V2:
- when loading from disk cache also binary insert into memory cache.
- check that the binary loaded from disk is the correct size. If not
delete the cache item and skip loading from cache.
V3:
- remove unrequired variable
Tested-by: Michel Dänzer <[email protected]>
---
src/gallium/drivers/radeonsi/si_state_shaders.c | 67
++++++++++++++++++++++---
1 file changed, 60 insertions(+), 7 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c
b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 750cdd6..a82e38e 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -29,20 +29,23 @@
#include "sid.h"
#include "radeon/r600_cs.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_ureg.h"
#include "util/hash_table.h"
#include "util/crc32.h"
#include "util/u_memory.h"
#include "util/u_prim.h"
+#include "util/disk_cache.h"
+#include "util/mesa-sha1.h"
+
/* SHADER_CACHE */
/**
* Return the TGSI binary in a buffer. The first 4 bytes contain its
size as
* integer.
*/
static void *si_get_tgsi_binary(struct si_shader_selector *sel)
{
unsigned tgsi_size = tgsi_num_tokens(sel->tokens) *
sizeof(struct tgsi_token);
@@ -175,54 +178,104 @@ static bool si_load_shader_binary(struct
si_shader *shader, void *binary)
}
/**
* Insert a shader into the cache. It's assumed the shader is not in
the cache.
* Use si_shader_cache_load_shader before calling this.
*
* Returns false on failure, in which case the tgsi_binary should be
freed.
*/
static bool si_shader_cache_insert_shader(struct si_screen *sscreen,
void *tgsi_binary,
- struct si_shader *shader)
+ struct si_shader *shader,
+ bool insert_into_disk_cache)
{
void *hw_binary;
struct hash_entry *entry;
+ uint8_t key[CACHE_KEY_SIZE];
entry = _mesa_hash_table_search(sscreen->shader_cache,
tgsi_binary);
if (entry)
return false; /* already added */
hw_binary = si_get_shader_binary(shader);
if (!hw_binary)
return false;
if (_mesa_hash_table_insert(sscreen->shader_cache, tgsi_binary,
hw_binary) == NULL) {
FREE(hw_binary);
return false;
}
+ if (sscreen->b.disk_shader_cache && insert_into_disk_cache) {
+ _mesa_sha1_compute(tgsi_binary, *((uint32_t *)tgsi_binary),
key);
+ disk_cache_put(sscreen->b.disk_shader_cache, key, hw_binary,
+ *((uint32_t *) hw_binary));
+ }
+
return true;
}
static bool si_shader_cache_load_shader(struct si_screen *sscreen,
void *tgsi_binary,
struct si_shader *shader)
{
struct hash_entry *entry =
_mesa_hash_table_search(sscreen->shader_cache, tgsi_binary);
- if (!entry)
- return false;
+ if (!entry) {
+ if (sscreen->b.disk_shader_cache) {
+ unsigned char sha1[CACHE_KEY_SIZE];
+ size_t tg_size = *((uint32_t *) tgsi_binary);
+
+ _mesa_sha1_compute(tgsi_binary, tg_size, sha1);
+
+ size_t binary_size;
+ uint8_t *buffer =
+ disk_cache_get(sscreen->b.disk_shader_cache,
+ sha1, &binary_size);
+ if (!buffer)
+ return false;
- if (!si_load_shader_binary(shader, entry->data))
- return false;
+ if (binary_size < sizeof(uint32_t) ||
+ *((uint32_t*)buffer) != binary_size) {
+ /* Something has gone wrong discard the item
+ * from the cache and rebuild/link from
+ * source.
+ */
+ assert(!"Invalid radeonsi shader disk cache "
+ "item!");
+
+ disk_cache_remove(sscreen->b.disk_shader_cache,
+ sha1);
+ free(buffer);
+
+ return false;
+ }
+
+ if (!si_load_shader_binary(shader, buffer)) {
+ free(buffer);
+ return false;
+ }
+ free(buffer);
+ if (!si_shader_cache_insert_shader(sscreen, tgsi_binary,
+ shader, false))
+ FREE(tgsi_binary);
+ } else {
+ return false;
+ }
+ } else {
+ if (si_load_shader_binary(shader, entry->data))
+ FREE(tgsi_binary);
+ else
+ return false;
+ }
p_atomic_inc(&sscreen->b.num_shader_cache_hits);
return true;
}
static uint32_t si_shader_cache_key_hash(const void *key)
{
/* The first dword is the key size. */
return util_hash_crc32(key, *(uint32_t*)key);
}
@@ -244,20 +297,21 @@ static void
si_destroy_shader_cache_entry(struct hash_entry *entry)
FREE(entry->data);
}
bool si_init_shader_cache(struct si_screen *sscreen)
{
pipe_mutex_init(sscreen->shader_cache_mutex);
sscreen->shader_cache =
_mesa_hash_table_create(NULL,
si_shader_cache_key_hash,
si_shader_cache_key_equals);
+
return sscreen->shader_cache != NULL;
}
void si_destroy_shader_cache(struct si_screen *sscreen)
{
if (sscreen->shader_cache)
_mesa_hash_table_destroy(sscreen->shader_cache,
si_destroy_shader_cache_entry);
pipe_mutex_destroy(sscreen->shader_cache_mutex);
}
@@ -1400,37 +1454,36 @@ void si_init_shader_selector_async(void *job,
int thread_index)
shader->selector = sel;
si_parse_next_shader_property(&sel->info, &shader->key);
tgsi_binary = si_get_tgsi_binary(sel);
/* Try to load the shader from the shader cache. */
pipe_mutex_lock(sscreen->shader_cache_mutex);
if (tgsi_binary &&
si_shader_cache_load_shader(sscreen, tgsi_binary,
shader)) {
- FREE(tgsi_binary);
pipe_mutex_unlock(sscreen->shader_cache_mutex);
} else {
pipe_mutex_unlock(sscreen->shader_cache_mutex);
/* Compile the shader if it hasn't been loaded from
the cache. */
if (si_compile_tgsi_shader(sscreen, tm, shader, false,
debug) != 0) {
FREE(shader);
FREE(tgsi_binary);
fprintf(stderr, "radeonsi: can't compile a main
shader part\n");
return;
}
if (tgsi_binary) {
pipe_mutex_lock(sscreen->shader_cache_mutex);
- if (!si_shader_cache_insert_shader(sscreen,
tgsi_binary, shader))
+ if (!si_shader_cache_insert_shader(sscreen,
tgsi_binary, shader, true))
FREE(tgsi_binary);
pipe_mutex_unlock(sscreen->shader_cache_mutex);
}
}
*si_get_main_shader_part(sel, &shader->key) = shader;
/* Unset "outputs_written" flags for outputs converted to
* DEFAULT_VAL, so that later inter-shader optimizations don't
* try to eliminate outputs that don't exist in the final
