https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121536
Tamar Christina <tnfchris at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Last reconfirmed| |2025-08-13
CC| |tnfchris at gcc dot gnu.org
Status|UNCONFIRMED |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |tnfchris at gcc dot
gnu.org
--- Comment #1 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
The main loop originally picks VNx16QI
and the costing of them looks the same:
: note: Original vector body cost = 6
: note: Vector loop iterates at most 0 times
: note: Scalar issue estimate:
: note: load operations = 0
: note: store operations = 0
: note: general operations = 2
: note: reduction latency = 2
: note: estimated min cycles per iteration = 2.000000
: note: estimated cycles per vector iteration (for VF 2) = 4.000000
: note: SVE issue estimate:
: note: load operations = 0
: note: store operations = 0
: note: general operations = 2
: note: predicate operations = 2
: note: reduction latency = 4
: note: estimated cycles per iteration to rename = 1.000000
: note: estimated min cycles per iteration without predication = 4.000000
: note: estimated min cycles per iteration for predication = 1.000000
: note: estimated min cycles per iteration = 4.000000
: note: Cost model analysis:
Vector inside of loop cost: 6
Vector prologue cost: 2
Vector epilogue cost: 6
Scalar iteration cost: 2
Scalar outside cost: 2
Vector outside cost: 8
prologue iterations: 0
epilogue iterations: 0
vs
: note: Original vector body cost = 6
: note: Vector loop iterates at most 0 times
: note: Scalar issue estimate:
: note: load operations = 0
: note: store operations = 0
: note: general operations = 2
: note: reduction latency = 2
: note: estimated min cycles per iteration = 2.000000
: note: estimated cycles per vector iteration (for VF 2) = 4.000000
: note: SVE issue estimate:
: note: load operations = 0
: note: store operations = 0
: note: general operations = 2
: note: predicate operations = 2
: note: reduction latency = 4
: note: estimated cycles per iteration to rename = 1.000000
: note: estimated min cycles per iteration without predication = 4.000000
: note: estimated min cycles per iteration for predication = 1.000000
: note: estimated min cycles per iteration = 4.000000
: note: Cost model analysis:
Vector inside of loop cost: 6
Vector prologue cost: 2
Vector epilogue cost: 6
Scalar iteration cost: 4
Scalar outside cost: 2
Vector outside cost: 8
prologue iterations: 0
epilogue iterations: 0
Minimum number of vector iterations: 4
Calculated minimum iters for profitability: 8
but the conclusion is different.
The old code still thinks the vector loop iterates 4 times whereas the new one
thinks it doesn't...
But this seems like a backend issue so I'll take a look.
Mine.
