Separate out EH handler lowering into its own pass #21720
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
kshyatt
added
error handling
ararslan
reviewed
May 6, 2017
src/llvm-lower-handlers.cpp
Outdated
| @@ -0,0 +1,225 @@ | |||
| // This file is a part of Julia. License is MIT: http://julialang.org/license | |||
yuyichao
approved these changes
May 6, 2017
JeffBezanson
added
compiler:codegen
|
that segfault during bootstrap on win64 appveyor looks legit |
|
Indeed. I'll take a look. |
Currently it's being done as part of the GC root lowering pass. However, I'm working on replacing that pass, so this functionality needed to be separated out anyway. While I'm here: 1. Simplify the algorithm to a simple DFS numbering 2. Have it insert lifetime intrinisics to allow the optimizer to do stack slot coloring later 3. Add a test for this pass
|
Fixed, I think. |
yuyichao
approved these changes
May 15, 2017
|
This seems to be causing segfaults when compiling on FreeBSD. |
|
Do you have a backtrace? |
|
Can you attach a debugger? "make VERBOSE=1" will show you the exact command it was trying to run. |
|
The warning at https://gist.github.com/ararslan/392feece5a054804c57ed181fa6cfad0#file-backtrace-L62 seems real btw |
|
Regarding the warning, I'm getting that on macOS as well. Perhaps the function should have a I can run make in |
|
Yes, it's missing a |
|
The backtrace almost looks like the compiler assume the return will not be reachable and let the control flow fall into the next function, which might be a valid optimization according to C++ standard... Should |
|
@ararslan can you add that |
|
Just adding the |
|
It seems that that's actually the code clang produce for int g1(bool);
bool f(bool a)
{
if (a)
return false;
g1(a);
}
bool k()
{
return true;
}_Z1fb: # @_Z1fb
.cfi_startproc
# BB#0:
pushq %rax
.Lcfi0:
.cfi_def_cfa_offset 16
testb %dil, %dil
je .LBB0_2
# BB#1:
xorl %eax, %eax
popq %rcx
retq
.LBB0_2:
xorl %edi, %edi
callq _Z2g1b
.Lfunc_end0:
.size _Z1fb, .Lfunc_end0-_Z1fb
.cfi_endproc
.globl _Z1kv
.p2align 4, 0x90
.type _Z1kv,@function
_Z1kv: # @_Z1kv
.cfi_startproc
# BB#0:
movb $1, %al
retq
.Lfunc_end1:
.size _Z1kv, .Lfunc_end1-_Z1kv
.cfi_endproc |
|
Cool. A little surprised that this doesn't trigger a crash on macOS.... |
|
Huh, I was under the impression that it'd just do a |
|
That's what gcc does..... Apparently llvm (4.0 at least) doesn't....... not even a |
|
Actually, it does emit a |
|
@Keno Bisection suggests that this is causing an assertion error with LLVM-svn. cc @ranjanan |
|
Yes, I had already given @ranjanan a patch for this. |
|
Yes but after building I got another segfault at Base.precompile Perhaps it's something @yuyichao already mentioned in #21879 |
|
Yes it is. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Currently it's being done as part of the GC root lowering pass. However, I'm working
on replacing that pass, so this functionality needed to be separated out anyway.
While I'm here: