llvm/llvm/docs/HistoricalNotes/2000-12-06-EncodingIdea.txt

From: Chris Lattner [mailto:[email protected]]
Sent: Wednesday, December 06, 2000 6:41 PM
To: Vikram S. Adve
Subject: Additional idea with respect to encoding

Here's another idea with respect to keeping the common case instruction
size down (less than 32 bits ideally):

Instead of encoding an instruction to operate on two register numbers,
have it operate on two negative offsets based on the current register
number.  Therefore, instead of using:

r57 = add r55, r56  (r57 is the implicit dest register, of course)

We could use:

r57 = add -2, -1

My guess is that most SSA references are to recent values (especially if
they correspond to expressions like (x+y*z+p*q/ ...), so the negative
numbers would tend to stay small, even at the end of the procedure (where
the implicit register destination number could be quite large).  Of course
the negative sign is reduntant, so you would be storing small integers
almost all of the time, and 5-6 bits worth of register number would be
plenty for most cases...

What do you think?

-Chris