[Bug 1545479] Review Request: hcc - Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479

Elliott Sales de Andrade  changed:

   What|Removed |Added

 Status|POST|CLOSED
 CC||quantum.anal...@gmail.com
   Fixed In Version||hcc-1.3.18505-2.rocm2.0.0.f
   ||c31
 Resolution|--- |CURRENTRELEASE
Last Closed||2020-08-03 00:36:45



--- Comment #25 from Elliott Sales de Andrade  ---
Please close your Review Requests when they are complete.


-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: 
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc - Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479
Bug 1545479 depends on bug 1666836, which changed state.

Bug 1666836 Summary: Review Request: rocminfo -  ROCm system info utility
https://bugzilla.redhat.com/show_bug.cgi?id=1666836

   What|Removed |Added

 Status|ASSIGNED|CLOSED
 Resolution|--- |RAWHIDE



-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: 
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc - Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #24 from Gwyn Ciesla  ---
(fedscm-admin):  The Pagure repository was created at
https://src.fedoraproject.org/rpms/hcc

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc - Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479

Tom Stellard  changed:

   What|Removed |Added

Summary|Review Request: hcc-|Review Request: hcc -
   |Heterogeneous C++ Compiler  |Heterogeneous C++ Compiler



-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479

Robert-André Mauchin  changed:

   What|Removed |Added

 Status|NEW |POST
   Assignee|nob...@fedoraproject.org|zebo...@gmail.com
  Flags||fedora-review+



--- Comment #23 from Robert-André Mauchin  ---
LGTM, package is approved.






Package Review
==

Legend:
[x] = Pass, [!] = Fail, [-] = Not applicable, [?] = Not evaluated
[ ] = Manual review needed


Issues:
===
- Header files in -devel subpackage, if present.
  Note: hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_builtin_vars.h
  hcc : /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_cmath.h hcc
  :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_complex_builtins.h
  hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_device_functions.h
  hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_intrinsics.h hcc
  :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_libdevice_declares.h
  hcc :
 
/usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_math_forward_declares.h
  hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__clang_cuda_runtime_wrapper.h
  hcc : /usr/libexec/hcc/lib64/clang/8.0.0/include/__stddef_max_align_t.h
  hcc : /usr/libexec/hcc/lib64/clang/8.0.0/include/__wmmintrin_aes.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/__wmmintrin_pclmul.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/adxintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/altivec.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/ammintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/arm64intr.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/arm_acle.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/arm_fp16.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/arm_neon.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/armintr.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx2intrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512bitalgintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512bwintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512cdintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512dqintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512erintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512fintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512ifmaintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512ifmavlintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512pfintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vbmi2intrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vbmiintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vbmivlintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vlbitalgintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vlbwintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vlcdintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vldqintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vlintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vlvbmi2intrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vlvnniintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vnniintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vpopcntdqintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/avx512vpopcntdqvlintrin.h hcc
  : /usr/libexec/hcc/lib64/clang/8.0.0/include/avxintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/bmi2intrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/bmiintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/cetintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/cldemoteintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/clflushoptintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/clwbintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/clzerointrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/cpuid.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/emmintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/f16cintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/float.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/fma4intrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/fmaintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/fxsrintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/gfniintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/htmintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/htmxlintrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/ia32intrin.h hcc :
  /usr/libexec/hcc/lib64/clang/8.0.0/include/immintrin.h hcc :
  

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #22 from Tom Stellard  ---
(In reply to Felix Schwarz from comment #17)
> Two more things: Tom, some of your SRPM URLs are 404.
> 

Yes, COPR cleans up the older builds after a certain amount of time.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #21 from Tom Stellard  ---
(In reply to Felix Schwarz from comment #16)
> (In reply to Robert-André Mauchin from comment #14)
> >  - %{_includedir}/* in %files
> > 
> > Includes should go to a devel package
> > 
> > %{_libdir}/cmake/hcc/ too.
> 
> Not sure if I understood you correctly but hcc is basically a compiler. As
> far as I know there are no outside users of these header files so probably
> it does not make sense to split these in a -devel package.
> 
> (In reply to Robert-André Mauchin from comment #15)
> > This takes forever to compile.
> 
> It builds its bundled copy of llvm :-) Takes about 45 minutes on my machine
> (from the top of my head, might not be accurate). However you need to kill
> the "licensecheck" process when using fedora-review. That script will take
> many hours (only using a single core).
> 
> Btw Tom: I noticed that AMD just deprecated HCC with the ROCm 2.2 release:
> > Deprecation Notice
> >
> > AMD is deprecating HCC to put more focus on HIP development and on other 
> > languages supporting
> > heterogeneous compute. We will no longer develop any new feature in HCC and 
> > we will stop
> > maintaining HCC after its final release, which is planned for June 2019. If 
> > your application
> > was developed with the hc C++ API, we would encourage you to transition it 
> > to other languages
> > supported by AMD, such as HIP or OpenCL. HIP and hc language share the same 
> > compiler technology,
> > so many hc kernel language features (including inline assembly) are also 
> > available through
> > the HIP compilation path.
> https://github.com/RadeonOpenCompute/hcc/
> 
> I assume it still makes sense because hcc is needed right now for the rest
> of the ROCm stack? (Also the compiler code needs to go somewhere so either
> they move it into hip or keep hcc around but only as an internal component?)

It is still a build dependency for HIP, which is really the only reason why I'm
trying to package it.  If HIP ever drops hcc as a dependency then we should
be able to deprecate hcc in Fedora as well, but I'm not sure if and when this
will happen.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #20 from Tom Stellard  ---
SPEC URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00870303-hcc/hcc.spec
SRPM URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00870303-hcc/hcc-1.3.18505-2.rocm2.0.0.fc31.src.rpm

- Dropped libunwind-devel dependency.
- Moved headers/cmake files to -devel package.
- Used %{_lib} macro where needed.
- Replaced /usr with %{_prefix}

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #19 from Tom Stellard  ---
(In reply to Felix Schwarz from comment #17)
> Two more things: Tom, some of your SRPM URLs are 404.
> 
> Also in your last revision of the spec file you added "libunwind-devel" with
> no other changes. Would you mind commenting why you did that?

Do you mean 'removed' libunwind-devel?  I removed it, because hcc
dropped it as a dependency prior to the ROCm-2.0.0 version.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #18 from Tom Stellard  ---
(In reply to Felix Schwarz from comment #16)
> (In reply to Robert-André Mauchin from comment #14)
> >  - %{_includedir}/* in %files
> > 
> > Includes should go to a devel package
> > 
> > %{_libdir}/cmake/hcc/ too.
> 
> Not sure if I understood you correctly but hcc is basically a compiler. As
> far as I know there are no outside users of these header files so probably
> it does not make sense to split these in a -devel package.
> 

I took a closer look and most of the header files in /usr/include/ are actually
for
the runtime library, so I think putting them in the -devel package makes sense.
The compiler internal headers are still part of the hcc package, but are
installed
into /usr/libexec/hcc/

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #17 from Felix Schwarz  ---
Two more things: Tom, some of your SRPM URLs are 404.

Also in your last revision of the spec file you added "libunwind-devel" with no
other changes. Would you mind commenting why you did that?

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #16 from Felix Schwarz  ---
(In reply to Robert-André Mauchin from comment #14)
>  - %{_includedir}/* in %files
> 
> Includes should go to a devel package
> 
> %{_libdir}/cmake/hcc/ too.

Not sure if I understood you correctly but hcc is basically a compiler. As far
as I know there are no outside users of these header files so probably it does
not make sense to split these in a -devel package.

(In reply to Robert-André Mauchin from comment #15)
> This takes forever to compile.

It builds its bundled copy of llvm :-) Takes about 45 minutes on my machine
(from the top of my head, might not be accurate). However you need to kill the
"licensecheck" process when using fedora-review. That script will take many
hours (only using a single core).

Btw Tom: I noticed that AMD just deprecated HCC with the ROCm 2.2 release:
> Deprecation Notice
>
> AMD is deprecating HCC to put more focus on HIP development and on other 
> languages supporting
> heterogeneous compute. We will no longer develop any new feature in HCC and 
> we will stop
> maintaining HCC after its final release, which is planned for June 2019. If 
> your application
> was developed with the hc C++ API, we would encourage you to transition it to 
> other languages
> supported by AMD, such as HIP or OpenCL. HIP and hc language share the same 
> compiler technology,
> so many hc kernel language features (including inline assembly) are also 
> available through
> the HIP compilation path.
https://github.com/RadeonOpenCompute/hcc/

I assume it still makes sense because hcc is needed right now for the rest of
the ROCm stack? (Also the compiler code needs to go somewhere so either they
move it into hip or keep hcc around but only as an internal component?)

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #15 from Robert-André Mauchin  ---
This takes forever to compile.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #14 from Robert-André Mauchin  ---
%if 0%{?__isa_bits} == 64
%global build_libdir lib64/
%else
%global build_libdir lib/
%endif

 - Why not use %{_lib}? It resolves to either lib or lib64 depending on the
arch.

 -
-DCMAKE_INSTALL_PREFIX=%{buildroot}/usr \

→

-DCMAKE_INSTALL_PREFIX=%{buildroot}%{_prefix} \

 - %{_includedir}/* in %files

Includes should go to a devel package

%{_libdir}/cmake/hcc/ too.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #13 from Tom Stellard  ---
SPEC URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00863201-hcc/hcc.spec
SRPM URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00863201-hcc/hcc-1.3.18505-2.rocm2.0.0.fc30.src.rpm

I added back the cmake-tests target.  It's not used by our build, but that's
one less patch that we have to apply.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #12 from Tom Stellard  ---
SPEC URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00862882-hcc/hcc.spec
SRPM URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00862882-hcc/hcc-1.3.18505-2.rocm2.0.0.fc30.src.rpm

(In reply to Felix Schwarz from comment #11)
> I'm mostly done with the review (still hoping that someone more experienced
> will join here :-). Some questions+bikeshedding and one bigger issue.
> 

It's up to you, but there is no experience requirement for doing a review.  I'd
say this has been a very good review so far.

> Let's start with the most important one:
> .so files in /usr/libexec/hcc/lib64/ are not stripped (manual stripping
> trims the size from 1.5 GB to 47.2 MB). Unless there is a really good reason
> not to strip these I think that needs to be taken care of.
> 

This is fixed.

> Some questions/notes:
> 1. hcc depends on "hcc-runtime-devel" but I was wondering why "hcc-runtime"
> is not sufficient. "hcc-runtime-devel" just contains unversioned .so files.

hcc's linker invocation references the unversioned libraries, which is why the
hcc-runtime-devel files are needed.

> 2. hcc also requires "rocminfo" but I was wondering where this is used or
> why this is a dependency. If I grep for "rocminfo" in the sources it is only
> mentioned when building a docker container.

hcc needs rocm_agent_enumerator which is shipped with the rocminfo package.
It is used to determine the GPU target when one isn't specified.  It's
referenced
from the clang source code.

> 3. "compiler-rt" and "cmake-tests" are removed. Would you mind providing
> some rationale for this? Did I see correctly that "cmake-tests" requires
> that "hc::accelerator()" returns something? (which might not work on build
> machines)

It doesn't seem like compiler-rt is actually needed, I think libgcc is enough.
I also wanted hcc to use the system compiler-rt.  I don't recall why I removed
cmake-tests, probably because it wasn't working on the builders, but I can look
at that again.

> 4. rpmlint complains about "executable stack" (.so files in /usr/lib64/). Is
> this necessary for hcc?
> (https://fedoraproject.org/wiki/Packaging_tricks#Executable_stack)

I filed an issue on hcc github for this, I'm not sure why the stack is being
marked as electable.  I fixed this in the specfile too.

> 5. rpmlint complains that "/usr/libexec/hcc/lib64/libLLVM-8-rocm.so"
> contains an invalid soname but AFAIK this is just an internal library which
> has no soname at all.

Yes, it's just an internal library.

> 6. The upstream "hcc" package declares a "NCSA" license but it contains some
> files with a different license:
>Apache:
> - stl-test/*.pl.in
> - tests/Conformance
> 
>Expat license:
> - lib/hsa/unpinned_copy_engine.*
> - hc2/external/elfio 
> 
>3-clause BSD:
> - utils/gtest/
> 
>Expat is probably similar enough to NCSA but I'm not sure what to do with
> the others (only test code as far as I can see).
> 

I've updated the license in the spec file.
> 
> Bikeshedding:
> - changelog: "1.3.18505-2.rocm2.0.0" is the latest but
> "1.3.19020-1.rocm2.0.0" comes before (higher version number). Is that just
> actual recording of history or due to some mishap?
> - "HCC": I think you should use consistent upper-case spelling of HCC
> - spelling in spec file: "trunck", "lincense"
> - rpmlint says: "summary-not-capitalized" for various RPMs

I've fixed all these too.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #11 from Felix Schwarz  ---
I'm mostly done with the review (still hoping that someone more experienced
will join here :-). Some questions+bikeshedding and one bigger issue.

Let's start with the most important one:
.so files in /usr/libexec/hcc/lib64/ are not stripped (manual stripping trims
the size from 1.5 GB to 47.2 MB). Unless there is a really good reason not to
strip these I think that needs to be taken care of.

Some questions/notes:
1. hcc depends on "hcc-runtime-devel" but I was wondering why "hcc-runtime" is
not sufficient. "hcc-runtime-devel" just contains unversioned .so files.
2. hcc also requires "rocminfo" but I was wondering where this is used or why
this is a dependency. If I grep for "rocminfo" in the sources it is only
mentioned when building a docker container.
3. "compiler-rt" and "cmake-tests" are removed. Would you mind providing some
rationale for this? Did I see correctly that "cmake-tests" requires that
"hc::accelerator()" returns something? (which might not work on build machines)
4. rpmlint complains about "executable stack" (.so files in /usr/lib64/). Is
this necessary for hcc?
(https://fedoraproject.org/wiki/Packaging_tricks#Executable_stack)
5. rpmlint complains that "/usr/libexec/hcc/lib64/libLLVM-8-rocm.so" contains
an invalid soname but AFAIK this is just an internal library which has no
soname at all.
6. The upstream "hcc" package declares a "NCSA" license but it contains some
files with a different license:
   Apache:
- stl-test/*.pl.in
- tests/Conformance

   Expat license:
- lib/hsa/unpinned_copy_engine.*
- hc2/external/elfio 

   3-clause BSD:
- utils/gtest/

   Expat is probably similar enough to NCSA but I'm not sure what to do with
the others (only test code as far as I can see).


Bikeshedding:
- changelog: "1.3.18505-2.rocm2.0.0" is the latest but "1.3.19020-1.rocm2.0.0"
comes before (higher version number). Is that just actual recording of history
or due to some mishap?
- "HCC": I think you should use consistent upper-case spelling of HCC
- spelling in spec file: "trunck", "lincense"
- rpmlint says: "summary-not-capitalized" for various RPMs

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #10 from Tom Stellard  ---
Spec URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00857850-hcc/hcc.spec
SRPM URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00857850-hcc/hcc-1.3.18505-2.rocm2.0.0.fc30.src.rpm

(In reply to Felix Schwarz from comment #9)
> Hi Tom, fixing all the small issues I brought up (I had hoped someone more
> experienced than me would have reviewed the package by now).
> 
> I'm still hesitant to formally reviewing this package but I think I found
> some more issues:
> 
> hcc package
> ===
> 
> - contains a folder "/usr/include/experimental" which should probably be
> stored as "/usr/include/hcc/experimental"

I've fixed this.
> e.g. "algorithm" (no extension?) refers to "../hcc.hpp"
> - empty "/lib" folder?

There are hidden build-id files in /lib that were generated by RPM.

> 
> - hcc-config.cmake contains references to "/opt/rocm" (should be
> /usr/include/hcc?)
>   - also there it contains something like
>   set_and_check( hcc_INCLUDE_DIR "${PACKAGE_PREFIX_DIR}/include" )
>   - I'm not sure where this cmake file is actually used but I suspect this
> should point to "${PACKAGE_PREFIX_DIR}/include/hcc" as well

I've fixed this.

>   - are the "cmake" files actually required by hcc directly?

These are meant to be used by other programs, but I'm not sure what actually
uses them.

> 
> - hcc.spec:
>   for f in clamp-device clamp-embed clamp-assemble clamp-link
> hc-kernel-assemble hc-host-assemble error-check; do
>   mv %{buildroot}{/usr,%{_libexecdir}/hcc}/bin/$f
>   done
>   I'd prefer "install --preserve-timestamps"
> 

I updated the spec so these are installed by 'make install' now.

> 
> hcc-runtime package
> =
> - empty "/lib" folder?

Has build-id files in it.

> - does not install a license file!

I've fixed this.
> 
> 
> rocm-device-libs
> =
> - does not install a license file!
> 

Fixed this.
> 
> hcc-runtime-devel
> =
> 
> Note: Package has .a files: hcc-runtime-devel. Does not provide -static:
> hcc-runtime-devel.
>  -> I assume you are aware of the static library and it is actually required
> by some part of the ROCm stack
>  -> Fedora guidelines say that the static library MUST be in a -static
> package
> (https://docs.fedoraproject.org/en-US/packaging-guidelines/#packaging-static-
> libraries)
> 

I've replaced the static library with a shared library.

> 
> --
> I was unable to check the license of all sources (due to the huge number of
> sources files + the perl check script taking forever) so far. Also several
> other points which I could not check yet.
> 
> I'm also not sure if llvm counts as a bundled library and must be declared
> as such
> (https://docs.fedoraproject.org/en-US/packaging-guidelines/#bundling), see
> also Fedora's "rust" package.

I've added Provides: bundled(llvm)

> Tom: I assume the rocm stack currently does not work against any upstream
> version of llvm, right? Is there an official way of trying to build with a
> vanilla llvm?

I think it would likely work with upstream LLVM, however the ROCm release
don't line up with upstream LLVM releases.  So what is packaged now in hcc
is a snapshot from LLVM SVN, and we only ship official llvm release in Fedora.

> 
> My plan is to work through the rest of the review items (though that will
> take some time) and also check your "hip" package in parallel. That way I
> hope to validate the packaging decisions made in the "hcc" package.

Ok, thanks for the review.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #9 from Felix Schwarz  ---
Hi Tom, fixing all the small issues I brought up (I had hoped someone more
experienced than me would have reviewed the package by now).

I'm still hesitant to formally reviewing this package but I think I found some
more issues:

hcc package
===

- contains a folder "/usr/include/experimental" which should probably be stored
as "/usr/include/hcc/experimental"
e.g. "algorithm" (no extension?) refers to "../hcc.hpp"
- empty "/lib" folder?

- hcc-config.cmake contains references to "/opt/rocm" (should be
/usr/include/hcc?)
  - also there it contains something like
  set_and_check( hcc_INCLUDE_DIR "${PACKAGE_PREFIX_DIR}/include" )
  - I'm not sure where this cmake file is actually used but I suspect this
should point to "${PACKAGE_PREFIX_DIR}/include/hcc" as well
  - are the "cmake" files actually required by hcc directly?

- hcc.spec:
  for f in clamp-device clamp-embed clamp-assemble clamp-link
hc-kernel-assemble hc-host-assemble error-check; do
mv %{buildroot}{/usr,%{_libexecdir}/hcc}/bin/$f
  done
  I'd prefer "install --preserve-timestamps"


hcc-runtime package
=
- empty "/lib" folder?
- does not install a license file!


rocm-device-libs
=
- does not install a license file!


hcc-runtime-devel
=

Note: Package has .a files: hcc-runtime-devel. Does not provide -static:
hcc-runtime-devel.
 -> I assume you are aware of the static library and it is actually required by
some part of the ROCm stack
 -> Fedora guidelines say that the static library MUST be in a -static package
(https://docs.fedoraproject.org/en-US/packaging-guidelines/#packaging-static-libraries)


--
I was unable to check the license of all sources (due to the huge number of
sources files + the perl check script taking forever) so far. Also several
other points which I could not check yet.

I'm also not sure if llvm counts as a bundled library and must be declared as
such (https://docs.fedoraproject.org/en-US/packaging-guidelines/#bundling), see
also Fedora's "rust" package.
Tom: I assume the rocm stack currently does not work against any upstream
version of llvm, right? Is there an official way of trying to build with a
vanilla llvm?

My plan is to work through the rest of the review items (though that will take
some time) and also check your "hip" package in parallel. That way I hope to
validate the packaging decisions made in the "hcc" package.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #8 from Tom Stellard  ---
Spec URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00849663-hcc/hcc.spec
SRPM URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00849663-hcc/hcc-1.3.18505-1.rocm2.0.0.fc30.src.rpm

(In reply to Felix Schwarz from comment #7)
> I noticed some things:
> The hcc package contains multiple copies of the same header file, e.g.
> kalmar_aligned_alloc.h is available via /usr/include/ and /usr/include/hcc/.
> I saw a patch in the SRPM which tried to move everything to the "hcc"
> subdirectory but I guess you missed at least on spot.
> 

I've fixed this now so that everything is in /usr/include/hcc.  The previous
build had all the headers symlinked into /usr/include because this is what some
apps, like hip, expect.  I've also updated the output of hcc-config --cxxflags
to add /usr/include/hcc as an include directory to support applications that
don't use the hcc prefix in their #include directives.

> Also there are some rpmlint errors (which I guess you are aware of already),
> e.g. that "hcc" requires "hcc-runtime-devel". As hcc is a compiler I think
> that is basically ok (compiler requires header files for its "standard
> library") but maybe these files are just part of the "hcc-runtime"?

This is ok for a compiler.  It's possible to use the hcc-runtime without hcc, I
think hip may do this, so this is why I separated the headers and libraries
into separate packages.

> rpmlint also complained about some rpaths in binaries like hcc and libraries
> like "LLVMPromotePointerKernArgsToGlobal.so".
> 

hcc uses and depends on internal libraries, so this is OK according to
https://docs.fedoraproject.org/en-US/packaging-guidelines/#_rpath_for_internal_libraries
 

> "hcc-config --cxxflags" still returns "-I/opt/rocm/include" but lacks
> "-I/usr/include/hcc/". Same applies to "clamp-config" (and I was wondering

These include paths are fixed now.

> why "clamp-config" was available at all as I read that ROCm 2.0 drop C++ AMP
> support - but that is likely due to my limited understanding/an upstream
> issue).

I read that too, but upstream still ships this so it is an upstream issue

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479

Tom Stellard  changed:

   What|Removed |Added

 Blocks||1668010




Referenced Bugs:

https://bugzilla.redhat.com/show_bug.cgi?id=1668010
[Bug 1668010] Review Request: hip- Tool for porting CUDA to Portable C++ Code
-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #7 from Felix Schwarz  ---
I rebuilt your updated hcc package on Fedora 29 which worked fine. The
resulting "hcc" binary seems to work. I was able to run "saxpy" and several
example programs from
https://github.com/ROCm-Developer-Tools/HCC-Example-Application .

Just for reference: I'm able to run the "saxpy" example from
https://github.com/RadeonOpenCompute/hcc/wiki#how-to-use-hcc when invoking hcc
like this: hcc `hcc-config --cxxflags --ldflags` -I/usr/include/hcc saxpy.cpp
-o saxpy

I noticed some things:
The hcc package contains multiple copies of the same header file, e.g.
kalmar_aligned_alloc.h is available via /usr/include/ and /usr/include/hcc/. I
saw a patch in the SRPM which tried to move everything to the "hcc"
subdirectory but I guess you missed at least on spot.

Also there are some rpmlint errors (which I guess you are aware of already),
e.g. that "hcc" requires "hcc-runtime-devel". As hcc is a compiler I think that
is basically ok (compiler requires header files for its "standard library") but
maybe these files are just part of the "hcc-runtime"?
rpmlint also complained about some rpaths in binaries like hcc and libraries
like "LLVMPromotePointerKernArgsToGlobal.so".

"hcc-config --cxxflags" still returns "-I/opt/rocm/include" but lacks
"-I/usr/include/hcc/". Same applies to "clamp-config" (and I was wondering why
"clamp-config" was available at all as I read that ROCm 2.0 drop C++ AMP
support - but that is likely due to my limited understanding/an upstream
issue).

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479

Tom Stellard  changed:

   What|Removed |Added

 Depends On||1666836



--- Comment #6 from Tom Stellard  ---
Spec URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00847787-hcc/hcc.spec
SRPM URL:
https://copr-be.cloud.fedoraproject.org/results/tstellar/rocm-2.0/fedora-rawhide-x86_64/00847787-hcc/hcc-1.3.18505-1.rocm2.0.0.fc30.src.rpm

Updated builds for ROCm 2.0.  There is a new dependency on the rocminfo
package, which is currently in review (bug 1666836).  For now you can get
rocminfo from this copr:
https://copr.fedorainfracloud.org/coprs/tstellar/rocm-2.0/


Referenced Bugs:

https://bugzilla.redhat.com/show_bug.cgi?id=1666836
[Bug 1666836] Review Request: rocminfo -  ROCm system info utility
-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #5 from Tom Stellard  ---
(In reply to Felix Schwarz from comment #4)
> > I'm not going to push a ROCm 1.6 version of hcc into Fedora. I have updated
> > the spec file locally to ROCm 1.7, but I would need to update rocm-runtime 
> > and
> > libhsakmt in Fedora to ROCm 1.7, before I can it to Fedora.
> 
> So this review request here is currently outdated (as you started to pacakge
> ROCm 1.7)? Or is a package review pointless until ROCm works with mainline
> kernel out of the box? (libhsakmt/thunk interface)
> 

Yes, it is out-dated now.

> > ROCm 1.8 hcc will probably not happen in Fedora because AMD have not 
> > released
> > a special version of libhsakmt that works with the upstream kernel like they
> > did for 1.6 and 1.7.
> 
> Just out of curiosity/to get a more complete picture: This problem should be
> solved once enough features are merged in amdkfd so the user space ROCm
> stack works with a vanilla upstream kernel, right? Or is there some special
> "secret sauce" which is unlikely to be upstreamed?
> 

There are already enough features in upstream amdkfd, the issue now is getting
a new thunk (hsakmt in Fedora) that is compatible with what is upstream.  AMD
has been publishing code for an upstream compatible thunk in this branch:
https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/commits/fxkamd/drm-next-wip.
 The current hsamkt Fedora was built from an early snapshot of this branch and
the current HEAD is
compatible with ROCm 1.7.

Now that everything is upstream AMD is going to start working on making their
'official' thunk compatible with upstream, which I'm hoping will happen by ROCm
1.9.

> Anyway, I don't want to turn this ticket into a personal support forum so
> the more important question is: What should happen to this review request?
> - Review it and push an (incomplete) ROCm 1.6 stack in Fedora?
> - Close the request (because 1.7 will be submitted for merging at a later
> time)?
> - Keep it open because at some point in the future you submit a updated spec
> for ROCM 1.7?

Let's just keep it open for now in case I have to time to push the ROCm 1.7
packaging.  The current 1.6 spec is old and has several bugs that I fixed with
the 1.7 packaging.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org/message/3S5PKA6534IBNOQQAS6ZD67EBNEAHUL2/

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #4 from Felix Schwarz  ---
> I'm not going to push a ROCm 1.6 version of hcc into Fedora. I have updated
> the spec file locally to ROCm 1.7, but I would need to update rocm-runtime and
> libhsakmt in Fedora to ROCm 1.7, before I can it to Fedora.

So this review request here is currently outdated (as you started to pacakge
ROCm 1.7)? Or is a package review pointless until ROCm works with mainline
kernel out of the box? (libhsakmt/thunk interface)

> ROCm 1.8 hcc will probably not happen in Fedora because AMD have not released
> a special version of libhsakmt that works with the upstream kernel like they
> did for 1.6 and 1.7.

Just out of curiosity/to get a more complete picture: This problem should be
solved once enough features are merged in amdkfd so the user space ROCm stack
works with a vanilla upstream kernel, right? Or is there some special "secret
sauce" which is unlikely to be upstreamed?

Anyway, I don't want to turn this ticket into a personal support forum so the
more important question is: What should happen to this review request?
- Review it and push an (incomplete) ROCm 1.6 stack in Fedora?
- Close the request (because 1.7 will be submitted for merging at a later
time)?
- Keep it open because at some point in the future you submit a updated spec
for ROCM 1.7?

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org/message/AAZLSOIFQP3CG6GMSY5QTVMAQAE6D355/

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #3 from Tom Stellard  ---
I'm not going to push a ROCm 1.6 version of hcc into Fedora.  I have updated
the spec file locally to ROCm 1.7, but I would need to update rocm-runtime and
libhsakmt in Fedora to ROCm 1.7, before I can it to Fedora.

ROCm 1.8 hcc will probably not happen in Fedora because AMD have not released a
special version of libhsakmt that works with the upstream kernel like they did
for 1.6 and 1.7.

If you are interested in testing, I have a copr with most of the ROCm 1.7
packages (it's basically hiptensorflow plus all its dependencies) if you want
to take a look:

https://copr.fedorainfracloud.org/coprs/tstellar/rocm-1.7-userspace/

Note that since I haven't packaged the upstream thunk yet for ROCm 1.7, you
will need to install the 'official' AMD kernel and thunk from this copr to get
it to work:

https://copr.fedorainfracloud.org/coprs/tstellar/rocm-1.7-amd/

But on the plus side, with the AMD kernel you will get dGPU and Vega support.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org/message/OCIY6GRBM5TLJ6WQ4U7AU43WASW5KMO5/

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479



--- Comment #2 from Felix Schwarz  ---
Just a random note: Do you plan to update the spec to the latest HCC
version/roc 1.7/1.8 in this review request or would you like to do that after
the package got reviewed? (Thanks for working on this btw :-)

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/package-review@lists.fedoraproject.org/message/7CCFS5SSX33SHBEINJYD6GCJO5GOTYRF/

[Bug 1545479] Review Request: hcc- Heterogeneous C++ Compiler

[bugzilla]

https://bugzilla.redhat.com/show_bug.cgi?id=1545479

Robert-André Mauchin  changed:

   What|Removed |Added

 CC||zebo...@gmail.com



--- Comment #1 from Robert-André Mauchin  ---
%description
Heterogeneous C++ Compiler

   Could you expand the description a little bit.

 - Is this expected that these libraries aren't versionned:

%{_libdir}/libhc_am.so
%{_libdir}/libmcwamp_hsa.so
%{_libdir}/libmcwamp.so

See
https://fedoraproject.org/wiki/Packaging:Guidelines#Downstream_.so_name_versioning

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are always notified about changes to this product and component
___
package-review mailing list -- package-review@lists.fedoraproject.org
To unsubscribe send an email to package-review-le...@lists.fedoraproject.org