Experimental AVR target improvements #1

gergoerdi · 2017-05-07T05:09:52Z

Noob alert: Please review very carefully -- I haven't even looked at LLVM's internals until a week ago.

The commits should be fairly self-explanatory; the fixes are for the following issues:

"Relocation truncated to fit" from branches in long functions rust-legacy-fork#44
Switch lookup gets turned into a branch to a wrong address rust-legacy-fork#46
Switch lookup tables are not properly placed in program memory rust-legacy-fork#47

shepmaster · 2017-05-07T14:07:47Z

Thanks for the patches! I think our preference is to land everything in upstream LLVM before it finds its way into this repository, but this seems like a good place to track those patches before then (and also to allow @dylanmckay to grab them and review them)

Said another way: I think this PR can stay open, but not be merged until each of the commits has been merged into LLVM.

gergoerdi · 2017-05-07T14:11:16Z

Oh, I thought it was exactly the other way round -- that this repo is to stage dubious changes until they can be upstreamed to LLVM.

BTW, most definitely don't merge this PR, since the last three commits break structure unpacking: avr-rust/rust-legacy-fork#48

gergoerdi · 2017-05-11T11:15:19Z

This branch now has a ton of useful fixes with which I was able to compile a non-trivial Rust app to AVR (I hope to write a blog post about it shortly). Can I leave it in your hands to upstream it to LLVM proper?

shepmaster · 2017-05-11T15:40:45Z

@gergoerdi each separate fix to LLVM should have one or more tests associated with it; otherwise it's highly likely that any fixes will regress with time.

@dylanmckay any other things you'd like to see?

gergoerdi · 2017-05-12T02:12:04Z

@shepmaster OK I'll try to find the time to add tests for the new fixes. My question still stands though -- can you or @dylanmckay take care of the upstreaming? I'm not yet invested enough in this to want to take on that hassle.

gergoerdi · 2017-05-12T02:22:01Z

Also, I think there's three categories of my commits here:

Definite bugfixes, no reason not to merge them:

c7d5217 (Final continuation jump missing from MBB rust-legacy-fork#49)
9fe62d7 (CPI, and who knows what other instructions, are defined with wrong operand registers rust-legacy-fork#50)
910429c (Potential bug in shl operator implementation rust-legacy-fork#52)

Definite improvement, but implementation could be better

383e429 (Switch lookup tables are not properly placed in program memory rust-legacy-fork#47)

Desperate struggling piling hacks upon hacks fighting the Harvard architecture

I don't think commits from the third category should be merged; I'll open a separate ticket where we can hopefully come up with a proper solution to PROGMEM-related issues.

dylanmckay · 2017-05-13T00:04:19Z

I'm happy to take on upstreaming @gergoerdi, will keep you updated.

dylanmckay · 2017-05-11T22:29:26Z

lib/Target/AVR/AVRInstrInfo.td

@@ -2057,6 +2109,10 @@ def : Pat<(store i8:$src, (i16 (AVRWrapper tglobaladdr:$dst))),
 def : Pat<(store i16:$src, (i16 (AVRWrapper tglobaladdr:$dst))),
          (STSWKRr tglobaladdr:$dst, i16:$src)>;

+// Temporary workaround -- this one is BOGUS!


Why is this temporary? What is it working around? Can you pop a comment somewhere with an explanation?

… same spot Contributed by Dr. Gergő Érdi. Fixes a bug. Raised from (avr-rust/rust-legacy-fork#49). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@302973 91177308-0d34-0410-b5e6-96231b3b80d8

(avr-rust/rust-legacy-fork#50) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304283 91177308-0d34-0410-b5e6-96231b3b80d8

When generating code for a shift loop, check the shift amount against the literal value 0, not R0 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304284 91177308-0d34-0410-b5e6-96231b3b80d8

Summary: CPI does not read the status register, but only writes it. Reviewers: dylanmckay Reviewed By: dylanmckay Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D33223 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@304116 91177308-0d34-0410-b5e6-96231b3b80d8

…e same spot as the original MBB (avr-rust/rust-legacy-fork#62)

(avr-rust/rust-legacy-fork#63 avr-rust/rust-legacy-fork#64) [AVR] Use 1-byte alignment for all data types (avr-rust/rust-legacy-fork#64)

For now, these relaxed branches are used unconditionally, even when the branch target is nearby. avr-rust/rust-legacy-fork#44

FIXME: implementation is mostly copy-pasted from LDWRdPtr, so we should refactor a bit and unify the two

LPMRdZ (avr-rust/rust-legacy-fork#47, avr-rust/rust-legacy-fork#48)

to ProgMem

avr-rust/rust-legacy-fork#58

dylanmckay · 2017-10-04T09:27:18Z

It looks like there's some stuff here that still needs to be merged

I will change the base branch to see more clearly what is missing that should go into master

dylanmckay · 2017-10-04T09:28:34Z

That didn't help, the histories are different!

dylanmckay · 2017-10-04T09:36:12Z

Stuff to merge

"[AVR] Elaborate LDWRdPtr into ld r, X++; ld r+1, X"
Grab mayLoad change from the diff out of "[AVR] Add ISel pattern which maches load-from-global, compile it to LPMRdZ"
"[AVR] Implement LPMWRdZ pseudo-instruction's expansion."

I believe everything else is already upstreamed, or were temporary hacks

dylanmckay · 2017-10-04T09:51:30Z

Ran tests locally, all pass

dylanmckay · 2017-10-04T10:12:59Z

Wrong merge request! did not mean to merge, nor did I run tests for this PR.

Reopening now, have reverted

dylanmckay · 2017-10-04T10:13:43Z

Looks like there's no reopen button, will just cherry-pick stuff now

dylanmckay · 2017-10-04T10:35:59Z

Upstreamed ""[AVR] Elaborate LDWRdPtr into ld r, X++; ld r+1, X"" in r314896.

I also fixed a bunch of tests before committing

dylanmckay · 2017-10-04T10:38:11Z

Upsteamed mayLoad factoring out in r314897

dylanmckay · 2017-10-04T10:39:34Z

Upsteamed "[AVR] Implement LPMWRdZ pseudo-instruction's expansion." in r314898

dylanmckay · 2017-10-04T10:44:56Z

Cherry-picked the upstreamed commits to avr-rust/llvm in

06d0f6b [AVR] Implement LPMWRdZ pseudo-instruction's expansion.
f6d00a7 [AVR] Factor out mayLoad in tablegen patterns
d4274bd [AVR] Elaborate LDWRdPtr into ld r, X++; ld r+1, X

dylanmckay · 2017-10-04T10:46:41Z

Updated version of LLVM in for Rust in avr-rust/rust-legacy-fork@a22dd46

dylanmckay · 2017-10-04T10:49:09Z

Added these commits to the current AVR 5.0.1 merge request PR34829

dylanmckay · 2017-10-04T10:52:25Z

Have double checked all the commits in this PR and everything looks good.

Thanks for the patches @gergoerdi, sorry for taking so long, completely forgot about this PR!

gergoerdi force-pushed the avr-rust-experiments branch 2 times, most recently from 05e9b65 to 5e80213 Compare May 7, 2017 06:32

gergoerdi force-pushed the avr-rust-experiments branch 2 times, most recently from bf415da to 9fe62d7 Compare May 10, 2017 07:44

gergoerdi force-pushed the avr-rust-experiments branch from 65f3f2a to 4e1e3f0 Compare May 12, 2017 12:40

dylanmckay reviewed May 13, 2017

View reviewed changes

gergoerdi force-pushed the avr-rust-experiments branch 2 times, most recently from 9b6cd32 to f4d1897 Compare May 20, 2017 16:12

gergoerdi force-pushed the avr-rust-experiments branch 2 times, most recently from 5e0418c to e65b858 Compare May 28, 2017 10:40

dylanmckay and others added 12 commits May 31, 2017 18:26

[AVR] Implement LPMWRdZ pseudo-instruction's expansion.

cb05e5a

FIXME: implementation is mostly copy-pasted from LDWRdPtr, so we should refactor a bit and unify the two

[AVR] Add ISel pattern which maches load-from-global, compile it to

2bd500f

LPMRdZ (avr-rust/rust-legacy-fork#47, avr-rust/rust-legacy-fork#48)

[AVR] Add 16-bit version of load-from-global

afb23e5

[AVR] Use separate ProgMemWrapper for GlobalAddresses of constants

d2a94a0

#TEMP add temporary instrument selector pattern for creating pointers

e71bfa6

to ProgMem

[AVR] Elaborate LDWRdPtr into ld r, X++; ld r+1, X

f3dcdf8

avr-rust/rust-legacy-fork#58

gergoerdi force-pushed the avr-rust-experiments branch from e65b858 to f3dcdf8 Compare June 17, 2017 02:03

dylanmckay changed the base branch from avr-rust-2017-05-03 to avr-rust-llvm-release-4-0-1 October 4, 2017 09:27

dylanmckay changed the base branch from avr-rust-llvm-release-4-0-1 to avr-rust-2017-05-03 October 4, 2017 09:28

dylanmckay merged commit 634f5a7 into avr-rust:avr-rust-2017-05-03 Oct 4, 2017

dylanmckay mentioned this pull request Oct 4, 2017

Revert "Experimental AVR target improvements" #7

Merged

dylanmckay added a commit to avr-rust/rust-legacy-fork that referenced this pull request Oct 4, 2017

Cherry-pick fixes from avr-rust/llvm#1

a22dd46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Experimental AVR target improvements #1

Experimental AVR target improvements #1

gergoerdi commented May 7, 2017

shepmaster commented May 7, 2017

gergoerdi commented May 7, 2017

gergoerdi commented May 11, 2017

shepmaster commented May 11, 2017

gergoerdi commented May 12, 2017

gergoerdi commented May 12, 2017 •

edited by dylanmckay

Loading

dylanmckay commented May 13, 2017

dylanmckay May 11, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017 •

edited

Loading

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017 •

edited

Loading

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

Experimental AVR target improvements #1

Experimental AVR target improvements #1

Conversation

gergoerdi commented May 7, 2017

shepmaster commented May 7, 2017

gergoerdi commented May 7, 2017

gergoerdi commented May 11, 2017

shepmaster commented May 11, 2017

gergoerdi commented May 12, 2017

gergoerdi commented May 12, 2017 • edited by dylanmckay Loading

dylanmckay commented May 13, 2017

dylanmckay May 11, 2017

Choose a reason for hiding this comment

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017 • edited Loading

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017 • edited Loading

dylanmckay commented Oct 4, 2017

dylanmckay commented Oct 4, 2017

gergoerdi commented May 12, 2017 •

edited by dylanmckay

Loading

dylanmckay commented Oct 4, 2017 •

edited

Loading

dylanmckay commented Oct 4, 2017 •

edited

Loading