add convolution for nspin_4

Author: A-006

source/source_basis/module_pw/module_fft/fft_cuda.cpp

@@ -16,13 +16,13 @@ template <>
 void FFT_CUDA<float>::setupFFT()
 {
     cufftPlan3d(&c_handle, this->nx, this->ny, this->nz, CUFFT_C2C);
-    resmem_cd_op()(this->c_auxr_3d, this->nx * this->ny * this->nz);
+    resmem_cd_op()(this->c_auxr_3d, 2*this->nx * this->ny * this->nz);
 }
 template <>
 void FFT_CUDA<double>::setupFFT()
 {
     cufftPlan3d(&z_handle, this->nx, this->ny, this->nz, CUFFT_Z2Z);
-    resmem_zd_op()(this->z_auxr_3d, this->nx * this->ny * this->nz);
+    resmem_zd_op()(this->z_auxr_3d, 2*this->nx * this->ny * this->nz);
 }
 template <>
 void FFT_CUDA<float>::cleanFFT()

source/source_basis/module_pw/pw_transform_convolution.cpp

@@ -189,14 +189,18 @@ void PW_Basis_K::convolution_gpu(const int ik,
     assert(this->gamma_only == false);
     base_device::DEVICE_GPU* gpu_ctx;
     // memset the auxr of 0 in the auxr,here the len of the auxr is nxyz
-    base_device::memory::set_memory_op<std::complex<FPTYPE>, base_device::DEVICE_GPU>()(
-        tmp,
-        0,
-        2*this->nxyz);
     const int startig = ik * this->npwk_max;
     const int npw_k = this->npwk[ik];
-    auto *auxg = tmp;
-    auto *auxg1 = &tmp[size];
+    auto *auxg = this->fft_bundle.get_auxr_3d_data<FPTYPE>();
+    auto *auxg1 = &this->fft_bundle.get_auxr_3d_data<FPTYPE>()[size];
+    base_device::memory::set_memory_op<std::complex<FPTYPE>, base_device::DEVICE_GPU>()(
+        auxg,
+        0,
+        this->nxyz);
+    base_device::memory::set_memory_op<std::complex<FPTYPE>, base_device::DEVICE_GPU>()(
+        auxg1,
+        0,
+        this->nxyz);
     set_3d_fft_box_op<FPTYPE, base_device::DEVICE_GPU>()(npw_k,
                                                          this->ig2ixyz_k + startig,
                                                          input,
@@ -215,14 +219,14 @@ void PW_Basis_K::convolution_gpu(const int ik,
     // use 3d fft backward
     set_real_to_recip_output_op<FPTYPE, base_device::DEVICE_GPU>()(npw_k,
                                                                    this->nxyz,
-                                                                   add,
+                                                                   true,
                                                                    factor,
                                                                    this->ig2ixyz_k + startig,
                                                                    auxg,
                                                                    output);
     set_real_to_recip_output_op<FPTYPE, base_device::DEVICE_GPU>()(npw_k,
                                                                    this->nxyz,
-                                                                   add,
+                                                                   true,
                                                                    factor,
                                                                    this->ig2ixyz_k + startig,
                                                                    auxg1,

source/source_basis/module_pw/pw_transform_k.cpp

@@ -2,7 +2,6 @@
 #include "source_basis/module_pw/kernels/pw_op.h"
 #include "pw_basis_k.h"
 #include "pw_gatherscatter.h"
-#include "source_pw/module_pwdft/kernels/veff_op.h"
 namespace ModulePW
 {