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

zink: add a "compact" descriptor mode 

this compacts the ubo+ssbo and samplerview+image sets, enabling zink
to run fully featured with only 4 descriptor sets used

fixes #4964

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16645>
This commit is contained in:
Mike Blumenkrantz 2022-05-16 17:28:59 -04:00 committed by Marge Bot
parent 0adf2dc875
commit 8636717270
5 changed files with 283 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

115
src/gallium/drivers/zink/zink_descriptors.c
View File

@ -60,6 +60,7 @@ struct zink_descriptor_set {
bool invalid;
bool punted;
bool recycled;
uint8_t compacted; //bitmask of zink_descriptor_type
struct zink_descriptor_state_key key;
struct zink_batch_usage *batch_uses;
#ifndef NDEBUG
@ -730,6 +731,37 @@ populate_zds_key(struct zink_context *ctx, enum zink_descriptor_type type, bool
}
}
static void
populate_zds_key_compact(struct zink_context *ctx, enum zink_descriptor_type type, bool is_compute,
struct zink_descriptor_state_key *key, uint32_t push_usage)
{
if (is_compute) {
for (unsigned i = 1; i < ZINK_SHADER_COUNT; i++)
key->exists[i] = false;
key->exists[0] = true;
if (type == ZINK_DESCRIPTOR_TYPES)
key->state[0] = ctx->dd->push_state[is_compute];
else {
assert(ctx->dd->compact_descriptor_states[is_compute].valid[type]);
key->state[0] = ctx->dd->compact_descriptor_states[is_compute].state[type];
}
} else if (type == ZINK_DESCRIPTOR_TYPES) {
/* gfx only */
for (unsigned i = 0; i < ZINK_SHADER_COUNT; i++) {
if (push_usage & BITFIELD_BIT(i)) {
key->exists[i] = true;
key->state[i] = ctx->dd->gfx_push_state[i];
} else
key->exists[i] = false;
}
} else {
for (unsigned i = 0; i < ZINK_SHADER_COUNT; i++) {
key->exists[i] = ctx->dd->compact_gfx_descriptor_states[i].valid[type];
key->state[i] = ctx->dd->compact_gfx_descriptor_states[i].state[type];
}
}
}
static void
punt_invalid_set(struct zink_descriptor_set *zds, struct hash_entry *he)
{
@ -748,6 +780,7 @@ zink_descriptor_set_get(struct zink_context *ctx,
bool *cache_hit)
{
*cache_hit = false;
struct zink_screen *screen = zink_screen(ctx->base.screen);
struct zink_descriptor_set *zds;
struct zink_program *pg = is_compute ? (struct zink_program *)ctx->curr_compute : (struct zink_program *)ctx->curr_program;
struct zink_batch *batch = &ctx->batch;
@ -757,7 +790,10 @@ zink_descriptor_set_get(struct zink_context *ctx,
assert(type <= ZINK_DESCRIPTOR_TYPES);
assert(pool->key->layout->num_bindings);
assert(!screen->compact_descriptors || (type != ZINK_DESCRIPTOR_TYPE_SSBO && type != ZINK_DESCRIPTOR_TYPE_IMAGE));
uint32_t hash = push_set ? ctx->dd->push_state[is_compute] :
screen->compact_descriptors ?
ctx->dd->compact_descriptor_states[is_compute].state[type] :
ctx->dd->descriptor_states[is_compute].state[type];
struct zink_descriptor_set *last_set = push_set ? ctx->dd->last_set[is_compute] : pdd_cached(pg)->last_set[type];
@ -766,12 +802,17 @@ zink_descriptor_set_get(struct zink_context *ctx,
* early return here can be done safely and with no locking
*/
if (last_set && ((push_set && !ctx->dd->changed[is_compute][ZINK_DESCRIPTOR_TYPES]) ||
(!push_set && !ctx->dd->changed[is_compute][type]))) {
(!push_set && (screen->compact_descriptors ?
!ctx->dd->changed[is_compute][type] && !ctx->dd->changed[is_compute][type+2] :
!ctx->dd->changed[is_compute][type])))) {
*cache_hit = true;
return last_set;
}
struct zink_descriptor_state_key key;
if (screen->compact_descriptors)
populate_zds_key_compact(ctx, type, is_compute, &key, pg->dd->push_usage);
else
populate_zds_key(ctx, type, is_compute, &key, pg->dd->push_usage);
simple_mtx_lock(&pool->mtx);
@ -860,10 +901,26 @@ out:
_mesa_hash_table_num_entries(pool->free_desc_sets) < ZINK_DEFAULT_MAX_DESCS - ZINK_DEFAULT_DESC_CLAMP))
ctx->oom_flush = ctx->oom_stall = true;
zds->hash = hash;
if (screen->compact_descriptors)
populate_zds_key_compact(ctx, type, is_compute, &zds->key, pg->dd->push_usage);
else
populate_zds_key(ctx, type, is_compute, &zds->key, pg->dd->push_usage);
zds->recycled = false;
_mesa_hash_table_insert_pre_hashed(pool->desc_sets, hash, &zds->key, zds);
quick_out:
if (!push_set) {
if (screen->compact_descriptors) {
if (zink_desc_type_from_vktype(pool->key->sizes[0].type) == type)
zds->compacted |= BITFIELD_BIT(type);
for (unsigned i = 0; i < pool->key->num_type_sizes; i++) {
if (zink_desc_type_from_vktype(pool->key->sizes[0].type) == type + 2) {
zds->compacted |= BITFIELD_BIT(type + 2);
break;
}
}
} else
zds->compacted |= BITFIELD_BIT(type);
}
zds->punted = zds->invalid = false;
batch_add_desc_set(batch, zds);
if (push_set)
@ -1282,7 +1339,13 @@ update_push_ubo_descriptors(struct zink_context *ctx, struct zink_descriptor_set
static void
set_descriptor_set_refs(struct zink_context *ctx, struct zink_descriptor_set *zds, struct zink_program *pg, bool cache_hit)
{
enum zink_descriptor_type type = zds->pool->type;
const bool compact_descriptors = zink_screen(ctx->base.screen)->compact_descriptors;
STATIC_ASSERT(ZINK_DESCRIPTOR_TYPE_UBO + 2 == ZINK_DESCRIPTOR_TYPE_SSBO);
STATIC_ASSERT(ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW + 2 == ZINK_DESCRIPTOR_TYPE_IMAGE);
const enum zink_descriptor_type types[] = {zds->pool->type, zds->pool->type + 2};
unsigned num_types = compact_descriptors ? 2 : 1;
for (unsigned n = 0; n < num_types; n++) {
const enum zink_descriptor_type type = types[n];
for (unsigned i = 0; i < pdd_cached(pg)->num_refs[type]; i++) {
switch (type) {
case ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW:
@ -1298,6 +1361,7 @@ set_descriptor_set_refs(struct zink_context *ctx, struct zink_descriptor_set *zd
}
}
}
}
static void
update_descriptors_internal(struct zink_context *ctx, enum zink_descriptor_type type, struct zink_descriptor_set *zds, struct zink_program *pg, bool cache_hit)
@ -1326,6 +1390,11 @@ update_descriptors_internal(struct zink_context *ctx, enum zink_descriptor_type
VkWriteDescriptorSet wds[ZINK_MAX_DESCRIPTORS_PER_TYPE];
unsigned num_wds = 0;
const enum zink_descriptor_type types[2] = {type, type + 2};
for (unsigned n = 0; n < ARRAY_SIZE(types); n++) {
if (!(zds->compacted & BITFIELD_BIT(types[n])))
continue;
type = types[n];
for (int i = 0; i < num_stages; i++) {
struct zink_shader *shader = stages[i];
if (!shader)
@ -1386,6 +1455,7 @@ update_descriptors_internal(struct zink_context *ctx, enum zink_descriptor_type
num_wds = init_write_descriptor(shader, zds->desc_set, type, j, &wds[num_wds], num_wds);
}
}
}
if (num_wds)
VKSCR(UpdateDescriptorSets)(screen->dev, num_wds, wds, 0, NULL);
}
@ -1556,6 +1626,11 @@ calc_descriptor_state_hash_ssbo(struct zink_context *ctx, struct zink_shader *zs
struct pipe_shader_buffer *ssbo = &ctx->ssbos[shader][idx];
hash = XXH32(&ssbo->buffer_offset, sizeof(ssbo->buffer_offset), hash);
hash = XXH32(&ssbo->buffer_size, sizeof(ssbo->buffer_size), hash);
/* compacted sets need a way to differentiate between a buffer bound as a ubo vs ssbo */
if (zink_screen(ctx->base.screen)->compact_descriptors) {
uint32_t writable = ctx->writable_ssbos[shader] & BITFIELD_BIT(idx);
hash = XXH32(&writable, sizeof(writable), hash);
}
}
return hash;
}
@ -1705,9 +1780,45 @@ zink_context_update_descriptor_states(struct zink_context *ctx, struct zink_prog
}
for (unsigned i = 0; i < ZINK_DESCRIPTOR_TYPES; i++) {
if (pdd_cached(pg)->pool[screen->desc_set_id[i] - 1] && pdd_cached(pg)->cache_misses[i] < MAX_CACHE_MISSES &&
ctx->dd->changed[pg->is_compute][i] &&
!ctx->dd->descriptor_states[pg->is_compute].valid[i])
update_descriptor_state(ctx, i, pg->is_compute);
}
if (!screen->compact_descriptors)
return;
for (unsigned n = 0; n < 2; n++) {
ctx->dd->compact_descriptor_states[pg->is_compute].valid[n] = ctx->dd->descriptor_states[pg->is_compute].valid[n] |
ctx->dd->descriptor_states[pg->is_compute].valid[n + 2];
if (ctx->dd->compact_descriptor_states[pg->is_compute].valid[n]) {
if (pg->is_compute) {
ctx->dd->compact_descriptor_states[pg->is_compute].state[n] = ctx->dd->descriptor_states[pg->is_compute].state[n] ^
ctx->dd->descriptor_states[pg->is_compute].state[n + 2];
} else {
uint32_t hash = 0;
bool first = true;
for (unsigned i = 0; i < ZINK_SHADER_COUNT; i++) {
ctx->dd->compact_gfx_descriptor_states[i].valid[n] = ctx->dd->gfx_descriptor_states[i].valid[n] |
ctx->dd->gfx_descriptor_states[i].valid[n + 2];
if (ctx->dd->compact_gfx_descriptor_states[i].valid[n]) {
ctx->dd->compact_gfx_descriptor_states[i].state[n] = ctx->dd->gfx_descriptor_states[i].state[n] ^
ctx->dd->gfx_descriptor_states[i].state[n + 2];
if (first)
hash = ctx->dd->compact_gfx_descriptor_states[i].state[n];
else
hash = XXH32(&ctx->dd->compact_gfx_descriptor_states[i].state[n], sizeof(uint32_t), hash);
first = false;
} else {
ctx->dd->compact_gfx_descriptor_states[i].state[n] = 0;
}
}
ctx->dd->compact_descriptor_states[pg->is_compute].state[n] = hash;
}
} else {
ctx->dd->compact_descriptor_states[pg->is_compute].state[n] = 0;
}
}
}
void

50
src/gallium/drivers/zink/zink_descriptors.h
View File

@ -73,6 +73,15 @@ enum zink_descriptor_size_index {
ZDS_INDEX_STORAGE_TEXELS,
};
enum zink_descriptor_size_index_compact {
ZDS_INDEX_COMP_UBO,
ZDS_INDEX_COMP_STORAGE_BUFFER,
ZDS_INDEX_COMP_COMBINED_SAMPLER,
ZDS_INDEX_COMP_UNIFORM_TEXELS,
ZDS_INDEX_COMP_STORAGE_IMAGE,
ZDS_INDEX_COMP_STORAGE_TEXELS,
};
struct hash_table;
struct zink_context;
@ -103,7 +112,7 @@ struct zink_descriptor_pool_key {
unsigned use_count;
unsigned num_type_sizes;
struct zink_descriptor_layout_key *layout;
VkDescriptorPoolSize sizes[2];
VkDescriptorPoolSize sizes[4];
};
struct zink_descriptor_reference {
@ -114,6 +123,8 @@ struct zink_descriptor_reference {
struct zink_descriptor_data {
struct zink_descriptor_state gfx_descriptor_states[ZINK_SHADER_COUNT]; // keep incremental hashes here
struct zink_descriptor_state descriptor_states[2]; // gfx, compute
struct zink_descriptor_state compact_gfx_descriptor_states[ZINK_SHADER_COUNT]; // keep incremental hashes here
struct zink_descriptor_state compact_descriptor_states[2]; // gfx, compute
struct hash_table *descriptor_pools[ZINK_DESCRIPTOR_TYPES];
struct zink_descriptor_layout_key *push_layout_keys[2]; //gfx, compute
@ -178,6 +189,28 @@ zink_vktype_to_size_idx(VkDescriptorType type)
unreachable("unknown type");
}
static inline enum zink_descriptor_size_index_compact
zink_vktype_to_size_idx_comp(VkDescriptorType type)
{
switch (type) {
case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
return ZDS_INDEX_COMP_UBO;
case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
return ZDS_INDEX_COMP_COMBINED_SAMPLER;
case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
return ZDS_INDEX_COMP_UNIFORM_TEXELS;
case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
return ZDS_INDEX_COMP_STORAGE_BUFFER;
case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
return ZDS_INDEX_COMP_STORAGE_IMAGE;
case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
return ZDS_INDEX_COMP_STORAGE_TEXELS;
default: break;
}
unreachable("unknown type");
}
static inline enum zink_descriptor_size_index
zink_descriptor_type_to_size_idx(enum zink_descriptor_type type)
{
@ -194,6 +227,21 @@ zink_descriptor_type_to_size_idx(enum zink_descriptor_type type)
}
unreachable("unknown type");
}
static inline enum zink_descriptor_size_index_compact
zink_descriptor_type_to_size_idx_comp(enum zink_descriptor_type type)
{
switch (type) {
case ZINK_DESCRIPTOR_TYPE_UBO:
return ZDS_INDEX_COMP_UBO;
case ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW:
return ZDS_INDEX_COMP_COMBINED_SAMPLER;
case ZINK_DESCRIPTOR_TYPE_SSBO:
case ZINK_DESCRIPTOR_TYPE_IMAGE:
default: break;
}
unreachable("unknown type");
}
bool
zink_descriptor_layouts_init(struct zink_context *ctx);

34
src/gallium/drivers/zink/zink_descriptors_lazy.c
View File

@ -141,12 +141,30 @@ descriptor_program_num_sizes(VkDescriptorPoolSize *sizes, enum zink_descriptor_t
unreachable("unknown type");
}
static uint16_t
descriptor_program_num_sizes_compact(VkDescriptorPoolSize *sizes, unsigned desc_set)
{
switch (desc_set) {
case ZINK_DESCRIPTOR_TYPE_UBO:
return !!sizes[ZDS_INDEX_COMP_UBO].descriptorCount + !!sizes[ZDS_INDEX_COMP_STORAGE_BUFFER].descriptorCount;
case ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW:
return !!sizes[ZDS_INDEX_COMP_COMBINED_SAMPLER].descriptorCount +
!!sizes[ZDS_INDEX_COMP_UNIFORM_TEXELS].descriptorCount +
!!sizes[ZDS_INDEX_COMP_STORAGE_IMAGE].descriptorCount +
!!sizes[ZDS_INDEX_COMP_STORAGE_TEXELS].descriptorCount;
case ZINK_DESCRIPTOR_TYPE_SSBO:
case ZINK_DESCRIPTOR_TYPE_IMAGE:
default: break;
}
unreachable("unknown type");
}
bool
zink_descriptor_program_init_lazy(struct zink_context *ctx, struct zink_program *pg)
{
struct zink_screen *screen = zink_screen(ctx->base.screen);
VkDescriptorSetLayoutBinding bindings[ZINK_DESCRIPTOR_TYPES][PIPE_SHADER_TYPES * 32];
VkDescriptorUpdateTemplateEntry entries[ZINK_DESCRIPTOR_TYPES][PIPE_SHADER_TYPES * 32];
VkDescriptorSetLayoutBinding bindings[ZINK_DESCRIPTOR_TYPES][PIPE_SHADER_TYPES * 64];
VkDescriptorUpdateTemplateEntry entries[ZINK_DESCRIPTOR_TYPES][PIPE_SHADER_TYPES * 64];
unsigned num_bindings[ZINK_DESCRIPTOR_TYPES] = {0};
uint8_t has_bindings = 0;
unsigned push_count = 0;
@ -200,14 +218,17 @@ zink_descriptor_program_init_lazy(struct zink_context *ctx, struct zink_program
binding->stageFlags = stage_flags;
binding->pImmutableSamplers = NULL;
enum zink_descriptor_size_index idx = zink_vktype_to_size_idx(shader->bindings[j][k].type);
unsigned idx = screen->compact_descriptors ? zink_vktype_to_size_idx_comp(shader->bindings[j][k].type) :
zink_vktype_to_size_idx(shader->bindings[j][k].type);
sizes[idx].descriptorCount += shader->bindings[j][k].size;
sizes[idx].type = shader->bindings[j][k].type;
init_template_entry(shader, j, k, &entries[desc_set][entry_idx[desc_set]], &entry_idx[desc_set], screen->descriptor_mode == ZINK_DESCRIPTOR_MODE_LAZY);
num_bindings[desc_set]++;
has_bindings |= BITFIELD_BIT(desc_set);
}
num_type_sizes[desc_set] = descriptor_program_num_sizes(sizes, j);
num_type_sizes[desc_set] = screen->compact_descriptors ?
descriptor_program_num_sizes_compact(sizes, desc_set) :
descriptor_program_num_sizes(sizes, j);
}
pg->dd->bindless |= shader->bindless;
}
@ -630,9 +651,12 @@ zink_context_invalidate_descriptor_state_lazy(struct zink_context *ctx, enum pip
{
if (type == ZINK_DESCRIPTOR_TYPE_UBO && !start)
dd_lazy(ctx)->push_state_changed[shader == PIPE_SHADER_COMPUTE] = true;
else
else {
if (zink_screen(ctx->base.screen)->compact_descriptors && type > ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW)
type -= 2;
dd_lazy(ctx)->state_changed[shader == PIPE_SHADER_COMPUTE] |= BITFIELD_BIT(type);
}
}
void
zink_batch_descriptor_deinit_lazy(struct zink_screen *screen, struct zink_batch_state *bs)

12
src/gallium/drivers/zink/zink_screen.c
View File

@ -79,6 +79,7 @@ zink_debug_options[] = {
{ "tgsi", ZINK_DEBUG_TGSI, "Dump TGSI during program compile" },
{ "validation", ZINK_DEBUG_VALIDATION, "Dump Validation layer output" },
{ "sync", ZINK_DEBUG_SYNC, "Force synchronization before draws/dispatches" },
{ "compact", ZINK_DEBUG_COMPACT, "Use only 4 descriptor sets" },
DEBUG_NAMED_VALUE_END
};
@ -2180,12 +2181,23 @@ zink_internal_create_screen(const struct pipe_screen_config *config)
zink_verify_device_extensions(screen);
if ((zink_debug & ZINK_DEBUG_COMPACT) ||
screen->info.props.limits.maxBoundDescriptorSets < ZINK_MAX_DESCRIPTOR_SETS) {
screen->desc_set_id[ZINK_DESCRIPTOR_TYPES] = 0;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_UBO] = 1;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_SSBO] = 1;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW] = 2;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_IMAGE] = 2;
screen->desc_set_id[ZINK_DESCRIPTOR_BINDLESS] = 3;
screen->compact_descriptors = true;
} else {
screen->desc_set_id[ZINK_DESCRIPTOR_TYPES] = 0;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_UBO] = 1;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_SAMPLER_VIEW] = 2;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_SSBO] = 3;
screen->desc_set_id[ZINK_DESCRIPTOR_TYPE_IMAGE] = 4;
screen->desc_set_id[ZINK_DESCRIPTOR_BINDLESS] = 5;
}
if (screen->info.have_EXT_calibrated_timestamps && !check_have_device_time(screen))
goto fail;

6
src/gallium/drivers/zink/zink_screen.h
View File

@ -67,6 +67,7 @@ enum zink_descriptor_type;
#define ZINK_DEBUG_TGSI 0x4
#define ZINK_DEBUG_VALIDATION 0x8
#define ZINK_DEBUG_SYNC 0x10
#define ZINK_DEBUG_COMPACT (1<<5)
#define NUM_SLAB_ALLOCATORS 3
#define MIN_SLAB_ORDER 8
@ -78,8 +79,12 @@ enum zink_descriptor_mode {
ZINK_DESCRIPTOR_MODE_LAZY,
ZINK_DESCRIPTOR_MODE_NOFALLBACK,
ZINK_DESCRIPTOR_MODE_NOTEMPLATES,
ZINK_DESCRIPTOR_MODE_COMPACT,
};
//keep in sync with zink_descriptor_type since headers can't be cross-included
#define ZINK_MAX_DESCRIPTOR_SETS 6
struct zink_modifier_prop {
uint32_t drmFormatModifierCount;
VkDrmFormatModifierPropertiesEXT* pDrmFormatModifierProperties;
@ -164,6 +169,7 @@ struct zink_screen {
struct vk_dispatch_table vk;
bool compact_descriptors;
uint8_t desc_set_id[ZINK_MAX_DESCRIPTOR_SETS];
bool (*descriptor_program_init)(struct zink_context *ctx, struct zink_program *pg);
void (*descriptor_program_deinit)(struct zink_context *ctx, struct zink_program *pg);