3 files changed, 38 insertions, 3 deletions

diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
index f597218cf67..6c2734eac29 100644
--- a/src/gallium/drivers/radeonsi/si_hw_context.c
+++ b/src/gallium/drivers/radeonsi/si_hw_context.c
@@ -155,4 +155,5 @@ void si_begin_new_cs(struct si_context *ctx)
 	r600_postflush_resume_features(&ctx->b);
 
 	ctx->b.initial_gfx_cs_size = ctx->b.rings.gfx.cs->cdw;
+	si_invalidate_draw_sh_constants(ctx);
 }
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 15ca9a559e5..e3cc09a0b85 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -34,6 +34,11 @@
 #define SI_BIG_ENDIAN 0
 #endif
 
+/* The base vertex can be any number, but we must pick one which
+ * will mean "unknown" for the purpose of state tracking and the number
+ * shouldn't be a commonly-used one. */
+#define SI_BASE_VERTEX_UNKNOWN INT_MIN
+
 #define SI_TRACE_CS 0
 #define SI_TRACE_CS_DWORDS		6
 
@@ -169,6 +174,11 @@ struct si_context {
 	bool			db_depth_clear;
 	bool			db_depth_disable_expclear;
 	unsigned		ps_db_shader_control;
+
+	/* Draw state. */
+	int			last_base_vertex;
+	int			last_start_instance;
+	int			last_sh_base_reg;
 };
 
 /* si_blit.c */
@@ -227,4 +237,12 @@ si_resource_create_custom(struct pipe_screen *screen,
 		PIPE_BIND_CUSTOM, usage, size));
 }
 
+static INLINE void
+si_invalidate_draw_sh_constants(struct si_context *sctx)
+{
+	sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
+	sctx->last_start_instance = -1; /* reset to an unknown value */
+	sctx->last_sh_base_reg = -1; /* reset to an unknown value */
+}
+
 #endif
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index a69d5962503..f9ff0e34c17 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -242,13 +242,29 @@ static void si_emit_draw_packets(struct si_context *sctx,
 	}
 
 	if (!info->indirect) {
+		int base_vertex;
+
 		radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
 		radeon_emit(cs, info->instance_count);
 
-		si_write_sh_reg_seq(cs, sh_base_reg + SI_SGPR_BASE_VERTEX * 4, 2);
-		radeon_emit(cs, info->indexed ? info->index_bias : info->start);
-		radeon_emit(cs, info->start_instance);
+		/* Base vertex and start instance. */
+		base_vertex = info->indexed ? info->index_bias : info->start;
+
+		if (base_vertex != sctx->last_base_vertex ||
+		    sctx->last_base_vertex == SI_BASE_VERTEX_UNKNOWN ||
+		    info->start_instance != sctx->last_start_instance ||
+		    sh_base_reg != sctx->last_sh_base_reg) {
+			si_write_sh_reg_seq(cs, sh_base_reg + SI_SGPR_BASE_VERTEX * 4, 2);
+			radeon_emit(cs, base_vertex);
+			radeon_emit(cs, info->start_instance);
+
+			sctx->last_base_vertex = base_vertex;
+			sctx->last_start_instance = info->start_instance;
+			sctx->last_sh_base_reg = sh_base_reg;
+		}
 	} else {
+		si_invalidate_draw_sh_constants(sctx);
+
 		r600_context_bo_reloc(&sctx->b, &sctx->b.rings.gfx,
 				      (struct r600_resource *)info->indirect,
 				      RADEON_USAGE_READ, RADEON_PRIO_MIN);