<html dir="ltr"><head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body style="text-align:left; direction:ltr;"><div>Am Montag, den 04.11.2019, 17:28 +0100 schrieb Lukas Sommer:</div><blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<p>Hi Jonas,</p>
<p>thanks for your help!</p>
<p>I tried the suggested command, unfortunately it still fails:</p>
<p>
</p><blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">clang -fopenmp
-fopenmp-targets=x86_64-unknown-linux-gnu app.o</blockquote>
yields:
<p>
</p><blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">/usr/bin/ld: /tmp/app-c48732.o: relocation
R_X86_64_32 against `.data' can not be used when making a shared
object; recompile with -fPIC<br>
/usr/bin/ld: final link failed: nonrepresentable section on
output<br>
/usr/bin/ld: cannot find /tmp/app-23d932.out inside /<br>
/usr/bin/ld: cannot find /tmp/app-23d932.out inside /<br>
clang-9: error: linker command failed with exit code 1 (use -v
to see invocation)<br>
clang-9: error: linker command failed with exit code 1 (use -v
to see invocation)</blockquote>
I also tried to generate and bundle assembly (.asm) and run that
through clang (with openmp-flags as above), but that gave the same
error.<br></blockquote><div><br></div><div>Hmm, the target image is built as a shared library. To make this work, Clang adds -shared -fPIC in Generic_GCC::TranslateArgs.</div><div>You're replacing this step with a manual invocation of `opt', so we need to pass some extra flags. I'm currently looking into what's needed, opt --relocation-model=pic might be a good start but I'm not sure that's all.</div><div><br></div><div>Jonas</div><div><br></div><blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<p>Lukas<br>
</p>
<pre>Lukas Sommer, M.Sc.</pre><pre>TU Darmstadt</pre><pre>Embedded Systems and Applications Group (ESA)</pre><pre>Hochschulstr. 10, 64289 Darmstadt, Germany</pre><pre>Phone: +49 6151 1622429</pre><a class="moz-txt-link-abbreviated" href="http://www.esa.informatik.tu-darmstadt.de"><pre>www.esa.informatik.tu-darmstadt.de</pre></a>
<div class="moz-cite-prefix">On 11/4/19 5:01 PM, Jonas Hahnfeld
wrote:<br>
</div>
<blockquote type="cite" cite="mid:79be0a771d7cf25e2a3444808c5ebcef59856ded.camel@hahnjo.de" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<div>Hi Lukas,</div>
<div><br>
</div>
<div>Am Montag, den 28.10.2019, 13:42 +0100 schrieb Lukas Sommer
via Openmp-dev:</div>
<blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<pre>Hello,</pre>
<pre>I'm trying to use the clang-offload-bundler to merge the results of the</pre>
<pre>compilation for device and host for an application using OpenMP target</pre>
<pre>pragmas.</pre>
<pre>My goal is to run a custom pass on the device code and then bundle an</pre>
<pre>application from the modified device code and the host code. For testing</pre>
<pre>purposes, I use a very simple vector addition with target offloading</pre>
<pre>pragmas and the "offloading" to x86_64 (LLVM/Clang release 9.0).</pre>
<pre>I use clang to generate LLVM IR from the application code. Then I use</pre>
<pre>the clang-offload-bundler to unbundle the LLVM IR for device and host. I</pre>
<pre>run opt and llc separately on the LLVM IR for host and device.</pre>
<pre>To this end, on my application code (code.c), I run the following steps:</pre>
<blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<pre>clang -fopenmp -fopenmp-targets=x86_64-unknown-linux-gnu -S -emit-llvm</pre>
<pre>-O3 -o bundled.ll code.c</pre>
<pre>clang-offload-bundler --inputs=bundled.ll</pre>
<pre>--outputs=host.in.ll,device.in.ll</pre>
<pre>--targets=host-x86_64-unknown-linux-gnu,openmp-x86_64-unknown-linux-gnu</pre>
<pre>--type=ll --unbundle</pre>
<pre>opt -o host.opt.ll -O3 host.in.ll</pre>
<pre>llc -o host.o -filetype=obj host.opt.ll</pre>
<pre>opt -o device.opt.ll -O3 device.in.ll (custom pass should be added</pre>
<pre>here later on)</pre>
<pre>llc -o device.o -filetype=obj device.opt.ll</pre>
</blockquote>
<pre>Afterwards, I tried to bundle the object files for host & device:</pre>
<blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<pre>clang-offload-bundler --inputs=host.o,device.o --outputs=app.o</pre>
<pre>--targets=host-x86_64-unknown-linux-gnu,openmp-x86_64-unknown-linux-gnu</pre>
<pre>--type=o</pre>
</blockquote>
<pre>This command completes without any warning or error. </pre>
<pre>I tried to invoke gcc on the output to generate the executable from the</pre>
<pre>object file (app.o):</pre>
<blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<pre>gcc app.o -lomp -lomptarget</pre>
</blockquote>
<pre>However, I get the following error:</pre>
<blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<pre>/usr/bin/ld:</pre>
<pre>app.o:(.rodata..omp_offloading.device_images[.omp_offloading.descriptor_reg.x86_64-unknown-linux-gnu]+0x10):</pre>
<pre>undefined reference to `.omp_offloading.entries_begin'</pre>
<pre>/usr/bin/ld:</pre>
<pre>app.o:(.rodata..omp_offloading.device_images[.omp_offloading.descriptor_reg.x86_64-unknown-linux-gnu]+0x18):</pre>
<pre>undefined reference to `.omp_offloading.entries_end'</pre>
<pre>/usr/bin/ld:</pre>
<pre>app.o:(.rodata..omp_offloading.descriptor[.omp_offloading.descriptor_reg.x86_64-unknown-linux-gnu]+0x10):</pre>
<pre>undefined reference to `.omp_offloading.entries_begin'</pre>
<pre>/usr/bin/ld:</pre>
<pre>app.o:(.rodata..omp_offloading.descriptor[.omp_offloading.descriptor_reg.x86_64-unknown-linux-gnu]+0x18):</pre>
<pre>undefined reference to `.omp_offloading.entries_end'</pre>
</blockquote>
<pre>Do I need to run any additional steps before going through gcc or pass</pre>
<pre>additional flags? Is there another way of accomplishing the desired</pre>
<pre>behavior?</pre>
</blockquote>
<div><br>
</div>
<div>I think you should use clang with the full -fopenmp
-fopenmp-targets=x86_64-unknown-linux-gnu to build the final
executable, ie</div>
<div> $ clang -fopenmp -fopenmp-targets=x86_64-unknown-linux-gnu
app.o</div>
<div>If I remember correctly, the above symbols are provided by a
linker script in Clang 9.0, which gcc has no clue about.</div>
<div><br>
</div>
<div>Please note that there have been recent changes in that
area and trunk now relies on a new tool clang-offload-wrapper
AFAICS (see <a href="https://github.com/llvm/llvm-project/commit/5836c356fa6e17d0e10a2f9e0e111b7236dc15fb" moz-do-not-send="true">https://github.com/llvm/llvm-project/commit/5836c356fa6e17d0e10a2f9e0e111b7236dc15fb</a>).</div>
<div>As before this should be called transparently by clang, as
long as it's correctly invoked with a consistent value for
-fopenmp-targets.</div>
<div><br>
</div>
<div>I haven't tested above command, but I hope it works. Please
let me know if it doesn't and I'll take a closer look!</div>
<div>Jonas</div>
<div><br>
</div>
<blockquote type="cite" style="margin:0 0 0 .8ex; border-left:2px #729fcf solid;padding-left:1ex">
<pre>Thanks a lot in advance,</pre>
<pre>Best regards</pre>
<pre>Lukas</pre>
</blockquote>
</blockquote>
</blockquote></body></html>