[Refactor] Remove some dependencies in module_base (#7311)

* Remove useless include in math_ylm_op

* Remove incorrect include file in parallel_global.cpp

* Remove some parameter.h dependency

Author: Critsium-xy

source/source_base/kernels/math_ylm_op.h

@@ -1,7 +1,7 @@
 #ifndef MODULE_BASE_MATH_MULTI_DEVICE_H
 #define MODULE_BASE_MATH_MULTI_DEVICE_H
 
-#include "source_psi/psi.h"
+#include "source_base/module_device/types.h"
 #include <complex>
 
 namespace ModuleBase {

source/source_base/module_device/device.h

@@ -31,8 +31,10 @@ int get_device_num(std::string device_flag);
 /**
  * @brief Output the device information
  * for source_esolver
+ * @param output  output stream.
+ * @param device  device flag, "cpu" / "gpu" / "dsp".
  */
-void output_device_info(std::ostream& output);
+void output_device_info(std::ostream& output, const std::string& device);
 
 /**
  * @brief Safely probes for GPU availability without exiting on error.

source/source_base/module_device/output_device.cpp

@@ -3,7 +3,6 @@
 #include "gpu_runtime.h"
 
 #include "source_base/tool_quit.h"
-#include "source_io/module_parameter/parameter.h"
 
 #include <base/macros/macros.h>
 #include <cstring>
@@ -92,7 +91,7 @@ int get_device_num(std::string device_flag)
   return 0;
 }
 
-void output_device_info(std::ostream &output)
+void output_device_info(std::ostream &output, const std::string& device)
 {
 #ifdef __MPI
     int world_rank, world_size;
@@ -105,7 +104,7 @@ void output_device_info(std::ostream &output)
     // Get local hardware info
     int local_gpu_count = 0;
     #if defined(__CUDA) || defined(__ROCM)
-    if(PARAM.inp.device == "gpu" && local_rank == 0)
+    if(device == "gpu" && local_rank == 0)
     {
         local_gpu_count = get_device_num("gpu");
     }
@@ -129,7 +128,7 @@ void output_device_info(std::ostream &output)
         std::string cpu_name = get_device_name("cpu");
         std::string gpu_name;
         #if defined(__CUDA) || defined(__ROCM)
-        if(PARAM.inp.device == "gpu" && total_gpus > 0)
+        if(device == "gpu" && total_gpus > 0)
         {
             gpu_name = get_device_name("gpu");
         }
@@ -139,7 +138,7 @@ void output_device_info(std::ostream &output)
         output << " RUNNING WITH DEVICE  : " << "CPU" << " / "
                   << cpu_name << " (x" << total_cpus << ")" << std::endl;
         #if defined(__CUDA) || defined(__ROCM)
-        if(PARAM.inp.device == "gpu" && total_gpus > 0)
+        if(device == "gpu" && total_gpus > 0)
         {
             output << "                        " << "GPU" << " / "
                   << gpu_name << " (x" << total_gpus << ")" << std::endl;
@@ -152,7 +151,7 @@ void output_device_info(std::ostream &output)
         output << " RUNNING WITH DEVICE  : " << "CPU" << " / "
                    << cpu_name << " (x" << cpu_sockets << ")" << std::endl;
         #if defined(__CUDA) || defined(__ROCM)
-        if(PARAM.inp.device == "gpu")
+        if(device == "gpu")
         {
             int gpu_count = get_device_num("gpu");
             if(gpu_count > 0)

source/source_base/module_fft/fft_bundle.cpp

@@ -74,7 +74,9 @@ void FFT_Bundle::initfft(int nx_in,
         {
             ModuleBase::WARNING_QUIT("device", "now dsp fft is not supported for the float type");
         }
-        fft_double = make_unique<FFT_DSP<double>>();
+        auto dsp_fft = make_unique<FFT_DSP<double>>();
+        dsp_fft->cluster_id = this->dsp_cluster_id_;
+        fft_double = std::move(dsp_fft);
         fft_double->initfft(nx_in, ny_in, nz_in);
     }else
 #endif

source/source_base/module_fft/fft_bundle.h

@@ -32,6 +32,14 @@ class FFT_Bundle
      */
     void setfft(std::string device_in, std::string precision_in);
 
+    /**
+     * @brief Set the DSP cluster id for the FFT_DSP backend.
+     * @param id  cluster id, typically computed as (MPI rank % dsp_count).
+     *
+     * Caller-injected DSP routing info; only used when device == "dsp".
+     */
+    void set_dsp_cluster_id(int id) { this->dsp_cluster_id_ = id; }
+
     /**
      * @brief Initialize the fft parameters.
      * @param nx_in  number of grid points in x direction.
@@ -202,6 +210,7 @@ class FFT_Bundle
 
     std::string device = "cpu";
     std::string precision = "double";
+    int dsp_cluster_id_ = 0;
 };
 // Use RAII (Resource Acquisition Is Initialization) to 
 // control the resources used by hthread when setting the DSP

source/source_base/module_fft/fft_dsp.cpp

@@ -1,8 +1,6 @@
 #include "fft_dsp.h"
 
-#include "source_base/global_variable.h"
 #include "source_base/global_function.h"
-#include "source_io/module_parameter/parameter.h"
 
 #include <iostream>
 #include <string.h>
@@ -15,7 +13,6 @@ void FFT_DSP<double>::initfft(int nx_in, int ny_in, int nz_in)
     this->nx = nx_in;
     this->ny = ny_in;
     this->nz = nz_in;
-    cluster_id = GlobalV::MY_RANK % PARAM.inp.dsp_count;
     nxyz = this->nx * this->ny * this->nz;
 }
 template <>

source/source_base/parallel_global.cpp

@@ -14,7 +14,7 @@
 #include "source_base/global_function.h"
 #include "source_base/parallel_common.h"
 #include "source_base/parallel_reduce.h"
-#include "source_io/module_parameter/parameter.h"
+#include "source_base/global_variable.h"
 #include "source_base/tool_quit.h"
 #include "source_main/version.h"
 

source/source_basis/module_pw/pw_basis_k.cpp

@@ -1,6 +1,7 @@
 #include "pw_basis_k.h"
 
 #include "source_base/constants.h"
+#include "source_base/global_variable.h"
 #include "source_base/memory.h"
 #include "source_base/timer.h"
 #include "source_io/module_parameter/parameter.h"
@@ -215,6 +216,7 @@ void PW_Basis_K::setuptransform()
     std::string fft_device = this->device;
 #if defined(__DSP)
     fft_device = "dsp";
+    this->fft_bundle.set_dsp_cluster_id(GlobalV::MY_RANK % PARAM.inp.dsp_count);
 #endif
     this->fft_bundle.setfft(fft_device, this->precision);
     if (this->xprime)

source/source_esolver/esolver.cpp

@@ -115,8 +115,8 @@ std::string determine_type()
             c = std::toupper(c);
         }
     }
-    base_device::information::output_device_info(std::cout);
-    base_device::information::output_device_info(GlobalV::ofs_running);
+    base_device::information::output_device_info(std::cout, PARAM.inp.device);
+    base_device::information::output_device_info(GlobalV::ofs_running, PARAM.inp.device);
     /***auto end_time = std::chrono::high_resolution_clock::now();
     auto duration = std::chrono::duration_cast<std::chrono::duration<double>>(end_time - start_time);
     std::cout << "hipGetDeviceInfo took " << duration.count() << " seconds" << std::endl;***/