Some settings like PRECISE_F32 don't work in wasm in general (since wasm
has native f32s). That one just silently does nothing. Other flags that
don't work should throw an error. If not that's a bug.

If nothing seems obvious, if you can provide a testcase hitting that error
we can look into it.

On Tue, May 21, 2019 at 3:58 PM Soeren Balko <[email protected]> wrote:

> Thanks for pointing out that it's supposed to be __EMSCRIPTEN__, not
> EMSCRIPTEN. Will change that in our code base.
>
> Other than that, I did use the "latest-upstream" emsdk version. The only
> oddity is that I am not using the Emscripten version from emsdk, but a
> recent fork of Emscripten's incoming branch with some patches on top. None
> of these patches should affect the compiler backend though, most of them
> are in preamble.js.
>
> Is there anything else I could try to get the LLVM backend to work? Are
> some settings that I use (PRECISE_F32, DOUBLE_MODE, etc.) not available
> outside of fastcomp?
>
> Soeren
>
> On Wednesday, May 22, 2019 at 3:15:26 AM UTC+10, Alon Zakai wrote:
>>
>> This may be due to an out of date LLVM. You say you got LLVM from the
>> emsdk - did you get "latest-upstream"? The upstream version in the fastcomp
>> build is not new enough, in particular, only very recent LLVM from upstream
>> master will work (with emscripten incoming).
>>
>> Regarding that symbol, I think __EMSCRIPTEN__ is defined automatically.
>> It's possible fastcomp also defined it without underscores - if so, we
>> should probably either support both, or document this better.
>>
>>
>> On Mon, May 20, 2019 at 4:16 PM Soeren Balko <[email protected]> wrote:
>>
>>> I have some trouble getting the LLVM backend to work. My FFmpeg build
>>> links a couple of libraries, which works fine with the "fastcomp" backend.
>>> However, for some of the linked-in libraries (eg., libvpx, libopus),
>>> llvm-nm fails with:
>>>
>>>  llvm-nm failed on file libvpx.bc: return code 1, error:
>>> emsdk/upstream/6308/bin/llvm-nm: error: libvpx.bc Relocations not in offset
>>> order
>>>
>>> This causes wasm-ld to fail subsequently when linking the entire build.
>>> Other libraries (eg., libavfilter from the FFmpeg build) do not seem to
>>> exhibit this problem.
>>>
>>> My build settings are this:
>>>
>>> -s USE_OGG=1 \
>>>     -s USE_VORBIS=1 \
>>>     -s USE_ZLIB=1 \
>>>     -s PRECISE_F32=2 \
>>>     -s DOUBLE_MODE=0 \
>>>     -s ASSERTIONS=0 \
>>>     -s INVOKE_RUN=0 \
>>>     -s NODEJS_CATCH_EXIT=0 \
>>>     --memory-init-file 0
>>>
>>> Not sure what causes this and how to fix it. As a side note, I've used
>>> Emscripten from its incoming branch (commit id is
>>> 3a9424299cdda8157e0ccf38ebef951a1b7b5fe4), whereas LLVM is retrieved by
>>> means of emsdk. I normally build the "fastcomp" backend myself, cloning the
>>> emscripten-fastcomp and emscripten-fastcomp-clang repos and keeping them in
>>> lockstep with emscripten. However, I could not find the instructions on how
>>> to compile emscripten-fastcomp to include the LLVM backend (I tried adding
>>> "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=WebAssembly") to the cmake command,
>>> but that does not seem to generate the "wasm-ld" command (I get this: emcc:
>>> WASM_BACKEND selected but could not find lld (wasm-ld):
>>> /home/balko/dev/libav.js/emscripten-fastcomp/build/bin/wasm-ld).
>>>
>>> Other than that, I had to manually add a "-DEMSCRIPTEN=1" compiler
>>> define in order to make the "EMSCRIPTEN" symbol known throughout the code,
>>> which I found surprising.
>>>
>>> Any clues on how to get this to work from anyone who has successfully
>>> used the LLVM backend in a similar setting would be greatly appreciated.
>>>
>>> Thanks,
>>> Soeren
>>>
>>>
>>> --
>>> You received this message because you are subscribed to the Google
>>> Groups "emscripten-discuss" group.
>>> To unsubscribe from this group and stop receiving emails from it, send
>>> an email to [email protected].
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/emscripten-discuss/f3247662-0409-4d67-bdfc-d187fd659fbb%40googlegroups.com
>>> <https://groups.google.com/d/msgid/emscripten-discuss/f3247662-0409-4d67-bdfc-d187fd659fbb%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>> For more options, visit https://groups.google.com/d/optout.
>>>
>> --
> You received this message because you are subscribed to the Google Groups
> "emscripten-discuss" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/emscripten-discuss/0271178d-d75d-4b40-b3e2-85a897e2227b%40googlegroups.com
> <https://groups.google.com/d/msgid/emscripten-discuss/0271178d-d75d-4b40-b3e2-85a897e2227b%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>

-- 
You received this message because you are subscribed to the Google Groups 
"emscripten-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/emscripten-discuss/CAEX4NpQTgxQcRi-pOhHq83L7Dsx_c2qQzLVVc9GatYLNn6nv8g%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to