llvm dev - Sep 2010 - [LLVMdev] Fwd: bitcode / bytecode

On Sep 29, 2010, at 3:12 PM, Bill Wendling wrote:
> Chris rewrote the original LLVM bytecode into it's present bitcode
form. The newer code is much smaller than the previous form. That would be my
guess for why he chose to call it bitcode. Of course, he can say for himself.
:-)
The format is literally a bitstream, not a bytestream.  See this document for
more details:
http://llvm.org/docs/BitCodeFormat.html

-Chris

Am 30.09.2010 00:34, schrieb Chris Lattner:
>
> The format is literally a bitstream, not a bytestream.  See this document
for more details:
> http://llvm.org/docs/BitCodeFormat.html
Ah... something I always meant to ask: How does the LLVM infrastructure 
deal with the added overhead of packing/unpacking a bitstream?
Ignore it since modern CPUs can deal with that?
Some ingenious algorithm?

Last time I heard about a bitstream format was with the i432, where 
hindsight lore has it that bitstreams are just too complicated to 
maintain, but I guess some things might have changed since then.

Regards,
Jo

On Sep 30, 2010, at 5:26 AM, Joachim Durchholz wrote:
> Am 30.09.2010 00:34, schrieb Chris Lattner:
>> 
>> The format is literally a bitstream, not a bytestream.  See this
document for more details:
>> http://llvm.org/docs/BitCodeFormat.html
> 
> Ah... something I always meant to ask: How does the LLVM infrastructure 
> deal with the added overhead of packing/unpacking a bitstream?
> Ignore it since modern CPUs can deal with that?
> Some ingenious algorithm?
We don't do anything tricky, feel free to look at the source in
llvm/lib/Bitcode.

-Chris