radeonsi:optimizing SET_CONTEXT_REG for shaders Tessellation

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>

4 files changed, 26 insertions, 5 deletions

diff --git a/src/gallium/drivers/radeonsi/si_gfx_cs.c b/src/gallium/drivers/radeonsi/si_gfx_cs.c
index 5f1c6987009..532a6365bf1 100644
--- a/src/gallium/drivers/radeonsi/si_gfx_cs.c
+++ b/src/gallium/drivers/radeonsi/si_gfx_cs.c
@@ -376,6 +376,7 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
 		ctx->tracked_regs.reg_value[SI_TRACKED_SPI_SHADER_Z_FORMAT]  = 0x00000000;
 		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;
 
 		/* 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 109c70a9f91..49b1ccd5823 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -686,6 +686,9 @@ struct si_shader {
 			unsigned	cb_shader_mask;
 		} ps;
 	} ctx_reg;
+
+	/*For save precompute registers value */
+	unsigned vgt_tf_param; /* VGT_TF_PARAM */
 };
 
 struct si_shader_part {
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 878b67f0ed3..54b03e0992f 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -312,6 +312,7 @@ enum si_tracked_reg {
 	SI_TRACKED_SPI_SHADER_COL_FORMAT,
 
 	SI_TRACKED_CB_SHADER_MASK,
+	SI_TRACKED_VGT_TF_PARAM,
 
 	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 b074214bbd3..e493f991a1b 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -396,11 +396,11 @@ static void si_set_tesseval_regs(struct si_screen *sscreen,
 	} else
 		distribution_mode = V_028B6C_DISTRIBUTION_MODE_NO_DIST;
 
-	si_pm4_set_reg(pm4, R_028B6C_VGT_TF_PARAM,
-		       S_028B6C_TYPE(type) |
-		       S_028B6C_PARTITIONING(partitioning) |
-		       S_028B6C_TOPOLOGY(topology) |
-		       S_028B6C_DISTRIBUTION_MODE(distribution_mode));
+	assert(pm4->shader);
+	pm4->shader->vgt_tf_param = S_028B6C_TYPE(type) |
+				    S_028B6C_PARTITIONING(partitioning) |
+				    S_028B6C_TOPOLOGY(topology) |
+				    S_028B6C_DISTRIBUTION_MODE(distribution_mode);
 }
 
 /* Polaris needs different VTX_REUSE_DEPTH settings depending on
@@ -568,6 +568,12 @@ static void si_emit_shader_es(struct si_context *sctx)
 	radeon_opt_set_context_reg(sctx, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
 				   SI_TRACKED_VGT_ESGS_RING_ITEMSIZE,
 				   shader->selector->esgs_itemsize / 4);
+
+	if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
+		radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM,
+					   SI_TRACKED_VGT_TF_PARAM,
+					   shader->vgt_tf_param);
+
 }
 
 static void si_shader_es(struct si_screen *sscreen, struct si_shader *shader)
@@ -802,6 +808,11 @@ static void si_emit_shader_gs(struct si_context *sctx)
 		radeon_opt_set_context_reg(sctx, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
 					   SI_TRACKED_VGT_ESGS_RING_ITEMSIZE,
 					   shader->ctx_reg.gs.vgt_esgs_ring_itemsize);
+
+		if (shader->key.part.gs.es->type == PIPE_SHADER_TESS_EVAL)
+			radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM,
+						   SI_TRACKED_VGT_TF_PARAM,
+						   shader->vgt_tf_param);
 	}
 }
 
@@ -965,6 +976,11 @@ static void si_emit_shader_vs(struct si_context *sctx)
 	radeon_opt_set_context_reg(sctx, R_028818_PA_CL_VTE_CNTL,
 				   SI_TRACKED_PA_CL_VTE_CNTL,
 				   shader->ctx_reg.vs.pa_cl_vte_cntl);
+
+	if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
+		radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM,
+					   SI_TRACKED_VGT_TF_PARAM,
+					   shader->vgt_tf_param);
 }
 
 /**