Ondřej Hošek wrote:
Quick run-down:
1. Reports surface that the Orange Box SrcDS (OBSrcDS from here on)
fails on older processors.
2. Circumstantial evidence is shown that lack of SSE2 might lead to
these errors.
3. Chris Green (Valve) states that SSE2 is not required (only plain SSE,
available on P3s, is) as no compiler flags have been used to force it.
What a weird argument. It's not like the compiler is the only part that
can introduce SSEn.
Most likely it's some inline asm or SSE. intrinsics.
I really wonder how things like these can be hard to debug. And let's
assume for a moment
that it is actually caused by a missing command on the CPU, why are
there no checks in place?
You can't just go around and call random functions that aren't there :P
4. Chris returns with a message that he has been able to reproduce this
and will have to look down into the guts of the assembly code to get the
bug fixed. (No response from him since.)
5. Reports continue.
6. Alfred Reynolds (Valve, mailing list legend) says missing FCMOV
extensions mean epic fail.
7. I read Intel and AMD processor manuals. They say that if feature
flags "CMOV" and "FPU" are both enabled, the processor can do FCMOV.
8. People report that even though both flags are set on their machines,
bad bone weight errors still occur for them.
9. I write an assembler program that exits with zero if one of the FCMOV
instructions is executed, and, by definition, fails if the instruction
is unknown to the processor. People try it out on their machines, and
even though no Unknown Instruction error occurs, OBSrcDS still has bad
bone weights.
And that's where the story pretty much ends.
The culprit is probably lack of SSE2 and badly generated code. It's not
known if it's a Valve or a GCC bug.
~~ Ondra
On 06.11.07 20:35 Uhr, Some Llama wrote:
--
[ Picked text/plain from multipart/alternative ]
Hey All,
So I have a P3 linux server I run CS:S on and thought i'd dedicate it
to TF2, read through some of the old postings to his mailing list and
it looks like the P3 doesn't support some processor function the
newer TF2 engine needs in order to run on Linux?
Is that correct? Can someone please confirm?
I'm getting the bad bone weights messages along with the errors at
the end of the debug.log but looking through the postings it looks
like the conversation died off after talk of support for FCMOV?
Thanks for your time.
llama
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives,
please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
