In the s318 and s3113 kernels, the integer variant of the abs() function was to determined the absolute value. It operates on a float value, and the result is stored in a variable max which has the float type, which makes me think the intention here was to use fabs() variant, not the integer abs() variant.
Details
Diff Detail
- Repository
- rT test-suite
Event Timeline
MultiSource/Benchmarks/TSVC/types.h | ||
---|---|---|
12 | This is not entirely ideal to always use the double precision variant in the else-case, but we only generate float and double variants of the kernels (not integers), so didn't feel like making it more complicated than this. |
LGTM, thanks!
MultiSource/Benchmarks/TSVC/tsc.inc | ||
---|---|---|
3778 | UFF! This should really be CLOCKS_PER_SEC... But it's ok to make that a sweeping mechanical change on another patch. |
MultiSource/Benchmarks/TSVC/tsc.inc | ||
---|---|---|
3778 | Wowsers!!! Good catch. But while I am at it, I could do the find and replace as a follow up. |
UFF! This should really be CLOCKS_PER_SEC... But it's ok to make that a sweeping mechanical change on another patch.