On Wed, Oct 13, 2010 at 7:19 AM, Paolo Bonzini <pbonz...@redhat.com> wrote: > On 10/12/2010 09:04 PM, Blue Swirl wrote: >>> >>> diff --git a/tests/Makefile b/tests/Makefile >>> index ff7f787..a789e2d 100644 >>> --- a/tests/Makefile >>> +++ b/tests/Makefile >>> @@ -64,11 +64,21 @@ linux-test: linux-test.c >>> $(CC) $(CFLAGS) $(LDFLAGS) -o $@ $< -lm >>> >>> # speed test >>> +ifeq ($(shell uname -m), x86_64) >>> +CC_I386 = $(CC) -m32 >>> +else >>> +ifeq ($(shell uname -m), i386) >>> +CC_I386 = $(CC) >>> +else >>> +CC_I386 = i386-pc-linux-gnu-$(CC) >>> +endif >>> +endif >> >> I'd move this logic to configure. > > I can do that, but it seemed overengineered for a variable that's only > needed only in one makefile rule, and furthermore a rule that will be > invoked manually by the user. In fact I was tempted to hardcode > "CC_I386=$(CC) -m32" which would work on x86_64 and i386 (and leave the task > of overriding it to the user).
That part in configure could be extended to handle other targets as well. If Alpha, ARM, CRIS, or MIPS (cross-)compilers are available, we can easily build tests for those too. In the original Makefile location, this would be a mess.
