From b86bd6947c7e556753de4b937d256a2cbf91076b Mon Sep 17 00:00:00 2001
From: Markus Schwarz <markus.schwarz@kit.edu>
Date: Fri, 14 Feb 2025 16:29:22 +0100
Subject: [PATCH] fix for --optimize compile option and ARM chips
The --optimize compile option included some x86 specific instruction sets. These lead to errors when an ARM-based CPU is used. This fix checks for ARM CPU and then ignores the x86 specific instruction.
---
blond/compile.py | 36 +++++++++++++++++++-----------------
1 file changed, 19 insertions(+), 17 deletions(-)
diff --git a/blond/compile.py b/blond/compile.py
index 63162b08..822f011a 100644
--- a/blond/compile.py
+++ b/blond/compile.py
@@ -22,6 +22,7 @@ from __future__ import print_function
import argparse
import ctypes
import os
+import platform
import subprocess
import sys
@@ -211,21 +212,23 @@ def compile_cpp_library(args, cflags, float_flags, libs, cpp_files):
# Format the output list
stdout = ret.stdout.replace('#define ', '').replace(
'__ 1', '').replace('__', '').split('\n')
- # Add the appropriate vectorization flag (not use avx512)
- if 'AVX2' in stdout:
- cflags += ['-mavx2']
- elif 'AVX' in stdout:
- cflags += ['-mavx']
- elif 'SSE4_2' in stdout or 'SSE4_1' in stdout:
- cflags += ['-msse4']
- elif 'SSE3' in stdout:
- cflags += ['-msse3']
- else:
- cflags += ['-msse']
-
- # Add FMA if supported
- if 'FMA' in stdout:
- cflags += ['-mfma']
+ # following options exist only on x86 processors
+ if 'arm' not in platform.machine():
+ # Add the appropriate vectorization flag (not use avx512)
+ if 'AVX2' in stdout:
+ cflags += ['-mavx2']
+ elif 'AVX' in stdout:
+ cflags += ['-mavx']
+ elif 'SSE4_2' in stdout or 'SSE4_1' in stdout:
+ cflags += ['-msse4']
+ elif 'SSE3' in stdout:
+ cflags += ['-msse3']
+ else:
+ cflags += ['-msse']
+
+ # Add FMA if supported
+ if 'FMA' in stdout:
+ cflags += ['-mfma']
root, ext = os.path.splitext(args['libname'])
if not ext:
@@ -312,10 +315,9 @@ def compile_cuda_library(args, nvccflags, float_flags, cuda_files, nvcc):
# Compile the GPU library
# print('\n' + ''.join(['='] * 80))
print('\nCompiling the CUDA library')
- import cupy as cp
-
if args['gpu'] == 'discover':
print('Discovering the device compute capability..')
+ import cupy as cp
dev = cp.cuda.Device(0)
dev_name = cp.cuda.runtime.getDeviceProperties(dev)['name']
--
GitLab