llvm dev - Mar 2019 - ORC JIT fails with standard math librrary

Hi,

I still can't get IR functions to JIT compile with the ORC JIT when they 
contain a call to the standard math library. Attached is a minimal exploit.

The program uses the KaleidoscopeJIT.h that ships with LLVM 8 (except 
that I had to expose the Datalayout). It reads from the filesystem an IR 
file (filename "func_works.ll" or "func_cos_fails.ll) and asks
the ORC
JIT first for the symbol "func_ir" and then for the address.

In case the file "func_works.ll" was read the program succeeds with:

symbol found!
address found!

In case the file "func_cos_fails.ll" was read the program fails with:

symbol found!
Failure value returned from cantFail wrapped call
UNREACHABLE executed at 
install/llvm-8.0-x86-debug/include/llvm/Support/Error.h:732!
Stack dump:

This was tested on an x86 Linux Ubuntu system.

Does anyone see what's going on or is it time to file a bug report?


Regards,

Frank




-------------- next part --------------
source_filename = "module"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"

declare float @cosf(float)

define float @func_ir(float %arg0) {
entrypoint:
  %0 = call float @cosf(float %arg0)
  ret float %0
}
-------------- next part --------------
source_filename = "module"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"


define float @func_ir(float %arg0) {
entrypoint:
  ret float %arg0
}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: KaleidoscopeJIT.h
Type: text/x-chdr
Size: 5065 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190326/10780d64/attachment.h>
-------------- next part --------------

LLVM_CONFIG = /home/fwinter/toolchain/install/llvm-8.0-x86-debug/bin/llvm-config

LLVM_CXXFLAGS = $(shell $(LLVM_CONFIG) --cxxflags)
LLVM_LDFLAGS  = $(shell $(LLVM_CONFIG) --ldflags)
LLVM_LIBS     = $(shell $(LLVM_CONFIG) --libs)

all: test

test.o: test.cc KaleidoscopeJIT.h
	g++ -c -o $@ $< $(LLVM_CXXFLAGS)

test: test.o
	g++ -o $@ $< $(LLVM_LDFLAGS) $(LLVM_LIBS)

clean:
	rm -f *.o
	rm -f test
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.cc
Type: text/x-c++src
Size: 2060 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190326/10780d64/attachment.cc>

I see the problem: The KaleidoscopeJIT in this example isn't searching
process symbols. This leads to a "symbol not found" failure when we
try to
look up math functions, but due to the way legacy ORC lookup works this
manifests as a "failure to materialize symbols" error when we try to
materialize a function that uses these symbols, rather than a plain "symbol
not found" error.

I should be able to fix this up tomorrow, and the issue will go away
entirely once the tutorials are moved on to concurrent ORC.

-- Lang.

On Tue, Mar 26, 2019 at 1:37 PM Frank Winter via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hi,
>
> I still can't get IR functions to JIT compile with the ORC JIT when
they
> contain a call to the standard math library. Attached is a minimal exploit.
>
> The program uses the KaleidoscopeJIT.h that ships with LLVM 8 (except
> that I had to expose the Datalayout). It reads from the filesystem an IR
> file (filename "func_works.ll" or "func_cos_fails.ll) and
asks the ORC
> JIT first for the symbol "func_ir" and then for the address.
>
> In case the file "func_works.ll" was read the program succeeds
with:
>
> symbol found!
> address found!
>
> In case the file "func_cos_fails.ll" was read the program fails
with:
>
> symbol found!
> Failure value returned from cantFail wrapped call
> UNREACHABLE executed at
> install/llvm-8.0-x86-debug/include/llvm/Support/Error.h:732!
> Stack dump:
>
> This was tested on an x86 Linux Ubuntu system.
>
> Does anyone see what's going on or is it time to file a bug report?
>
>
> Regards,
>
> Frank
>
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190506/800d6cd3/attachment.html>

Hi Frank,

It looks like this was fixed already in r354598. The solution was to
resolve symbols by calling findMangledSymbol (which searches the process
symbols via RTDyldMemoryManager::getSymbolAddressInProcess) instead of
ObjLayer.findSymbol (which only searches JIT'd code).

Cheers,
Lang.

On Mon, May 6, 2019 at 6:56 PM Lang Hames <lhames at gmail.com> wrote:
> I see the problem: The KaleidoscopeJIT in this example isn't searching
> process symbols. This leads to a "symbol not found" failure when
we try to
> look up math functions, but due to the way legacy ORC lookup works this
> manifests as a "failure to materialize symbols" error when we try
to
> materialize a function that uses these symbols, rather than a plain
"symbol
> not found" error.
>
> I should be able to fix this up tomorrow, and the issue will go away
> entirely once the tutorials are moved on to concurrent ORC.
>
> -- Lang.
>
> On Tue, Mar 26, 2019 at 1:37 PM Frank Winter via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> Hi,
>>
>> I still can't get IR functions to JIT compile with the ORC JIT when
they
>> contain a call to the standard math library. Attached is a minimal
>> exploit.
>>
>> The program uses the KaleidoscopeJIT.h that ships with LLVM 8 (except
>> that I had to expose the Datalayout). It reads from the filesystem an
IR
>> file (filename "func_works.ll" or "func_cos_fails.ll)
and asks the ORC
>> JIT first for the symbol "func_ir" and then for the address.
>>
>> In case the file "func_works.ll" was read the program
succeeds with:
>>
>> symbol found!
>> address found!
>>
>> In case the file "func_cos_fails.ll" was read the program
fails with:
>>
>> symbol found!
>> Failure value returned from cantFail wrapped call
>> UNREACHABLE executed at
>> install/llvm-8.0-x86-debug/include/llvm/Support/Error.h:732!
>> Stack dump:
>>
>> This was tested on an x86 Linux Ubuntu system.
>>
>> Does anyone see what's going on or is it time to file a bug report?
>>
>>
>> Regards,
>>
>> Frank
>>
>>
>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190508/1eb30cc3/attachment.html>