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

zink: expand unordered_exec 

track read+write

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17667>
This commit is contained in:
Mike Blumenkrantz 2022-07-14 21:52:44 -04:00 committed by Marge Bot
parent b1781ec9ab
commit ca03e35821
7 changed files with 67 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/gallium/drivers/zink/zink_batch.c
View File

@ -38,7 +38,7 @@ zink_reset_batch_state(struct zink_context *ctx, struct zink_batch_state *bs)
set_foreach_remove(bs->resources, entry) {
struct zink_resource_object *obj = (struct zink_resource_object *)entry->key;
if (!zink_resource_object_usage_unset(obj, bs)) {
obj->unordered_exec = false;
obj->unordered_read = obj->unordered_write = false;
obj->access = 0;
obj->access_stage = 0;
}

4
src/gallium/drivers/zink/zink_blit.c
View File

@ -372,8 +372,8 @@ zink_blit(struct pipe_context *pctx,
util_blitter_blit(ctx->blitter, info);
}
end:
src->obj->unordered_exec = false;
dst->obj->unordered_exec = false;
src->obj->unordered_read = false;
dst->obj->unordered_read = dst->obj->unordered_write = false;
if (needs_present_readback)
zink_kopper_present_readback(ctx, src);
}

2
src/gallium/drivers/zink/zink_clear.c
View File

@ -498,7 +498,7 @@ zink_clear_buffer(struct pipe_context *pctx,
zink_batch_reference_resource_rw(batch, res, true);
util_range_add(&res->base.b, &res->valid_buffer_range, offset, offset + size);
zink_resource_buffer_barrier(ctx, res, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT);
res->obj->unordered_exec = false;
res->obj->unordered_read = res->obj->unordered_write = false;
VKCTX(CmdFillBuffer)(batch->state->cmdbuf, res->obj->buffer, offset, size, *(uint32_t*)clear_value);
return;
}

70
src/gallium/drivers/zink/zink_context.c
View File

@ -1144,7 +1144,7 @@ zink_set_vertex_buffers(struct pipe_context *pctx,
/* always barrier before possible rebind */
zink_resource_buffer_barrier(ctx, res, VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
} else {
enabled_buffers &= ~BITFIELD_BIT(start_slot + i);
}
@ -1296,7 +1296,7 @@ zink_set_constant_buffer(struct pipe_context *pctx,
zink_batch_resource_usage_set(&ctx->batch, new_res, false);
zink_resource_buffer_barrier(ctx, new_res, VK_ACCESS_UNIFORM_READ_BIT,
new_res->gfx_barrier);
new_res->obj->unordered_exec = false;
new_res->obj->unordered_read = false;
}
update |= ((index || zink_descriptor_mode == ZINK_DESCRIPTOR_MODE_LAZY) && ctx->ubos[shader][index].buffer_offset != offset) ||
!!res != !!buffer || (res && res->obj->buffer != new_res->obj->buffer) ||
@ -1417,7 +1417,10 @@ zink_set_shader_buffers(struct pipe_context *pctx,
update = true;
max_slot = MAX2(max_slot, start_slot + i);
update_descriptor_state_ssbo(ctx, p_stage, start_slot + i, new_res);
new_res->obj->unordered_exec = false;
if (zink_resource_access_is_write(access))
new_res->obj->unordered_read = new_res->obj->unordered_write = false;
else
new_res->obj->unordered_read = false;
} else {
update = !!res;
ssbo->buffer_offset = 0;
@ -1646,7 +1649,10 @@ zink_set_shader_images(struct pipe_context *pctx,
zink_resource_access_is_write(access));
update = true;
update_descriptor_state_image(ctx, p_stage, start_slot + i, res);
res->obj->unordered_exec = false;
if (zink_resource_access_is_write(access))
res->obj->unordered_read = res->obj->unordered_write = false;
else
res->obj->unordered_read = false;
res->image_binds[p_stage] |= BITFIELD_BIT(start_slot + i);
} else if (image_view->base.resource) {
update = true;
@ -1769,7 +1775,7 @@ zink_set_sampler_views(struct pipe_context *pctx,
}
res->sampler_binds[shader_type] |= BITFIELD_BIT(start_slot + i);
zink_batch_resource_usage_set(&ctx->batch, res, false);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
} else if (a) {
unbind_samplerview(ctx, shader_type, start_slot + i);
update = true;
@ -1934,7 +1940,7 @@ zink_make_texture_handle_resident(struct pipe_context *pctx, uint64_t handle, bo
util_dynarray_append(&ctx->di.bindless[0].resident, struct zink_bindless_descriptor *, bd);
uint32_t h = is_buffer ? handle + ZINK_MAX_BINDLESS_HANDLES : handle;
util_dynarray_append(&ctx->di.bindless[0].updates, uint32_t, h);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
} else {
zero_bindless_descriptor(ctx, handle, is_buffer, false);
util_dynarray_delete_unordered(&ctx->di.bindless[0].resident, struct zink_bindless_descriptor *, bd);
@ -2053,7 +2059,10 @@ zink_make_image_handle_resident(struct pipe_context *pctx, uint64_t handle, unsi
util_dynarray_append(&ctx->di.bindless[1].resident, struct zink_bindless_descriptor *, bd);
uint32_t h = is_buffer ? handle + ZINK_MAX_BINDLESS_HANDLES : handle;
util_dynarray_append(&ctx->di.bindless[1].updates, uint32_t, h);
res->obj->unordered_exec = false;
if (zink_resource_access_is_write(access))
res->obj->unordered_read = res->obj->unordered_write = false;
else
res->obj->unordered_read = false;
} else {
zero_bindless_descriptor(ctx, handle, is_buffer, true);
util_dynarray_delete_unordered(&ctx->di.bindless[1].resident, struct zink_bindless_descriptor *, bd);
@ -2440,7 +2449,7 @@ zink_prep_fb_attachment(struct zink_context *ctx, struct zink_surface *surf, uns
layout = zink_render_pass_attachment_get_barrier_info(&rt, i < ctx->fb_state.nr_cbufs, &pipeline, &access);
}
zink_resource_image_barrier(ctx, res, layout, access, pipeline);
res->obj->unordered_exec = false;
res->obj->unordered_read = res->obj->unordered_write = false;
if (i == ctx->fb_state.nr_cbufs && res->sampler_bind_count[0])
update_res_sampler_layouts(ctx, res);
return surf->image_view;
@ -2571,7 +2580,10 @@ update_resource_refs_for_stage(struct zink_context *ctx, enum pipe_shader_type s
continue;
bool is_write = zink_resource_access_is_write(get_access_flags_for_binding(ctx, i, stage, j));
zink_batch_resource_usage_set(batch, res, is_write);
res->obj->unordered_exec = false;
if (is_write)
res->obj->unordered_read = res->obj->unordered_write = false;
else
res->obj->unordered_read = false;
struct zink_sampler_view *sv = zink_sampler_view(ctx->sampler_views[stage][j]);
struct zink_sampler_state *sampler_state = ctx->sampler_states[stage][j];
@ -2616,7 +2628,7 @@ zink_update_descriptor_refs(struct zink_context *ctx, bool compute)
struct zink_resource *res = zink_resource(ctx->vertex_buffers[i].buffer.resource);
if (res) {
zink_batch_resource_usage_set(batch, res, false);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
}
}
if (ctx->curr_program)
@ -2628,7 +2640,10 @@ zink_update_descriptor_refs(struct zink_context *ctx, bool compute)
util_dynarray_foreach(&ctx->di.bindless[i].resident, struct zink_bindless_descriptor*, bd) {
struct zink_resource *res = zink_descriptor_surface_resource(&(*bd)->ds);
zink_batch_resource_usage_set(&ctx->batch, res, (*bd)->access & PIPE_IMAGE_ACCESS_WRITE);
res->obj->unordered_exec = false;
if ((*bd)->access & PIPE_IMAGE_ACCESS_WRITE)
res->obj->unordered_read = res->obj->unordered_write = false;
else
res->obj->unordered_read = false;
}
}
}
@ -3123,18 +3138,30 @@ resource_check_defer_buffer_barrier(struct zink_context *ctx, struct zink_resour
}
static inline bool
unordered_res_exec(const struct zink_context *ctx, const struct zink_resource *res)
unordered_res_exec(const struct zink_context *ctx, const struct zink_resource *res, bool is_write)
{
return !zink_resource_usage_matches(res, ctx->batch.state) || res->obj->unordered_exec;
/* if all usage is unordered, keep unordered */
if (res->obj->unordered_read && res->obj->unordered_write)
return true;
/* if testing write access but have any ordered read access, cannot promote */
if (is_write && zink_batch_usage_matches(res->obj->bo->reads, ctx->batch.state) && !res->obj->unordered_read)
return false;
/* if write access is unordered or nonexistent, always promote */
return res->obj->unordered_write || !zink_batch_usage_matches(res->obj->bo->writes, ctx->batch.state);
}
static inline VkCommandBuffer
get_cmdbuf(struct zink_context *ctx, struct zink_resource *res, struct zink_resource *res_2)
get_cmdbuf(struct zink_context *ctx, struct zink_resource *src, struct zink_resource *dst)
{
bool unordered_exec = unordered_res_exec(ctx, res) && (!res_2 || unordered_res_exec(ctx, res_2)) && (zink_debug & ZINK_DEBUG_NOREORDER) == 0;
res->obj->unordered_exec = unordered_exec;
if (res_2)
res_2->obj->unordered_exec = unordered_exec;
bool unordered_exec = (zink_debug & ZINK_DEBUG_NOREORDER) == 0;
if (src)
unordered_exec &= unordered_res_exec(ctx, src, false);
if (dst)
unordered_exec &= unordered_res_exec(ctx, dst, true);
if (src)
src->obj->unordered_read = unordered_exec;
if (dst)
dst->obj->unordered_write = unordered_exec;
if (unordered_exec) {
ctx->batch.state->has_barriers = true;
return ctx->batch.state->barrier_cmdbuf;
@ -3180,7 +3207,8 @@ zink_resource_image_barrier(struct zink_context *ctx, struct zink_resource *res,
if (!zink_resource_image_barrier_init(&imb, res, new_layout, flags, pipeline))
return;
/* only barrier if we're changing layout or doing something besides read -> read */
VkCommandBuffer cmdbuf = get_cmdbuf(ctx, res, NULL);
bool is_write = zink_resource_access_is_write(imb.dstAccessMask);
VkCommandBuffer cmdbuf = is_write ? get_cmdbuf(ctx, NULL, res) : get_cmdbuf(ctx, res, NULL);
assert(new_layout);
if (!res->obj->access_stage)
imb.srcAccessMask = 0;
@ -3277,7 +3305,8 @@ zink_resource_buffer_barrier(struct zink_context *ctx, struct zink_resource *res
bmb.dstAccessMask = flags;
if (!res->obj->access_stage)
bmb.srcAccessMask = 0;
VkCommandBuffer cmdbuf = get_cmdbuf(ctx, res, NULL);
bool is_write = zink_resource_access_is_write(flags);
VkCommandBuffer cmdbuf = is_write ? get_cmdbuf(ctx, NULL, res) : get_cmdbuf(ctx, res, NULL);
/* only barrier if we're changing layout or doing something besides read -> read */
VKCTX(CmdPipelineBarrier)(
cmdbuf,
@ -3677,7 +3706,6 @@ zink_set_stream_output_targets(struct pipe_context *pctx,
if (so) {
so->so_bind_count++;
update_res_bind_count(ctx, so, false, false);
so->obj->unordered_exec = false;
}
}
for (unsigned i = num_targets; i < ctx->num_so_targets; i++) {

17
src/gallium/drivers/zink/zink_draw.cpp
View File

@ -40,7 +40,7 @@ zink_emit_xfb_counter_barrier(struct zink_context *ctx)
stage |= VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT;
}
zink_resource_buffer_barrier(ctx, res, access, stage);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
}
}
@ -86,7 +86,7 @@ check_buffer_barrier(struct zink_context *ctx, struct pipe_resource *pres, VkAcc
{
struct zink_resource *res = zink_resource(pres);
zink_resource_buffer_barrier(ctx, res, flags, pipeline);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
}
ALWAYS_INLINE static void
@ -345,7 +345,10 @@ update_barriers(struct zink_context *ctx, bool is_compute,
if (layout != res->layout)
zink_resource_image_barrier(ctx, res, layout, res->barrier_access[is_compute], pipeline);
}
res->obj->unordered_exec = false;
if (zink_resource_access_is_write(res->barrier_access[is_compute]))
res->obj->unordered_read = res->obj->unordered_write = false;
else
res->obj->unordered_read = false;
/* always barrier on draw if this resource has either multiple image write binds or
* image write binds and image read binds
*/
@ -484,7 +487,7 @@ zink_draw(struct pipe_context *pctx,
struct zink_resource *res = zink_resource(t->base.buffer);
zink_resource_buffer_barrier(ctx, res,
VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT, VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT);
res->obj->unordered_exec = false;
res->obj->unordered_read = res->obj->unordered_write = false;
}
}
}
@ -502,7 +505,7 @@ zink_draw(struct pipe_context *pctx,
zink_resource_buffer_barrier(ctx, res,
VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
}
zink_query_update_gs_states(ctx, dinfo->was_line_loop);
@ -775,7 +778,7 @@ zink_draw(struct pipe_context *pctx,
struct zink_resource *res = zink_resource(t->counter_buffer);
t->stride = ctx->last_vertex_stage->sinfo.so_info.stride[i] * sizeof(uint32_t);
zink_batch_reference_resource_rw(batch, res, true);
res->obj->unordered_exec = false;
res->obj->unordered_read = res->obj->unordered_write = false;
if (t->counter_buffer_valid) {
counter_buffers[i] = res->obj->buffer;
counter_buffer_offsets[i] = t->counter_buffer_offset;
@ -892,7 +895,7 @@ zink_draw_vertex_state(struct pipe_context *pctx,
struct zink_resource *res = zink_resource(vstate->input.vbuffer.buffer.resource);
zink_resource_buffer_barrier(ctx, res, VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT);
res->obj->unordered_exec = false;
res->obj->unordered_read = false;
struct zink_vertex_elements_hw_state *hw_state = ctx->gfx_pipeline_state.element_state;
ctx->gfx_pipeline_state.element_state = &((struct zink_vertex_state*)vstate)->velems.hw_state;

4
src/gallium/drivers/zink/zink_query.c
View File

@ -732,7 +732,7 @@ copy_pool_results_to_buffer(struct zink_context *ctx, struct zink_query *query,
zink_resource_buffer_barrier(ctx, res, VK_ACCESS_TRANSFER_WRITE_BIT, 0);
util_range_add(&res->base.b, &res->valid_buffer_range, offset, offset + result_size);
assert(query_id < NUM_QUERIES);
res->obj->unordered_exec = false;
res->obj->unordered_read = res->obj->unordered_write = false;
VKCTX(CmdCopyQueryPoolResults)(batch->state->cmdbuf, pool, query_id, num_results, res->obj->buffer,
offset, base_result_size, flags);
}
@ -1135,7 +1135,7 @@ zink_start_conditional_render(struct zink_context *ctx)
begin_info.sType = VK_STRUCTURE_TYPE_CONDITIONAL_RENDERING_BEGIN_INFO_EXT;
begin_info.buffer = ctx->render_condition.query->predicate->obj->buffer;
begin_info.flags = begin_flags;
ctx->render_condition.query->predicate->obj->unordered_exec = false;
ctx->render_condition.query->predicate->obj->unordered_read = false;
VKCTX(CmdBeginConditionalRenderingEXT)(batch->state->cmdbuf, &begin_info);
zink_batch_reference_resource_rw(batch, ctx->render_condition.query->predicate, false);
ctx->render_condition.active = true;

3
src/gallium/drivers/zink/zink_resource.h
View File

@ -60,7 +60,8 @@ struct zink_resource_object {
VkPipelineStageFlagBits access_stage;
VkAccessFlags access;
bool unordered_exec;
bool unordered_read;
bool unordered_write;
unsigned persistent_maps; //if nonzero, requires vkFlushMappedMemoryRanges during batch use
struct zink_descriptor_refs desc_set_refs;