pan_perf.c

/*
 * Copyright © 2021 Collabora, Ltd.
 * Author: Antonio Caggiano <[email protected]>
 *
 * 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
 * AUTHORS OR COPYRIGHT HOLDERS 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 "pan_perf.h"

#include <pan_perf_metrics.h>
#include <lib/pan_device.h>
#include <drm-uapi/panfrost_drm.h>

#define PAN_COUNTERS_PER_CATEGORY 64
#define PAN_SHADER_CORE_INDEX 2

uint32_t
panfrost_perf_counter_read(const struct panfrost_perf_counter *counter,
                           const struct panfrost_perf *perf)
{
   assert(counter->offset < perf->n_counter_values);
   uint32_t ret = perf->counter_values[counter->offset];

   // If counter belongs to shader core, accumulate values for all other cores
   if (counter->category == &perf->cfg->categories[PAN_SHADER_CORE_INDEX]) {
      for (uint32_t core = 1; core < perf->dev->core_count; ++core) {
         ret += perf->counter_values[counter->offset + PAN_COUNTERS_PER_CATEGORY * core];
      }
   }

   return ret;
}

static const struct panfrost_perf_config*
get_perf_config(unsigned int gpu_id)
{
   switch (gpu_id) {
   case 0x720:
      return &panfrost_perf_config_t72x;
   case 0x750:
      return &panfrost_perf_config_t76x;
   case 0x820:
      return &panfrost_perf_config_t82x;
   case 0x830:
      return &panfrost_perf_config_t83x;
   case 0x860:
      return &panfrost_perf_config_t86x;
   case 0x880:
      return &panfrost_perf_config_t88x;
   case 0x6221:
      return &panfrost_perf_config_thex;
   case 0x7093:
      return &panfrost_perf_config_tdvx;
   case 0x7212:
   case 0x7402:
      return &panfrost_perf_config_tgox;
   default:
      unreachable("Invalid GPU ID");
   }
}

void
panfrost_perf_init(struct panfrost_perf *perf, struct panfrost_device *dev)
{
   perf->dev = dev;
   perf->cfg = get_perf_config(dev->gpu_id);

   // Generally counter blocks are laid out in the following order:
   // Job manager, tiler, L2 cache, and one or more shader cores.
   uint32_t n_blocks = 3 + dev->core_count;
   perf->n_counter_values = PAN_COUNTERS_PER_CATEGORY * n_blocks;
   perf->counter_values = ralloc_array(perf, uint32_t, perf->n_counter_values);
}

static int
panfrost_perf_query(struct panfrost_perf *perf, uint32_t enable)
{
   struct drm_panfrost_perfcnt_enable perfcnt_enable = {enable, 0};
   return drmIoctl(perf->dev->fd, DRM_IOCTL_PANFROST_PERFCNT_ENABLE, &perfcnt_enable);
}

int
panfrost_perf_enable(struct panfrost_perf *perf)
{
   return panfrost_perf_query(perf, 1 /* enable */);
}

int
panfrost_perf_disable(struct panfrost_perf *perf)
{
   return panfrost_perf_query(perf, 0 /* disable */);
}

int
panfrost_perf_dump(struct panfrost_perf *perf)
{
   // Dump performance counter values to the memory buffer pointed to by counter_values
   struct drm_panfrost_perfcnt_dump perfcnt_dump = {(uint64_t)(uintptr_t)perf->counter_values};
   return drmIoctl(perf->dev->fd, DRM_IOCTL_PANFROST_PERFCNT_DUMP, &perfcnt_dump);
}