Thorkil Naur wrote:
On Thursday 02 August 2007 14:04, Thorkil Naur wrote:
On Thursday 02 August 2007 13:17, Simon Marlow wrote:
Thorkil Naur wrote:
This is about the x86 linux build: The error (" Prologue junk?: .globl
stg_ap_p_ret" etc.) is reported again in tonight's build
http://darcs.haskell.org/buildbot/head/tnaur%20x86%20Linux%20head/builds/41/step-stage1/0.
And, as mentioned below, the error disappears when unpulling all the way
back
past the pointer tagging patch. And it re-appears when the pointer
tagging
patch is pulled back in.
I honestly can't imagine how I introduced a bug in this area with the
pointer-tagging patch!
Can you generate AutoApply.raw_s before and after the pointer-tagging
patch, and send me the files please? Use -keep-raw-s-file, and presumably
you need to recompile stage1 after applying the patch, but not the
libraries.
Sure I will.
Here they are.
gcc seems to be putting these instructions in the prologue for certain
functions in AutoApply.hc:
pushl %ebp
movl %esp, %ebp
but it shouldn't be, because we're using -fomit-frame-pointer. I don't
understand why gcc is doing this for only some of the functions, e.g. it
happens for stg_ap_pppv_ret but not stg_ap_ppp_ret. Bizarre. I can't wait
until we can drop -fvia-C.
One workaround might be to compile AutoApply using -fasm. Can you check
whether this works for you?
If you have other versions of gcc available, could you test them too?
Cheers,
Simon
_______________________________________________
Cvs-ghc mailing list
Cvs-ghc@haskell.org
http://www.haskell.org/mailman/listinfo/cvs-ghc
