change teh cmake file

Author: A-006

CMakeLists.txt

@@ -257,9 +257,8 @@ target_link_libraries(${ABACUS_BIN_NAME} ${SCALAPACK_LIBRARY_DIR})
 
 if (USE_DSP)
   add_compile_definitions(__DSP)
-  target_link_libraries(${ABACUS_BIN_NAME} ${MTBLAS_FFT_LIBRARY_DIR})
   target_link_libraries(${ABACUS_BIN_NAME} ${OMPI_LIBRARY1})
-  include_directories(${MTBLAS_FFT_LIBRARY_DIR}/include)
+  include_directories(${MTBLAS_FFT_DIR}/libmtblas/include)
   include_directories(${MT_HOST_DIR}/include)
   target_link_libraries(${ABACUS_BIN_NAME} ${MT_HOST_DIR}/hthreads/lib/libhthread_device.a)
   target_link_libraries(${ABACUS_BIN_NAME} ${MT_HOST_DIR}/hthreads/lib/libhthread_host.a)

source/module_base/CMakeLists.txt

@@ -66,8 +66,8 @@ add_library(
 
 target_link_libraries(base PUBLIC container)
 if (USE_DSP)
-  target_link_libraries(base PUBLIC ${MTBLAS_FFT_LIBRARY_DIR}/lib/libmtblas.a)
-  target_link_libraries(base PUBLIC ${MTBLAS_FFT_LIBRARY_DIR}/lib/libmtblasdev.a)
+  target_link_libraries(base PUBLIC ${MTBLAS_FFT_DIR}/libmtblas/lib/libmtblas.a)
+  target_link_libraries(base PUBLIC ${MTBLAS_FFT_DIR}/libmtblas/lib/libmtblasdev.a)
 endif()
 add_subdirectory(module_container)
 

source/module_basis/module_pw/CMakeLists.txt

@@ -44,7 +44,9 @@ add_library(
 
 if (USE_DSP)
 target_link_libraries(planewave  PRIVATE
-${MTBLAS_FFT_LIBRARY_DIR}/lib/libmtfft.a)
+${MTBLAS_FFT_DIR}/libmtblas/lib/libmtfft.a)
+target_compile_definitions( planewave PUBLIC 
+FFT_DAT_DIR="${MTBLAS_FFT_DIR}/datfile/mt_fft_blas.dat") 
 endif()
 if(ENABLE_COVERAGE)
   add_coverage(planewave)

source/module_basis/module_pw/module_fft/fft_dsp.cpp

@@ -23,8 +23,7 @@ void FFT_DSP<double>::setupFFT()
     PLAN* ptr_plan_backward;
     INT   num_thread=8;
     INT   size;
-
-    hthread_dat_load(cluster_id, "/vol8/home/dptech_zyz1/develop/blasfft/mtfftblas/datfile/mt_fft_blas.dat");
+    hthread_dat_load(cluster_id, FFT_DAT_DIR);
 
     //compute the size of and malloc thread
     size = nx*ny*nz*2*sizeof(E);

source/module_basis/module_pw/pw_basis.cpp

@@ -15,7 +15,7 @@ PW_Basis::PW_Basis()
 
 PW_Basis::PW_Basis(std::string device_, std::string precision_) : device(std::move(device_)), precision(std::move(precision_)) {
     classname="PW_Basis";
-    this->fft_bundle.setfft("cpu",this->precision);
+    this->fft_bundle.setfft(this->device,this->precision);
 }
 
 PW_Basis:: ~PW_Basis()

source/module_basis/module_pw/pw_basis_k.cpp

@@ -22,9 +22,6 @@ PW_Basis_K::~PW_Basis_K()
     delete[] igl2isz_k;
     delete[] igl2ig_k;
     delete[] gk2;
-#if defined(__DSP)
-    delete[] ig2ixyz_k_cpu;
-#endif
 #if defined(__CUDA) || defined(__ROCM)
     if (this->device == "gpu") {
         if (this->precision == "single") {
@@ -148,7 +145,7 @@ void PW_Basis_K::setupIndGk()
 
     //get igl2isz_k and igl2ig_k
     if(this->npwk_max <= 0) { return;}
-    
+
     delete[] igl2isz_k; this->igl2isz_k = new int [this->nks * this->npwk_max];
     delete[] igl2ig_k; this->igl2ig_k = new int [this->nks * this->npwk_max];
     for (int ik = 0; ik < this->nks; ik++)
@@ -188,7 +185,11 @@ void PW_Basis_K::setuptransform()
     this->getstartgr();
     this->setupIndGk();
     this->fft_bundle.clear();
-    this->fft_bundle.setfft("dsp",this->precision);
+    #if defined(__DSP)
+        this->fft_bundle.setfft("dsp",this->precision);
+    #else
+        this->fft_bundle.setfft("cpu",this->precision);
+    #endif
     if(this->xprime){
         this->fft_bundle.initfft(this->nx,this->ny,this->nz,this->lix,this->rix,this->nst,this->nplane,this->poolnproc,this->gamma_only, this->xprime);
     }else{     
@@ -337,12 +338,12 @@ int& PW_Basis_K::getigl2ig(const int ik, const int igl) const
 
 void PW_Basis_K::get_ig2ixyz_k()
 {
-    if (this->device != "gpu")
-    {
-        //only GPU need to get ig2ixyz_k
-        return;
-    }
-    int * ig2ixyz_k_cpu = new int [this->npwk_max * this->nks];
+    // if (this->device != "gpu")
+    // {
+    //     //only GPU need to get ig2ixyz_k
+    //     return;
+    // }
+    ig2ixyz_k_cpu.resize(this->npwk_max * this->nks);
     ModuleBase::Memory::record("PW_B_K::ig2ixyz", sizeof(int) * this->npwk_max * this->nks);
     assert(gamma_only == false); //We only finish non-gamma_only fft on GPU temperarily.
     for(int ik = 0; ik < this->nks; ++ik)
@@ -359,10 +360,7 @@ void PW_Basis_K::get_ig2ixyz_k()
         }
     }
     resmem_int_op()(ig2ixyz_k, this->npwk_max * this->nks);
-    syncmem_int_h2d_op()(this->ig2ixyz_k, ig2ixyz_k_cpu, this->npwk_max * this->nks);
-    #if not defined (__DSP)
-        delete[] this->ig2ixyz_k_cpu;
-    #endif
+    syncmem_int_h2d_op()(this->ig2ixyz_k, ig2ixyz_k_cpu.data(), this->npwk_max * this->nks);
 }
 
 std::vector<int> PW_Basis_K::get_ig2ix(const int ik) const

source/module_basis/module_pw/pw_basis_k.h

@@ -87,7 +87,7 @@ class PW_Basis_K : public PW_Basis
     int *igl2isz_k=nullptr, * d_igl2isz_k = nullptr; //[npwk_max*nks] map (igl,ik) to (is,iz)
     int *igl2ig_k=nullptr;//[npwk_max*nks] map (igl,ik) to ig
     int *ig2ixyz_k=nullptr; ///< [npw] map ig to ixyz
-    int *ig2ixyz_k_cpu = nullptr; /// [npw] map ig to ixyz,which is used in dsp fft.
+    std::vector<int> ig2ixyz_k_cpu; /// [npw] map ig to ixyz,which is used in dsp fft.
     double *gk2=nullptr; // modulus (G+K)^2 of G vectors [npwk_max*nks]
 
     // liuyu add 2023-09-06

source/module_basis/module_pw/pw_transform_k.cpp

@@ -308,7 +308,6 @@ void PW_Basis_K::real_to_recip(const base_device::DEVICE_CPU* /*dev*/,
                                const double factor) const
 {
     #if defined(__DSP)
-        printf("beforce the real_to_recip\n");
         this->real2recip_dsp(in,out,ik,add,factor);
     #else
         this->real2recip(in, out, ik, add, factor);
@@ -334,7 +333,6 @@ void PW_Basis_K::recip_to_real(const base_device::DEVICE_CPU* /*dev*/,
                                const double factor) const
 {
     #if defined(__DSP)
-        printf("beforce the recip_to_real\n");
         this->recip2real_dsp(in,out,ik,add,factor);
     #else
         this->recip2real(in, out, ik, add, factor);

source/module_basis/module_pw/pw_transform_k_dsp.cpp

@@ -35,7 +35,7 @@ namespace ModulePW
                                                             this->nxyz,
                                                             add,
                                                             factor,
-                                                            this->ig2ixyz_k_cpu + startig,
+                                                            this->ig2ixyz_k_cpu.data() + startig,
                                                             auxr,
                                                             out);
     }
@@ -49,23 +49,19 @@ namespace ModulePW
         assert(this->gamma_only == false);
         const base_device::DEVICE_CPU* ctx;
         const base_device::DEVICE_GPU* gpux;
-        printf("beforce the recip2real_dsp\n");
         // memset the auxr of 0 in the auxr,here the len of the auxr is nxyz
         auto * auxr = this->fft_bundle.get_auxr_3d_data<double>();
         memset(auxr,0,this->nxyz*2*8);
 
         const int startig = ik * this->npwk_max;
         const int npw_k   = this->npwk[ik];
-        printf("beforce the set_3d_fft_box_op\n");
         //copy the mapping form the type of stick to the 3dfft
         set_3d_fft_box_op<double,base_device::DEVICE_CPU>()
         (
-            ctx,npw_k,this->ig2ixyz_k_cpu+startig,in,auxr
+            ctx,npw_k,this->ig2ixyz_k_cpu.data()+startig,in,auxr
         );
-        printf("beforce the fft3D_backward\n");
         // use 3d fft backward
         this->fft_bundle.fft3D_backward(gpux,auxr,auxr);
-        printf("beforce the add\n");
         if(add) 
         {
             const int one =1;
@@ -76,7 +72,6 @@ namespace ModulePW
         {
             memcpy(out,auxr,nrxx*2*8);
         }
-        printf("after the add\n");
     }
     template <>
     void PW_Basis_K::convolution(const base_device::DEVICE_CPU* ctx,
@@ -114,7 +109,7 @@ namespace ModulePW
         //copy the mapping form the type of stick to the 3dfft
         set_3d_fft_box_op<double,base_device::DEVICE_CPU>()
         (
-            ctx,npw_k,this->ig2ixyz_k_cpu+startig,input,auxr
+            ctx,npw_k,this->ig2ixyz_k_cpu.data()+startig,input,auxr
         );
 
         // use 3d fft backward
@@ -135,7 +130,7 @@ namespace ModulePW
                                                             this->nxyz,
                                                             add,
                                                             factor,
-                                                            this->ig2ixyz_k_cpu + startig,
+                                                            this->ig2ixyz_k_cpu.data() + startig,
                                                             auxr,
                                                             output);
         ModuleBase::timer::tick(this->classname,"convolution");