Unverified Commit 4cbe0954 authored by Axel Kohlmeyer's avatar Axel Kohlmeyer Committed by GitHub
Browse files

Merge pull request #2234 from akohlmey/collected-small-fixes 

Collected small fixes and unittest updates
parents 9068d17a b37b45c8

cmake/Modules/Packages/GPU.cmake

+50 −28
Original line number Diff line number Diff line
@@ -66,29 +66,40 @@ if(GPU_API STREQUAL "CUDA") 
  # --arch translates directly instead of JIT, so this should be for the preferred or most common architecture

  set(GPU_CUDA_GENCODE "-arch=${GPU_ARCH}")

  # Fermi (GPU Arch 2.x) is supported by CUDA 3.2 to CUDA 8.0

  if((CUDA_VERSION VERSION_GREATER "3.1") AND (CUDA_VERSION VERSION_LESS "9.0"))

  if((CUDA_VERSION VERSION_GREATER_EQUAL "3.2") AND (CUDA_VERSION VERSION_LESS "9.0"))

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_20,code=[sm_20,compute_20] ")

  endif()

  # Kepler (GPU Arch 3.x) is supported by CUDA 5 and later

  if(CUDA_VERSION VERSION_GREATER "4.9")

  # Kepler (GPU Arch 3.0) is supported by CUDA 5 to CUDA 10.2

  if((CUDA_VERSION VERSION_GREATER_EQUAL "5.0") AND (CUDA_VERSION VERSION_LESS "11.0"))

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_30,code=[sm_30,compute_30] ")

  endif()

  # Kepler (GPU Arch 3.5) is supported by CUDA 5 to CUDA 11

  if((CUDA_VERSION VERSION_GREATER_EQUAL "5.0") AND (CUDA_VERSION VERSION_LESS "12.0"))

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_30,code=[sm_30,compute_30] -gencode arch=compute_35,code=[sm_35,compute_35]")

  endif()

  # Maxwell (GPU Arch 5.x) is supported by CUDA 6 and later

  if(CUDA_VERSION VERSION_GREATER "5.9")

  if(CUDA_VERSION VERSION_GREATER_EQUAL "6.0")

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_50,code=[sm_50,compute_50] -gencode arch=compute_52,code=[sm_52,compute_52]")

  endif()

  # Pascal (GPU Arch 6.x) is supported by CUDA 8 and later

  if(CUDA_VERSION VERSION_GREATER "7.9")

  if(CUDA_VERSION VERSION_GREATER_EQUAL "8.0")

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_60,code=[sm_60,compute_60] -gencode arch=compute_61,code=[sm_61,compute_61]")

  endif()

  # Volta (GPU Arch 7.0) is supported by CUDA 9 and later

  if(CUDA_VERSION VERSION_GREATER "8.9")

  if(CUDA_VERSION VERSION_GREATER_EQUAL "9.0")

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_70,code=[sm_70,compute_70]")

  endif()

  # Turing (GPU Arch 7.5) is supported by CUDA 10 and later

  if(CUDA_VERSION VERSION_GREATER "9.9")

  if(CUDA_VERSION VERSION_GREATER_EQUAL "10.0")

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_75,code=[sm_75,compute_75]")

  endif()

  # Ampere (GPU Arch 8.0) is supported by CUDA 11 and later

  if(CUDA_VERSION VERSION_GREATER_EQUAL "11.0")

    string(APPEND GPU_CUDA_GENCODE " -gencode arch=compute_80,code=[sm_80,compute_80]")

  endif()

  if(CUDA_VERSION VERSION_GREATER_EQUAL "12.0")

    message(WARNING "Unsupported CUDA version. Use at your own risk.")

  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})
@@ -228,29 +239,40 @@ elseif(GPU_API STREQUAL "HIP") 
    # --arch translates directly instead of JIT, so this should be for the preferred or most common architecture

    set(HIP_CUDA_GENCODE "-arch=${HIP_ARCH}")

    # Fermi (GPU Arch 2.x) is supported by CUDA 3.2 to CUDA 8.0

    if((CUDA_VERSION VERSION_GREATER "3.1") AND (CUDA_VERSION VERSION_LESS "9.0"))

    if((CUDA_VERSION VERSION_GREATER_EQUAL "3.2") AND (CUDA_VERSION VERSION_LESS "9.0"))

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_20,code=[sm_20,compute_20]")

    endif()

    # Kepler (GPU Arch 3.x) is supported by CUDA 5 and later

    if(CUDA_VERSION VERSION_GREATER "4.9")

      string(APPEND HIP_CUDA_GENCODE "-gencode arch=compute_30,code=[sm_30,compute_30] -gencode arch=compute_35,code=[sm_35,compute_35] ")

    # Kepler (GPU Arch 3.0) is supported by CUDA 5 to CUDA 10.2

    if((CUDA_VERSION VERSION_GREATER_EQUAL "5.0") AND (CUDA_VERSION VERSION_LESS "11.0"))

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_30,code=[sm_30,compute_30]")

    endif()

    # Kepler (GPU Arch 3.5) is supported by CUDA 5 to CUDA 11.0

    if((CUDA_VERSION VERSION_GREATER_EQUAL "5.0") AND (CUDA_VERSION VERSION_LESS "12.0"))

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_35,code=[sm_35,compute_35]")

    endif()

    # Maxwell (GPU Arch 5.x) is supported by CUDA 6 and later

    if(CUDA_VERSION VERSION_GREATER "5.9")

    if(CUDA_VERSION VERSION_GREATER_EQUAL "6.0")

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_50,code=[sm_50,compute_50] -gencode arch=compute_52,code=[sm_52,compute_52]")

    endif()

    # Pascal (GPU Arch 6.x) is supported by CUDA 8 and later

    if(CUDA_VERSION VERSION_GREATER "7.9")

    if(CUDA_VERSION VERSION_GREATER_EQUAL "8.0")

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_60,code=[sm_60,compute_60] -gencode arch=compute_61,code=[sm_61,compute_61]")

    endif()

    # Volta (GPU Arch 7.0) is supported by CUDA 9 and later

    if(CUDA_VERSION VERSION_GREATER "8.9")

    if(CUDA_VERSION VERSION_GREATER_EQUAL "9.0")

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_70,code=[sm_70,compute_70]")

    endif()

    # Turing (GPU Arch 7.5) is supported by CUDA 10 and later

    if(CUDA_VERSION VERSION_GREATER "9.9")

    if(CUDA_VERSION VERSION_GREATER_EQUAL "10.0")

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_75,code=[sm_75,compute_75]")

    endif()

    # Ampere (GPU Arch 8.0) is supported by CUDA 11 and later

    if(CUDA_VERSION VERSION_GREATER_EQUAL "11.0")

      string(APPEND HIP_CUDA_GENCODE " -gencode arch=compute_80,code=[sm_80,compute_80]")

    endif()

    if(CUDA_VERSION VERSION_GREATER_EQUAL "12.0")

      message(WARNING "Unsupported CUDA version. Use at your own risk.")

    endif()

  endif()



  file(GLOB GPU_LIB_CU ${LAMMPS_LIB_SOURCE_DIR}/gpu/[^.]*.cu ${CMAKE_CURRENT_SOURCE_DIR}/gpu/[^.]*.cu)

doc/src/Build_extras.rst

+11 −2
Original line number Diff line number Diff line
@@ -105,10 +105,10 @@ CMake build 
                                # generic (default) or intel (Intel CPU) or fermi, kepler, cypress (NVIDIA)

   -D GPU_ARCH=value            # primary GPU hardware choice for GPU_API=cuda

                                # value = sm_XX, see below

                                # default is sm_30

                                # default is sm_50

   -D HIP_ARCH=value            # primary GPU hardware choice for GPU_API=hip

                                # value depends on selected HIP_PLATFORM

                                # default is 'gfx906' for HIP_PLATFORM=hcc and 'sm_30' for HIP_PLATFORM=nvcc

                                # default is 'gfx906' for HIP_PLATFORM=hcc and 'sm_50' for HIP_PLATFORM=nvcc

   -D HIP_USE_DEVICE_SORT=value # enables GPU sorting

                                # value = yes (default) or no

   -D CUDPP_OPT=value           # optimization setting for GPU_API=cuda
@@ -1255,6 +1255,15 @@ also typically :ref:`install the USER-OMP package <user-omp>`, as it can be 
used in tandem with the USER-INTEL package to good effect, as explained

on the :doc:`Speed intel <Speed_intel>` doc page.



When using Intel compilers version 16.0 or later is required.  You can

also use the GNU or Clang compilers and they will provide performance

improvements over regular styles and USER-OMP styles, but less so than

with the Intel compilers.  Please also note, that some compilers have

been found to apply memory alignment constraints incompletely or

incorrectly and thus can cause segmentation faults in otherwise correct

code when using features from the USER-INTEL package.





CMake build

^^^^^^^^^^^

doc/src/Packages_details.rst

+2 −1
Original line number Diff line number Diff line
@@ -306,7 +306,8 @@ gpu" or "-suffix gpu" :doc:`command-line switches <Run_options>`. See 
also the :ref:`KOKKOS <PKG-KOKKOS>` package, which has GPU-enabled styles.



**Authors:** Mike Brown (Intel) while at Sandia and ORNL and Trung Nguyen

(Northwestern U) while at ORNL.

(Northwestern U) while at ORNL and later. AMD HIP support by Evgeny

Kuznetsov, Vladimir Stegailov, and Vsevolod Nikolskiy (HSE University).



**Install:**

doc/src/Speed_gpu.rst

+4 −0
Original line number Diff line number Diff line
@@ -50,6 +50,10 @@ but this can be overridden using the device option of the :doc:`package <package 
command. run lammps/lib/gpu/ocl_get_devices to get a list of available

platforms and devices with a suitable ICD available.



To compute and use this package in HIP mode, you have to have the AMD ROCm

software installed. Versions of ROCm older than 3.5 are currently deprecated

by AMD.



**Building LAMMPS with the GPU package:**



See the :ref:`Build extras <gpu>` doc page for

doc/src/Speed_intel.rst

+3 −3
Original line number Diff line number Diff line
@@ -138,10 +138,10 @@ For Intel Xeon Phi co-processors (Offload): 


**Required hardware/software:**



When using Intel compilers version 16.0 or later is required.



In order to use offload to co-processors, an Intel Xeon Phi

co-processor and an Intel compiler are required. For this, the

recommended version of the Intel compiler is 14.0.1.106 or

versions 15.0.2.044 and higher.

co-processor and an Intel compiler are required.



Although any compiler can be used with the USER-INTEL package,

currently, vectorization directives are disabled by default when

CRA Git | Maintained and supported by SUSTech CRA and CCSE