add omp_sz_hacker_ForLargeSystems

.github/workflows/main.yml

@@ -1,6 +1,10 @@
 name: CI
 
-on: [push]
+on:
+  push:
+  pull_request:
+  schedule:
+    - cron: '0 0 1,15 * *' # JST 9:00 on 1st and 15th every month
 
 jobs:
   ctest:
@@ -15,9 +19,6 @@ jobs:
           - os: ubuntu-24.04
             mpisize: 1
             ompsize: 3
-          - os: macos-13
-            mpisize: 1
-            ompsize: 1
           - os: macos-latest
             mpisize: 1
             ompsize: 1
@@ -58,7 +59,7 @@ jobs:
         shell: bash
         run: |
           if [ ${{ runner.os }} = "macOS" ] ; then
-            export FC=gfortran-12
+            export FC=gfortran-15
           fi
           cmake -DCMAKE_VERBOSE_MAKEFILE=ON $GITHUB_WORKSPACE
 

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 2.8.0 FATAL_ERROR)
+cmake_minimum_required(VERSION 3.0...3.10)
 enable_testing()
 
 project(HPhi NONE)
@@ -24,7 +24,7 @@ enable_language(C Fortran)
 #set(CMAKE_C_FLAGS "-lscalapack -lgfortran -lm")
 set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/lib")
 set(CMAKE_SKIP_BUILD_RPATH FALSE)
-set(CMAKE_BUILD_WITH_INSTALL_RPATH FALSE) 
+set(CMAKE_BUILD_WITH_INSTALL_RPATH FALSE)
 set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)
 set(CMAKE_MACOSX_RPATH 1)
 
@@ -56,9 +56,9 @@ endif()
 
 if(CMAKE_C_COMPILER_ID STREQUAL "Intel")
   if("${CMAKE_C_COMPILER_VERSION}" VERSION_LESS "15.0.0.20140528")
-     set(OMP_FLAG_Intel "-openmp")
+    set(OMP_FLAG_Intel "-openmp")
   else()
-     set(OMP_FLAG_Intel "-qopenmp")
+    set(OMP_FLAG_Intel "-qopenmp")
   endif()
   set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OMP_FLAG_Intel}")
 else()

doc/en/source/filespecification/expertmode_en/ModPara_file_en.rst

@@ -153,11 +153,22 @@ Common parameters
 
    **Type :** Int (positive integer)
 
-   **Description :** If ``CalcHS=1``, an efficient algorithm for generating
-   the restricted Hilbert space with the specified quantum number is
-   used (Details of algorithm is shown in 
-   http://www.pasums.issp.u-tokyo.ac.jp/wp-content/themes/HPhi/media/develop/tips.pdf [in Japanese]). Default
-   value is 1 and the efficient algorithm is used.
+   **Description :** Selects the algorithm used to enumerate the
+   restricted Hilbert space with the specified quantum number.
+   If ``CalcHS=1``, an efficient algorithm based on bit manipulation
+   is used (details are given in
+   http://www.pasums.issp.u-tokyo.ac.jp/wp-content/themes/HPhi/media/develop/tips.pdf [in Japanese]).
+   If ``CalcHS=2``, an alternative enumeration path is taken that can
+   be useful for some large-sector cases by avoiding the full
+   ``sdim = 2^Nsite`` outer loop of ``CalcHS=1``; note, however, that
+   the current implementation still allocates the auxiliary index
+   arrays (``list_2_1``, ``list_2_2``, ``list_jb``) at size ``O(sdim)``
+   and uses an additional ``O(idim_max)`` temporary sort buffer for
+   the Hubbard branch, so it is not a memory-only large-system fix.
+   ``CalcHS=2`` is currently available only for the ``Hubbard`` and
+   ``HubbardNConserved`` models; passing it with any other model
+   (e.g. ``Kondo``, ``Spin``) results in an explicit error. Default
+   value is 1.
 
 
 

doc/ja/source/filespecification/expertmode_ja/ModPara_file_ja.rst

@@ -137,8 +137,16 @@ ModParaファイル
    **形式 :** int型 (自然数)
 
    **説明 :**
-   ``CalcHS=1``\ で量子数を指定したときのヒルベルト空間生成が高速化されます(詳細は http://www.pasums.issp.u-tokyo.ac.jp/wp-content/themes/HPhi/media/develop/tips.pdfを参照)。
-   デフォルトの値は1で、高速化アルゴリズムが使用されます。
+   ``CalcHS=1``\ ではビット操作による高速化アルゴリズムを使用します
+   (詳細は http://www.pasums.issp.u-tokyo.ac.jp/wp-content/themes/HPhi/media/develop/tips.pdf を参照)。
+   ``CalcHS=2``\ は ``CalcHS=1`` の ``sdim = 2^Nsite`` 外側ループを回避する
+   代替の列挙経路で、一部の大きな sector で有効です。ただし現在の実装では
+   補助の index 配列 (``list_2_1`` / ``list_2_2`` / ``list_jb``) は依然 ``O(sdim)`` で
+   確保され、Hubbard 分岐では ``O(idim_max)`` の一時 sort buffer も使用するため、
+   メモリ面だけで大規模系を解決するものではありません。
+   ``CalcHS=2``\ は現時点で ``Hubbard`` と ``HubbardNConserved`` モデルでのみ利用可能で、
+   それ以外のモデル (例: ``Kondo``, ``Spin``) で指定した場合はエラーになります。
+   デフォルト値は 1 です。
 
 
 

src/CMakeLists.txt

@@ -1,7 +1,6 @@
 # include guard
-cmake_minimum_required(VERSION 2.8)
 if(${CMAKE_PROJECT_NAME} STREQUAL "Project")
-  message(FATAL_ERROR "cmake should be executed not for 'src' subdirectory, but for the top directory of HPhi.")
+  message(FATAL_ERROR "cmake should not be executed directly in subdirectory.")
 endif(${CMAKE_PROJECT_NAME} STREQUAL "Project")
 
 add_subdirectory(komega)

src/include/sz.h

@@ -135,6 +135,26 @@ int sz(
     long unsigned int *list_2_2_
 );
 
+unsigned long int sz_hacker_for_large_systems(
+                        long unsigned int ihfbit,
+                        struct BindStruct *X,
+                        long unsigned int *list_1_,
+                        long unsigned int *list_2_1_,
+                        long unsigned int *list_2_2_,
+                        long unsigned int *list_jb_
+);
+
+long unsigned int make_true_spin(long unsigned int spin, int shift_offset, int Nsite);
+
+void update_lists(
+    long unsigned int i, long unsigned int ihfbit,
+    long unsigned int *list_1_, long unsigned int *list_2_1_, long unsigned int *list_2_2_,
+    long unsigned int *ja, long unsigned int *jb,
+    long unsigned int *prev_ib, long unsigned int Nsite
+);
+
+int compare_ulong(const void *a, const void *b);
+
 int Read_sz(
     struct BindStruct *X,
     const long unsigned int irght,

src/komega/CMakeLists.txt

@@ -1,7 +1,6 @@
 # include guard
-cmake_minimum_required(VERSION 2.8)
 if(${CMAKE_PROJECT_NAME} STREQUAL "Project")
-  message(FATAL_ERROR "cmake should be executed not for 'src' subdirectory, but for the top directory of HPhi.")
+  message(FATAL_ERROR "cmake should not be executed directly in subdirectory.")
 endif(${CMAKE_PROJECT_NAME} STREQUAL "Project")
 
 add_definitions(-D__NO_ZDOT)