radeonsi: optimize viewport states

same as scissors

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

6 files changed, 54 insertions, 26 deletions

diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index b2f342f4fa1..c28b2a80088 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -65,9 +65,7 @@ static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
 		util_blitter_save_sample_mask(sctx->blitter,
 					      sctx->queued.named.sample_mask->sample_mask);
 	}
-	if (sctx->queued.named.viewport[0]) {
-		util_blitter_save_viewport(sctx->blitter, &sctx->queued.named.viewport[0]->viewport);
-	}
+	util_blitter_save_viewport(sctx->blitter, &sctx->viewports.states[0]);
 	util_blitter_save_scissor(sctx->blitter, &sctx->scissors.states[0]);
 	util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
 	util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
index 873a4727976..8284306f30a 100644
--- a/src/gallium/drivers/radeonsi/si_hw_context.c
+++ b/src/gallium/drivers/radeonsi/si_hw_context.c
@@ -195,7 +195,9 @@ void si_begin_new_cs(struct si_context *ctx)
 	si_all_descriptors_begin_new_cs(ctx);
 
 	ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
+	ctx->viewports.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
 	si_mark_atom_dirty(ctx, &ctx->scissors.atom);
+	si_mark_atom_dirty(ctx, &ctx->viewports.atom);
 
 	r600_postflush_resume_features(&ctx->b);
 
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 9060f948971..a1845ba893d 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -135,6 +135,12 @@ struct si_scissors {
 	struct pipe_scissor_state	states[SI_MAX_VIEWPORTS];
 };
 
+struct si_viewports {
+	struct r600_atom		atom;
+	unsigned			dirty_mask;
+	struct pipe_viewport_state	states[SI_MAX_VIEWPORTS];
+};
+
 #define SI_NUM_ATOMS(sctx) (sizeof((sctx)->atoms)/sizeof((sctx)->atoms.array[0]))
 
 struct si_context {
@@ -163,6 +169,7 @@ struct si_context {
 			struct r600_atom *clip_regs;
 			struct r600_atom *shader_userdata;
 			struct r600_atom *scissors;
+			struct r600_atom *viewports;
 		} s;
 		struct r600_atom *array[0];
 	} atoms;
@@ -191,6 +198,7 @@ struct si_context {
 	unsigned			border_color_offset;
 
 	struct si_scissors		scissors;
+	struct si_viewports		viewports;
 	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 8bd35a8422e..940aaa0b247 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -578,29 +578,52 @@ static void si_set_viewport_states(struct pipe_context *ctx,
                                    const struct pipe_viewport_state *state)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
-	struct si_state_viewport *viewport;
-	struct si_pm4_state *pm4;
 	int i;
 
-	for (i = start_slot; i < start_slot + num_viewports; i++) {
-		int idx = i - start_slot;
-		int offset = i * 4 * 6;
+	for (i = 0; i < num_viewports; i++)
+		sctx->viewports.states[start_slot + i] = state[i];
 
-		viewport = CALLOC_STRUCT(si_state_viewport);
-		if (!viewport)
-			return;
-		pm4 = &viewport->pm4;
+	sctx->viewports.dirty_mask |= ((1 << num_viewports) - 1) << start_slot;
+	si_mark_atom_dirty(sctx, &sctx->viewports.atom);
+}
+
+static void si_emit_viewports(struct si_context *sctx, struct r600_atom *atom)
+{
+	struct radeon_winsys_cs *cs = sctx->b.rings.gfx.cs;
+	struct pipe_viewport_state *states = sctx->viewports.states;
+	unsigned mask = sctx->viewports.dirty_mask;
+
+	/* The simple case: Only 1 viewport is active. */
+	if (mask & 1 &&
+	    !si_get_vs_info(sctx)->writes_viewport_index) {
+		r600_write_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE, 6);
+		radeon_emit(cs, fui(states[0].scale[0]));
+		radeon_emit(cs, fui(states[0].translate[0]));
+		radeon_emit(cs, fui(states[0].scale[1]));
+		radeon_emit(cs, fui(states[0].translate[1]));
+		radeon_emit(cs, fui(states[0].scale[2]));
+		radeon_emit(cs, fui(states[0].translate[2]));
+		sctx->viewports.dirty_mask &= ~1; /* clear one bit */
+		return;
+	}
 
-		viewport->viewport = state[idx];
-		si_pm4_set_reg(pm4, R_02843C_PA_CL_VPORT_XSCALE + offset, fui(state[idx].scale[0]));
-		si_pm4_set_reg(pm4, R_028440_PA_CL_VPORT_XOFFSET + offset, fui(state[idx].translate[0]));
-		si_pm4_set_reg(pm4, R_028444_PA_CL_VPORT_YSCALE + offset, fui(state[idx].scale[1]));
-		si_pm4_set_reg(pm4, R_028448_PA_CL_VPORT_YOFFSET + offset, fui(state[idx].translate[1]));
-		si_pm4_set_reg(pm4, R_02844C_PA_CL_VPORT_ZSCALE + offset, fui(state[idx].scale[2]));
-		si_pm4_set_reg(pm4, R_028450_PA_CL_VPORT_ZOFFSET + offset, fui(state[idx].translate[2]));
+	while (mask) {
+		int start, count, i;
 
-		si_pm4_set_state(sctx, viewport[i], viewport);
+		u_bit_scan_consecutive_range(&mask, &start, &count);
+
+		r600_write_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE +
+					       start * 4 * 6, count * 6);
+		for (i = start; i < start+count; i++) {
+			radeon_emit(cs, fui(states[i].scale[0]));
+			radeon_emit(cs, fui(states[i].translate[0]));
+			radeon_emit(cs, fui(states[i].scale[1]));
+			radeon_emit(cs, fui(states[i].translate[1]));
+			radeon_emit(cs, fui(states[i].scale[2]));
+			radeon_emit(cs, fui(states[i].translate[2]));
+		}
 	}
+	sctx->viewports.dirty_mask = 0;
 }
 
 /*
@@ -3011,6 +3034,7 @@ void si_init_state_functions(struct si_context *sctx)
 	si_init_atom(&sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state, 10);
 	si_init_atom(&sctx->clip_regs, &sctx->atoms.s.clip_regs, si_emit_clip_regs, 6);
 	si_init_atom(&sctx->scissors.atom, &sctx->atoms.s.scissors, si_emit_scissors, 16*4);
+	si_init_atom(&sctx->viewports.atom, &sctx->atoms.s.viewports, si_emit_viewports, 16*8);
 
 	sctx->b.b.create_blend_state = si_create_blend_state;
 	sctx->b.b.bind_blend_state = si_bind_blend_state;
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 34dbba48050..32147831699 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -48,11 +48,6 @@ struct si_state_sample_mask {
 	uint16_t		sample_mask;
 };
 
-struct si_state_viewport {
-	struct si_pm4_state		pm4;
-	struct pipe_viewport_state	viewport;
-};
-
 struct si_state_rasterizer {
 	struct si_pm4_state	pm4;
 	bool			flatshade;
@@ -91,7 +86,6 @@ union si_state {
 		struct si_pm4_state		*blend_color;
 		struct si_pm4_state		*clip;
 		struct si_state_sample_mask	*sample_mask;
-		struct si_state_viewport	*viewport[16];
 		struct si_state_rasterizer	*rasterizer;
 		struct si_state_dsa		*dsa;
 		struct si_pm4_state		*fb_rs;
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 5a9ef29a549..4ca9aa50814 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -775,6 +775,8 @@ static void si_update_viewports_and_scissors(struct si_context *sctx)
 
 	if (sctx->scissors.dirty_mask)
 	    si_mark_atom_dirty(sctx, &sctx->scissors.atom);
+	if (sctx->viewports.dirty_mask)
+	    si_mark_atom_dirty(sctx, &sctx->viewports.atom);
 }
 
 static void si_bind_vs_shader(struct pipe_context *ctx, void *state)