On 9/24/26 5:48 AM, Samarth Tandale via Sourceware Forge wrote:
From: Samarth Tandale <[email protected]>
Fold (a != 0) | (popcount(a) == n) into (a != 0) | (n == 0), and
(a == 0) & (popcount(a) != n) into (a == 0) & (n != 0).
This follows from x == 0 => popcount(x) == 0: when a == 0 the
popcount comparison reduces to a comparison against zero, and when
a != 0 the outer OR/AND already determines the result.
PR tree-optimization/112094
gcc/ChangeLog:
* match.pd: Add fold for (a != 0) | (popcount(a) == n) and
(a == 0) & (popcount(a) != n).
gcc/testsuite/ChangeLog:
* gcc.dg/fold-popcount-12.c: New test.
* gcc.dg/fold-popcount-13.c: New test.
Signed-off-by: Samarth Tandale <[email protected]>
---
Fold (a != 0) | (popcount(a) == n) into (a != 0) | (n == 0), and
(a == 0) & (popcount(a) != n) into (a == 0) & (n != 0).
This follows from x == 0 => popcount(x) == 0: when a == 0 the
popcount comparison reduces to a comparison against zero, and when
a != 0 the outer OR/AND already determines the result.
PR tree-optimization/112094
```
gcc/ChangeLog:
* match.pd: Add fold for (a != 0) | (popcount(a) == n) and
(a == 0) & (popcount(a) != n).
gcc/testsuite/ChangeLog:
* gcc.dg/fold-popcount-12.c: New test.
* gcc.dg/fold-popcount-13.c: New test.
```
Signed-off-by: Samarth Tandale <[email protected]>
This looks good to me (ignoring a couple potential whitespace nits).
But let's give 24hr for others to chime in.
Thanks!
jeff
