On 15/11/24 11:43 -0500, Jason Merrill wrote:
On 11/15/24 10:21 AM, Jonathan Wakely wrote:
On Fri, 15 Nov 2024 at 15:20, Jonathan Wakely <jwak...@redhat.com> wrote:
On Fri, 15 Nov 2024 at 15:12, Jason Merrill <ja...@redhat.com> wrote:
On 11/15/24 9:20 AM, Jonathan Wakely wrote:
On 15/11/24 08:53 -0500, Jason Merrill wrote:
On 11/15/24 6:04 AM, Jonathan Wakely wrote:
On 14/11/24 23:09 -0500, Jason Merrill wrote:
Since bits/stdc++.h also intends to include the whole standard
library, I
include it rather than duplicate it. But stdc++.h comments out
<execution>,
so I include it separately. Alternatively, we could uncomment it in
stdc++.h.
It's not included by default because it pulls in a dependency on
libtbb.so even if you don't use the parallel algos:
$ g++ -include algorithm -include execution -x c++ - <<< 'int main()
{ }'
/usr/bin/ld: /tmp/ccA76qrz.o: in function
`tbb::detail::d1::execution_slot(tbb::detail::d1::execution_data
const&)':
<stdin>:
(.text._ZN3tbb6detail2d114execution_slotERKNS1_14execution_dataE[_ZN3tbb6detail2d114execution_slotERKNS1_14execution_dataE]+0x14):
undefined reference to
`tbb::detail::r1::execution_slot(tbb::detail::d1::execution_data const*)'
/usr/bin/ld: /tmp/ccA76qrz.o: in function
`tbb::detail::d1::current_thread_index()':
<stdin>:
(.text._ZN3tbb6detail2d120current_thread_indexEv[_ZN3tbb6detail2d120current_thread_indexEv]+0xe):
undefined reference to
`tbb::detail::r1::execution_slot(tbb::detail::d1::execution_data const*)'
collect2: error: ld returned 1 exit status
The way we arranged the PSTL inclusion is that you only get a
dependeny on TBB if you include a header containing a parallel algo
(<algorithm>, <numeric>) *and* include <execution>. Since including
<execution> is formally required by the standard in order to pass one
of std::execution::par, std::execution::seq etc. to a parallel algo,
we switch on whether <execution> has been included or not to detect
whether the user actually wants to use parallel algos.
Is this not a problem for 'import std;'?
Hmm, I haven't been seeing that in my testing, std.o only contains the
reference to ios_base_library_init.
But then, I don't get that error with your testcase above, either.
With trunk, the compiler output for that only contains main for me.
Then I assume you don't have tbb-devel installed. If the tbb headers
aren't present, then <execution> doesn't include them.
Could you do 'dnf install tbb-devel' and try again?
Indeed, that's what the difference was. With tbb-devel installed,
compiling std.cc fails with an error about TBB code exposing
task_ptr_or_nullptr in the anonymous namespace.
tbb-devel also isn't installed on tor/torsion, and I don't seem to have
sudo on them anymore.
Maybe we should figure out how to fix it for #include <execution>
anyway. When we eventually support senders/receivers in <execution>
for C++26, users are going to want to include that for reasons other
than using the parallel algos and their execution policies.
Is there a way I can suppress pstl with a macro? If not, I guess I'll
just leave <execution> out for now.
Defining _GLIBCXX_USE_TBB_PAR_BACKEND to 0 will work.
That only suppresses the TBB backend, so you can still use the
execution policies in <execution> and call parallel algos ... they
just don't run in parallel :-)
All the declarations required by the standard will be present.
Hmm, I'm not sure whether that is preferable to leaving them out
entirely.
That would be possible, but I don't think we currently have any
mechanism to do that.
But here's a v3 that does that, along with your other
suggestions:
Looks great. OK for trunk, if you think it's ready.
