[RESEND PATCH 2/5] drm: Add Reusable task barrier.

[Andrey Grodzovsky]

On 12/11/19 11:04 PM, Ma, Le wrote:
>
> [AMD Official Use Only - Internal Distribution Only]
>
> -----Original Message-----
> From: Andrey Grodzovsky <andrey.grodzovsky at amd.com>
> Sent: Thursday, December 12, 2019 4:39 AM
> To: dri-devel at lists.freedesktop.org; amd-gfx at lists.freedesktop.org
> Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Ma, Le 
> <Le.Ma at amd.com>; Zhang, Hawking <Hawking.Zhang at amd.com>; Quan, Evan 
> <Evan.Quan at amd.com>; Grodzovsky, Andrey <Andrey.Grodzovsky at amd.com>
> Subject: [RESEND PATCH 2/5] drm: Add Reusable task barrier.
>
> It is used to synchronize N threads at a rendevouz point before 
> execution of critical code that has to be started by all the threads 
> at approximatly the same time.
>
> Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky at amd.com 
> <mailto:andrey.grodzovsky at amd.com>>
>
> ---
>
> include/drm/task_barrier.h | 106 
> +++++++++++++++++++++++++++++++++++++++++++++
>
> 1 file changed, 106 insertions(+)
>
> create mode 100644 include/drm/task_barrier.h
>
> diff --git a/include/drm/task_barrier.h b/include/drm/task_barrier.h 
> new file mode 100644 index 0000000..81fb0f7
>
> --- /dev/null
>
> +++ b/include/drm/task_barrier.h
>
> @@ -0,0 +1,106 @@
>
> +/*
>
> + * Copyright 2019 Advanced Micro Devices, Inc.
>
> + *
>
> + * Permission is hereby granted, free of charge, to any person
>
> +obtaining a
>
> + * copy of this software and associated documentation files (the
>
> +"Software"),
>
> + * to deal in the Software without restriction, including without
>
> +limitation
>
> + * the rights to use, copy, modify, merge, publish, distribute,
>
> +sublicense,
>
> + * and/or sell copies of the Software, and to permit persons to whom
>
> +the
>
> + * Software is furnished to do so, subject to the following conditions:
>
> + *
>
> + * The above copyright notice and this permission notice shall be
>
> +included in
>
> + * all copies or substantial portions of the Software.
>
> + *
>
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
>
> +EXPRESS OR
>
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
>
> +MERCHANTABILITY,
>
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT
>
> +SHALL
>
> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM,
>
> +DAMAGES OR
>
> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
>
> +OTHERWISE,
>
> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE
>
> +OR
>
> + * OTHER DEALINGS IN THE SOFTWARE.
>
> + *
>
> + */
>
> +#include <linux/semaphore.h>
>
> +#include <linux/atomic.h>
>
> +
>
> +/*
>
> + * Reusable 2 PHASE task barrier (randevouz point) implementation for 
> N tasks.
>
> + * Based on the Little book of sempahores -
>
> +https://greenteapress.com/wp/semaphores/
>
> + */
>
> +
>
> +
>
> +
>
> +#ifndef DRM_TASK_BARRIER_H_
>
> +#define DRM_TASK_BARRIER_H_
>
> +
>
> [Le]: It might be better to prefix “drm_” to the functions and 
> structure below, even this header file name.
>

I am not sure about this - see the example of spsc_queue we added for 
GPU scheduler use. I just followed it as an example of where to place 
the structure. There is nothing DRM specific about spsc_queue or 
task_barrier, they are generic constructs that we place in DRM subsystem 
for common use.


> +/*
>
> + * Represents an instance of a task barrier.
>
> + */
>
> +struct task_barrier {
>
> +          unsigned int n;
>
> [Le]: We can define it as signed type here for more common use.
>

This is a counter of number of tasks/threads to synchronize in the 
barrier it cannot go bellow 0

Andrey


> +          atomic_t count;
>
> +          struct semaphore enter_turnstile;
>
> +          struct semaphore exit_turnstile;
>
> +};
>
> +
>
> +static inline void task_barrier_signal_turnstile(struct semaphore 
> *turnstile,
>
> + unsigned int n)
>
> +{
>
> +          int i;
>
> +
>
> +          for (i = 0 ; i < n; i++)
>
> +                      up(turnstile);
>
> +}
>
> +
>
> +static inline void task_barrier_init(struct task_barrier *tb) {
>
> +          tb->n = 0;
>
> +          atomic_set(&tb->count, 0);
>
> + sema_init(&tb->enter_turnstile, 0);
>
> + sema_init(&tb->exit_turnstile, 0);
>
> +}
>
> +
>
> +static inline void task_barrier_add_task(struct task_barrier *tb) {
>
> +          tb->n++;
>
> +}
>
> +
>
> +static inline void task_barrier_rem_task(struct task_barrier *tb) {
>
> +          tb->n--;
>
> +}
>
> +
>
> +/*
>
> + * Lines up all the threads BEFORE the critical point.
>
> + *
>
> + * When all thread passed this code the entry barrier is back to 
> locked state.
>
> + */
>
> +static inline void task_barrier_enter(struct task_barrier *tb) {
>
> +          if (atomic_inc_return(&tb->count) == tb->n)
>
> + task_barrier_signal_turnstile(&tb->enter_turnstile, tb->n);
>
> +
>
> + down(&tb->enter_turnstile);
>
> +}
>
> +
>
> +/*
>
> + * Lines up all the threads AFTER the critical point.
>
> + *
>
> + * This function is used to avoid any one thread running ahead of the
>
> +reset if
>
> [Le]: No need to mention “reset” here.
>
> With the above addressed, Acked-by: Le Ma Le.Ma at amd.com 
> <mailto:Le.Ma at amd.com>
>
> Regards,
>
> Ma Le
>
> + * the barrier is used in a loop (repeatedly) .
>
> + */
>
> +static inline void task_barrier_exit(struct task_barrier *tb) {
>
> +          if (atomic_dec_return(&tb->count) == 0)
>
> + task_barrier_signal_turnstile(&tb->exit_turnstile, tb->n);
>
> +
>
> + down(&tb->exit_turnstile);
>
> +}
>
> +
>
> +static inline void task_barrier_full(struct task_barrier *tb) {
>
> +          task_barrier_enter(tb);
>
> +          task_barrier_exit(tb);
>
> +}
>
> +
>
> +#endif
>
> --
>
> 2.7.4
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20191212/9cd0c169/attachment-0001.htm>