magnusuMET
commented
1 year ago | Perf assesment for a case with bomb of 100kt/1e6 particles | Pre PR (master) | This PR | impact (this PR/pre PR) |
|---|---|---|---|
| 02:54 | 02:59 | 103% (slower) |
Closed magnusuMET closed 1 year ago
magnusuMET
commented
1 year ago | Perf assesment for a case with bomb of 100kt/1e6 particles | Pre PR (master) | This PR | impact (this PR/pre PR) |
|---|---|---|---|
| 02:54 | 02:59 | 103% (slower) |
heikoklein
commented
1 year ago 3% performance reduction is acceptable. Do you know if this is mainly caused by the repeated 'float comparsion' vs bool-value in is_active, or is this general function-overhead (I would expect these functions are inlined)? General overhead would be acceptable, but if it is the float-comparison, I would revert to extra boolean.
magnusuMET
commented
1 year ago The inlining is very poor, does not seem to inline even within the same module for the ifort we use.