fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 07:20:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

panvk: fix calculate_task_axis_and_increment

Browse source 
task_axis selects the dim of the global workgroup, not the dim of the
local workgroup.

v2: fix assert for dEQP-VK.compute.pipeline.basic.empty_workgroup*

Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Tested-by: Yiwei Zhang <zzyiwei@chromium.org> (v1)
Reviewed-by: Christoph Pillmayer <christoph.pillmayer@arm.com> (v1)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37988>
This commit is contained in:
Chia-I Wu 2025-10-20 15:06:31 -07:00 • committed by Marge Bot
parent 546d73721b
commit 5fd32d79ee
3 changed files with 15 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
src/panfrost/vulkan/csf/panvk_cmd_buffer.h
View file

@ -672,8 +672,8 @@ void panvk_per_arch(cmd_inherit_render_state)(
static inline void
panvk_per_arch(calculate_task_axis_and_increment)(
const struct panvk_shader_variant *shader,
struct panvk_physical_device *phys_dev, unsigned *task_axis,
unsigned *task_increment)
struct panvk_physical_device *phys_dev, const struct pan_compute_dim *wg_dim,
unsigned *task_axis, unsigned *task_increment)
{
/* Pick the task_axis and task_increment to maximize thread
* utilization. */
@ -682,14 +682,17 @@ panvk_per_arch(calculate_task_axis_and_increment)(
unsigned max_thread_cnt = pan_compute_max_thread_count(
&phys_dev->kmod.props, shader->info.work_reg_count);
unsigned threads_per_task = threads_per_wg;
unsigned local_size[3] = {
shader->cs.local_size.x,
shader->cs.local_size.y,
shader->cs.local_size.z,
};
const unsigned wg_count[3] = {wg_dim->x, wg_dim->y, wg_dim->z};
const unsigned total_wgs = wg_dim->x * wg_dim->y * wg_dim->z;
if (!total_wgs) {
*task_axis = MALI_TASK_AXIS_X;
*task_increment = 1;
return;
}
for (unsigned i = 0; i < 3; i++) {
if (threads_per_task * local_size[i] >= max_thread_cnt) {
if (threads_per_task * wg_count[i] >= max_thread_cnt) {
/* We reached out thread limit, stop at the current axis and
* calculate the increment so it doesn't exceed the per-core
* thread capacity.
@ -701,11 +704,11 @@ panvk_per_arch(calculate_task_axis_and_increment)(
* threads. Pick the current axis grid size as our increment
* as there's no point using something bigger.
*/
*task_increment = local_size[i];
*task_increment = wg_count[i];
break;
}
threads_per_task *= local_size[i];
threads_per_task *= wg_count[i];
(*task_axis)++;
}

2
src/panfrost/vulkan/csf/panvk_vX_cmd_dispatch.c
View file

@ -296,7 +296,7 @@ cmd_dispatch(struct panvk_cmd_buffer *cmdbuf, struct panvk_dispatch_info *info)
unsigned task_axis = MALI_TASK_AXIS_X;
unsigned task_increment = 0;
panvk_per_arch(calculate_task_axis_and_increment)(
cs, phys_dev, &task_axis, &task_increment);
cs, phys_dev, &dim, &task_axis, &task_increment);
cs_trace_run_compute(b, tracing_ctx, cs_scratch_reg_tuple(b, 0, 4),
task_increment, task_axis,
cs_shader_res_sel(0, 0, 0, 0));

2
src/panfrost/vulkan/csf/panvk_vX_cmd_precomp.c
View file

@ -117,7 +117,7 @@ panvk_per_arch(dispatch_precomp)(struct panvk_precomp_ctx *ctx,
unsigned task_axis = MALI_TASK_AXIS_X;
unsigned task_increment = 0;
panvk_per_arch(calculate_task_axis_and_increment)(
shader, phys_dev, &task_axis, &task_increment);
shader, phys_dev, &dim, &task_axis, &task_increment);
cs_trace_run_compute(b, tracing_ctx, cs_scratch_reg_tuple(b, 0, 4),
task_increment, task_axis,
cs_shader_res_sel(0, 0, 0, 0));