[Mesa-dev] [PATCH 2/7] radeonsi: require that compiler threads are enabled

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

threaded gallium can't use pipe_context's LLVM target machine, because
create_shader_selector can be called from a non-driver thread.
---
 src/gallium/drivers/radeonsi/si_pipe.c          | 21 ++++++++++++---------
 src/gallium/drivers/radeonsi/si_state_shaders.c |  3 +--
 2 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index a9a7c8f..a3e001c 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -683,22 +683,21 @@ static void si_destroy_screen(struct pipe_screen* pscreen)
 		sscreen->ps_epilogs
 	};
 	unsigned i;
 
 	if (!sscreen)
 		return;
 
 	if (!sscreen->b.ws->unref(sscreen->b.ws))
 		return;
 
-	if (util_queue_is_initialized(&sscreen->shader_compiler_queue))
-		util_queue_destroy(&sscreen->shader_compiler_queue);
+	util_queue_destroy(&sscreen->shader_compiler_queue);
 
 	for (i = 0; i < ARRAY_SIZE(sscreen->tm); i++)
 		if (sscreen->tm[i])
 			LLVMDisposeTargetMachine(sscreen->tm[i]);
 
 	/* Free shader parts. */
 	for (i = 0; i < ARRAY_SIZE(parts); i++) {
 		while (parts[i]) {
 			struct si_shader_part *part = parts[i];
 
@@ -790,20 +789,31 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 
 	si_init_screen_state_functions(sscreen);
 
 	if (!r600_common_screen_init(&sscreen->b, ws) ||
 	    !si_init_gs_info(sscreen) ||
 	    !si_init_shader_cache(sscreen)) {
 		FREE(sscreen);
 		return NULL;
 	}
 
+	/* Only enable as many threads as we have target machines and CPUs. */
+	num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
+	num_compiler_threads = MIN2(num_cpus, ARRAY_SIZE(sscreen->tm));
+
+	if (!util_queue_init(&sscreen->shader_compiler_queue, "si_shader",
+			     32, num_compiler_threads)) {
+		si_destroy_shader_cache(sscreen);
+		FREE(sscreen);
+		return NULL;
+	}
+
 	si_handle_env_var_force_family(sscreen);
 
 	if (!debug_get_bool_option("RADEON_DISABLE_PERFCOUNTERS", false))
 		si_init_perfcounters(sscreen);
 
 	/* Hawaii has a bug with offchip buffers > 256 that can be worked
 	 * around by setting 4K granularity.
 	 */
 	sscreen->tess_offchip_block_dw_size =
 		sscreen->b.family == CHIP_HAWAII ? 4096 : 8192;
@@ -834,28 +844,21 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
 		(sscreen->b.debug_flags & DBG_MONOLITHIC_SHADERS) != 0;
 
 	sscreen->b.barrier_flags.cp_to_L2 = SI_CONTEXT_INV_SMEM_L1 |
 					    SI_CONTEXT_INV_VMEM_L1 |
 					    SI_CONTEXT_INV_GLOBAL_L2;
 	sscreen->b.barrier_flags.compute_to_L2 = SI_CONTEXT_CS_PARTIAL_FLUSH;
 
 	if (debug_get_bool_option("RADEON_DUMP_SHADERS", false))
 		sscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
 
-	/* Only enable as many threads as we have target machines and CPUs. */
-	num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
-	num_compiler_threads = MIN2(num_cpus, ARRAY_SIZE(sscreen->tm));
-
 	for (i = 0; i < num_compiler_threads; i++)
 		sscreen->tm[i] = si_create_llvm_target_machine(sscreen);
 
-	util_queue_init(&sscreen->shader_compiler_queue, "si_shader",
-                        32, num_compiler_threads);
-
 	/* Create the auxiliary context. This must be done last. */
 	sscreen->b.aux_context = sscreen->b.b.context_create(&sscreen->b.b, NULL, 0);
 
 	if (sscreen->b.debug_flags & DBG_TEST_DMA)
 		r600_test_dma(&sscreen->b);
 
 	return &sscreen->b.b;
 }
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 9cde0aa..3596f5a 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1762,22 +1762,21 @@ static void *si_create_shader_selector(struct pipe_context *ctx,
 	} else {
 		/* Case 1. */
 		sel->db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
 	}
 
 	(void) mtx_init(&sel->mutex, mtx_plain);
 	util_queue_fence_init(&sel->ready);
 
 	if ((sctx->b.debug.debug_message && !sctx->b.debug.async) ||
 	    sctx->is_debug ||
-	    r600_can_dump_shader(&sscreen->b, sel->info.processor) ||
-	    !util_queue_is_initialized(&sscreen->shader_compiler_queue))
+	    r600_can_dump_shader(&sscreen->b, sel->info.processor))
 		si_init_shader_selector_async(sel, -1);
 	else
 		util_queue_add_job(&sscreen->shader_compiler_queue, sel,
                                    &sel->ready, si_init_shader_selector_async,
                                    NULL);
 
 	return sel;
 }
 
 static void si_bind_vs_shader(struct pipe_context *ctx, void *state)
-- 
2.7.4