On Thu, 18 Sep 2003, Atsushi Nemoto wrote:
> macro> I wanted to avoid that as the resulting code would be ugly. I
> macro> guess there is no other choice, although I think that's a bug
> macro> in gcc.
> I have no idea that is a gcc bug, but I think align_mod() inline
> function is not so beautiful because it can not be compiled anyway if
> non-constant value was passed.
Well, the asm statement requires immediates, so if macros are used
variables won't work anyway, but the the code will look more obscurely.
> macro> Can you quote the exact command line used for building the
> macro> file?
> mips64el-linux-gcc -D__KERNEL__ -I/home/anemo/work/linux/include -Wall
> -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common
> -fomit-frame-pointer -I /home/anemo/work/linux/include/asm/gcc -mabi=64 -G 0
> -mno-abicalls -fno-pic -Wa,--trap -pipe -march=r4600 -nostdinc
> -iwithprefix include -DKBUILD_BASENAME=cpu_probe -c -o cpu-probe.o
> I also tried with -finline-limit=100000 but no luck.
It looks like gcc insists on forcing the constants into registers. The
following patch should work for gcc 3.x. A few warnings will still be
emitted, but the code will get build properly.
I'm applying the change to the CVS as it's good anyway.
+ Maciej W. Rozycki, Technical University of Gdansk, Poland +
+ e-mail: email@example.com, PGP key available +
diff -up --recursive --new-file
2003-09-12 01:18:01.000000000 +0000
+++ linux-mips-2.4.22-20030916/arch/mips64/kernel/cpu-probe.c 2003-09-18
@@ -113,7 +113,7 @@ static inline void check_wait(void)
-static inline void align_mod(int align, int mod)
+static inline void align_mod(const int align, const int mod)
@@ -124,11 +124,11 @@ static inline void align_mod(int align,
- : "i" (align), "i" (mod));
+ : "n" (align), "n" (mod));
static inline void mult_sh_align_mod(long *v1, long *v2, long *w,
- int align, int mod)
+ const int align, const int mod)
unsigned long flags;
int m1, m2;