Open delcypher opened 8 years ago
Note for shoc/s3d/qssab/kernel.bpl
the change appears to happen between snapshot-9 and snapshot-10. snapshot-10 adds support for non-aliasing symbolic indices which isn't supposed to matter for the GPU benchmarks (the pattern we used on worked on mathematical integers). Seeing as the time is close to the timeout it's likely that adding support for non-aliasing symbolic indicies added some overhead or that the machines I used were a little bit inconsistent in their performance.
Looks like for gpgpu-sim_ispass2009/RAY/renderPixel.bpl
and shoc/devicememory/readGlobalMemoryCoalesced/kernel.bpl
the regression happened between snapshot10-mapsymidx and snapshot-11-effcntclone
shoc/devicememory/readGlobalMemoryCoalesced/kernel.bpl
snapshot-10-mapsymidx/merged.yml : FinalResultType.FULLY_EXPLORED (/tmp/pbs.1302112.cx1b/wd/workdir-498) (607.015669389007 secs)
snapshot-11-effctclone/merged.yml : FinalResultType.UNKNOWN (/tmp/pbs.1296549.cx1b/wd/workdir-498) (900.0 secs)
gpgpu-sim_ispass2009/RAY/renderPixel.bpl
snapshot-10-mapsymidx/merged.yml : FinalResultType.FULLY_EXPLORED (/tmp/pbs.1296548.cx1b/wd/workdir-385) (14.141802517386774 secs)
snapshot-11-effctclone/merged.yml : FinalResultType.TIMED_OUT (/tmp/pbs.1296549.cx1b/wd/workdir-385) (900.0 secs)
Here are some notable changes for the fully explored benchmarks.