Go: cmd/compile: delete the legacy compiler back end
When all the SSA backends are done, delete the legacy compiler.
This is basically a tracking bug for the work on all the SSA backends.
amd64 - done
arm - #15365
arm64 - #16009
ppc64 - #16010 #16175
randall77
All 32 comments
386 - #16358
randall77
on 13 Jul 2016
mips64, @cherrymui ?
bradfitz
on 13 Jul 2016
MIPS64- #16359
cherrymui
on 13 Jul 2016
amd64p32 is now using the SSA backend.
randall77
on 9 Aug 2016
386 (including GO386=387) is now using the SSA backend.
randall77
on 11 Aug 2016
ARM is done.
randall77
on 12 Aug 2016
MIPS64 is done.
All that is left is s390x and ppc64(be).
randall77
on 26 Aug 2016
ppc64be lacks only a test machine.
Last time it was tested, it worked, or so I hear.
On Fri, Aug 26, 2016 at 4:51 PM, Keith Randall [email protected]
wrote:
MIPS64 is done.
All that is left is s390x and ppc64(be).
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/golang/go/issues/16357#issuecomment-242848247, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AB1vJ4p97fG7FUUW5Ww24JH4c5n0oeUkks5qj1HLgaJpZM4JLvsd
.
dr2chase
on 26 Aug 2016
working on s390x
billotosyr
on 26 Aug 2016
arm64 is done.
randall77
on 31 Aug 2016
s390x is done #16677
We now have the greenlight to start deleting the legacy compiler.
Any suggestions on how to do the equivalent of GOSSAHASH when there is no longer a legacy compiler are welcome.
randall77
on 13 Sep 2016
@randall77, could GOSSAHASH remain but instead of using the old backend, mean to disable optimizations (like -N)?
bradfitz
on 13 Sep 2016
That would work if you're hacking on an SSA pass that is disabled with -N. That covers some cases but not all.
I want something like "clone the cmd/compile/internal/ssa package. Develop a CL on one of them. Use GOSSAHASH to decide per-function whether to use the orignal or the modified version".
Maybe include cmd/compile/internal/gc/ssa.go and cmd/compile/internal/ARCH/ssa.go in that clone/swap process.
randall77
on 13 Sep 2016
How about a top level flag (-hash=) and a support function (func hashmatch(name string) bool) that reports whether it matches. Then you can conditionally enable/disable any chunk of code you want, anywhere in the compiler.
josharian
on 13 Sep 2016
An individual phase would be pretty easy to hack in by hand, yes.
You just have to duplicate the code in question and add a flag check. It gets daunting when your change is across 3 packages.
randall77
on 13 Sep 2016
I've done Josh's method before, and I'm pretty happy with that method.
Some way of doing it wholesale old/new would be more than a little interesting.
dr2chase
on 13 Sep 2016
Back to the exciting topic of deleting the old backend, Keith, it seems to me you should have the honor of doing the first big del, if you want it.
josharian
on 13 Sep 2016
Another, much more complicated GOSSAHASH idea: Teach the linker to be able to combine two sources of object files. Using the known good compiler, compile everything and stash the object files. At link time, use GOSSAHASH to merge them. Wrap up behind a tool, maybe via -toolexec?
Sounds complicated, but maybe that's ok...or maybe it'll inspire a better idea in someone else.
josharian
on 13 Sep 2016
I am licking my lips.
On Tue, Sep 13, 2016 at 1:52 PM, Josh Bleecher Snyder <
[email protected]> wrote:
Back to the exciting topic of deleting the old backend, Keith, it seems to
me you should have the honor of doing the first big del, if you want it.—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/golang/go/issues/16357#issuecomment-246820040, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AGkgIDR1ieJsp6xBnmCKp9h35xlfsMROks5qpw0dgaJpZM4JLvsd
.
randall77
on 13 Sep 2016
Given that we export all the phases as function pointers, this sort of old/new switchiness is perhaps not that magical. We might want to be more disciplined in our connections between gc/ssa.go and the ssa package.
dr2chase
on 13 Sep 2016
I like Josh's idea. Kind of like a per-function toolstash. A hacky Go
program that takes two pkg directories and merges each .o/.a using hashes
of the function names to decide which one to take.
On Tue, Sep 13, 2016 at 2:00 PM, dr2chase [email protected] wrote:
Given that we export all the phases as function pointers, this sort of
old/new switchiness is perhaps not that magical. We might want to be more
disciplined in our connections between gc/ssa.go and the ssa package.—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/golang/go/issues/16357#issuecomment-246822287, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AGkgIKxnfZYjtqspnk5Dpsd6ved-lgXWks5qpw7QgaJpZM4JLvsd
.
randall77
on 13 Sep 2016
Ah, I misunderstood. Is that going to be easy/possible to embed within make.bash or go test?
dr2chase
on 13 Sep 2016
Ah, I misunderstood. Is that going to be easy/possible to embed within make.bash or go test?
You would presumably write a wrapper, like what buildall does for toolstash -cmp.
josharian
on 14 Sep 2016
CL https://golang.org/cl/29168 mentions this issue.
gopherbot
on 15 Sep 2016
More deletes for the record:
https://github.com/golang/go/commit/24965bc96ce2fd1483b6b8d5759b0ea129688758
https://github.com/golang/go/commit/ad7732a9acf6447e4b70845085263772ea788a6c
https://github.com/golang/go/commit/f03855f40ef131de6c0881ec12996a747be05a83
https://github.com/golang/go/commit/82703f84e4e2ab5b68d43825e0f046a5ad990612
bradfitz
on 15 Sep 2016
More:
https://github.com/golang/go/commit/f13701bf2f32fe48f6dcf9149b35b9ce3effe022
https://github.com/golang/go/commit/ae7e0ad7b8cdc16b6a7d5043375bb92a9cf6190a
https://github.com/golang/go/commit/1f2930cb5cf78cab648e39b9a0cee2f7a1e98d99
https://github.com/golang/go/commit/544010a05a90e45763ec9a8c149fc5137e1ec461
https://github.com/golang/go/commit/1bd91d4ccc57d3dbb2e5452c16ff6281d53e9763
https://github.com/golang/go/commit/9658de32c6141ca036a741006da5b83b4b2c6af5
/cc @mdempsky
bradfitz
on 16 Sep 2016
Should I start including and suggesting others to add "Updates #16357" to cleanup CLs related to this?
mdempsky
on 16 Sep 2016
Yes, can't hurt.
On Fri, Sep 16, 2016 at 10:44 AM, Matthew Dempsky [email protected]
wrote:
Should I start including and suggesting others to add "Updates #16357
https://github.com/golang/go/issues/16357" to cleanup CLs related to
this?—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/golang/go/issues/16357#issuecomment-247662542, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AGkgIEe62Rip0btCljPzVpoPNfcNBnxBks5qqtVjgaJpZM4JLvsd
.
randall77
on 16 Sep 2016
Roger roger.
mdempsky
on 16 Sep 2016
CL https://golang.org/cl/29350 mentions this issue.
gopherbot
on 16 Sep 2016
CL https://golang.org/cl/29219 mentions this issue.
gopherbot
on 22 Sep 2016
Declaring this done.
randall77
on 7 Oct 2016
Related issues
bbodenmiller
·
3Comments
lkarlslund
·
3Comments
jayhuang75
·
3Comments
Miserlou
·
3Comments
rsc
·
3Comments
Most helpful comment
arm64 is done.