Commit 6b057f8e authored by Nicolai Hähnle's avatar Nicolai Hähnle

vbo: cache/memoize the result of vbo_get_minmax_indices (v3)

Some games developers are unaware that an index buffer in a VBO still needs
to be read by the CPU if some varying data comes from a user pointer (unless
glDrawRangeElements and friends are used). This is particularly bad when
they tell us that the index buffer should live in VRAM.

This cache helps, e.g. lifting This War Of Mine (a particularly bad
offender) from under 10fps to slightly over 20fps on a Carrizo.

Note that there is nothing prohibiting a user from rendering from multiple
threads simultaneously with the same index buffer, hence the locking. (The
internal buffer map taken for the buffer still leads to a race, but at least
the locks are a move in the right direction.)

v2: disable the cache on USAGE_TEXTURE_BUFFER as well (Chris Forbes)

v3:
- use bool instead of GLboolean for MinMaxCacheDirty (Ian Romanick)
- replace the sticky USAGE_PERSISTENT_WRITE_MAP bit by a direct
  AccessFlags check

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz> (v2)
Reviewed-by: 's avatarMarek Olšák <marek.olsak@amd.com>
parent 1a570d96
......@@ -458,6 +458,7 @@ _mesa_delete_buffer_object(struct gl_context *ctx,
{
(void) ctx;
vbo_delete_minmax_cache(bufObj);
_mesa_align_free(bufObj->Data);
/* assign strange values here to help w/ debugging */
......@@ -1528,6 +1529,7 @@ _mesa_buffer_storage(struct gl_context *ctx, struct gl_buffer_object *bufObj,
bufObj->Written = GL_TRUE;
bufObj->Immutable = GL_TRUE;
bufObj->MinMaxCacheDirty = true;
assert(ctx->Driver.BufferData);
if (!ctx->Driver.BufferData(ctx, target, size, data, GL_DYNAMIC_DRAW,
......@@ -1641,6 +1643,7 @@ _mesa_buffer_data(struct gl_context *ctx, struct gl_buffer_object *bufObj,
FLUSH_VERTICES(ctx, _NEW_BUFFER_OBJECT);
bufObj->Written = GL_TRUE;
bufObj->MinMaxCacheDirty = true;
#ifdef VBO_DEBUG
printf("glBufferDataARB(%u, sz %ld, from %p, usage 0x%x)\n",
......@@ -1753,6 +1756,7 @@ _mesa_buffer_sub_data(struct gl_context *ctx, struct gl_buffer_object *bufObj,
}
bufObj->Written = GL_TRUE;
bufObj->MinMaxCacheDirty = true;
assert(ctx->Driver.BufferSubData);
ctx->Driver.BufferSubData(ctx, offset, size, data, bufObj);
......@@ -1872,6 +1876,8 @@ _mesa_clear_buffer_sub_data(struct gl_context *ctx,
if (size == 0)
return;
bufObj->MinMaxCacheDirty = true;
if (data == NULL) {
/* clear to zeros, per the spec */
ctx->Driver.ClearBufferSubData(ctx, offset, size,
......@@ -2285,6 +2291,8 @@ _mesa_copy_buffer_sub_data(struct gl_context *ctx,
}
}
dst->MinMaxCacheDirty = true;
ctx->Driver.CopyBufferSubData(ctx, src, dst, readOffset, writeOffset, size);
}
......@@ -2489,8 +2497,10 @@ _mesa_map_buffer_range(struct gl_context *ctx,
assert(bufObj->Mappings[MAP_USER].AccessFlags == access);
}
if (access & GL_MAP_WRITE_BIT)
if (access & GL_MAP_WRITE_BIT) {
bufObj->Written = GL_TRUE;
bufObj->MinMaxCacheDirty = true;
}
#ifdef VBO_DEBUG
if (strstr(func, "Range") == NULL) { /* If not MapRange */
......
......@@ -1282,6 +1282,10 @@ struct gl_buffer_object
GLuint NumMapBufferWriteCalls;
struct gl_buffer_mapping Mappings[MAP_COUNT];
/** Memoization of min/max index computations for static index buffers */
struct hash_table *MinMaxCache;
bool MinMaxCacheDirty;
};
......
......@@ -180,6 +180,9 @@ vbo_sizeof_ib_type(GLenum type)
}
}
void
vbo_delete_minmax_cache(struct gl_buffer_object *bufferObj);
void
vbo_get_minmax_indices(struct gl_context *ctx, const struct _mesa_prim *prim,
const struct _mesa_index_buffer *ib,
......
......@@ -32,6 +32,167 @@
#include "main/macros.h"
#include "main/sse_minmax.h"
#include "x86/common_x86_asm.h"
#include "util/hash_table.h"
struct minmax_cache_key {
GLintptr offset;
GLuint count;
GLenum type;
};
struct minmax_cache_entry {
struct minmax_cache_key key;
GLuint min;
GLuint max;
};
static uint32_t
vbo_minmax_cache_hash(const struct minmax_cache_key *key)
{
return _mesa_hash_data(key, sizeof(*key));
}
static bool
vbo_minmax_cache_key_equal(const struct minmax_cache_key *a,
const struct minmax_cache_key *b)
{
return (a->offset == b->offset) && (a->count == b->count) && (a->type == b->type);
}
static void
vbo_minmax_cache_delete_entry(struct hash_entry *entry)
{
free(entry->data);
}
static GLboolean
vbo_use_minmax_cache(struct gl_buffer_object *bufferObj)
{
if (bufferObj->UsageHistory & (USAGE_TEXTURE_BUFFER |
USAGE_ATOMIC_COUNTER_BUFFER |
USAGE_SHADER_STORAGE_BUFFER |
USAGE_TRANSFORM_FEEDBACK_BUFFER |
USAGE_PIXEL_PACK_BUFFER))
return GL_FALSE;
if ((bufferObj->Mappings[MAP_USER].AccessFlags &
(GL_MAP_PERSISTENT_BIT | GL_MAP_WRITE_BIT)) ==
(GL_MAP_PERSISTENT_BIT | GL_MAP_WRITE_BIT))
return GL_FALSE;
return GL_TRUE;
}
void
vbo_delete_minmax_cache(struct gl_buffer_object *bufferObj)
{
_mesa_hash_table_destroy(bufferObj->MinMaxCache, vbo_minmax_cache_delete_entry);
bufferObj->MinMaxCache = NULL;
}
static GLboolean
vbo_get_minmax_cached(struct gl_buffer_object *bufferObj,
GLenum type, GLintptr offset, GLuint count,
GLuint *min_index, GLuint *max_index)
{
GLboolean found = GL_FALSE;
struct minmax_cache_key key;
uint32_t hash;
struct hash_entry *result;
if (!bufferObj->MinMaxCache)
return GL_FALSE;
if (!vbo_use_minmax_cache(bufferObj))
return GL_FALSE;
mtx_lock(&bufferObj->Mutex);
if (bufferObj->MinMaxCacheDirty) {
_mesa_hash_table_clear(bufferObj->MinMaxCache, vbo_minmax_cache_delete_entry);
bufferObj->MinMaxCacheDirty = false;
goto out;
}
key.type = type;
key.offset = offset;
key.count = count;
hash = vbo_minmax_cache_hash(&key);
result = _mesa_hash_table_search_pre_hashed(bufferObj->MinMaxCache, hash, &key);
if (result) {
struct minmax_cache_entry *entry = result->data;
*min_index = entry->min;
*max_index = entry->max;
found = GL_TRUE;
}
out:
mtx_unlock(&bufferObj->Mutex);
return found;
}
static void
vbo_minmax_cache_store(struct gl_context *ctx,
struct gl_buffer_object *bufferObj,
GLenum type, GLintptr offset, GLuint count,
GLuint min, GLuint max)
{
struct minmax_cache_entry *entry;
struct hash_entry *table_entry;
uint32_t hash;
if (!vbo_use_minmax_cache(bufferObj))
return;
mtx_lock(&bufferObj->Mutex);
if (!bufferObj->MinMaxCache) {
bufferObj->MinMaxCache =
_mesa_hash_table_create(NULL,
(uint32_t (*)(const void *))vbo_minmax_cache_hash,
(bool (*)(const void *, const void *))vbo_minmax_cache_key_equal);
if (!bufferObj->MinMaxCache)
goto out;
}
entry = MALLOC_STRUCT(minmax_cache_entry);
if (!entry)
goto out;
entry->key.offset = offset;
entry->key.count = count;
entry->key.type = type;
entry->min = min;
entry->max = max;
hash = vbo_minmax_cache_hash(&entry->key);
table_entry = _mesa_hash_table_search_pre_hashed(bufferObj->MinMaxCache,
hash, &entry->key);
if (table_entry) {
/* It seems like this could happen when two contexts are rendering using
* the same buffer object from multiple threads.
*/
_mesa_debug(ctx, "duplicate entry in minmax cache\n");
free(entry);
goto out;
}
table_entry = _mesa_hash_table_insert_pre_hashed(bufferObj->MinMaxCache,
hash, &entry->key, entry);
if (!table_entry)
free(entry);
out:
mtx_unlock(&bufferObj->Mutex);
}
/**
......@@ -56,6 +217,11 @@ vbo_get_minmax_index(struct gl_context *ctx,
indices = (char *) ib->ptr + prim->start * index_size;
if (_mesa_is_bufferobj(ib->obj)) {
GLsizeiptr size = MIN2(count * index_size, ib->obj->Size);
if (vbo_get_minmax_cached(ib->obj, ib->type, (GLintptr) indices, count,
min_index, max_index))
return;
indices = ctx->Driver.MapBufferRange(ctx, (GLintptr) indices, size,
GL_MAP_READ_BIT, ib->obj,
MAP_INTERNAL);
......@@ -139,6 +305,8 @@ vbo_get_minmax_index(struct gl_context *ctx,
}
if (_mesa_is_bufferobj(ib->obj)) {
vbo_minmax_cache_store(ctx, ib->obj, ib->type, prim->start, count,
*min_index, *max_index);
ctx->Driver.UnmapBuffer(ctx, ib->obj, MAP_INTERNAL);
}
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment