OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

iris: Enable threaded shader compilation

Browse Source 
There are a couple minor things that can be improved:

1. Eliminate (or reduce) the dynamic allocation of the
threaded_compile_job.

2. For apps like shader-db, improve the case where nr_threads=0.  Right
now this adds thread switching and mutex overhead.

3. Other performance improvements?  iris_uncompiled_shader::variants has
some special properties that make it ripe for replacement with a
lockless list.  Without gathering some data, it's hard to guess what
impact that could have.

v2: Fix whitespace and formatting issues.  Noticed by Ken.
s/threaded_compile_job/iris_threaded_compile_job/g.  Suggested by Ken.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11229>
This commit is contained in:
Ian Romanick
2021-06-04 14:17:42 -07:00
committed by Marge Bot
parent 9011cc7405
commit 42c34e1ac8
5 changed files with 135 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
src/gallium/drivers/iris/iris_program.c
View File

@@ -38,6 +38,7 @@
#include "util/u_atomic.h"
#include "util/u_upload_mgr.h"
#include "util/debug.h"
#include "util/u_async_debug.h"
#include "compiler/nir/nir.h"
#include "compiler/nir/nir_builder.h"
#include "compiler/nir/nir_serialize.h"
@@ -54,6 +55,14 @@
.base.tex.compressed_multisample_layout_mask = ~0, \
.base.tex.msaa_16 = (gen >= 9 ? ~0 : 0)
struct iris_threaded_compile_job {
struct iris_screen *screen;
struct u_upload_mgr *uploader;
struct pipe_debug_callback *dbg;
struct iris_uncompiled_shader *ish;
struct iris_compiled_shader *shader;
};
static unsigned
get_new_program_id(struct iris_screen *screen)
{
@@ -1174,6 +1183,42 @@ find_or_add_variant(const struct iris_screen *screen,
return variant;
}
static void
iris_threaded_compile_job_delete(void *_job, UNUSED void *_gdata,
UNUSED int thread_index)
{
free(_job);
}
static void
iris_schedule_compile(struct iris_screen *screen,
struct util_queue_fence *ready_fence,
struct pipe_debug_callback *dbg,
struct iris_threaded_compile_job *job,
util_queue_execute_func execute)
{
util_queue_fence_init(ready_fence);
struct util_async_debug_callback async_debug;
if (dbg) {
u_async_debug_init(&async_debug);
job->dbg = &async_debug.base;
}
util_queue_add_job(&screen->shader_compiler_queue, job, ready_fence, execute,
iris_threaded_compile_job_delete, 0);
if (screen->driconf.sync_compile || dbg)
util_queue_fence_wait(ready_fence);
if (dbg) {
u_async_debug_drain(&async_debug, dbg);
u_async_debug_cleanup(&async_debug);
}
}
/**
* Compile a vertex shader, and upload the assembly.
*/
@@ -2457,12 +2502,17 @@ iris_create_compute_state(struct pipe_context *ctx,
}
static void
iris_compile_shader(struct iris_screen *screen,
struct u_upload_mgr *uploader,
struct pipe_debug_callback *dbg,
struct iris_uncompiled_shader *ish,
struct iris_compiled_shader *shader)
iris_compile_shader(void *_job, UNUSED void *_gdata, UNUSED int thread_index)
{
const struct iris_threaded_compile_job *job =
(struct iris_threaded_compile_job *) _job;
struct iris_screen *screen = job->screen;
struct u_upload_mgr *uploader = job->uploader;
struct pipe_debug_callback *dbg = job->dbg;
struct iris_uncompiled_shader *ish = job->ish;
struct iris_compiled_shader *shader = job->shader;
switch (ish->nir->info.stage) {
case MESA_SHADER_VERTEX:
iris_compile_vs(screen, uploader, dbg, ish, shader);
@@ -2615,7 +2665,17 @@ iris_create_shader_state(struct pipe_context *ctx,
if (!iris_disk_cache_retrieve(screen, uploader, ish, shader,
&key, key_size)) {
iris_compile_shader(screen, uploader, &ice->dbg, ish, shader);
assert(!util_queue_fence_is_signalled(&shader->ready));
struct iris_threaded_compile_job *job = calloc(1, sizeof(*job));
job->screen = screen;
job->uploader = uploader;
job->ish = ish;
job->shader = shader;
iris_schedule_compile(screen, &ish->ready, &ice->dbg, job,
iris_compile_shader);
}
}
@@ -2643,6 +2703,7 @@ iris_destroy_shader_state(struct pipe_context *ctx, void *state)
}
simple_mtx_destroy(&ish->lock);
util_queue_fence_destroy(&ish->ready);
ralloc_free(ish->nir);
free(ish);