On 2013/09/10 16:47:34, Yang wrote:
I don't see results being verified in either of the benchmarks. Why is
this
necessary?
Some SunSpider benchmarks do internal data validation, for instance lines
171-173 of access-nbody.js.
If I run the test like this on Ubuntu 12.04 LTS:
$ make arm.release -j8
$ tools/run-tests.py -j8 --arch=arm --mode=release --timeout=240 --time
--no-presubmit --report benchmarks/sunspider
I get 3 errors, the first one is this:
=== benchmarks/sunspider/access-nbody ===
/home/paul/v8/gv8/test/benchmarks/sunspider/access-nbody.js:173: ERROR: bad
result: expected -1.3524862408537381 but got -1.352486240853738
throw "ERROR: bad result: expected " + expected + " but got " + ret;
The problem is that the x86 defaults to 80-bit precision for results that
stay
inside the FP unit. So we do not get the answer expected if all operations
had
64-bit precision as they do on real ARM or MIPS cores.
Note that this does NOT happen on Mac running 10.8.4, so apparently the FPU
defaults are different on Linux & OS-X. We could put this change behind a
platform check, I suppose.
https://codereview.chromium.org/24067002/
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
