radeonsi:optimizing SET_CONTEXT_REG for shaders vgt_vertex_reuse

Signed-off-by: Sonny Jiang <[email protected]>
Signed-off-by: Marek Olšák <[email protected]>

4 files changed, 18 insertions, 2 deletions

diff --git a/src/gallium/drivers/radeonsi/si_gfx_cs.c b/src/gallium/drivers/radeonsi/si_gfx_cs.c
index 532a6365bf1..3ddd7864d1c 100644
--- a/src/gallium/drivers/radeonsi/si_gfx_cs.c
+++ b/src/gallium/drivers/radeonsi/si_gfx_cs.c
@@ -377,6 +377,7 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
 		ctx->tracked_regs.reg_value[SI_TRACKED_SPI_SHADER_COL_FORMAT]  = 0x00000000;
 		ctx->tracked_regs.reg_value[SI_TRACKED_CB_SHADER_MASK]  = 0xffffffff;
 		ctx->tracked_regs.reg_value[SI_TRACKED_VGT_TF_PARAM]  = 0x00000000;
+		ctx->tracked_regs.reg_value[SI_TRACKED_VGT_VERTEX_REUSE_BLOCK_CNTL]  = 0x0000001e; /* From VI */
 
 		/* Set all saved registers state to saved. */
 		ctx->tracked_regs.reg_saved = 0xffffffffffffffff;
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 49b1ccd5823..09dd558d789 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -689,6 +689,7 @@ struct si_shader {
 
 	/*For save precompute registers value */
 	unsigned vgt_tf_param; /* VGT_TF_PARAM */
+	unsigned vgt_vertex_reuse_block_cntl; /* VGT_VERTEX_REUSE_BLOCK_CNTL */
 };
 
 struct si_shader_part {
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 54b03e0992f..fffc63680d0 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -313,6 +313,7 @@ enum si_tracked_reg {
 
 	SI_TRACKED_CB_SHADER_MASK,
 	SI_TRACKED_VGT_TF_PARAM,
+	SI_TRACKED_VGT_VERTEX_REUSE_BLOCK_CNTL,
 
 	SI_NUM_TRACKED_REGS,
 };
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index e493f991a1b..2bdac33586b 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -440,8 +440,8 @@ static void polaris_set_vgt_vertex_reuse(struct si_screen *sscreen,
 		    PIPE_TESS_SPACING_FRACTIONAL_ODD)
 			vtx_reuse_depth = 14;
 
-		si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
-			       vtx_reuse_depth);
+		assert(pm4->shader);
+		pm4->shader->vgt_vertex_reuse_block_cntl = vtx_reuse_depth;
 	}
 }
 
@@ -574,6 +574,10 @@ static void si_emit_shader_es(struct si_context *sctx)
 					   SI_TRACKED_VGT_TF_PARAM,
 					   shader->vgt_tf_param);
 
+	if (shader->vgt_vertex_reuse_block_cntl)
+		radeon_opt_set_context_reg(sctx, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
+					   SI_TRACKED_VGT_VERTEX_REUSE_BLOCK_CNTL,
+					   shader->vgt_vertex_reuse_block_cntl);
 }
 
 static void si_shader_es(struct si_screen *sscreen, struct si_shader *shader)
@@ -813,6 +817,10 @@ static void si_emit_shader_gs(struct si_context *sctx)
 			radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM,
 						   SI_TRACKED_VGT_TF_PARAM,
 						   shader->vgt_tf_param);
+		if (shader->vgt_vertex_reuse_block_cntl)
+			radeon_opt_set_context_reg(sctx, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
+						   SI_TRACKED_VGT_VERTEX_REUSE_BLOCK_CNTL,
+						   shader->vgt_vertex_reuse_block_cntl);
 	}
 }
 
@@ -981,6 +989,11 @@ static void si_emit_shader_vs(struct si_context *sctx)
 		radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM,
 					   SI_TRACKED_VGT_TF_PARAM,
 					   shader->vgt_tf_param);
+
+	if (shader->vgt_vertex_reuse_block_cntl)
+		radeon_opt_set_context_reg(sctx, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
+					   SI_TRACKED_VGT_VERTEX_REUSE_BLOCK_CNTL,
+					   shader->vgt_vertex_reuse_block_cntl);
 }
 
 /**