Merge pull request lammps#4009 from ndtrung81/gpu-updates-Dec23

GPU package updates Dec 2023
ndtrung81 · Dec 13, 2023 · 0c4a1cb · 0c4a1cb
2 parents 98feaa0 + 569c23a
commit 0c4a1cb
Show file tree

Hide file tree

Showing 71 changed files with 9,239 additions and 44 deletions.
diff --git a/cmake/Modules/Packages/GPU.cmake b/cmake/Modules/Packages/GPU.cmake
@@ -151,10 +151,10 @@ if(GPU_API STREQUAL "CUDA")
   endif()
 
   cuda_compile_fatbin(GPU_GEN_OBJS ${GPU_LIB_CU} OPTIONS ${CUDA_REQUEST_PIC}
-          -DUNIX -O3 --use_fast_math -Wno-deprecated-gpu-targets -DNV_KERNEL -DUCL_CUDADR ${GPU_CUDA_GENCODE} -D_${GPU_PREC_SETTING} -DLAMMPS_${LAMMPS_SIZES})
+          -DUNIX -O3 --use_fast_math -Wno-deprecated-gpu-targets -allow-unsupported-compiler -DNV_KERNEL -DUCL_CUDADR ${GPU_CUDA_GENCODE} -D_${GPU_PREC_SETTING} -DLAMMPS_${LAMMPS_SIZES})
 
   cuda_compile(GPU_OBJS ${GPU_LIB_CUDPP_CU} OPTIONS ${CUDA_REQUEST_PIC}
-          -DUNIX -O3 --use_fast_math -Wno-deprecated-gpu-targets -DUCL_CUDADR ${GPU_CUDA_GENCODE} -D_${GPU_PREC_SETTING} -DLAMMPS_${LAMMPS_SIZES})
+          -DUNIX -O3 --use_fast_math -Wno-deprecated-gpu-targets -allow-unsupported-compiler -DUCL_CUDADR ${GPU_CUDA_GENCODE} -D_${GPU_PREC_SETTING} -DLAMMPS_${LAMMPS_SIZES})
 
   foreach(CU_OBJ ${GPU_GEN_OBJS})
     get_filename_component(CU_NAME ${CU_OBJ} NAME_WE)

diff --git a/cmake/presets/gpu-cuda.cmake b/cmake/presets/gpu-cuda.cmake
@@ -0,0 +1,11 @@
+# preset that enables GPU and selects CUDA API
+
+set(PKG_GPU ON CACHE BOOL "Build GPU package" FORCE)
+set(GPU_API "cuda" CACHE STRING "APU used by GPU package" FORCE)
+set(GPU_PREC "mixed" CACHE STRING "" FORCE)
+
+set(CUDA_NVCC_FLAGS "-allow-unsupported-compiler" CACHE STRING "" FORCE)
+set(CUDA_NVCC_FLAGS_DEBUG "-allow-unsupported-compiler" CACHE STRING "" FORCE)
+set(CUDA_NVCC_FLAGS_MINSIZEREL "-allow-unsupported-compiler" CACHE STRING "" FORCE)
+set(CUDA_NVCC_FLAGS_RELWITHDEBINFO "-allow-unsupported-compiler" CACHE STRING "" FORCE)
+set(CUDA_NVCC_FLAGS_RELEASE "-allow-unsupported-compiler" CACHE STRING "" FORCE)
diff --git a/doc/src/Commands_pair.rst b/doc/src/Commands_pair.rst
@@ -87,7 +87,7 @@ OPT.
    * :doc:`coul/long/soft (o) <pair_fep_soft>`
    * :doc:`coul/msm (o) <pair_coul>`
    * :doc:`coul/slater/cut <pair_coul_slater>`
-   * :doc:`coul/slater/long <pair_coul_slater>`
+   * :doc:`coul/slater/long (g) <pair_coul_slater>`
    * :doc:`coul/shield <pair_coul_shield>`
    * :doc:`coul/streitz <pair_coul>`
    * :doc:`coul/tt <pair_coul_tt>`
@@ -110,7 +110,7 @@ OPT.
    * :doc:`eam/he <pair_eam>`
    * :doc:`edip (o) <pair_edip>`
    * :doc:`edip/multi <pair_edip>`
-   * :doc:`edpd <pair_mesodpd>`
+   * :doc:`edpd (g) <pair_mesodpd>`
    * :doc:`eff/cut <pair_eff>`
    * :doc:`eim (o) <pair_eim>`
    * :doc:`exp6/rx (k) <pair_exp6_rx>`
@@ -158,14 +158,14 @@ OPT.
    * :doc:`lj/cut (gikot) <pair_lj>`
    * :doc:`lj/cut/coul/cut (gko) <pair_lj_cut_coul>`
    * :doc:`lj/cut/coul/cut/dielectric (o) <pair_dielectric>`
-   * :doc:`lj/cut/coul/cut/soft (o) <pair_fep_soft>`
+   * :doc:`lj/cut/coul/cut/soft (go) <pair_fep_soft>`
    * :doc:`lj/cut/coul/debye (gko) <pair_lj_cut_coul>`
    * :doc:`lj/cut/coul/debye/dielectric (o) <pair_dielectric>`
    * :doc:`lj/cut/coul/dsf (gko) <pair_lj_cut_coul>`
    * :doc:`lj/cut/coul/long (gikot) <pair_lj_cut_coul>`
    * :doc:`lj/cut/coul/long/cs <pair_cs>`
    * :doc:`lj/cut/coul/long/dielectric (o) <pair_dielectric>`
-   * :doc:`lj/cut/coul/long/soft (o) <pair_fep_soft>`
+   * :doc:`lj/cut/coul/long/soft (go) <pair_fep_soft>`
    * :doc:`lj/cut/coul/msm (go) <pair_lj_cut_coul>`
    * :doc:`lj/cut/coul/msm/dielectric <pair_dielectric>`
    * :doc:`lj/cut/coul/wolf (o) <pair_lj_cut_coul>`
@@ -202,7 +202,7 @@ OPT.
    * :doc:`lubricate/poly (o) <pair_lubricate>`
    * :doc:`lubricateU <pair_lubricateU>`
    * :doc:`lubricateU/poly <pair_lubricateU>`
-   * :doc:`mdpd <pair_mesodpd>`
+   * :doc:`mdpd (g) <pair_mesodpd>`
    * :doc:`mdpd/rhosum <pair_mesodpd>`
    * :doc:`meam (k) <pair_meam>`
    * :doc:`meam/ms (k) <pair_meam>`
@@ -268,11 +268,11 @@ OPT.
    * :doc:`smtbq <pair_smtbq>`
    * :doc:`snap (ik) <pair_snap>`
    * :doc:`soft (go) <pair_soft>`
-   * :doc:`sph/heatconduction <pair_sph_heatconduction>`
+   * :doc:`sph/heatconduction (g) <pair_sph_heatconduction>`
    * :doc:`sph/idealgas <pair_sph_idealgas>`
-   * :doc:`sph/lj <pair_sph_lj>`
+   * :doc:`sph/lj (g) <pair_sph_lj>`
    * :doc:`sph/rhosum <pair_sph_rhosum>`
-   * :doc:`sph/taitwater <pair_sph_taitwater>`
+   * :doc:`sph/taitwater (g) <pair_sph_taitwater>`
    * :doc:`sph/taitwater/morris <pair_sph_taitwater_morris>`
    * :doc:`spin/dipole/cut <pair_spin_dipole>`
    * :doc:`spin/dipole/long <pair_spin_dipole>`

diff --git a/doc/src/pair_coul_slater.rst b/doc/src/pair_coul_slater.rst
@@ -1,6 +1,7 @@
 .. index:: pair_style coul/slater
 .. index:: pair_style coul/slater/cut
 .. index:: pair_style coul/slater/long
+.. index:: pair_style coul/slater/long/gpu
 
 pair_style coul/slater command
 ==============================
@@ -11,6 +12,8 @@ pair_style coul/slater/cut command
 pair_style coul/slater/long command
 ===================================
 
+Accelerator Variants: *coul/slater/long/gpu*
+
 Syntax
 """"""
 

diff --git a/doc/src/pair_fep_soft.rst b/doc/src/pair_fep_soft.rst
@@ -1,8 +1,10 @@
 .. index:: pair_style lj/cut/soft
 .. index:: pair_style lj/cut/soft/omp
 .. index:: pair_style lj/cut/coul/cut/soft
+.. index:: pair_style lj/cut/coul/cut/soft/gpu
 .. index:: pair_style lj/cut/coul/cut/soft/omp
 .. index:: pair_style lj/cut/coul/long/soft
+.. index:: pair_style lj/cut/coul/long/soft/gpu
 .. index:: pair_style lj/cut/coul/long/soft/omp
 .. index:: pair_style lj/cut/tip4p/long/soft
 .. index:: pair_style lj/cut/tip4p/long/soft/omp
@@ -27,12 +29,12 @@ Accelerator Variants: *lj/cut/soft/omp*
 pair_style lj/cut/coul/cut/soft command
 =======================================
 
-Accelerator Variants: *lj/cut/coul/cut/soft/omp*
+Accelerator Variants: *lj/cut/coul/cut/soft/gpu*, *lj/cut/coul/cut/soft/omp*
 
 pair_style lj/cut/coul/long/soft command
 ========================================
 
-Accelerator Variants: *lj/cut/coul/long/soft/omp*
+Accelerator Variants: *lj/cut/coul/long/soft/gpu*, *lj/cut/coul/long/soft/omp*
 
 pair_style lj/cut/tip4p/long/soft command
 =========================================

diff --git a/doc/src/pair_mesodpd.rst b/doc/src/pair_mesodpd.rst
@@ -1,14 +1,20 @@
 .. index:: pair_style edpd
+.. index:: pair_style edpd/gpu
 .. index:: pair_style mdpd
+.. index:: pair_style mdpd/gpu
 .. index:: pair_style mdpd/rhosum
 .. index:: pair_style tdpd
 
 pair_style edpd command
 =======================
 
+Accelerator Variants: *edpd/gpu*
+
 pair_style mdpd command
 =======================
 
+Accelerator Variants: *mdpd/gpu*
+
 pair_style mdpd/rhosum command
 ==============================
 

diff --git a/doc/src/pair_sph_heatconduction.rst b/doc/src/pair_sph_heatconduction.rst
@@ -1,8 +1,11 @@
 .. index:: pair_style sph/heatconduction
+.. index:: pair_style sph/heatconduction/gpu
 
 pair_style sph/heatconduction command
 =====================================
 
+Accelerator Variants: *sph/heatconduction/gpu*
+
 Syntax
 """"""
 

diff --git a/doc/src/pair_sph_lj.rst b/doc/src/pair_sph_lj.rst
@@ -1,8 +1,11 @@
 .. index:: pair_style sph/lj
+.. index:: pair_style sph/lj/gpu
 
 pair_style sph/lj command
 =========================
 
+Accelerator Variants: *sph/lj/gpu*
+
 Syntax
 """"""
 

diff --git a/doc/src/pair_sph_taitwater.rst b/doc/src/pair_sph_taitwater.rst
@@ -1,8 +1,11 @@
 .. index:: pair_style sph/taitwater
+.. index:: pair_style sph/taitwater/gpu
 
 pair_style sph/taitwater command
 ================================
 
+Accelerator Variants: *sph/taitwater/gpu*
+
 Syntax
 """"""
 

diff --git a/examples/PACKAGES/dpd-meso/mdpd/in.mdpd b/examples/PACKAGES/dpd-meso/mdpd/in.mdpd
@@ -16,6 +16,7 @@ neighbor	        0.3 bin
 neigh_modify        every 1 delay 0 check yes
 
 atom_style	        mdpd
+comm_modify vel yes
 
 region     mdpd     block -25 25 -10 10 -10 10 units box
 create_box          1 mdpd

diff --git a/lib/gpu/Makefile.linux_multi b/lib/gpu/Makefile.linux_multi
@@ -65,7 +65,7 @@ CUDA_PRECISION = -D_SINGLE_DOUBLE
 
 CUDA_INCLUDE = -I$(CUDA_HOME)/include
 CUDA_LIB = -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib64/stubs
-CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC) -Xcompiler -fPIC
+CUDA_OPTS = -DUNIX -O3 --use_fast_math $(LMP_INC) -Xcompiler -fPIC -allow-unsupported-compiler
 
 CUDR_CPP = mpicxx -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK -DOMPI_SKIP_MPICXX=1 -fPIC -std=c++11
 CUDR_OPTS = -O2 $(LMP_INC) # -xHost -no-prec-div -ansi-alias

diff --git a/lib/gpu/geryon/ocl_mat.h b/lib/gpu/geryon/ocl_mat.h
@@ -54,6 +54,6 @@ namespace ucl_opencl {
 #include "ucl_print.h"
 #undef UCL_PRINT_ALLOW
 
-} // namespace ucl_cudart
+} // namespace ucl_opencl
 
 #endif
diff --git a/lib/gpu/lal_amoeba.cpp b/lib/gpu/lal_amoeba.cpp
@@ -281,13 +281,7 @@ int AmoebaT::polar_real(const int eflag, const int vflag) {
 
   const int BX=this->block_size();
   const int GX=static_cast<int>(ceil(static_cast<double>(ainum)/(BX/this->_threads_per_atom)));
-  /*
-  const int cus = this->device->gpu->cus();
-  while (GX < cus && GX > 1) {
-    BX /= 2;
-    GX=static_cast<int>(ceil(static_cast<double>(ainum)/(BX/this->_threads_per_atom)));
-  }
-  */
+
   this->time_pair.start();
 
   // Build the short neighbor list if not done yet

diff --git a/lib/gpu/lal_base_dpd.cpp b/lib/gpu/lal_base_dpd.cpp
@@ -56,7 +56,8 @@ int BaseDPDT::init_atomic(const int nlocal, const int nall,
                           const int max_nbors, const int maxspecial,
                           const double cell_size, const double gpu_split,
                           FILE *_screen, const void *pair_program,
-                          const char *k_name, const int onetype) {
+                          const char *k_name, const int onetype,
+                          const int extra_fields) {
   screen=_screen;
 
   int gpu_nbor=0;
@@ -75,7 +76,8 @@ int BaseDPDT::init_atomic(const int nlocal, const int nall,
   bool charge = false;
   bool rot = false;
   bool vel = true;
-  int success=device->init(*ans,charge,rot,nlocal,nall,maxspecial,vel);
+  _extra_fields = extra_fields;
+  int success=device->init(*ans,charge,rot,nlocal,nall,maxspecial,vel,_extra_fields/4);
   if (success!=0)
     return success;
 

diff --git a/lib/gpu/lal_base_dpd.h b/lib/gpu/lal_base_dpd.h
@@ -53,7 +53,7 @@ class BaseDPD {
                   const int maxspecial, const double cell_size,
                   const double gpu_split, FILE *screen,
                   const void *pair_program, const char *k_name,
-                  const int onetype=0);
+                  const int onetype=0, const int extra_fields=0);
 
   /// Estimate the overhead for GPU context changes and CPU driver
   void estimate_gpu_overhead();
@@ -167,7 +167,6 @@ class BaseDPD {
   /// Atom Data
   Atom<numtyp,acctyp> *atom;
 
-
   // ------------------------ FORCE/ENERGY DATA -----------------------
 
   Answer<numtyp,acctyp> *ans;
@@ -199,7 +198,7 @@ class BaseDPD {
 
  protected:
   bool _compiled;
-  int _block_size, _threads_per_atom, _onetype;
+  int _block_size, _threads_per_atom, _onetype, _extra_fields;
   double  _max_bytes, _max_an_bytes;
   double _gpu_overhead, _driver_overhead;
   UCL_D_Vec<int> *_nbor_data;