[PATCH v2 0/4] s390: Fix and optimize __flogr() inline assembly

Heiko Carstens posted 4 patches 2 weeks, 1 day ago
arch/s390/include/asm/bitops.h | 21 ++++++++++++++-------
include/linux/compiler_types.h | 23 +++++++++++++++++++++++
init/Kconfig                   | 10 ++++++++++
3 files changed, 47 insertions(+), 7 deletions(-)
[PATCH v2 0/4] s390: Fix and optimize __flogr() inline assembly
Posted by Heiko Carstens 2 weeks, 1 day ago
v2:

- Split patches differently, so that the first patch is the actual
  fix, which addresses the reported warning / error. The subsequent
  patches are optimizations

v1:

https://lore.kernel.org/all/20250910151216.646600-1-hca@linux.ibm.com/

A recent optimization of the s390 specific ffs() and ffs64()
implementations leads to a new compiler warning. Instead of reverting the
optimization address this with the rather new assume attribute, which
generates even better code, if supported by compilers.

Since the assume attribute may be useful for others as well, add the
__assume macro to compiler attributes, so it is kernel wide available,
instead of adding an s390 specific optimization.

Heiko Carstens (4):
  s390/bitops: Limit return value range of __flogr()
  compiler_types: Add __assume macro
  s390/bitops: Use __assume() for __flogr() inline assembly return value
  s390/bitops: Cleanup __flogr()

 arch/s390/include/asm/bitops.h | 21 ++++++++++++++-------
 include/linux/compiler_types.h | 23 +++++++++++++++++++++++
 init/Kconfig                   | 10 ++++++++++
 3 files changed, 47 insertions(+), 7 deletions(-)

-- 
2.48.1
Re: [PATCH v2 0/4] s390: Fix and optimize __flogr() inline assembly
Posted by Alexander Gordeev 1 week, 6 days ago
On Tue, Sep 16, 2025 at 03:47:59PM +0200, Heiko Carstens wrote:
> A recent optimization of the s390 specific ffs() and ffs64()
> implementations leads to a new compiler warning. Instead of reverting the
> optimization address this with the rather new assume attribute, which
> generates even better code, if supported by compilers.
> 
> Since the assume attribute may be useful for others as well, add the
> __assume macro to compiler attributes, so it is kernel wide available,
> instead of adding an s390 specific optimization.
> 
> Heiko Carstens (4):
>   s390/bitops: Limit return value range of __flogr()
>   compiler_types: Add __assume macro
>   s390/bitops: Use __assume() for __flogr() inline assembly return value
>   s390/bitops: Cleanup __flogr()
> 
>  arch/s390/include/asm/bitops.h | 21 ++++++++++++++-------
>  include/linux/compiler_types.h | 23 +++++++++++++++++++++++
>  init/Kconfig                   | 10 ++++++++++
>  3 files changed, 47 insertions(+), 7 deletions(-)

Applied, thanks!