The first release of unified Khronos OpenCL headers, which unlike
old headers (which is what app-eselect/eselect-opencl installs)
work with all released OpenCL versions rather than having different
versions supported by files from different directories.
Note that in order to avoid file collisions with eselect-opencl this
package presently installs header files into
/usr/$(get_libdir)/OpenCL/vendors/opencl-icd-loader/include ,
which is why it needs multilib support and why it will likely not work
right when installed standalone (in contrast to being pulled in by
dev-libs/opencl-icd-loader).
Signed-off-by: Marek Szuba <marecki@gentoo.org>
The NDEBUG macro turns the assert() function into a noop. This gives a
small performance boost, but may allow subtle programming errors to go
unnoticed.
This code was added back in 2008, when we started passing
-DCMAKE_BUILD_TYPE=None instead of Release or Debug. It probably tries
to mimic a default behavior of Release type builds.
Other common build systems do not do this by default. For example,
autoconf's AC_HEADER_ASSERT macro only sets NDEBUG if --disable-assert
is passed to configure (it defaults to enabled).
It is better to let users add this to CPPFLAGS themselves if they really
want to save those few CPU cycles.
Signed-off-by: Mike Gilbert <floppym@gentoo.org>
I should have taken a closer look at ut2003-data. I would change how
we package this but I am unable to test it.
Package-Manager: Portage-2.3.96, Repoman-2.3.20
Signed-off-by: James Le Cuirot <chewi@gentoo.org>
This reverts commit f8363acd3f.
My bad for dropping sord from commits list for other reasons
without realizing sratom depends on it
Signed-off-by: Mart Raudsepp <leio@gentoo.org>
SLOT=7 does not mean much nowadays. Let's default to SLOT=0.
Reported-by: Arfrever Frehtes Taifersar Arahesis
Bug: https://bugs.gentoo.org/693250
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Unconditionally depends on virtual/opencl and we are in the process
of dropping support for 32-bit OpenCL from Gentoo. Approved by
zerochaos in #gentoo-dev.
Signed-off-by: Marek Szuba <marecki@gentoo.org>
1. Not sure if this is due to what upstream supports (like for both
Intel providers) or because it has not been implemented - but
either way, as of 2020-04-01 dev-libs/rocm-opencl-runtime neither is
keyworded x86 nor supports multilib on amd64;
2. Conversely, dev-libs/amdgpu-pro-opencl does both so in order for
multilib to work properly it should be passed $MULTILIB_USEDEP.
Signed-off-by: Marek Szuba <marecki@gentoo.org>
Advantages:
* keyworded x86
* supports multilib on amd64
* all OpenCL-aware software should compile and link fine against it
* should make it easier for user to deploy out-of-tree OpenCL
providers (side note: all actual OpenCL providers currently
in the tree except x11-drivers/nvidia-drivers actually *require*
this to work), should they choose to do so
Disadvantages:
* essentially a hack to ensure the integrity of the dependency tree
on x86 and amd64/abi_x86_32 - most of the actual OpenCL providers
currently in the tree are 64-bit only anyway
Signed-off-by: Marek Szuba <marecki@gentoo.org>