freedreno: support for using generic clear path

Since clears are more or less just normal draws, there isn't that much
benefit in having hand-rolled clear path.  Add support to use u_blitter
instead if gen specific backend doesn't implement ctx->clear().

Signed-off-by: Rob Clark <robdclark@gmail.com>

src/gallium/drivers/freedreno/freedreno_context.c

@@ -115,6 +115,9 @@ fd_context_destroy(struct pipe_context *pctx)
 	if (ctx->blitter)
 		util_blitter_destroy(ctx->blitter);
 
+	if (ctx->clear_rs_state)
+		pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state);
+
 	if (ctx->primconvert)
 		util_primconvert_destroy(ctx->primconvert);
 

src/gallium/drivers/freedreno/freedreno_context.h

@@ -117,6 +117,7 @@ struct fd_context {
 	struct util_queue flush_queue;
 
 	struct blitter_context *blitter;
+	void *clear_rs_state;
 	struct primconvert_context *primconvert;
 
 	/* slab for pipe_transfer allocations: */

src/gallium/drivers/freedreno/freedreno_draw.c

@@ -203,6 +203,72 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 	fd_batch_check_size(batch);
 }
 
+/* Generic clear implementation (partially) using u_blitter: */
+static void
+fd_blitter_clear(struct pipe_context *pctx, unsigned buffers,
+		const union pipe_color_union *color, double depth, unsigned stencil)
+{
+	struct fd_context *ctx = fd_context(pctx);
+	struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
+	struct blitter_context *blitter = ctx->blitter;
+
+	fd_blitter_pipe_begin(ctx, false, true, FD_STAGE_CLEAR);
+
+	util_blitter_common_clear_setup(blitter, pfb->width, pfb->height,
+			buffers, NULL, NULL);
+
+	struct pipe_stencil_ref sr = {
+		.ref_value = { stencil & 0xff }
+	};
+	pctx->set_stencil_ref(pctx, &sr);
+
+	struct pipe_constant_buffer cb = {
+		.buffer_size = 16,
+		.user_buffer = &color->ui,
+	};
+	pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 0, &cb);
+
+	if (!ctx->clear_rs_state) {
+		const struct pipe_rasterizer_state tmpl = {
+			.cull_face = PIPE_FACE_NONE,
+			.half_pixel_center = 1,
+			.bottom_edge_rule = 1,
+			.flatshade = 1,
+			.depth_clip = 1,
+		};
+		ctx->clear_rs_state = pctx->create_rasterizer_state(pctx, &tmpl);
+	}
+	pctx->bind_rasterizer_state(pctx, ctx->clear_rs_state);
+
+	struct pipe_viewport_state vp = {
+		.scale     = { 0.5f * pfb->width, -0.5f * pfb->height, depth },
+		.translate = { 0.5f * pfb->width,  0.5f * pfb->height, 0.0f },
+	};
+	pctx->set_viewport_states(pctx, 0, 1, &vp);
+
+	pctx->bind_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
+	pctx->set_vertex_buffers(pctx, blitter->vb_slot, 1,
+			&ctx->solid_vbuf_state.vertexbuf.vb[0]);
+	pctx->set_stream_output_targets(pctx, 0, NULL, NULL);
+	pctx->bind_vs_state(pctx, ctx->solid_prog.vp);
+	pctx->bind_fs_state(pctx, ctx->solid_prog.fp);
+
+	struct pipe_draw_info info = {
+		.mode = PIPE_PRIM_MAX,    /* maps to DI_PT_RECTLIST */
+		.count = 2,
+		.instance_count = 1,
+	};
+	ctx->draw_vbo(ctx, &info);
+
+	util_blitter_restore_constant_buffer_state(blitter);
+	util_blitter_restore_vertex_states(blitter);
+	util_blitter_restore_fragment_states(blitter);
+	util_blitter_restore_render_cond(blitter);
+	util_blitter_unset_running_flag(blitter);
+
+	fd_blitter_pipe_end(ctx);
+}
+
 /* TODO figure out how to make better use of existing state mechanism
  * for clear (and possibly gmem->mem / mem->gmem) so we can (a) keep
  * track of what state really actually changes, and (b) reduce the code
@@ -274,6 +340,14 @@ fd_clear(struct pipe_context *pctx, unsigned buffers,
 		util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 		util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 
+	/* if per-gen backend doesn't implement ctx->clear() generic
+	 * blitter clear:
+	 */
+	if (!ctx->clear) {
+		fd_blitter_clear(pctx, buffers, color, depth, stencil);
+		return;
+	}
+
 	fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_CLEAR);
 
 	ctx->clear(ctx, buffers, color, depth, stencil);

src/gallium/drivers/freedreno/freedreno_resource.c

@@ -107,16 +107,13 @@ realloc_bo(struct fd_resource *rsc, uint32_t size)
 	fd_bc_invalidate_resource(rsc, true);
 }
 
-static void fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond, bool discard);
-static void fd_blitter_pipe_end(struct fd_context *ctx);
-
 static void
 do_blit(struct fd_context *ctx, const struct pipe_blit_info *blit, bool fallback)
 {
 	/* TODO size threshold too?? */
 	if ((blit->src.resource->target != PIPE_BUFFER) && !fallback) {
 		/* do blit on gpu: */
-		fd_blitter_pipe_begin(ctx, false, true);
+		fd_blitter_pipe_begin(ctx, false, true, FD_STAGE_BLIT);
 		util_blitter_blit(ctx->blitter, blit);
 		fd_blitter_pipe_end(ctx);
 	} else {
@@ -939,7 +936,7 @@ fd_blitter_pipe_copy_region(struct fd_context *ctx,
 		return false;
 
 	/* TODO we could discard if dst box covers dst level fully.. */
-	fd_blitter_pipe_begin(ctx, false, false);
+	fd_blitter_pipe_begin(ctx, false, false, FD_STAGE_BLIT);
 	util_blitter_copy_texture(ctx->blitter,
 			dst, dst_level, dstx, dsty, dstz,
 			src, src_level, src_box);
@@ -1045,14 +1042,17 @@ fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
 		return;
 	}
 
-	fd_blitter_pipe_begin(ctx, info.render_condition_enable, discard);
+	fd_blitter_pipe_begin(ctx, info.render_condition_enable, discard, FD_STAGE_BLIT);
 	util_blitter_blit(ctx->blitter, &info);
 	fd_blitter_pipe_end(ctx);
 }
 
-static void
-fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond, bool discard)
+void
+fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond, bool discard,
+		enum fd_render_stage stage)
 {
+	util_blitter_save_fragment_constant_buffer_slot(ctx->blitter,
+			ctx->constbuf[PIPE_SHADER_FRAGMENT].cb);
 	util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
 	util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
 	util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
@@ -1078,12 +1078,12 @@ fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond, bool discard)
 			ctx->cond_query, ctx->cond_cond, ctx->cond_mode);
 
 	if (ctx->batch)
-		fd_hw_query_set_stage(ctx->batch, ctx->batch->draw, FD_STAGE_BLIT);
+		fd_hw_query_set_stage(ctx->batch, ctx->batch->draw, stage);
 
 	ctx->in_blit = discard;
 }
 
-static void
+void
 fd_blitter_pipe_end(struct fd_context *ctx)
 {
 	if (ctx->batch)

src/gallium/drivers/freedreno/freedreno_resource.h

@@ -154,6 +154,10 @@ fd_resource_offset(struct fd_resource *rsc, unsigned level, unsigned layer)
 	return offset;
 }
 
+void fd_blitter_pipe_begin(struct fd_context *ctx, bool render_cond, bool discard,
+		enum fd_render_stage stage);
+void fd_blitter_pipe_end(struct fd_context *ctx);
+
 void fd_resource_screen_init(struct pipe_screen *pscreen);
 void fd_resource_context_init(struct pipe_context *pctx);