Mesa (master): radeonsi: add radeonsi_sync_compile option

[GitLab Mirror]

Module: Mesa
Branch: master
Commit: 9445a4ab43939f78ecf1958db94bc80b276c4652
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=9445a4ab43939f78ecf1958db94bc80b276c4652

Author: Nicolai Hähnle <nicolai.haehnle at amd.com>
Date:   Mon Mar 25 15:44:45 2019 +0100

radeonsi: add radeonsi_sync_compile option

Force the driver thread to sync immediately with a compiler thread (but
compilation still happens in a separate thread).

This can be useful to simplify debugging compiler issues.

Reviewed-by: Marek Olšák <marek.olsak at amd.com>

---

 src/gallium/drivers/radeonsi/si_debug_options.h |  1 +
 src/gallium/drivers/radeonsi/si_state_shaders.c | 13 ++++++++++---
 2 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_debug_options.h b/src/gallium/drivers/radeonsi/si_debug_options.h
index f4c3e19ed95..019256ca1d1 100644
--- a/src/gallium/drivers/radeonsi/si_debug_options.h
+++ b/src/gallium/drivers/radeonsi/si_debug_options.h
@@ -1,5 +1,6 @@
 OPT_BOOL(clear_db_cache_before_clear, false, "Clear DB cache before fast depth clear")
 OPT_BOOL(enable_nir, false, "Enable NIR")
 OPT_BOOL(aux_debug, false, "Generate ddebug_dumps for the auxiliary context")
+OPT_BOOL(sync_compile, false, "Always compile synchronously (will cause stalls)")
 
 #undef OPT_BOOL
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 5bdfd4f6ac1..f57e7730905 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1952,6 +1952,10 @@ current_not_ready:
 		/* Use the default (unoptimized) shader for now. */
 		memset(&key->opt, 0, sizeof(key->opt));
 		mtx_unlock(&sel->mutex);
+
+		if (sscreen->options.sync_compile)
+			util_queue_fence_wait(&shader->ready);
+
 		goto again;
 	}
 
@@ -2164,12 +2168,12 @@ void si_schedule_initial_compile(struct si_context *sctx, unsigned processor,
 	util_queue_fence_init(ready_fence);
 
 	struct util_async_debug_callback async_debug;
-	bool wait =
+	bool debug =
 		(sctx->debug.debug_message && !sctx->debug.async) ||
 		sctx->is_debug ||
 		si_can_dump_shader(sctx->screen, processor);
 
-	if (wait) {
+	if (debug) {
 		u_async_debug_init(&async_debug);
 		compiler_ctx_state->debug = async_debug.base;
 	}
@@ -2177,11 +2181,14 @@ void si_schedule_initial_compile(struct si_context *sctx, unsigned processor,
 	util_queue_add_job(&sctx->screen->shader_compiler_queue, job,
 			   ready_fence, execute, NULL);
 
-	if (wait) {
+	if (debug) {
 		util_queue_fence_wait(ready_fence);
 		u_async_debug_drain(&async_debug, &sctx->debug);
 		u_async_debug_cleanup(&async_debug);
 	}
+
+	if (sctx->screen->options.sync_compile)
+		util_queue_fence_wait(ready_fence);
 }
 
 /* Return descriptor slot usage masks from the given shader info. */