AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

broadcom/common: move CSD supergroup sizing to a common helper

Browse Source 
We want to use this in GL too.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10541>
This commit is contained in:
Iago Toral Quiroga
2021-04-29 09:23:28 +02:00
committed by Marge Bot
parent afc33a7430
commit 3ce249e65e
5 changed files with 125 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/broadcom/common/v3d_util.c
+81
View File
@@ -0,0 +1,81 @@
/*
* Copyright © 2021 Raspberry Pi
*
* 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 (including the next
* paragraph) 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 "v3d_util.h"
#include "util/macros.h"
/* Choose a number of workgroups per supergroup that maximizes
* lane occupancy. We can pack up to 16 workgroups into a supergroup.
*/
uint32_t
v3d_csd_choose_workgroups_per_supergroup(struct v3d_device_info *devinfo,
bool has_tsy_barrier,
uint32_t threads,
uint32_t num_wgs,
uint32_t wg_size)
{
/* Compute maximum number of batches in a supergroup for this workgroup size.
* Each batch is 16 elements, and we can have up to 16 work groups in a
* supergroup:
*
* max_batches_per_sg = (wg_size * max_wgs_per_sg) / elements_per_batch
* since max_wgs_per_sg = 16 and elements_per_batch = 16, we get:
* max_batches_per_sg = wg_size
*/
uint32_t max_batches_per_sg = wg_size;
/* QPU threads will stall at TSY barriers until the entire supergroup
* reaches the barrier. Limit the supergroup size to half the QPU threads
* available, so we can have at least 2 supergroups executing in parallel
* and we don't stall all our QPU threads when a supergroup hits a barrier.
*/
if (has_tsy_barrier) {
uint32_t max_qpu_threads = devinfo->qpu_count * threads;
max_batches_per_sg = MIN2(max_batches_per_sg, max_qpu_threads / 2);
}
uint32_t max_wgs_per_sg = max_batches_per_sg * 16 / wg_size;
uint32_t best_wgs_per_sg = 1;
uint32_t best_unused_lanes = 16;
for (uint32_t wgs_per_sg = 1; wgs_per_sg <= max_wgs_per_sg; wgs_per_sg++) {
/* Don't try to pack more workgroups per supergroup than the total amount
* of workgroups dispatched.
*/
if (wgs_per_sg > num_wgs)
return best_wgs_per_sg;
/* Compute wasted lines for this configuration and keep track of the
* config with less waste.
*/
uint32_t unused_lanes = (16 - ((wgs_per_sg * wg_size) % 16)) & 0x0f;
if (unused_lanes == 0)
return wgs_per_sg;
if (unused_lanes < best_unused_lanes) {
best_wgs_per_sg = wgs_per_sg;
best_unused_lanes = unused_lanes;
}
}
return best_wgs_per_sg;
}
src/broadcom/common/v3d_util.h
+36
View File
@@ -0,0 +1,36 @@
/*
* Copyright © 2021 Raspberry Pi
*
* 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 (including the next
* paragraph) 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.
*/
#ifndef V3D_UTIL_H
#define V3D_UTIL_H
#include "common/v3d_device_info.h"
uint32_t
v3d_csd_choose_workgroups_per_supergroup(struct v3d_device_info *devinfo,
bool has_tsy_barrier,
uint32_t threads,
uint32_t num_wgs,
uint32_t wg_size);
#endif
src/broadcom/meson.build
+1 -1
View File
@@ -53,7 +53,7 @@ endforeach
libbroadcom_v3d = static_library(
'libbroadcom_v3d',
[
files('common/v3d_debug.c', 'common/v3d_device_info.c', 'clif/clif_dump.c'),
files('common/v3d_debug.c', 'common/v3d_device_info.c', 'clif/clif_dump.c', 'common/v3d_util.c'),
v3d_xml_pack,
],
include_directories : [inc_include, inc_src, inc_mapi, inc_mesa, inc_gallium, inc_gallium_aux, inc_broadcom],
src/broadcom/vulkan/v3dv_cmd_buffer.c
+6 -58
View File
@@ -5268,62 +5268,6 @@ v3dv_cmd_buffer_rewrite_indirect_csd_job(
}
}
/* Choose a number of workgroups per supergroup that maximizes
* lane occupancy. We can pack up to 16 workgroups into a supergroup.
*/
static uint32_t
choose_workgroups_per_supergroup(struct v3d_device_info *devinfo,
struct v3dv_shader_variant *cs,
uint32_t num_wgs,
uint32_t wg_size)
{
/* Compute maximum number of batches in a supergroup for this workgroup size.
* Each batch is 16 elements, and we can have up to 16 work groups in a
* supergroup:
*
* max_batches_per_sg = (wg_size * max_wgs_per_sg) / elements_per_batch
* since max_wgs_per_sg = 16 and elements_per_batch = 16, we get:
* max_batches_per_sg = wg_size
*/
uint32_t max_batches_per_sg = wg_size;
/* QPU threads will stall at TSY barriers until the entire supergroup
* reaches the barrier. Limit the supergroup size to half the QPU threads
* available, so we can have at least 2 supergroups executing in parallel
* and we don't stall all our QPU threads when a supergroup hits a barrier.
*/
if (cs->prog_data.cs->base.has_control_barrier) {
uint32_t max_qpu_threads =
devinfo->qpu_count * cs->prog_data.cs->base.threads;
max_batches_per_sg = MIN2(max_batches_per_sg, max_qpu_threads / 2);
}
uint32_t max_wgs_per_sg = max_batches_per_sg * 16 / wg_size;
uint32_t best_wgs_per_sg = 1;
uint32_t best_unused_lanes = 16;
for (uint32_t wgs_per_sg = 1; wgs_per_sg <= max_wgs_per_sg; wgs_per_sg++) {
/* Don't try to pack more workgroups per supergroup than the total amount
* of workgroups dispatched.
*/
if (wgs_per_sg > num_wgs)
return best_wgs_per_sg;
/* Compute wasted lines for this configuration and keep track of the
* config with less waste.
*/
uint32_t unused_lanes = (16 - ((wgs_per_sg * wg_size) % 16)) & 0x0f;
if (unused_lanes == 0)
return wgs_per_sg;
if (unused_lanes < best_unused_lanes) {
best_wgs_per_sg = wgs_per_sg;
best_unused_lanes = unused_lanes;
}
}
return best_wgs_per_sg;
}
static struct v3dv_job *
cmd_buffer_create_csd_job(struct v3dv_cmd_buffer *cmd_buffer,
uint32_t group_count_x,
@@ -5367,8 +5311,12 @@ cmd_buffer_create_csd_job(struct v3dv_cmd_buffer *cmd_buffer,
cpd->local_size[2];
uint32_t wgs_per_sg =
choose_workgroups_per_supergroup(&cmd_buffer->device->devinfo,
cs_variant, num_wgs, wg_size);
v3d_csd_choose_workgroups_per_supergroup(
&cmd_buffer->device->devinfo,
cs_variant->prog_data.cs->base.has_control_barrier,
cs_variant->prog_data.cs->base.threads,
num_wgs, wg_size);
uint32_t batches_per_sg = DIV_ROUND_UP(wgs_per_sg * wg_size, 16);
uint32_t whole_sgs = num_wgs / wgs_per_sg;
uint32_t rem_wgs = num_wgs - whole_sgs * wgs_per_sg;
src/broadcom/vulkan/v3dv_private.h
+1
View File
@@ -56,6 +56,7 @@
#include "common/v3d_device_info.h"
#include "common/v3d_limits.h"
#include "common/v3d_util.h"
#include "compiler/shader_enums.h"
#include "compiler/spirv/nir_spirv.h"