freedreno/a6xx: Emit streamout state on every draw

If stream-output is active, we can't skip it's state-emit, even if the
SO buffer(s) have not changed.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9581>

src/gallium/drivers/freedreno/a6xx/fd6_draw.c

@@ -224,6 +224,10 @@ fd6_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
 	if (emit.vs->need_driver_params || fd6_ctx->has_dp_state)
 		emit.dirty_groups |= BIT(FD6_GROUP_VS_DRIVER_PARAMS);
 
+	/* If we are doing xfb, we need to emit the xfb state on every draw: */
+	if (emit.prog->stream_output)
+		emit.dirty_groups |= BIT(FD6_GROUP_SO);
+
 	if (unlikely(ctx->stats_users > 0)) {
 		ctx->stats.vs_regs += ir3_shader_halfregs(emit.vs);
 		ctx->stats.hs_regs += COND(emit.hs, ir3_shader_halfregs(emit.hs));

src/gallium/drivers/freedreno/a6xx/fd6_emit.c

@@ -890,12 +890,12 @@ fd6_emit_streamout(struct fd_ringbuffer *ring, struct fd6_emit *emit)
 {
 	struct fd_context *ctx = emit->ctx;
 	const struct fd6_program_state *prog = fd6_emit_get_prog(emit);
-	struct ir3_stream_output_info *info = &fd6_last_shader(prog)->shader->stream_output;
+	struct ir3_stream_output_info *info = prog->stream_output;;
 	struct fd_streamout_stateobj *so = &ctx->streamout;
 
 	emit->streamout_mask = 0;
 
-	if (!info->num_outputs)
+	if (!info)
 		return;
 
 	for (unsigned i = 0; i < so->num_targets; i++) {

src/gallium/drivers/freedreno/a6xx/fd6_program.c

@@ -1094,6 +1094,11 @@ fd6_program_create(void *data, struct ir3_shader_variant *bs,
 	setup_stateobj(state->stateobj, ctx, state, key, false);
 	state->interp_stateobj = create_interp_stateobj(ctx, state);
 
+	struct ir3_stream_output_info *stream_output =
+			&fd6_last_shader(state)->shader->stream_output;
+	if (stream_output->num_outputs > 0)
+		state->stream_output = stream_output;
+
 	return &state->base;
 }
 

src/gallium/drivers/freedreno/a6xx/fd6_program.h

@@ -50,6 +50,8 @@ struct fd6_program_state {
 	struct fd_ringbuffer *streamout_stateobj;
 	struct fd_ringbuffer *stateobj;
 
+	struct ir3_stream_output_info *stream_output;
+
 	/**
 	 * Output components from frag shader.  It is possible to have
 	 * a fragment shader that only writes a subset of the bound