radeonsi: convert clip state into an atom

Reducing calloc overhead.

Reviewed-by: Alex Deucher <[email protected]>
Acked-by: Christian König <[email protected]>

4 files changed, 19 insertions, 14 deletions

diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
index 561378196b5..689f8166b65 100644
--- a/src/gallium/drivers/radeonsi/si_hw_context.c
+++ b/src/gallium/drivers/radeonsi/si_hw_context.c
@@ -189,6 +189,7 @@ void si_begin_new_cs(struct si_context *ctx)
 	si_mark_atom_dirty(ctx, &ctx->framebuffer.atom);
 
 	si_mark_atom_dirty(ctx, &ctx->clip_regs);
+	si_mark_atom_dirty(ctx, &ctx->clip_state.atom);
 	si_mark_atom_dirty(ctx, &ctx->msaa_sample_locs);
 	si_mark_atom_dirty(ctx, &ctx->msaa_config);
 	si_mark_atom_dirty(ctx, &ctx->db_render_state);
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 9be4aa7d5b5..829acf735f6 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -131,6 +131,11 @@ struct si_framebuffer {
 	bool				dirty_zsbuf;
 };
 
+struct si_clip_state {
+	struct r600_atom		atom;
+	struct pipe_clip_state		state;
+};
+
 struct si_scissors {
 	struct r600_atom		atom;
 	unsigned			dirty_mask;
@@ -184,6 +189,7 @@ struct si_context {
 
 	struct si_scissors		scissors;
 	struct si_viewports		viewports;
+	struct si_clip_state		clip_state;
 	struct r600_atom		clip_regs;
 	struct r600_atom		msaa_sample_locs;
 	struct r600_atom		msaa_config;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 3c250484e3a..d3e05b5e024 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -473,22 +473,13 @@ static void si_set_clip_state(struct pipe_context *ctx,
 			      const struct pipe_clip_state *state)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
-	struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
 	struct pipe_constant_buffer cb;
 
-	if (pm4 == NULL)
+	if (memcmp(&sctx->clip_state.state, state, sizeof(*state)) == 0)
 		return;
 
-	for (int i = 0; i < 6; i++) {
-		si_pm4_set_reg(pm4, R_0285BC_PA_CL_UCP_0_X + i * 16,
-			       fui(state->ucp[i][0]));
-		si_pm4_set_reg(pm4, R_0285C0_PA_CL_UCP_0_Y + i * 16,
-			       fui(state->ucp[i][1]));
-		si_pm4_set_reg(pm4, R_0285C4_PA_CL_UCP_0_Z + i * 16,
-			       fui(state->ucp[i][2]));
-		si_pm4_set_reg(pm4, R_0285C8_PA_CL_UCP_0_W + i * 16,
-			       fui(state->ucp[i][3]));
-        }
+	sctx->clip_state.state = *state;
+	si_mark_atom_dirty(sctx, &sctx->clip_state.atom);
 
 	cb.buffer = NULL;
 	cb.user_buffer = state->ucp;
@@ -496,8 +487,14 @@ static void si_set_clip_state(struct pipe_context *ctx,
 	cb.buffer_size = 4*4*8;
 	ctx->set_constant_buffer(ctx, PIPE_SHADER_VERTEX, SI_DRIVER_STATE_CONST_BUF, &cb);
 	pipe_resource_reference(&cb.buffer, NULL);
+}
+
+static void si_emit_clip_state(struct si_context *sctx, struct r600_atom *atom)
+{
+	struct radeon_winsys_cs *cs = sctx->b.rings.gfx.cs;
 
-	si_pm4_set_state(sctx, clip, pm4);
+	r600_write_context_reg_seq(cs, R_0285BC_PA_CL_UCP_0_X, 6*4);
+	radeon_emit_array(cs, (uint32_t*)sctx->clip_state.state.ucp, 6*4);
 }
 
 #define SIX_BITS 0x3F
@@ -3060,6 +3057,7 @@ void si_init_state_functions(struct si_context *sctx)
 	si_init_atom(sctx, &sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state, 10);
 	si_init_atom(sctx, &sctx->msaa_config, &sctx->atoms.s.msaa_config, si_emit_msaa_config, 10);
 	si_init_atom(sctx, &sctx->clip_regs, &sctx->atoms.s.clip_regs, si_emit_clip_regs, 6);
+	si_init_atom(sctx, &sctx->clip_state.atom, &sctx->atoms.s.clip_state, si_emit_clip_state, 2+6*4);
 	si_init_atom(sctx, &sctx->scissors.atom, &sctx->atoms.s.scissors, si_emit_scissors, 16*4);
 	si_init_atom(sctx, &sctx->viewports.atom, &sctx->atoms.s.viewports, si_emit_viewports, 16*8);
 
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 8cf308d14da..a8a22eb5dfc 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -84,7 +84,6 @@ union si_state {
 	struct {
 		struct si_state_blend		*blend;
 		struct si_pm4_state		*blend_color;
-		struct si_pm4_state		*clip;
 		struct si_state_sample_mask	*sample_mask;
 		struct si_state_rasterizer	*rasterizer;
 		struct si_state_dsa		*dsa;
@@ -115,6 +114,7 @@ union si_state_atoms {
 		struct r600_atom *db_render_state;
 		struct r600_atom *msaa_config;
 		struct r600_atom *clip_regs;
+		struct r600_atom *clip_state;
 		struct r600_atom *shader_userdata;
 		struct r600_atom *scissors;
 		struct r600_atom *viewports;