staging: media: atomisp: Simplyfy masking bit logic
Subsystems (4)
RFC 1/1 staging: media: atomisp: Simplyfy masking bit logic
Simplified masking logic in pci/hive_isp_css_common/host/vmem.c.
---
I have tested this change on whole range of *valid* inputs, and it gives the same results as before, but this function seems to be little counter-intuitive as far as start is a (bit index) but end is (bit index + 1).
It is follow up to: https://lore.kernel.org/linux-staging/[email protected]/
--- .../staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +2 -2
@@ -22,14 +22,14 @@ typedef hive_uedge *hive_wide;static inline hive_uedge subword (hive_uedge w,unsigned int start,unsigned int end) { -return (w & (((1ULL << (end -1 )) -1 ) <<1 |1 )) >> start; +return (w &__GENMASK_ULL (end-1 ,0 )) >> start; }/* inverse subword bits move like this: MSB[xxxx____xxxx]LSB -> MSB[xxxx0000xxxx]LSB */ static inline hive_uedge inv_subword (hive_uedge w,unsigned int start,unsigned int end) { -return w & (~(((1ULL << (end -1 )) -1 ) <<1 |1 ) | ((1ULL << start) -1 )); +return w & (~__GENMASK_ULL (end-1 , start)); }#define uedge_bits (8 * sizeof(hive_uedge))
Signature · 1 line
2.51.0.318.gd7df087d1a-goog
Quoted 8 lines · Adrian Barnaś
Simplified masking logic in
pci/hive_isp_css_common/host/vmem.c.---
I have tested this change on whole range of *valid* inputs, and it gives the same results as before, but this function seems to be little counter-intuitive as far as start is a (bit index) but end is (bit index + 1).
Yeah. The "end - 1" is unfortunate. I guess we accidentally started counting from 1... I looked at how to remove it but got lost.
Quoted 15 lines
It is follow up to: https://lore.kernel.org/linux-staging/[email protected]/
--- .../staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +0
-1@@ -22,14 +22,14 @@ typedef hive_uedge *hive_wide;static inline hive_uedge subword (hive_uedge w,unsigned int start,unsigned int end) { -return (w & (((1ULL << (end -1 )) -1 ) <<1 |1 )) >> start;
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c typedef hive_uedge *hive_wide; · line 33
Quoted 10 lines
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +2
-1+return (w &__GENMASK_ULL (end-1 ,0 )) >> start; }/* inverse subword bits move like this: MSB[xxxx____xxxx]LSB -> MSB[xxxx0000xxxx]LSB */ static inline hive_uedge inv_subword (hive_uedge w,unsigned int start,unsigned int end) { -return w & (~(((1ULL << (end -1 )) -1 ) <<1 |1 ) | ((1ULL << start) -1 )); +return w & (~__GENMASK_ULL (end-1 , start)); }
nit: white space. Add spaces. Remove parentheses.
These are supposed to be opposites, right? Subword and inverse Subword. You could dress them up to make them look more opposite.
return (w & __GENMASK_ULL(end - 1, start)) >> start; return w & ~__GENMASK_ULL(end - 1, start);
regards,
dan carpenter
Quoted 1 line · Dan Carpenter
On Tue, Sep 02, 2025 at 07:38:40AM +0000, Adrian Barnaś wrote:
...
Quoted 9 lines
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c typedef hive_uedge *hive_wide; · line 33
Quoted 6 lines
static inline
hive_uedge
inv_subword(hive_uedge w, unsigned int start, unsigned int end){drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +1
-1-return w & (~(((1ULL << (end -1 )) -1 ) <<1 |1 ) | ((1ULL << start) -1 )); +return w & (~__GENMASK_ULL (end-1 , start)); }nit: white space. Add spaces. Remove parentheses.
Quoted 5 lines
These are supposed to be opposites, right? Subword and inverse Subword. You could dress them up to make them look more opposite.
return (w & __GENMASK_ULL(end - 1, start)) >> start; return w & ~__GENMASK_ULL(end - 1, start);
The problem is (and actually with the (end-1, start) above that it might generate a really bad code on some CPUs, so I really, really prefer the way when _at least_ one of the parameters is constant.
That said, using
GENMASK_ULL(end - 1, 0)
in both cases makes also them look more similar (and opposition comes on how start is being used).
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 17 lines · Andy Shevchenko
Quoted 1 line · Dan Carpenter
On Tue, Sep 02, 2025 at 07:38:40AM +0000, Adrian Barnaś wrote:
...
Quoted 12 lines
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c typedef hive_uedge *hive_wide; · line 33
Quoted 6 lines
static inline
hive_uedge
inv_subword(hive_uedge w, unsigned int start, unsigned int end){drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +1
-1-return w & (~(((1ULL << (end -1 )) -1 ) <<1 |1 ) | ((1ULL << start) -1 )); +return w & (~__GENMASK_ULL (end-1 , start)); }nit: white space. Add spaces. Remove parentheses.
These are supposed to be opposites, right? Subword and inverse Subword. You could dress them up to make them look more opposite.
Quoted 6 lines
Quoted 2 lines
return (w & __GENMASK_ULL(end - 1, start)) >> start; return w & ~__GENMASK_ULL(end - 1, start);The problem is (and actually with the (end-1, start) above that it might generate a really bad code on some CPUs, so I really, really prefer the way when
_atleast_one of the parameters is constant.
Ah. Okay. That makes sense. Thanks.
regards,
dan carpenter
Quoted 2 lines
These are supposed to be opposites, right? Subword and inverse Subword. You could dress them up to make them look more opposite.
In fact this name is misleading as well. For me it should be named clear_subword(). It is zero-ing the "subword" provided with low and high bit boundary (start, end), to be then easily replaced with binary or operation.
And this operation is not an inverse of extracting a subword using the subword function.
Quoted 3 lines
The problem is (and actually with the (end-1, start) above that it might generate a really bad code on some CPUs, so I really, really prefer the way when
_atleast_one of the parameters is constant.
It would be easier to create a bitmask sing GENMASK_ULL and just reverse it but if it is not safe, this is what looks fine for me and works the same as previous function:
// Added a helper to create a mask
static inline unsigned long long subword_bitmask(unsigned int start, unsigned int end)
{
return GENMASK_ULL(end - 1 , 0) & ~(GENMASK_ULL(start, 0) >> 1);
}/* inverse subword bits move like this: MSB[xxxx____xxxx]LSB -> MSB[xxxx0000xxxx]LSB */
static inline unsigned long longclear_subword(unsigned long long w, unsigned int start, unsigned int end)
{
return w & ~subword_bitmask(start, end);
}Best regards
Adrian Barnaś
Quoted 1 line · Adrian Barnaś
Simplified masking logic in
pci/hive_isp_css_common/host/vmem.c.
...
Quoted 6 lines
---
I have tested this change on whole range of *valid* inputs, and it gives the same results as before, but this function seems to be little counter-intuitive as far as start is a (bit index) but end is (bit index + 1).
We can't change it without changing the callers at the same time. So, better to change just helpers and later on, if working, change the semantics of the parameter.
Quoted 1 line
---
...
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c typedef hive_uedge *hive_wide; · line 25
Quoted 4 lines
subword(hive_uedge w, unsigned int start, unsigned int end){drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +1
-1-return (w & (((1ULL << (end -1 )) -1 ) <<1 |1 )) >> start; +return (w &__GENMASK_ULL (end-1 ,0 )) >> start;
Why __ variant?
Quoted 1 line
}
...
drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c typedef hive_uedge *hive_wide; · line 33
Quoted 5 lines
inv_subword(hive_uedge w, unsigned int start, unsigned int end){drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c +1
-1-return w & (~(((1ULL << (end -1 )) -1 ) <<1 |1 ) | ((1ULL << start) -1 )); +return w & (~__GENMASK_ULL (end-1 , start)); }
Ditto.
Signature · 2 lines
With Best Regards, Andy Shevchenko