[Mesa-dev] [PATCH 20/21] r600g: accelerate buffer copying

[Marek Olšák]

This will be useful for efficient handling of the DISCARD transfer flags.
---
 src/gallium/drivers/r600/r600_blit.c |   78 +++++++++++++++++++++++----------
 1 files changed, 54 insertions(+), 24 deletions(-)

diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
index 9ab85c8..0d9618d 100644
--- a/src/gallium/drivers/r600/r600_blit.c
+++ b/src/gallium/drivers/r600/r600_blit.c
@@ -27,18 +27,21 @@
 
 enum r600_blitter_op /* bitmask */
 {
-	R600_SAVE_TEXTURES      = 1,
-	R600_SAVE_FRAMEBUFFER   = 2,
-	R600_DISABLE_RENDER_COND = 4,
+	R600_SAVE_FRAGMENT_STATE = 1,
+	R600_SAVE_TEXTURES       = 2,
+	R600_SAVE_FRAMEBUFFER    = 4,
+	R600_DISABLE_RENDER_COND = 8,
 
-	R600_CLEAR         = 0,
+	R600_CLEAR         = R600_SAVE_FRAGMENT_STATE,
 
-	R600_CLEAR_SURFACE = R600_SAVE_FRAMEBUFFER,
+	R600_CLEAR_SURFACE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER,
 
-	R600_COPY          = R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
+	R600_COPY_BUFFER   = R600_DISABLE_RENDER_COND,
+
+	R600_COPY_TEXTURE  = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
 			     R600_DISABLE_RENDER_COND,
 
-	R600_DECOMPRESS    = R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
+	R600_DECOMPRESS    = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
 };
 
 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
@@ -47,23 +50,26 @@ static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op
 
 	r600_suspend_nontimer_queries(rctx);
 
-	util_blitter_save_blend(rctx->blitter, rctx->states[R600_PIPE_STATE_BLEND]);
-	util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->states[R600_PIPE_STATE_DSA]);
-	if (rctx->states[R600_PIPE_STATE_STENCIL_REF]) {
-		util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref);
-	}
-	util_blitter_save_rasterizer(rctx->blitter, rctx->states[R600_PIPE_STATE_RASTERIZER]);
-	util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
-	util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
-	util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_elements);
-	if (rctx->states[R600_PIPE_STATE_VIEWPORT]) {
-		util_blitter_save_viewport(rctx->blitter, &rctx->viewport);
-	}
 	util_blitter_save_vertex_buffers(rctx->blitter,
 					 rctx->vbuf_mgr->nr_vertex_buffers,
 					 rctx->vbuf_mgr->vertex_buffer);
+	util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_elements);
+	util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
 	util_blitter_save_so_targets(rctx->blitter, rctx->num_so_targets,
 				     (struct pipe_stream_output_target**)rctx->so_targets);
+	util_blitter_save_rasterizer(rctx->blitter, rctx->states[R600_PIPE_STATE_RASTERIZER]);
+
+	if (op & R600_SAVE_FRAGMENT_STATE) {
+		if (rctx->states[R600_PIPE_STATE_VIEWPORT]) {
+			util_blitter_save_viewport(rctx->blitter, &rctx->viewport);
+		}
+		util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
+		util_blitter_save_blend(rctx->blitter, rctx->states[R600_PIPE_STATE_BLEND]);
+		util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->states[R600_PIPE_STATE_DSA]);
+		if (rctx->states[R600_PIPE_STATE_STENCIL_REF]) {
+			util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref);
+		}
+	}
 
 	if (op & R600_SAVE_FRAMEBUFFER)
 		util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer);
@@ -240,7 +246,32 @@ static void r600_clear_depth_stencil(struct pipe_context *ctx,
 	r600_blitter_end(ctx);
 }
 
-
+static void r600_copy_buffer(struct pipe_context *ctx,
+			     struct pipe_resource *dst,
+			     unsigned dstx,
+			     struct pipe_resource *src,
+			     const struct pipe_box *src_box)
+{
+	struct r600_context *rctx = (struct r600_context*)ctx;
+
+	if (rctx->screen->info.r600_has_streamout &&
+	    /* Require dword alignment. */
+	    dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
+#if 0
+		printf("copy_buffer in hw (streamout=%i, dstx = %i, srcx = %i, size = %i\n",
+		       rctx->screen->info.r600_has_streamout, dstx, src_box->x, src_box->width);
+#endif
+		r600_blitter_begin(ctx, R600_COPY_BUFFER);
+		util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
+		r600_blitter_end(ctx);
+	} else {
+#if 0
+		printf("copy_buffer in sw (streamout=%i, dstx = %i, srcx = %i, size = %i\n",
+		       rctx->screen->info.r600_has_streamout, dstx, src_box->x, src_box->width);
+#endif
+		util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
+	}
+}
 
 /* Copy a block of pixels from one surface to another using HW. */
 static void r600_hw_copy_region(struct pipe_context *ctx,
@@ -253,7 +284,7 @@ static void r600_hw_copy_region(struct pipe_context *ctx,
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
 
-	r600_blitter_begin(ctx, R600_COPY);
+	r600_blitter_begin(ctx, R600_COPY_TEXTURE);
 	util_blitter_copy_texture(rctx->blitter, dst, dst_level, dstx, dsty, dstz,
 				  src, src_level, src_box, TRUE);
 	r600_blitter_end(ctx);
@@ -334,10 +365,9 @@ static void r600_resource_copy_region(struct pipe_context *ctx,
 
 	memset(orig_info, 0, sizeof(orig_info));
 
-	/* Fallback for buffers. */
+	/* Handle buffers first. */
 	if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
-		util_resource_copy_region(ctx, dst, dst_level, dstx, dsty, dstz,
-                                          src, src_level, src_box);
+		r600_copy_buffer(ctx, dst, dstx, src, src_box);
 		return;
 	}
 
-- 
1.7.5.4