[Ping x4] Re: [PATCH, nvptx, 1/2] Reimplement libgomp barriers for nvptx

[Chung-Lin Tang via Gcc-patches]

Ping x4

On 2022/11/8 12:34 AM, Chung-Lin Tang wrote:
> Ping x3.
> 
> On 2022/10/31 10:18 PM, Chung-Lin Tang wrote:
>> Ping x2.
>>
>> On 2022/10/17 10:29 PM, Chung-Lin Tang wrote:
>>> Ping.
>>>
>>> On 2022/9/21 3:45 PM, Chung-Lin Tang via Gcc-patches wrote:
>>>> Hi Tom,
>>>> I had a patch submitted earlier, where I reported that the current way of 
>>>> implementing
>>>> barriers in libgomp on nvptx created a quite significant performance drop 
>>>> on some SPEChpc2021
>>>> benchmarks:
>>>> https://gcc.gnu.org/pipermail/gcc-patches/2022-September/600818.html>>>>>
>>>> That previous patch wasn't accepted well (admittedly, it was kind of a 
>>>> hack).
>>>> So in this patch, I tried to (mostly) re-implement team-barriers for NVPTX.
>>>>
>>>> Basically, instead of trying to have the GPU do CPU-with-OS-like things 
>>>> that it isn't suited for,
>>>> barriers are implemented simplistically with bar.* synchronization 
>>>> instructions.
>>>> Tasks are processed after threads have joined, and only if 
>>>> team->task_count != 0
>>>>
>>>> (arguably, there might be a little bit of performance forfeited where 
>>>> earlier arriving threads
>>>> could've been used to process tasks ahead of other threads. But that again 
>>>> falls into requiring
>>>> implementing complex futex-wait/wake like behavior. Really, that kind of 
>>>> tasking is not what target
>>>> offloading is usually used for)
>>>>
>>>> Implementation highlight notes:
>>>> 1. gomp_team_barrier_wake() is now an empty function (threads never "wake" 
>>>> in the usual manner)
>>>> 2. gomp_team_barrier_cancel() now uses the "exit" PTX instruction.
>>>> 3. gomp_barrier_wait_last() now is implemented using "bar.arrive"
>>>>
>>>> 4. gomp_team_barrier_wait_end()/gomp_team_barrier_wait_cancel_end():
>>>>     The main synchronization is done using a 'bar.red' instruction. This 
>>>> reduces across all threads
>>>>     the condition (team->task_count != 0), to enable the task processing 
>>>> down below if any thread
>>>>     created a task. (this bar.red usage required the need of the second 
>>>> GCC patch in this series)
>>>>
>>>> This patch has been tested on x86_64/powerpc64le with nvptx offloading, 
>>>> using libgomp, ovo, omptests,
>>>> and sollve_vv testsuites, all without regressions. Also verified that the 
>>>> SPEChpc 2021 521.miniswp_t
>>>> and 534.hpgmgfv_t performance regressions that occurred in the GCC12 cycle 
>>>> has been restored to
>>>> devel/omp/gcc-11 (OG11) branch levels. Is this okay for trunk?
>>>>
>>>> (also suggest backporting to GCC12 branch, if performance regression can 
>>>> be considered a defect)
>>>>
>>>> Thanks,
>>>> Chung-Lin
>>>>
>>>> libgomp/ChangeLog:
>>>>
>>>> 2022-09-21  Chung-Lin Tang  <clt...@codesourcery.com>
>>>>
>>>>    * config/nvptx/bar.c (generation_to_barrier): Remove.
>>>>    (futex_wait,futex_wake,do_spin,do_wait): Remove.
>>>>    (GOMP_WAIT_H): Remove.
>>>>    (#include "../linux/bar.c"): Remove.
>>>>    (gomp_barrier_wait_end): New function.
>>>>    (gomp_barrier_wait): Likewise.
>>>>    (gomp_barrier_wait_last): Likewise.
>>>>    (gomp_team_barrier_wait_end): Likewise.
>>>>    (gomp_team_barrier_wait): Likewise.
>>>>    (gomp_team_barrier_wait_final): Likewise.
>>>>    (gomp_team_barrier_wait_cancel_end): Likewise.
>>>>    (gomp_team_barrier_wait_cancel): Likewise.
>>>>    (gomp_team_barrier_cancel): Likewise.
>>>>    * config/nvptx/bar.h (gomp_team_barrier_wake): Remove
>>>>    prototype, add new static inline function.
>