[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Lightning] sse instructions and gcc warnings

From: Paulo César Pereira de Andrade
Subject: Re: [Lightning] sse instructions and gcc warnings
Date: Fri, 30 Jul 2010 19:08:59 -0300
User-agent: SquirrelMail/1.4.19

Paulo César Pereira de Andrade wrote:
> Paulo César Pereira de Andrade wrote:
>>   Updated version attached, to check load and store. Tested in i686
>> and latest git, pass all tests, no output (no news is good news :-)
>   Sorry for replying to myself, but I updated the test to allow using
> any register as index, pointer or data, as I actually use it this way
> in some cases in my current code, as the only distinction should be
> caller or callee save I believe. And now there are failures, ending
> with a crash on i686. If the tests are wrong please let me know...
>   Now it is truly brute force testing all combinations for load/store :-)
>   Attached as a gzip file now to avoid waste of bandwidth.

  Updated testcase to pass unsigned long and double arguments as
pointers in x86_64 to have the stack aligned (I am not sure how it is
expected to work as arguments on stack in x86_64, possibly another test
case for later :-). In this test, in x86_64, if #define'ing sign_extend
to 0 [1], it will pass all tests, but i686 still crashes on
"V0 R0 V1 V2 F0", with failures on previous tests.

  Later on I can write some other "brute force" test on all combinations
for other operations, as the full set of tests should take like less
then one second to run, but possibly a few seconds to compile...


[1] But I would really prefer to have it properly sign extending, or,
have something like jit_ldxl_c "matching" jit_ldxi_c, or some other
approach to tell it to use 64 bits when loading a small int, so that
store of long would work as expected for negative values.

Attachment: lightning-test.c.gz
Description: GNU Zip compressed data

reply via email to

[Prev in Thread] Current Thread [Next in Thread]