The current way of getting the number of cpus uses different assumptions which may or may not be true in the future: * /proc/cpuinfo exists * cpuinfo provides a list of all processing units * all processing units are available for the current process and its childs * the word 'processor' is only used in the cpu index and 'model name' lines * cpu index is strict monotonic increasing * lowest cpu index is 0 and highest cpu index is n-1 (n is the number of processing units)
At least the number of available processing units for the current process can be shown to be wrong. on current systems by assigning a singe cpu to a process and its childs:
$ numactl --physcpubind=0 nproc 1
$ numactl --physcpubind=0 sh -c 'NUM_CPUS=`cat /proc/cpuinfo | grep -v "model name" | grep processor | tail -1 | awk -F" " '''{print $$3}'''`;echo `expr $NUM_CPUS + 1`' 4
nproc is part of coreutils since version 8.1. On systems without nproc it will just assume that a single processing unit is available.
Signed-off-by: Sven Eckelmann sven.eckelmann@gmx.de --- vis/Makefile | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/vis/Makefile b/vis/Makefile index 675dcd8..a119758 100644 --- a/vis/Makefile +++ b/vis/Makefile @@ -49,7 +49,7 @@ REVISION_VERSION="\ rv$(REVISION)"
VIS_VERSION= $(shell grep "^#define SOURCE_VERSION " $(SOURCE_VERSION_HEADER) | sed -e '1p' -n | awk -F '"' '{print $$2}' | awk '{print $$1}') FILE_NAME= $(PACKAGE_NAME)_$(VIS_VERSION)-rv$(REVISION)_$@ -NUM_CPUS = $(shell NUM_CPUS=`cat /proc/cpuinfo | grep -v 'model name' | grep processor | tail -1 | awk -F' ' '{print $$3}'`;echo `expr $$NUM_CPUS + 1`) +NUM_CPUS = $(shell nproc 2> /dev/null || echo 1)
all: