[GHC] #7198: New codegen more than doubles compile time of T3294

[GHC]

#7198: New codegen more than doubles compile time of T3294
-------------------------------------+-------------------------------------
        Reporter:  simonmar          |                Owner:  simonmar
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.4.2
  (CodeGen)                          |
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #4258             |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by michalt):

 So I was trying to learn more about the codegen using this ticket as the
 motivation and I think I’m missing something.

 I was looking at modifying `StgCmmBind.thunkCode` where we unconditionally
 load all the free variables into registers (before generating the code for
 the body of the thunk). Instead, I thought we could simply populate the
 environment with expressions containing the right memory accesses (using
 `addBindC`). My thinking was that: `node` is already a local register (so
 shouldn’t change), so anything relative to it should be stable and thus
 safe to use. Unfortunately, this segfaults `ghc-stage2` (i.e., `ghc-
 stage1` creates something bad). :-/

 I tried using `devel1` flavor but that didn't help find anything (i.e., no
 assertion failures). Trivial programs compiled by `ghc-stage1` seem to
 work. So it seems that GHC is large enough to trigger some silly bug in my
 code, but I’m a bit stuck at what that might be. Any ideas what I’m
 missing? I’m happy to dig further if you have no ideas (or time), but
 thought I’d ask if there’s something that’s obviously wrong with the idea
 or code. Thanks!

 Code: https://github.com/michalt/ghc/tree/t7198/1

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/7198#comment:11>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler