r600g: propogate resource usage flags to winsys, use to choose bo domains

This opens the question of what interface the winsys layer should
really have for talking about these concepts.

For now I'm using the existing gallium resource usage concept, but
there is no reason not use terms closer to what the hardware
understands - eg. the domains themselves.

7 files changed, 39 insertions, 13 deletions

diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 62d983269f5..5ec607bbd85 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -112,7 +112,8 @@ struct r600_tiling_info *r600_get_tiling_info(struct radeon *radeon);
 /* r600_bo.c */
 struct r600_bo;
 struct r600_bo *r600_bo(struct radeon *radeon,
-				  unsigned size, unsigned alignment, unsigned usage);
+                        unsigned size, unsigned alignment,
+                        unsigned binding, unsigned usage);
 struct r600_bo *r600_bo_handle(struct radeon *radeon,
 			       unsigned handle, unsigned *array_mode);
 void *r600_bo_map(struct radeon *radeon, struct r600_bo *bo, unsigned usage, void *ctx);
diff --git a/src/gallium/drivers/r600/r600_buffer.c b/src/gallium/drivers/r600/r600_buffer.c
index 455aa2e81f6..3c45d782a2c 100644
--- a/src/gallium/drivers/r600/r600_buffer.c
+++ b/src/gallium/drivers/r600/r600_buffer.c
@@ -86,7 +86,7 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 	rbuffer->r.base.vtbl = &r600_buffer_vtbl;
 	rbuffer->r.size = rbuffer->r.base.b.width0;
 	rbuffer->r.domain = r600_domain_from_usage(rbuffer->r.base.b.bind);
-	bo = r600_bo((struct radeon*)screen->winsys, rbuffer->r.base.b.width0, alignment, rbuffer->r.base.b.bind);
+	bo = r600_bo((struct radeon*)screen->winsys, rbuffer->r.base.b.width0, alignment, rbuffer->r.base.b.bind, rbuffer->r.base.b.usage);
 	if (bo == NULL) {
 		FREE(rbuffer);
 		return NULL;
@@ -156,8 +156,9 @@ static void *r600_buffer_transfer_map(struct pipe_context *pipe,
 				r600_bo_reference((struct radeon*)pipe->winsys, &rbuffer->r.bo, NULL);
 				rbuffer->num_ranges = 0;
 				rbuffer->r.bo = r600_bo((struct radeon*)pipe->winsys,
-							     rbuffer->r.base.b.width0, 0,
-							     rbuffer->r.base.b.bind);
+                                                        rbuffer->r.base.b.width0, 0,
+                                                        rbuffer->r.base.b.bind,
+                                                        rbuffer->r.base.b.usage);
 				break;
 			}
 		}
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 4106587398b..1a0b35d9bf5 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -218,7 +218,7 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
 
 	/* copy new shader */
 	if (shader->bo == NULL) {
-		shader->bo = r600_bo(rctx->radeon, rshader->bc.ndw * 4, 4096, 0);
+		shader->bo = r600_bo(rctx->radeon, rshader->bc.ndw * 4, 4096, 0, 0);
 		if (shader->bo == NULL) {
 			return -ENOMEM;
 		}
diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/r600/r600_texture.c
index 7222b43af65..9a52cfa46e2 100644
--- a/src/gallium/drivers/r600/r600_texture.c
+++ b/src/gallium/drivers/r600/r600_texture.c
@@ -294,7 +294,7 @@ r600_texture_create_object(struct pipe_screen *screen,
 	resource->size = rtex->size;
 
 	if (!resource->bo) {
-		resource->bo = r600_bo(radeon, rtex->size, 4096, 0);
+		resource->bo = r600_bo(radeon, rtex->size, 4096, base->bind, base->usage);
 		if (!resource->bo) {
 			FREE(rtex);
 			return NULL;
diff --git a/src/gallium/winsys/r600/drm/r600_bo.c b/src/gallium/winsys/r600/drm/r600_bo.c
index 7d54ff18fc2..9b9aec5f64c 100644
--- a/src/gallium/winsys/r600/drm/r600_bo.c
+++ b/src/gallium/winsys/r600/drm/r600_bo.c
@@ -29,23 +29,37 @@
 #include "radeon_drm.h"
 #include "r600_priv.h"
 #include "r600d.h"
+#include "radeon_drm.h"
 
 struct r600_bo *r600_bo(struct radeon *radeon,
-				  unsigned size, unsigned alignment, unsigned usage)
+			unsigned size, unsigned alignment,
+			unsigned binding, unsigned usage)
 {
 	struct r600_bo *ws_bo = calloc(1, sizeof(struct r600_bo));
 	struct pb_desc desc;
 	struct pb_manager *man;
 
 	desc.alignment = alignment;
-	desc.usage = usage;
+	desc.usage = (PB_USAGE_CPU_READ_WRITE | PB_USAGE_GPU_READ_WRITE);
 	ws_bo->size = size;
 
-	if (usage & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER))
+	if (binding & (PIPE_BIND_CONSTANT_BUFFER | PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER))
 		man = radeon->cman;
 	else
 		man = radeon->kman;
 
+	/* Staging resources particpate in transfers and blits only
+	 * and are used for uploads and downloads from regular
+	 * resources.  We generate them internally for some transfers.
+	 */
+	if (usage == PIPE_USAGE_STAGING)
+                ws_bo->domains = RADEON_GEM_DOMAIN_CPU | RADEON_GEM_DOMAIN_GTT;
+        else
+                ws_bo->domains = (RADEON_GEM_DOMAIN_CPU |
+                                  RADEON_GEM_DOMAIN_GTT |
+                                  RADEON_GEM_DOMAIN_VRAM);
+
+
 	ws_bo->pb = man->create_buffer(man, size, &desc);
 	if (ws_bo->pb == NULL) {
 		free(ws_bo);
@@ -69,6 +83,10 @@ struct r600_bo *r600_bo_handle(struct radeon *radeon,
 	}
 	bo = radeon_bo_pb_get_bo(ws_bo->pb);
 	ws_bo->size = bo->size;
+	ws_bo->domains = (RADEON_GEM_DOMAIN_CPU |
+			  RADEON_GEM_DOMAIN_GTT |
+			  RADEON_GEM_DOMAIN_VRAM);
+
 	pipe_reference_init(&ws_bo->reference, 1);
 
 	radeon_bo_get_tiling_flags(radeon, bo, &ws_bo->tiling_flags,
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 82d5deaad28..8b4521b86cd 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -44,7 +44,7 @@
 int r600_context_init_fence(struct r600_context *ctx)
 {
 	ctx->fence = 1;
-	ctx->fence_bo = r600_bo(ctx->radeon, 4096, 0, 0);
+	ctx->fence_bo = r600_bo(ctx->radeon, 4096, 0, 0, 0);
 	if (ctx->fence_bo == NULL) {
 		return -ENOMEM;
 	}
@@ -787,8 +787,8 @@ void r600_context_bo_reloc(struct r600_context *ctx, u32 *pm4, struct r600_bo *r
 	bo->reloc = &ctx->reloc[ctx->creloc];
 	bo->reloc_id = ctx->creloc * sizeof(struct r600_reloc) / 4;
 	ctx->reloc[ctx->creloc].handle = bo->handle;
-	ctx->reloc[ctx->creloc].read_domain = RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM;
-	ctx->reloc[ctx->creloc].write_domain = RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM;
+	ctx->reloc[ctx->creloc].read_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
+	ctx->reloc[ctx->creloc].write_domain = rbo->domains & (RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM);
 	ctx->reloc[ctx->creloc].flags = 0;
 	radeon_bo_reference(ctx->radeon, &ctx->bo[ctx->creloc], bo);
 	ctx->creloc++;
@@ -1306,7 +1306,12 @@ struct r600_query *r600_context_query_create(struct r600_context *ctx, unsigned
 	query->type = query_type;
 	query->buffer_size = 4096;
 
-	query->buffer = r600_bo(ctx->radeon, query->buffer_size, 1, 0);
+	/* As of GL4, query buffers are normally read by the CPU after
+	 * being written by the gpu, hence staging is probably a good
+	 * usage pattern.
+	 */
+	query->buffer = r600_bo(ctx->radeon, query->buffer_size, 1, 0,
+				PIPE_USAGE_STAGING);
 	if (!query->buffer) {
 		free(query);
 		return NULL;
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index efecfebc2b1..97c582397a7 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -79,6 +79,7 @@ struct r600_bo {
 	unsigned			size;
 	unsigned			tiling_flags;
 	unsigned                        kernel_pitch;
+	unsigned			domains;
 };