> In other words, the current implementation for --partial-loads-ok=yes
> models exactly the semantics of real machines:  It is OK to do an
> aligned load past the end of the block, provided you do not use the
> invalid data for anything.  If we just fix it for 16-byte loads -- and
> fix the validity bit propagation for, e.g., pmovmskb -- then all of
> these problems will go away.

Alas, we will also have to do 32 byte loads.  Maybe not right now, but
as soon as Intel's ever-growing instruction set sprouts 256-bit
PCMPxSTRx instructions, we will do.

The pmovmskb definedness thing should be fixed sooner, and is relatively
easy to do so -- add a new IR op that steers the data.  It already causes
problems not just in strlen code but also in SSE2 based graphics code, at
least the stuff that is in Firefox.

J

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Valgrind-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/valgrind-users

Reply via email to