radeonsi/gfx10: keep track of whether NGG is used

We always use NGG by default, except when tessellation is enabled with
extreme geometry shader amplification.

Acked-by: Bas Nieuwenhuizen <[email protected]>

4 files changed, 31 insertions, 1 deletions

diff --git a/src/gallium/drivers/radeonsi/si_debug_options.h b/src/gallium/drivers/radeonsi/si_debug_options.h
index d6cb3157632..9fdb0bc5dad 100644
--- a/src/gallium/drivers/radeonsi/si_debug_options.h
+++ b/src/gallium/drivers/radeonsi/si_debug_options.h
@@ -7,5 +7,6 @@ OPT_BOOL(debug_disassembly, false, "Report shader disassembly as part of driver
 OPT_BOOL(halt_shaders, false, "Halt shaders at the start (will hang)")
 OPT_BOOL(vs_fetch_always_opencode, false, "Always open code vertex fetches (less efficient, purely for testing)")
 OPT_BOOL(prim_restart_tri_strips_only, false, "Only enable primitive restart for triangle strips")
+OPT_BOOL(disable_ngg, false, "Unconditionally disable NGG (gfx10+)")
 
 #undef OPT_BOOL
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index a290d16b02d..9ea08c8fb26 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -486,6 +486,9 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen,
 	if (!sctx->border_color_map)
 		goto fail;
 
+	if (sctx->chip_class >= GFX10)
+		sctx->ngg = !sscreen->options.disable_ngg;
+
 	/* Initialize context functions used by graphics and compute. */
 	if (sctx->chip_class >= GFX10)
 		sctx->emit_cache_flush = gfx10_emit_cache_flush;
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 588de59e0cd..d7edc7be9ae 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -1041,6 +1041,7 @@ struct si_context {
 	bool			gs_tri_strip_adj_fix:1;
 	bool			ls_vgpr_fix:1;
 	bool			prim_discard_cs_instancing:1;
+	bool			ngg:1;
 	int			last_index_size;
 	int			last_base_vertex;
 	int			last_start_instance;
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 7a82126a0ad..a55a6d139d4 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -2588,6 +2588,27 @@ static void si_update_tess_uses_prim_id(struct si_context *sctx)
 		 sctx->ps_shader.cso->info.uses_primid);
 }
 
+static bool si_update_ngg(struct si_context *sctx)
+{
+	if (sctx->chip_class <= GFX9 ||
+	    sctx->screen->options.disable_ngg)
+		return false;
+
+	bool new_ngg = true;
+
+	/* EN_MAX_VERT_OUT_PER_GS_INSTANCE does not work with tesselation. */
+	if (sctx->gs_shader.cso && sctx->tes_shader.cso &&
+	    sctx->gs_shader.cso->gs_num_invocations * sctx->gs_shader.cso->gs_max_out_vertices > 256)
+		new_ngg = false;
+
+	if (new_ngg != sctx->ngg) {
+		sctx->ngg = new_ngg;
+		sctx->last_rast_prim = -1; /* reset this so that it gets updated */
+		return true;
+	}
+	return false;
+}
+
 static void si_bind_gs_shader(struct pipe_context *ctx, void *state)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
@@ -2595,6 +2616,7 @@ static void si_bind_gs_shader(struct pipe_context *ctx, void *state)
 	struct si_shader *old_hw_vs_variant = si_get_vs_state(sctx);
 	struct si_shader_selector *sel = state;
 	bool enable_changed = !!sctx->gs_shader.cso != !!sel;
+	bool ngg_changed;
 
 	if (sctx->gs_shader.cso == sel)
 		return;
@@ -2606,8 +2628,10 @@ static void si_bind_gs_shader(struct pipe_context *ctx, void *state)
 	si_update_common_shader_state(sctx);
 	sctx->last_rast_prim = -1; /* reset this so that it gets updated */
 
-	if (enable_changed) {
+	ngg_changed = si_update_ngg(sctx);
+	if (ngg_changed || enable_changed)
 		si_shader_change_notify(sctx);
+	if (enable_changed) {
 		if (sctx->ia_multi_vgt_param_key.u.uses_tess)
 			si_update_tess_uses_prim_id(sctx);
 	}
@@ -2659,6 +2683,7 @@ static void si_bind_tes_shader(struct pipe_context *ctx, void *state)
 	sctx->last_rast_prim = -1; /* reset this so that it gets updated */
 
 	if (enable_changed) {
+		si_update_ngg(sctx);
 		si_shader_change_notify(sctx);
 		sctx->last_tes_sh_base = -1; /* invalidate derived tess state */
 	}