	`746`	`FPTYPE fi = this->stofunc.fd(FPTYPE(en[ib]));`
`755`	`747`	`expmtmf_fact[ib] = 1 - fi;`
`756`	`748`	`expmtf_fact[ib] = fi;`
-Original file line number
+Diff line change
@@ @@ -0,0 +1,4 @@ @@
 +K_POINTS
 +0
 +Gamma
 +1 1 1 0 0 0