trans: always use a memcpy for ABI argument/return casts. #34141

eddyb · 2016-06-07T15:31:06Z

When storing incoming arguments or values returned by call/invoke, always do a memcpy from a temporary of the cast type, if there is an ABI cast.
While Clang has gotten smarter (store vs memcpy), a memcpy will always work.
This is what @dotdash has wanted to do all along, and it fixes #32049.

rust-highfive · 2016-06-07T15:31:21Z

r? @nrc

(rust_highfive has picked a reviewer for you, use r? to override)

eddyb · 2016-06-07T15:33:51Z

Nominated for backporting to beta because it reverts #33872's effect of triggering #32049 in the image crate. #33872 was already accepted, and this PR can be seen as a "more correct" alternative to that.

nikomatsakis · 2016-06-07T15:40:26Z

src/librustc_trans/abi.rs

+
+                // We instead thus allocate some scratch space...
+                let llscratch = AllocaFcx(bcx.fcx(), ty, "abi_cast");
+                base::Lifetime::End.call(bcx, llscratch);


shouldn't this be lifetime start?

nikomatsakis · 2016-06-07T15:44:35Z

@bors r+ p=1

bors · 2016-06-07T15:44:35Z

📌 Commit 2e68f31 has been approved by nikomatsakis

nikomatsakis · 2016-06-07T15:45:37Z

I'm inclined to beta-accept -- thoughts from @rust-lang/compiler team?

nagisa · 2016-06-07T15:59:52Z

I’m fine with it. My fix wasn’t supposed to be more than a short-term quick-hack anyway (for certain time-pressing reasons like trains rolling over the next day).

EDIT: or rather, please do accept it.

eddyb · 2016-06-07T16:04:58Z

@bors r=nikomatsakis p=2

I've rebased this on top of #34128 because it's likely to get in and because I won't be around to babysit it.

bors · 2016-06-07T16:04:59Z

📌 Commit 0d145c9 has been approved by nikomatsakis

dotdash · 2016-06-07T16:07:43Z

+1 on the backport

dotdash · 2016-06-07T16:11:39Z

src/librustc_trans/base.rs

-    })
+impl Lifetime {
+    pub fn call(self, b: &Builder, ptr: ValueRef) {
+        core_lifetime_emit(b.ccx, ptr, Lifetime::Start, |ccx, size, lifetime_intrinsic| {


I think this needs to be self instead of Lifetime::Start?

I... wrote this twice. It was fine the first time :/.

dotdash · 2016-06-07T16:14:43Z

Yep, that's much closer (not sure if 100% the same) to what I had in mind. Sorry for not getting that done myself earlier and causing some unnecessary churn :-/

That should also be useful for the outstanding transmute fix of mine that's still missing its MIR part.

Nice!

eddyb · 2016-06-07T16:17:57Z

@bors r=nikomatsakis

bors · 2016-06-07T16:17:58Z

📌 Commit d9f93b9 has been approved by nikomatsakis

bors · 2016-06-07T17:24:00Z

⌛ Testing commit d9f93b9 with merge ab5b316...

bors · 2016-06-07T18:50:29Z

💔 Test failed - auto-win-gnu-32-opt-rustbuild

arielb1 · 2016-06-07T19:05:25Z

Real failure:

C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\src/test\codegen\stores.rs:34:11: error: expected string not found in input
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[Y8]], i8* [[TMP8]], i64 4, i32 1, i1 false)
          ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\obj\build\i686-pc-windows-gnu\test\codegen\stores.ll:19:2: note: scanning from here
 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %4, i32 4, i32 1, i1 false)
 ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\obj\build\i686-pc-windows-gnu\test\codegen\stores.ll:19:2: note: with variable "Y8" equal to "%3"
 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %4, i32 4, i32 1, i1 false)
 ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\obj\build\i686-pc-windows-gnu\test\codegen\stores.ll:19:2: note: with variable "TMP8" equal to "%4"
 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %4, i32 4, i32 1, i1 false)
 ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\src/test\codegen\stores.rs:49:11: error: expected string not found in input
// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[Y8]], i8* [[TMP8]], i64 4, i32 1, i1 false)
          ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\obj\build\i686-pc-windows-gnu\test\codegen\stores.ll:50:2: note: scanning from here
 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %4, i32 4, i32 1, i1 false)
 ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\obj\build\i686-pc-windows-gnu\test\codegen\stores.ll:50:2: note: with variable "Y8" equal to "%3"
 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %4, i32 4, i32 1, i1 false)
 ^
C:\bot\slave\auto-win-gnu-32-opt-rustbuild\build\obj\build\i686-pc-windows-gnu\test\codegen\stores.ll:50:2: note: with variable "TMP8" equal to "%4"
 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %3, i8* %4, i32 4, i32 1, i1 false)

Difference: @llvm.memcpy.p0i8.p0i8.i64 vs. @llvm.memcpy.p0i8.p0i8.i32

eddyb · 2016-06-07T19:17:28Z

@arielb1 Who needs 32-bit anyway /s.

eddyb · 2016-06-07T21:35:26Z

@bors r=nikomatsakis

bors · 2016-06-07T21:35:26Z

📌 Commit e252865 has been approved by nikomatsakis

bors · 2016-06-07T21:45:39Z

⌛ Testing commit e252865 with merge ec872dc...

@dotdash

trans: always use a memcpy for ABI argument/return casts. When storing incoming arguments or values returned by call/invoke, always do a `memcpy` from a temporary of the cast type, if there is an ABI cast. While Clang has gotten smarter ([store](https://godbolt.org/g/EphFuK) vs [memcpy](https://godbolt.org/g/5dikH9)), a `memcpy` will always work. This is what @dotdash has wanted to do all along, and it fixes #32049.

bors · 2016-06-08T00:35:35Z

☀️ Test successful - auto-linux-32-nopt-t, auto-linux-32-opt, auto-linux-32cross-opt, auto-linux-64-cargotest, auto-linux-64-cross-armhf, auto-linux-64-cross-armsf, auto-linux-64-cross-freebsd, auto-linux-64-cross-netbsd, auto-linux-64-debug-opt, auto-linux-64-nopt-t, auto-linux-64-opt, auto-linux-64-opt-mir, auto-linux-64-opt-rustbuild, auto-linux-64-x-android-t, auto-linux-cross-opt, auto-linux-musl-64-opt, auto-mac-32-opt, auto-mac-64-nopt-t, auto-mac-64-opt, auto-mac-64-opt-rustbuild, auto-mac-cross-ios-opt, auto-win-gnu-32-nopt-t, auto-win-gnu-32-opt, auto-win-gnu-32-opt-rustbuild, auto-win-gnu-64-nopt-t, auto-win-gnu-64-opt, auto-win-msvc-32-cross-opt, auto-win-msvc-32-opt, auto-win-msvc-64-cargotest, auto-win-msvc-64-opt, auto-win-msvc-64-opt-mir, auto-win-msvc-64-opt-rustbuild

Backport #34141 to beta

rust-highfive assigned nrc Jun 7, 2016

eddyb added the beta-nominated Nominated for backporting to the compiler in the beta channel. label Jun 7, 2016

nikomatsakis reviewed Jun 7, 2016
View reviewed changes

eddyb force-pushed the trans-abi-memcpy branch from d43abf6 to 2e68f31 Compare June 7, 2016 15:41

eddyb force-pushed the trans-abi-memcpy branch from 2e68f31 to 0d145c9 Compare June 7, 2016 16:04

dotdash reviewed Jun 7, 2016
View reviewed changes

eddyb force-pushed the trans-abi-memcpy branch from 0d145c9 to d9f93b9 Compare June 7, 2016 16:17

trans: always use a memcpy for ABI argument/return casts.

e252865

eddyb force-pushed the trans-abi-memcpy branch from d9f93b9 to e252865 Compare June 7, 2016 21:35

bors merged commit e252865 into rust-lang:master Jun 8, 2016

eddyb deleted the trans-abi-memcpy branch June 8, 2016 00:36

arielb1 added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Jun 9, 2016

nikomatsakis added the beta-accepted Accepted for backporting to the compiler in the beta channel. label Jun 9, 2016

MagaTailor mentioned this pull request Jun 10, 2016

LTO segfaults in llvm #32272

Closed

alexcrichton mentioned this pull request Jun 28, 2016

Backport #34141 to beta #34533

Merged

brson removed the beta-nominated Nominated for backporting to the compiler in the beta channel. label Jun 28, 2016

brson added a commit that referenced this pull request Jun 28, 2016

Merge pull request #34533 from alexcrichton/beta-next

77c9c08

Backport #34141 to beta

trans: always use a memcpy for ABI argument/return casts. #34141

trans: always use a memcpy for ABI argument/return casts. #34141

Uh oh!

Conversation

eddyb commented Jun 7, 2016

Uh oh!

rust-highfive commented Jun 7, 2016

Uh oh!

eddyb commented Jun 7, 2016

Uh oh!

nikomatsakis Jun 7, 2016

Choose a reason for hiding this comment

Uh oh!

eddyb Jun 7, 2016

Choose a reason for hiding this comment

Uh oh!

nikomatsakis commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

nikomatsakis commented Jun 7, 2016

Uh oh!

nagisa commented Jun 7, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

eddyb commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

dotdash commented Jun 7, 2016

Uh oh!

dotdash Jun 7, 2016

Choose a reason for hiding this comment

Uh oh!

eddyb Jun 7, 2016

Choose a reason for hiding this comment

Uh oh!

dotdash commented Jun 7, 2016

Uh oh!

eddyb commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

arielb1 commented Jun 7, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

eddyb commented Jun 7, 2016

Uh oh!

eddyb commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

bors commented Jun 7, 2016

Uh oh!

bors commented Jun 8, 2016

Uh oh!

Uh oh!

nagisa commented Jun 7, 2016 •

edited

Loading

arielb1 commented Jun 7, 2016 •

edited

Loading