Merge branch 'develop' of github.com:deepmodeling/abacus-develop into HSolver

Author: dyzheng

CMakeLists.txt

@@ -20,6 +20,7 @@ option(USE_OPENMP " Enable OpenMP in abacus." ON)
 option(ENABLE_ASAN "Enable AddressSanitizer" OFF)
 option(BUILD_TESTING "Build ABACUS unit tests" OFF)
 option(GENERATE_TEST_REPORTS "Enable test report generation" OFF)
+option(INFO "Enable gathering of math library information" OFF)
 
 set(ABACUS_BIN_NAME abacus)
 list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/modules)
@@ -217,7 +218,7 @@ if(ENABLE_DEEPKS)
     HINTS ${libnpy_INCLUDE_DIR}
   )
   if(NOT libnpy_SOURCE_DIR)
-  include(FetchContent)
+    include(FetchContent)
     FetchContent_Declare(
       libnpy
       GIT_REPOSITORY https://github.com/llohse/libnpy.git
@@ -237,14 +238,22 @@ if(DEFINED Libxc_DIR)
   set(ENABLE_LIBXC ON)
 endif()
 if(ENABLE_LIBXC)
-  find_package(Libxc)
+  find_package(Libxc HINTS ${Libxc_DIR}/share/cmake/Libxc)
   if(${Libxc_FOUND})
     message("Using Libxc.")
     add_compile_definitions(USE_LIBXC)
     target_link_libraries(${ABACUS_BIN_NAME} Libxc::xc)
     include_directories(${Libxc_INCLUDE_DIRS})
   else()
-    message(WARNING "Will not use Libxc.")
+    include(FetchContent)
+    FetchContent_Declare(
+      Libxc
+      GIT_REPOSITORY https://gitlab.com/libxc/libxc.git
+      GIT_TAG "5.2.3"
+      GIT_SHALLOW TRUE
+      GIT_PROGRESS TRUE
+    )
+    FetchContent_MakeAvailable(Libxc)
   endif()
 endif()
 
@@ -261,6 +270,12 @@ add_compile_definitions(
   TEST_EXX_RADIAL=1
 )
 
+if(INFO)
+  message(STATUS "Will gather math lib info.")
+  add_compile_definitions(GATHER_INFO)
+  # modifications on blas_connector and lapack_connector
+endif()
+
 IF (BUILD_TESTING)
   set(CMAKE_CXX_STANDARD 14) # Required in orbital
   include(CTest)
@@ -319,7 +334,7 @@ target_link_libraries(${ABACUS_BIN_NAME}
     pw
     ri
     driver
-    xc
+    xc_
     hsolver
     elecstate
     hamilt

docs/input-main.md

@@ -597,9 +597,9 @@ This part of variables are used to control the parameters of stochastic DFT (SDF
 - **Description**:
   - Different method to do SDFT.
   - 1: SDFT calculates $T_n(\hat{h})\ket{\chi}$ twice, where $T_n(x)$ is the n-th order Chebyshev polynomial and $\hat{h}=\frac{\hat{H}-\bar{E}}{\Delta E}$ owning eigen-value $\in(-1,1)$. This method cost less memory but slow.
-  - 2: SDFT calculates $T_n(\hat{h})\ket{\chi}$ once but need much more memory. This method is fast but when memory is not enough, only method 1 can be used.
-  - other: use 1
-- **Default**: 1
+  - 2: SDFT calculates $T_n(\hat{h})\ket{\chi}$ once but need much more memory. This method is much faster. Besides, it calculate $N_e$ with $\bra{\chi}\sqrt{\hat f}\sqrt{\hat f}\ket{\chi}$, which needs smaller [nche_sto](#nche_sto). However, when memory is not enough, only method 1 can be used.
+  - other: use 2
+- **Default**: 2
 
 #### nbands_sto
 
@@ -642,6 +642,12 @@ This part of variables are used to control the parameters of stochastic DFT (SDF
 - **Description**: Frequency (once each initsto_freq steps) to generate new stochastic orbitals when running md.
 - **Default**:1000
 
+#### npart_sto
+
+- **Type**: Integer
+- **Description**: Make memory cost to 1/npart_sto times of previous one when running post process of SDFT like DOS with method_sto = 2.
+- **Default**:1
+
 ### Geometry relaxation
 
 This part of variables are used to control the geometry relaxation.

docs/install.md

@@ -34,14 +34,16 @@ To compile ABACUS, please make sure that the following prerequisites are present
 
 - C++ compiler, supporting C++11. You can use [Intel® C++ compiler](https://software.intel.com/enus/c-compilers) or [GCC](https://gcc.gnu.org/).
 - MPI compiler. The recommended version are [Intel MPI](https://software.intel.com/enus/mpi-library) or [MPICH](https://www.mpich.org/).
-- Fortran compiler for building `BLAS`, `LAPACK`, `ScaLAPACK` or `ELPA`. You can use[Intel® Fortran Compiler](https://www.intel.com/content/www/us/en/developer/tools/oneapi/fortran-compiler.html) [GFortran](https://gcc.gnu.org/fortran/).
+- Fortran compiler if you are building `BLAS`, `LAPACK`, `ScaLAPACK`, and `ELPA` from source file. You can use[Intel® Fortran Compiler](https://www.intel.com/content/www/us/en/developer/tools/oneapi/fortran-compiler.html) [GFortran](https://gcc.gnu.org/fortran/).
 - [BLAS](http://www.netlib.org/blas/). You can use [OpenBLAS](https://www.openblas.net/).
 - [LAPACK](http://www.netlib.org/lapack/).
 - [ScaLAPACK](http://www.netlib.org/scalapack/).
 - [FFTW3](http://www.fftw.org/).
 - [ELPA](https://elpa.mpcdf.mpg.de/) >= 2017.
 - [CEREAL](https://uscilab.github.io/cereal/).
 
+> GCC version 5.1 or later is required: Intel compilers also use GCC headers and libraries[(ref)](https://www.intel.com/content/www/us/en/develop/documentation/cpp-compiler-developer-guide-and-reference/top/compatibility-and-portability/gcc-compatibility-and-interoperability.html#gcc-compatibility-and-interoperability_GUID-52CB6FE0-83DA-4028-9EF4-0DFAF1652736). 
+
 These packages can be installed with popular package management system, such as `apt` and `yum`:
 
 ```bash
@@ -50,7 +52,7 @@ sudo apt update && sudo apt install -y libopenblas-dev liblapack-dev libscalapac
 
 > Installing ELPA by apt only matches requirements on Ubuntu 22.04. For earlier linux distributions, you may install elpa from source.
 
-Alternatively, you can choose [Intel® oneAPI toolkit](https://software.intel.com/content/www/us/en/develop/tools/oneapi/commercial-base-hpc.html) (former Parallel Studio) as toolchain. The [Intel® oneAPI Base Toolkit](https://software.intel.com/content/www/us/en/develop/tools/oneapi/all-toolkits.html#base-kit) contains Intel® oneAPI Math Kernel Library (aka `MKL`), including `BLAS`, `LAPACK`, `ScaLAPACK` and `FFTW3`,  - this means that no Fortran compiler required anymore. The [Intel® oneAPI HPC Toolkit](https://software.intel.com/content/www/us/en/develop/tools/oneapi/all-toolkits.html#hpc-kit) contains Intel® MPI Library, and C++ compiler(including MPI compiler). Please noted that building `elpa` with a different MPI library may cause conflict between MPI libraries. Don't forget to [set environment variables](https://software.intel.com/content/www/us/en/develop/documentation/get-started-with-intel-oneapi-render-linux/top/configure-your-system.html) before you start! `cmake` will use Intel MKL if the environment variable `MKLROOT` is set.
+Alternatively, you can choose [Intel® oneAPI toolkit](https://software.intel.com/content/www/us/en/develop/tools/oneapi/commercial-base-hpc.html) (former Parallel Studio) as toolchain. The [Intel® oneAPI Base Toolkit](https://software.intel.com/content/www/us/en/develop/tools/oneapi/all-toolkits.html#base-kit) contains Intel® oneAPI Math Kernel Library (aka `MKL`), including `BLAS`, `LAPACK`, `ScaLAPACK` and `FFTW3`. The [Intel® oneAPI HPC Toolkit](https://software.intel.com/content/www/us/en/develop/tools/oneapi/all-toolkits.html#hpc-kit) contains Intel® MPI Library, and C++ compiler(including MPI compiler). Please noted that building `elpa` with a different MPI library may cause conflict between MPI libraries. Don't forget to [set environment variables](https://software.intel.com/content/www/us/en/develop/documentation/get-started-with-intel-oneapi-render-linux/top/configure-your-system.html) before you start! `cmake` will use Intel MKL if the environment variable `MKLROOT` is set.
 
 > Please refer to our [guide](https://github.com/deepmodeling/abacus-develop/wiki/Building-and-Running-ABACUS) on requirements.
 
@@ -95,13 +97,13 @@ cmake -B build -DFFTW3_ROOT=/opt/fftw3
 
 If environment variable `MKLROOT` exists, `cmake` will take MKL as a preference, i.e. not using `LAPACK` and `ScaLAPACK`. To disable MKL, unset environment variable `MKLROOT`, or pass `-DMKLROOT=OFF` to `cmake`.
 
-You can also choose to build with which components.
+You can also choose to build with which components, e.g.:
 
 ```bash
 cmake -B build -DUSE_LIBXC=1 -DUSE_CUDA=1
 ```
 
-If Libxc is not installed in standard path (i.e. installed with a custom prefix path), you may add the installation prefix of `FindLibxc.cmake` to `CMAKE_MODULE_PATH` environment variable, or set `Libxc_DIR` to the directory containing the file.
+If Libxc is not installed in standard path (i.e. installed with a custom prefix path), you can set `Libxc_DIR` to the corresponding directory.
 
 ```bash
 cmake -B build -DLibxc_DIR=~/libxc
@@ -241,9 +243,9 @@ This part of installation is based on [Installation](#installation). If DeePKS f
 
 ### Extra prerequisites
 
-- C++ compiler, supporting **C++14**. For example, Intel C++ compiler 18
-- [LibTorch](https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-1.9.0%2Bcpu.zip) for cpu, with c++11 ABI;
-- [Libnpy](https://github.com.cnpmjs.org/llohse/libnpy/);
+- C++ compiler, supporting **C++14**
+- [LibTorch](https://pytorch.org/) with cxx11 ABI supporting CPU
+- [Libnpy](https://github.com/llohse/libnpy/)
 
 ### Extra settings for building
 

source/Makefile.Objects

@@ -245,6 +245,7 @@ OBJS_SURCHEM=H_correction_pw.o\
     corrected_energy.o\
     minimize_cg.o\
     efield.o\
+    sol_force.o\
 
 OBJS_XC=xc_funct_corr_gga.o \
 xc_funct_corr_lda.o \

source/input.cpp

@@ -144,7 +144,8 @@ void Input::Default(void)
     bndpar = 1;
     kpar = 1;
     initsto_freq = 1000;
-    method_sto = 1;
+    method_sto = 2;
+    npart_sto = 1;
     cal_cond = false;
     dos_nche = 100;
     cond_nche = 20;
@@ -407,6 +408,7 @@ void Input::Default(void)
     // test only
     //==========================================================
     test_just_neighbor = false;
+    test_skip_ewald = false;
 
     //==========================================================
     //    DFT+U     Xin Qu added on 2020-10-29
@@ -595,6 +597,10 @@ bool Input::Read(const std::string &fn)
         {
             read_value(ifs, method_sto);
         }
+        else if (strcmp("npart_sto", word) == 0)
+        {
+            read_value(ifs, npart_sto);
+        }
         else if (strcmp("cal_cond", word) == 0)
         {
             read_value(ifs, cal_cond);
@@ -1505,6 +1511,10 @@ bool Input::Read(const std::string &fn)
         {
             read_value(ifs, test_just_neighbor);
         }
+        else if (strcmp("test_skip_ewald", word) == 0)
+        {
+            read_value(ifs, test_skip_ewald);
+        }
         //--------------
         //----------------------------------------------------------------------------------
         //         Xin Qu added on 2020-10-29 for DFT+U
@@ -1940,6 +1950,10 @@ void Input::Default_2(void) // jiyy add 2019-08-04
     }
     if(calculation.substr(0,3) != "sto")    bndpar = 1;
     if(bndpar > GlobalV::NPROC) bndpar = GlobalV::NPROC;
+    if(method_sto != 1 && method_sto != 2) 
+    {
+        method_sto = 2;
+    }
 }
 #ifdef __MPI
 void Input::Bcast()
@@ -1974,6 +1988,7 @@ void Input::Bcast()
     Parallel_Common::bcast_double(emin_sto);
     Parallel_Common::bcast_int(initsto_freq);
     Parallel_Common::bcast_int(method_sto);
+    Parallel_Common::bcast_int(npart_sto);
     Parallel_Common::bcast_bool(cal_cond);
     Parallel_Common::bcast_int(cond_nche);
     Parallel_Common::bcast_double(cond_dw);
@@ -2203,6 +2218,7 @@ void Input::Bcast()
     Parallel_Common::bcast_int(td_vextout);
     Parallel_Common::bcast_int(td_dipoleout);
     Parallel_Common::bcast_bool(test_just_neighbor);
+    Parallel_Common::bcast_bool(test_skip_ewald);
     Parallel_Common::bcast_int(GlobalV::ocp);
     Parallel_Common::bcast_string(GlobalV::ocp_set);
     Parallel_Common::bcast_int(out_mul); // qifeng add 2019/9/10

source/input.h

@@ -72,6 +72,7 @@ class Input
     int bndpar; //parallel for stochastic/deterministic bands
     int initsto_freq; //frequency to init stochastic orbitals when running md
     int method_sto; //different methods for sdft, 1: slow, less memory  2: fast, more memory
+    int npart_sto; //for method_sto = 2, reduce memory
     bool cal_cond; //calculate electronic conductivities
     int cond_nche; //orders of Chebyshev expansions for conductivities
     double cond_dw; //d\omega for conductivities
@@ -421,7 +422,8 @@ class Input
     //==========================================================
     // variables for test only
     //==========================================================
-    bool test_just_neighbor;
+    bool test_just_neighbor = false;
+    bool test_skip_ewald = false;
 
   private:
     //==========================================================

source/input_conv.cpp

@@ -135,6 +135,7 @@ void Input_Conv::Convert(void)
     GlobalV::VION_IN_H = INPUT.vion_in_h;
     GlobalV::TEST_FORCE = INPUT.test_force;
     GlobalV::TEST_STRESS = INPUT.test_stress;
+    GlobalV::test_skip_ewald = INPUT.test_skip_ewald;
 
     //----------------------------------------------------------
     // iteration (1/3)

source/module_base/CMakeLists.txt

@@ -7,6 +7,7 @@ add_library(
     export.cpp
     integral.cpp
     inverse_matrix.cpp
+    gather_math_lib_info.cpp
     global_file.cpp
     global_function.cpp
     global_function_ddotreal.cpp
@@ -16,7 +17,7 @@ add_library(
     math_polyint.cpp
     math_sphbes.cpp
     math_ylmreal.cpp
-	math_bspline.cpp
+    math_bspline.cpp
     math_chebyshev.cpp
     mathzone.cpp
     mathzone_add1.cpp