Merge branch 'develop' into fft14

Author: mohanchen

.github/workflows/test.yml

@@ -48,9 +48,16 @@ jobs:
         run: |
           cmake --build build -j8
           cmake --install build
-      - name: Test
+      - name: Unit Test
         env:
           GTEST_COLOR: 'yes'
           OMP_NUM_THREADS: '2'
         run: |
-          cmake --build build --target test ARGS="-V --timeout 1700"
+          cmake --build build --target test ARGS="-V --timeout 1700 -E integrated_test"
+      - name: Integrated Test
+        env:
+          GTEST_COLOR: 'yes'
+          OMP_NUM_THREADS: '2'
+        run: |
+          cmake --build build --target test ARGS="-V --timeout 1700 -R integrated_test"
+

CMakeLists.txt

@@ -798,4 +798,4 @@ install(PROGRAMS ${ABACUS_BIN_PATH}
 
 if(ENABLE_COVERAGE)
   coverage_evaluate()
-endif()
+endif()

docs/advanced/input_files/input-main.md

@@ -1645,10 +1645,13 @@ These variables are used to control the output of properties.
 ### out_dm
 
 - **Type**: Boolean
-- **Availability**: Numerical atomic orbital basis (gamma-only algorithm)
+- **Availability**: Numerical atomic orbital basis
 - **Description**: Whether to output the density matrix of localized orbitals into files in the folder `OUT.${suffix}`. The files are named as:
-  - nspin = 1: SPIN1_DM;
-  - nspin = 2: SPIN1_DM, and SPIN2_DM.
+  - For gamma only case:
+    - nspin = 1: SPIN1_DM;
+    - nspin = 2: SPIN1_DM, and SPIN2_DM.
+  - For multi-k points case:
+    - SPIN\*_K\*_DM, where \* stands for index of spin and kpoints;
 - **Default**: False
 
 ### out_dm1

source/Makefile.Objects

@@ -126,7 +126,9 @@ OBJS_MAIN=main.o\
 
 OBJS_BASE=abfs-vector3_order.o\
     assoc_laguerre.o\
-    blas_connector.o\
+    blas_connector_base.o\
+    blas_connector_vector.o\
+    blas_connector_matrix.o\
     complexarray.o\
     complexmatrix.o\
     clebsch_gordan_coeff.o\

source/module_base/CMakeLists.txt

@@ -10,7 +10,9 @@ add_library(
     base
     OBJECT
     assoc_laguerre.cpp
-    blas_connector.cpp
+    blas_connector_base.cpp
+    blas_connector_vector.cpp
+    blas_connector_matrix.cpp
     clebsch_gordan_coeff.cpp
     complexarray.cpp
     complexmatrix.cpp

source/module_base/blas_connector.h

@@ -368,9 +368,26 @@ class BlasConnector
 
 #ifdef __CUDA
 
+#include <cuda_runtime.h>
+#include "cublas_v2.h"
+
+// If you want to use cublas, you need these functions to create and destroy the cublas/hipblas handle.
+// You also need to use these functions to translate the transpose parameter into cublas/hipblas datatype.
+
 namespace BlasUtils{
-	void createGpuBlasHandle();
-	void destoryBLAShandle();
+
+	static cublasHandle_t cublas_handle = nullptr;
+
+	void createGpuBlasHandle(); // Create a cublas/hipblas handle.
+
+	void destoryBLAShandle(); // Destroy the cublas/hipblas handle. Do this when the software is about to end.
+
+	cublasOperation_t judge_trans(bool is_complex, const char& trans, const char* name); // Translate a normal transpose parameter to a cublas/hipblas type.
+
+	cublasSideMode_t judge_side(const char& trans); // Translate a normal side parameter to a cublas/hipblas type.
+
+	cublasFillMode_t judge_fill(const char& trans); // Translate a normal fill parameter to a cublas/hipblas type.
+
 }
 
 #endif

source/module_base/blas_connector_base.cpp

@@ -0,0 +1,77 @@
+#include "blas_connector.h"
+#include "macros.h"
+
+#ifdef __CUDA
+#include <base/macros/macros.h>
+#include <cuda_runtime.h>
+#include "cublas_v2.h"
+#include "module_base/kernels/math_kernel_op.h"
+#include "module_base/module_device/memory_op.h"
+
+
+namespace BlasUtils{
+
+	void createGpuBlasHandle(){
+		if (cublas_handle == nullptr) {
+			cublasErrcheck(cublasCreate(&cublas_handle));
+		}
+	}
+
+	void destoryBLAShandle(){
+		if (cublas_handle != nullptr) {
+			cublasErrcheck(cublasDestroy(cublas_handle));
+			cublas_handle = nullptr;
+		}
+	}
+
+
+	cublasOperation_t judge_trans(bool is_complex, const char& trans, const char* name)
+	{
+		if (trans == 'N')
+		{
+			return CUBLAS_OP_N;
+		}
+		else if(trans == 'T')
+		{
+			return CUBLAS_OP_T;
+		}
+		else if(is_complex && trans == 'C')
+		{
+			return CUBLAS_OP_C;
+		}
+		return CUBLAS_OP_N;
+	}
+
+	cublasSideMode_t judge_side(const char& trans)
+	{
+		if (trans == 'L')
+		{
+			return CUBLAS_SIDE_LEFT;
+		}
+		else if (trans == 'R')
+		{
+			return CUBLAS_SIDE_RIGHT;
+		}
+		return CUBLAS_SIDE_LEFT;
+	}
+
+	cublasFillMode_t judge_fill(const char& trans)
+	{
+		if (trans == 'F')
+		{
+			return CUBLAS_FILL_MODE_FULL;
+		}
+		else if (trans == 'U')
+		{
+			return CUBLAS_FILL_MODE_UPPER;
+		}
+		else if (trans == 'D')
+		{
+			return CUBLAS_FILL_MODE_LOWER;
+		}
+		return CUBLAS_FILL_MODE_FULL;
+	}
+
+} // namespace BlasUtils
+
+#endif