Commits

Commits on May 14, 2021

scan: configurable number of runs
Oblomov
committedMay 14, 2021

Commits on May 13, 2021

reduction: configurable number of runs
Oblomov
committedMay 13, 2021

Commits on Mar 29, 2020

Use OpenCL keywords rather than equivalent bools
Oblomov
committedMar 29, 2020
reduction: be more verbose
Oblomov
committedMar 29, 2020

Commits on Mar 28, 2020

Update to support new headers
Oblomov
committedMar 28, 2020

Commits on Jul 31, 2016

scan: OP-based (and more compact) VECSCAN
Oblomov
committedJul 31, 2016
scan: typos and de-specialization
Oblomov
committedJul 31, 2016

Commits on Apr 13, 2016

More plugs
Oblomov
committedApr 13, 2016
common: plug leaks in read_file()
Oblomov
committedApr 13, 2016
scan: free host arrays at end
Oblomov
committedApr 13, 2016
common: check for error after getting each device info field
Oblomov
committedApr 13, 2016
Don't warn on deprecated declarations
Oblomov
committedApr 13, 2016

Commits on Dec 17, 2015

Reduction kernel support for more serial work per work-item in local part
Oblomov
committedDec 17, 2015
Use divisions insead of shifts
Oblomov
committedDec 17, 2015
VECREDUCE for othe ops than ADD is a TODO
Oblomov
committedDec 17, 2015

Commits on Nov 25, 2015

Scan: simplify global id increment in last step
Oblomov
committedNov 25, 2015
Scan: preload correction in last step
Oblomov
committedNov 25, 2015
Number of groups doesn't need to be a power-of-2
Oblomov
committedNov 25, 2015
Scan: reduce lmem usage
Oblomov
committedNov 25, 2015
Avoid unnecessary lmem usage
Oblomov
committedNov 25, 2015

Commits on Nov 24, 2015

Scan: use number of vector elements in local scan function
Oblomov
committedNov 24, 2015
Scan: operate on gi directly rather than offset
Oblomov
committedNov 24, 2015
Scan: smarter read/write lmem offset swap
Oblomov
committedNov 24, 2015
Default to 8 groups/CU
Oblomov
committedNov 24, 2015
Speed up last scan step with persistent groups and skipping the first
Oblomov
committedNov 24, 2015
Preliminary scan implementation
Oblomov
committedNov 24, 2015
Events should be in the specific source, not in common
Oblomov
committedNov 24, 2015
Even more refactoring
Oblomov
committedNov 24, 2015
More refactoring
Oblomov
committedNov 24, 2015
Move all preamble in a common header
Oblomov
committedNov 24, 2015

Commits on Oct 24, 2015

Print nice greppable summary
Oblomov
committedOct 24, 2015
Forgot to gobble the --unroll argument
Oblomov
committedOct 24, 2015

Commits on Oct 21, 2015

Use non-uniform init for sum test
Oblomov
committedOct 21, 2015
Oops, should read again after the unrolled reductions
Oblomov
committedOct 21, 2015
Option to unroll gmem loop
Oblomov
committedOct 21, 2015