radeonsi: remove dirty slot masks from scissor and viewport states

All registers in the array need to be updated if any of them is changed.

Only apps writing gl_ViewportIndex were affected by this bug.

6 files changed, 40 insertions, 93 deletions

diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index 610de289a20..9d3d7d3d27a 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -58,7 +58,7 @@ void si_blitter_begin(struct si_context *sctx, enum si_blitter_op op)
 		util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref.state);
 		util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader.cso);
 		util_blitter_save_sample_mask(sctx->blitter, sctx->sample_mask);
-		util_blitter_save_scissor(sctx->blitter, &sctx->scissors.states[0]);
+		util_blitter_save_scissor(sctx->blitter, &sctx->scissors[0]);
 		util_blitter_save_window_rectangles(sctx->blitter,
 						    sctx->window_rectangles_include,
 						    sctx->num_window_rectangles,
diff --git a/src/gallium/drivers/radeonsi/si_gfx_cs.c b/src/gallium/drivers/radeonsi/si_gfx_cs.c
index 3e369234597..d0d405c473f 100644
--- a/src/gallium/drivers/radeonsi/si_gfx_cs.c
+++ b/src/gallium/drivers/radeonsi/si_gfx_cs.c
@@ -315,9 +315,6 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
 	if (!has_clear_state || ctx->num_window_rectangles > 0)
 		si_mark_atom_dirty(ctx, &ctx->atoms.s.window_rectangles);
 
-	ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
-	ctx->viewports.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
-	ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.guardband);
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 1d26ca90219..d3ddb912245 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -672,14 +672,7 @@ struct si_signed_scissor {
 	enum si_quant_mode quant_mode;
 };
 
-struct si_scissors {
-	unsigned			dirty_mask;
-	struct pipe_scissor_state	states[SI_MAX_VIEWPORTS];
-};
-
 struct si_viewports {
-	unsigned			dirty_mask;
-	unsigned			depth_range_dirty_mask;
 	struct pipe_viewport_state	states[SI_MAX_VIEWPORTS];
 	struct si_signed_scissor	as_scissor[SI_MAX_VIEWPORTS];
 };
@@ -883,7 +876,7 @@ struct si_context {
 	struct si_clip_state		clip_state;
 	struct si_shader_data		shader_pointers;
 	struct si_stencil_ref		stencil_ref;
-	struct si_scissors		scissors;
+	struct pipe_scissor_state	scissors[SI_MAX_VIEWPORTS];
 	struct si_streamout		streamout;
 	struct si_viewports		viewports;
 	unsigned			num_window_rectangles;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index bc7e777ad73..2266b0a0824 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -1015,10 +1015,8 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
 	si_update_poly_offset_state(sctx);
 
 	if (!old_rs ||
-	    old_rs->scissor_enable != rs->scissor_enable) {
-		sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
+	    old_rs->scissor_enable != rs->scissor_enable)
 		si_mark_atom_dirty(sctx, &sctx->atoms.s.scissors);
-	}
 
 	if (!old_rs ||
 	    old_rs->line_width != rs->line_width ||
@@ -1027,10 +1025,8 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
 		si_mark_atom_dirty(sctx, &sctx->atoms.s.guardband);
 
 	if (!old_rs ||
-	    old_rs->clip_halfz != rs->clip_halfz) {
-		sctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
+	    old_rs->clip_halfz != rs->clip_halfz)
 		si_mark_atom_dirty(sctx, &sctx->atoms.s.viewports);
-	}
 
 	if (!old_rs ||
 	    old_rs->clip_plane_enable != rs->clip_plane_enable ||
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 4b60679484f..8e01e1b35e1 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -1487,10 +1487,9 @@ static void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *i
 
 		if (has_gfx9_scissor_bug &&
 		    (sctx->context_roll ||
-		     si_is_atom_dirty(sctx, &sctx->atoms.s.scissors))) {
-			sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
+		     si_is_atom_dirty(sctx, &sctx->atoms.s.scissors)))
 			sctx->atoms.s.scissors.emit(sctx);
-		}
+
 		sctx->dirty_atoms = 0;
 
 		si_emit_draw_packets(sctx, info, indexbuf, index_size, index_offset);
@@ -1519,10 +1518,9 @@ static void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *i
 
 		if (has_gfx9_scissor_bug &&
 		    (sctx->context_roll ||
-		     si_is_atom_dirty(sctx, &sctx->atoms.s.scissors))) {
-			sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
+		     si_is_atom_dirty(sctx, &sctx->atoms.s.scissors)))
 			sctx->atoms.s.scissors.emit(sctx);
-		}
+
 		sctx->dirty_atoms = 0;
 
 		si_emit_draw_packets(sctx, info, indexbuf, index_size, index_offset);
diff --git a/src/gallium/drivers/radeonsi/si_state_viewport.c b/src/gallium/drivers/radeonsi/si_state_viewport.c
index 6f348a9b58d..792d1c4efd1 100644
--- a/src/gallium/drivers/radeonsi/si_state_viewport.c
+++ b/src/gallium/drivers/radeonsi/si_state_viewport.c
@@ -37,13 +37,12 @@ static void si_set_scissor_states(struct pipe_context *pctx,
 	int i;
 
 	for (i = 0; i < num_scissors; i++)
-		ctx->scissors.states[start_slot + i] = state[i];
+		ctx->scissors[start_slot + i] = state[i];
 
 	if (!ctx->queued.named.rasterizer ||
 	    !ctx->queued.named.rasterizer->scissor_enable)
 		return;
 
-	ctx->scissors.dirty_mask |= ((1 << num_scissors) - 1) << start_slot;
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
 }
 
@@ -289,36 +288,27 @@ static void si_emit_guardband(struct si_context *ctx)
 static void si_emit_scissors(struct si_context *ctx)
 {
 	struct radeon_cmdbuf *cs = ctx->gfx_cs;
-	struct pipe_scissor_state *states = ctx->scissors.states;
-	unsigned mask = ctx->scissors.dirty_mask;
+	struct pipe_scissor_state *states = ctx->scissors;
 	bool scissor_enabled = ctx->queued.named.rasterizer->scissor_enable;
 
 	/* The simple case: Only 1 viewport is active. */
 	if (!ctx->vs_writes_viewport_index) {
 		struct si_signed_scissor *vp = &ctx->viewports.as_scissor[0];
 
-		if (!(mask & 1))
-			return;
-
 		radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL, 2);
 		si_emit_one_scissor(ctx, cs, vp, scissor_enabled ? &states[0] : NULL);
-		ctx->scissors.dirty_mask &= ~1; /* clear one bit */
 		return;
 	}
 
-	while (mask) {
-		int start, count, i;
-
-		u_bit_scan_consecutive_range(&mask, &start, &count);
-
-		radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL +
-					       start * 4 * 2, count * 2);
-		for (i = start; i < start+count; i++) {
-			si_emit_one_scissor(ctx, cs, &ctx->viewports.as_scissor[i],
-					    scissor_enabled ? &states[i] : NULL);
-		}
+	/* All registers in the array need to be updated if any of them is changed.
+	 * This is a hardware requirement.
+	 */
+	radeon_set_context_reg_seq(cs, R_028250_PA_SC_VPORT_SCISSOR_0_TL,
+				   SI_MAX_VIEWPORTS * 2);
+	for (unsigned i = 0; i < SI_MAX_VIEWPORTS; i++) {
+		si_emit_one_scissor(ctx, cs, &ctx->viewports.as_scissor[i],
+				    scissor_enabled ? &states[i] : NULL);
 	}
-	ctx->scissors.dirty_mask = 0;
 }
 
 static void si_set_viewport_states(struct pipe_context *pctx,
@@ -327,7 +317,6 @@ static void si_set_viewport_states(struct pipe_context *pctx,
 				   const struct pipe_viewport_state *state)
 {
 	struct si_context *ctx = (struct si_context *)pctx;
-	unsigned mask;
 	int i;
 
 	for (i = 0; i < num_viewports; i++) {
@@ -392,10 +381,6 @@ static void si_set_viewport_states(struct pipe_context *pctx,
 			scissor->quant_mode = SI_QUANT_MODE_16_8_FIXED_POINT_1_256TH;
 	}
 
-	mask = ((1 << num_viewports) - 1) << start_slot;
-	ctx->viewports.dirty_mask |= mask;
-	ctx->viewports.depth_range_dirty_mask |= mask;
-	ctx->scissors.dirty_mask |= mask;
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.guardband);
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
@@ -418,30 +403,21 @@ static void si_emit_viewports(struct si_context *ctx)
 {
 	struct radeon_cmdbuf *cs = ctx->gfx_cs;
 	struct pipe_viewport_state *states = ctx->viewports.states;
-	unsigned mask = ctx->viewports.dirty_mask;
 
 	/* The simple case: Only 1 viewport is active. */
 	if (!ctx->vs_writes_viewport_index) {
-		if (!(mask & 1))
-			return;
-
 		radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE, 6);
 		si_emit_one_viewport(ctx, &states[0]);
-		ctx->viewports.dirty_mask &= ~1; /* clear one bit */
 		return;
 	}
 
-	while (mask) {
-		int start, count, i;
-
-		u_bit_scan_consecutive_range(&mask, &start, &count);
-
-		radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE +
-					       start * 4 * 6, count * 6);
-		for (i = start; i < start+count; i++)
-			si_emit_one_viewport(ctx, &states[i]);
-	}
-	ctx->viewports.dirty_mask = 0;
+	/* All registers in the array need to be updated if any of them is changed.
+	 * This is a hardware requirement.
+	 */
+	radeon_set_context_reg_seq(cs, R_02843C_PA_CL_VPORT_XSCALE +
+				   0, SI_MAX_VIEWPORTS * 6);
+	for (unsigned i = 0; i < SI_MAX_VIEWPORTS; i++)
+		si_emit_one_viewport(ctx, &states[i]);
 }
 
 static inline void
@@ -460,41 +436,32 @@ static void si_emit_depth_ranges(struct si_context *ctx)
 {
 	struct radeon_cmdbuf *cs = ctx->gfx_cs;
 	struct pipe_viewport_state *states = ctx->viewports.states;
-	unsigned mask = ctx->viewports.depth_range_dirty_mask;
 	bool clip_halfz = ctx->queued.named.rasterizer->clip_halfz;
 	bool window_space = ctx->vs_disables_clipping_viewport;
 	float zmin, zmax;
 
 	/* The simple case: Only 1 viewport is active. */
 	if (!ctx->vs_writes_viewport_index) {
-		if (!(mask & 1))
-			return;
-
 		si_viewport_zmin_zmax(&states[0], clip_halfz, window_space,
 				      &zmin, &zmax);
 
 		radeon_set_context_reg_seq(cs, R_0282D0_PA_SC_VPORT_ZMIN_0, 2);
 		radeon_emit(cs, fui(zmin));
 		radeon_emit(cs, fui(zmax));
-		ctx->viewports.depth_range_dirty_mask &= ~1; /* clear one bit */
 		return;
 	}
 
-	while (mask) {
-		int start, count, i;
-
-		u_bit_scan_consecutive_range(&mask, &start, &count);
-
-		radeon_set_context_reg_seq(cs, R_0282D0_PA_SC_VPORT_ZMIN_0 +
-					   start * 4 * 2, count * 2);
-		for (i = start; i < start+count; i++) {
-			si_viewport_zmin_zmax(&states[i], clip_halfz, window_space,
-					      &zmin, &zmax);
-			radeon_emit(cs, fui(zmin));
-			radeon_emit(cs, fui(zmax));
-		}
+	/* All registers in the array need to be updated if any of them is changed.
+	 * This is a hardware requirement.
+	 */
+	radeon_set_context_reg_seq(cs, R_0282D0_PA_SC_VPORT_ZMIN_0,
+				   SI_MAX_VIEWPORTS * 2);
+	for (unsigned i = 0; i < SI_MAX_VIEWPORTS; i++) {
+		si_viewport_zmin_zmax(&states[i], clip_halfz, window_space,
+				      &zmin, &zmax);
+		radeon_emit(cs, fui(zmin));
+		radeon_emit(cs, fui(zmax));
 	}
-	ctx->viewports.depth_range_dirty_mask = 0;
 }
 
 static void si_emit_viewport_states(struct si_context *ctx)
@@ -527,8 +494,6 @@ void si_update_vs_viewport_state(struct si_context *ctx)
 
 	if (ctx->vs_disables_clipping_viewport != vs_window_space) {
 		ctx->vs_disables_clipping_viewport = vs_window_space;
-		ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
-		ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
 		si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
 		si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
 	}
@@ -541,15 +506,13 @@ void si_update_vs_viewport_state(struct si_context *ctx)
 	ctx->vs_writes_viewport_index = info->writes_viewport_index;
 	si_mark_atom_dirty(ctx, &ctx->atoms.s.guardband);
 
-	if (!ctx->vs_writes_viewport_index)
-		return;
-
-	if (ctx->scissors.dirty_mask)
+	/* Emit scissors and viewports that were enabled by having
+	 * the ViewportIndex output.
+	 */
+	if (info->writes_viewport_index) {
 	    si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
-
-	if (ctx->viewports.dirty_mask ||
-	    ctx->viewports.depth_range_dirty_mask)
 	    si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
+	}
 }
 
 static void si_emit_window_rectangles(struct si_context *sctx)