Building a kernel with CONFIG_CC_OPTIMISE_FOR_SIZE leads to
__ilog2_u32() being duplicated 50 times and __ilog2_u64() 3 times
in vmlinux on a tiny powerpc32 config.
__ilog2_u32() being 2 instructions it is not worth being kept out
of line, so force inlining. Allthough the u64 version is a bit bigger,
there is still a small benefit in keeping it inlined. On a 64 bits
config there's a real benefit.
With this change the size of vmlinux text is reduced by 1 kbytes,
which is approx 50% more than the size of the removed functions.
Before the patch there is for instance:
c00d2a94 <__ilog2_u32>:
c00d2a94: 7c 63 00 34 cntlzw r3,r3
c00d2a98: 20 63 00 1f subfic r3,r3,31
c00d2a9c: 4e 80 00 20 blr
c00d36d8 <__order_base_2>:
c00d36d8: 28 03 00 01 cmplwi r3,1
c00d36dc: 40 81 00 2c ble c00d3708 <__order_base_2+0x30>
c00d36e0: 94 21 ff f0 stwu r1,-16(r1)
c00d36e4: 7c 08 02 a6 mflr r0
c00d36e8: 38 63 ff ff addi r3,r3,-1
c00d36ec: 90 01 00 14 stw r0,20(r1)
c00d36f0: 4b ff f3 a5 bl c00d2a94 <__ilog2_u32>
c00d36f4: 80 01 00 14 lwz r0,20(r1)
c00d36f8: 38 63 00 01 addi r3,r3,1
c00d36fc: 7c 08 03 a6 mtlr r0
c00d3700: 38 21 00 10 addi r1,r1,16
c00d3704: 4e 80 00 20 blr
c00d3708: 38 60 00 00 li r3,0
c00d370c: 4e 80 00 20 blr
With the patch it has become:
c00d356c <__order_base_2>:
c00d356c: 28 03 00 01 cmplwi r3,1
c00d3570: 40 81 00 14 ble c00d3584 <__order_base_2+0x18>
c00d3574: 38 63 ff ff addi r3,r3,-1
c00d3578: 7c 63 00 34 cntlzw r3,r3
c00d357c: 20 63 00 20 subfic r3,r3,32
c00d3580: 4e 80 00 20 blr
c00d3584: 38 60 00 00 li r3,0
c00d3588: 4e 80 00 20 blr
No more need for __order_base_2() to setup a stack frame and
save/restore caller address. And the following 'add 1' is
merged in the substract.
Another typical use of it:
c080ff28 <hugepagesz_setup>:
...
c080fff8: 7f c3 f3 78 mr r3,r30
c080fffc: 4b 8f 81 f1 bl c01081ec <__ilog2_u32>
c0810000: 38 63 ff f2 addi r3,r3,-14
...
Becomes
c080ff1c <hugepagesz_setup>:
...
c080ffec: 7f c3 00 34 cntlzw r3,r30
c080fff0: 20 63 00 11 subfic r3,r3,17
...
Here no need to move r30 argument to r3 then substract 14 to result. Just work
on r30 and merge the 'sub 14' with the 'sub from 31'.
Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxxxxxx>
---
include/linux/log2.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/include/linux/log2.h b/include/linux/log2.h
index df0b155c2141..9f30d087a128 100644
--- a/include/linux/log2.h
+++ b/include/linux/log2.h
@@ -18,7 +18,7 @@
* - the arch is not required to handle n==0 if implementing the fallback
*/
#ifndef CONFIG_ARCH_HAS_ILOG2_U32
-static inline __attribute__((const))
+static __always_inline __attribute__((const))
int __ilog2_u32(u32 n)
{
return fls(n) - 1;
@@ -26,7 +26,7 @@ int __ilog2_u32(u32 n)
#endif
#ifndef CONFIG_ARCH_HAS_ILOG2_U64
-static inline __attribute__((const))
+static __always_inline __attribute__((const))
int __ilog2_u64(u64 n)
{
return fls64(n) - 1;
--
2.33.1
