{lib}[GCCcore/14.2.0] amdsmi v25.4.2 w/ ROCm 6.4.1

easybuild/easyconfigs/a/amdsmi/amdsmi-25.4.2-GCCcore-14.2.0-ROCm-6.4.1.eb

@@ -0,0 +1,72 @@
+easyblock = 'CMakeMake'
+
+name = 'amdsmi'
+version = '25.4.2'
+_rocm_version = '6.4.1'
+# amdsmi includes esmi as a component, which needs to be copied into the amdsmi sources.
+# Required version can be found in amdsmi/CMakeLists.txt by searching for current_esmi_tag.
+_esmi_version = '4.1.2'
+versionsuffix = f'-ROCm-{_rocm_version}'
+
+homepage = 'https://github.com/ROCm/amdsmi'
+description = """
+The AMD System Management Interface (AMD SMI) library offers a unified tool for managing and monitoring
+GPUs, particularly in high-performance computing environments. It provides a user-space interface that
+allows applications to control GPU operations, monitor performance, and retrieve information about the
+system's drivers and GPUs."""
+docurls = ['https://rocm.docs.amd.com/projects/amdsmi/en/latest/']
+
+toolchain = {'name': 'GCCcore', 'version': '14.2.0'}
+
+source_urls = [
+    'https://github.com/ROCm/amdsmi/archive/refs/tags/',
+    'https://github.com/amd/esmi_ib_library/archive/refs/tags/'
+]
+sources = [
+    f'rocm-{_rocm_version}.tar.gz',
+    f'esmi_pkg_ver-{_esmi_version}.tar.gz',
+]
+patches = ['amdsmi-25.4.2_handle-non-standard-rocm-paths.patch']
+checksums = [
+    {'rocm-6.4.1.tar.gz': '5e1030cebacf2c92e63a555db6433ce7bb4f91409910ec98947e459d36630401'},
+    {'esmi_pkg_ver-4.1.2.tar.gz': 'd95f12c1500c60dc01b500d8928864a276757bd3bbb4a6daa84fb8c028e2978b'},
+    {'amdsmi-25.4.2_handle-non-standard-rocm-paths.patch':
+     '97b9067a7702d8131b4c25fba990ebfa538a08de3180c5898271591470ea8d80'},
+]
+
+builddependencies = [
+    ('binutils', '2.42'),
+    ('CMake', '3.31.3'),
+    ('pkgconf', '2.3.0'),
+    # Needed as amdsmi tries to find git to determine commit hash it was built from
+    ('git', '2.49.0'),
+    ('libdrm', '2.4.125'),
+]
+
+dependencies = [
+    ('Python', '3.13.1'),  # Needed as amdsmi is a python script
+]
+
+preconfigopts = (f"cp -r %(builddir)s/esmi_ib_library-esmi_pkg_ver-{_esmi_version} "
+                 f"%(builddir)s/amdsmi-rocm-{_rocm_version}/esmi_ib_library && ")
+
+sanity_check_paths = {
+    'files': [f'lib/libamd_smi.{SHLIB_EXT}',
+              'include/amd_smi/amdsmi.h'],
+    'dirs': ['lib/cmake/amd_smi',
+             'share/doc/amd_smi']
+}
+
+# Sanity check commands require kernel drivers to be loaded.
+# The command itself will result in an exit code of non-zero, therefore try to grep
+# the tool version, which is always present.
+# Trying any amd-smi command without the correct kernel drivers loaded will fail with:
+# ERROR:root:Unable to get devices, driver not initialized (amdgpu not found in modules)
+# ERROR:root:Unable to detect any GPU devices, check amdgpu version and module status (sudo modprobe amdgpu)
+# ERROR:root:Unable to detect any CPU devices, check amd_hsmp version and module status (sudo modprobe amd_hsmp)
+# AMDSMI Tool: 25.4.2+unknown | AMDSMI Library version: 25.4.0 | ROCm version: N/A
+sanity_check_commands = [
+    f'amd-smi version | grep "AMDSMI Tool: {version}"',
+]
+
+moduleclass = 'lib'

easybuild/easyconfigs/a/amdsmi/amdsmi-25.4.2_handle-non-standard-rocm-paths.patch

@@ -0,0 +1,98 @@
+Adds an explicit check for esmi_ib_library instead of trying to download it via Git.
+Sources need to be moved in EasyConfig. Also point to EasyBuild paths first before
+trying ROCm default paths.
+
+Author: Jan Andre Reuter ([email protected])
+
+diff --color -Naur amdsmi-rocm-6.4.1.orig/CMakeLists.txt amdsmi-rocm-6.4.1/CMakeLists.txt
+--- amdsmi-rocm-6.4.1.orig/CMakeLists.txt       2025-05-09 19:06:29.000000000 +0200
++++ amdsmi-rocm-6.4.1/CMakeLists.txt    2025-06-19 09:57:50.521681766 +0200
+@@ -101,42 +101,7 @@
+     set(current_esmi_tag "esmi_pkg_ver-4.1.2")
+
+     if(NOT EXISTS ${PROJECT_SOURCE_DIR}/esmi_ib_library/src)
+-        # TODO: use ExternalProject_Add instead or a submodule
+-        message(STATUS "Adding esmi_ib_library...")
+-        execute_process(COMMAND git clone --depth=1 -b ${current_esmi_tag} https://github.com/amd/esmi_ib_library.git
+-                                ${PROJECT_SOURCE_DIR}/esmi_ib_library)
+-    else()
+-        message(STATUS "esmi_ib_library already installed, checking version...")
+-
+-        # Grab latest commit and get the tag
+-        execute_process(
+-            COMMAND git rev-list --tags --max-count=1
+-            WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}/esmi_ib_library
+-            OUTPUT_VARIABLE latest_commit
+-            OUTPUT_STRIP_TRAILING_WHITESPACE)
+-        execute_process(
+-            COMMAND git describe --tags ${latest_commit} --match "*pkg*"
+-            WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}/esmi_ib_library
+-            OUTPUT_VARIABLE latest_esmi_tag
+-            OUTPUT_STRIP_TRAILING_WHITESPACE)
+-
+-        # Update to latest tags if not matched
+-        if(NOT latest_esmi_tag STREQUAL current_esmi_tag)
+-            message(STATUS "Updating esmi_ib_library...")
+-            execute_process(
+-                COMMAND git clone --depth=1 -b ${current_esmi_tag} https://github.com/amd/esmi_ib_library.git
+-                        ${PROJECT_SOURCE_DIR}/esmi_ib_library_temp RESULT_VARIABLE clone_result)
+-            if(clone_result EQUAL 0)
+-                file(REMOVE_RECURSE ${PROJECT_SOURCE_DIR}/esmi_ib_library)
+-                file(RENAME ${PROJECT_SOURCE_DIR}/esmi_ib_library_temp ${PROJECT_SOURCE_DIR}/esmi_ib_library)
+-                message(STATUS "Successfully cloned updated esmi_ib_library")
+-            else()
+-                file(REMOVE_RECURSE ${PROJECT_SOURCE_DIR}/esmi_ib_library_temp)
+-                message(FATAL_ERROR "Failed to clone updated esmi_ib_library")
+-            endif()
+-        else()
+-            message(STATUS "esmi_ib_library is the latest version: ${current_esmi_tag}...")
+-        endif()
++        message(FATAL_ERROR "Expect esmi_ib_library_src to be present in sources. Please copy the required sources from esmi.")
+     endif()
+
+     # Make sure to update the amd_hsmp.h file with the corresponding esmi version
+diff --color -Naur amdsmi-rocm-6.4.1.orig/goamdsmi_shim/smiwrapper/amdsmi_go_shim.c amdsmi-rocm-6.4.1/goamdsmi_shim/smiwrapper/amdsmi_go_shim.c
+--- amdsmi-rocm-6.4.1.orig/goamdsmi_shim/smiwrapper/amdsmi_go_shim.c    2025-05-09 19:06:29.000000000 +0200
++++ amdsmi-rocm-6.4.1/goamdsmi_shim/smiwrapper/amdsmi_go_shim.c 2025-06-19 10:01:29.342064623 +0200
+@@ -24,7 +24,7 @@
+ #include <stdint.h>
+ #include <string.h>
+ #include "amdsmi_go_shim.h"
+-#include <amd_smi/amdsmi.h>
++#include "amd_smi/amdsmi.h"
+ #include <unistd.h>
+ #define nullptr ((void*)0)
+
+@@ -38,8 +38,6 @@
+ #define MAX_GPU_POWER_FROM_DRIVER      0xFFFF
+
+ #define AMDSMI_DRIVER_NAME     "AMDSMI"
+-#define AMDSMI_LIB_FILE        "/opt/rocm/lib/libamd_smi.so"
+-#define AMDSMI_LIB64_FILE      "/opt/rocm/lib64/libamd_smi.so"
+
+ #define AMDGPU_DRIVER_NAME     "AMDGPUDriver"
+ #define AMDGPU_INITSTATE_FILE  "/sys/module/amdgpu/initstate"
+diff --color -Naur amdsmi-rocm-6.4.1.orig/py-interface/amdsmi_wrapper.py amdsmi-rocm-6.4.1/py-interface/amdsmi_wrapper.py
+--- amdsmi-rocm-6.4.1.orig/py-interface/amdsmi_wrapper.py       2025-05-09 19:06:29.000000000 +0200
++++ amdsmi-rocm-6.4.1/py-interface/amdsmi_wrapper.py    2025-06-19 09:30:58.665549378 +0200
+@@ -179,6 +179,8 @@
+ def find_smi_library():
+     err = OSError("Could not load libamd_smi.so")
+     possible_locations = list()
++    # 0. Use EasyBuild variables
++    possible_locations.append(os.path.join(os.environ.get("EBROOTAMDSMI"), "lib/libamd_smi.so"))
+     # 1.
+     rocm_path = os.getenv("ROCM_HOME", os.getenv("ROCM_PATH"))
+     if rocm_path:
+diff --color -Naur amdsmi-rocm-6.4.1.orig/tools/generator.py amdsmi-rocm-6.4.1/tools/generator.py
+--- amdsmi-rocm-6.4.1.orig/tools/generator.py   2025-05-09 19:06:29.000000000 +0200
++++ amdsmi-rocm-6.4.1/tools/generator.py        2025-06-19 09:38:12.883141175 +0200
+@@ -194,6 +194,8 @@
+ def find_smi_library():
+     err = OSError("Could not load {library_name}")
+     possible_locations = list()
++    # 0. Try EasyBuild first
++    possible_locations.append(os.path.join(os.environ.get("EBROOTAMDSMI"), "lib/{library_name}"))
+     # 1.
+     rocm_path = os.getenv("ROCM_HOME", os.getenv("ROCM_PATH"))
+     if rocm_path: