[Intel-gfx] [PATCH 2/2] drm/i915: combine all small integers into one single bitfield

Daniel Vetter daniel.vetter at ffwll.ch
Wed Apr 28 11:02:32 CEST 2010


This saves a whooping 8 dwords. Zero functional changes. Because
some of the refcounts are rather tightly calculated, I've put
BUG_ONs in the code to check for overflows.

Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
---
 drivers/gpu/drm/i915/i915_drv.h |   87 ++++++++++++++++++++++++--------------
 drivers/gpu/drm/i915/i915_gem.c |    5 ++
 2 files changed, 60 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index e8783fd..070d557 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -663,19 +663,70 @@ struct drm_i915_gem_object {
 	 * (has pending rendering), and is not set if it's on inactive (ready
 	 * to be unbound).
 	 */
-	int active;
+	unsigned int active : 1;
 
 	/**
 	 * This is set if the object has been written to since last bound
 	 * to the GTT
 	 */
-	int dirty;
+	unsigned int dirty : 1;
+
+	/**
+	 * Fence register bits (if any) for this object.  Will be set
+	 * as needed when mapped into the GTT.
+	 * Protected by dev->struct_mutex.
+	 *
+	 * Size: 4 bits for 16 fences + sign (for FENCE_REG_NONE)
+	 */
+	int fence_reg : 5;
+
+	/**
+	 * Tiling changes can happen asynchronous to avoid gpu stalls.
+	 * This will be set if the current fencing may be invalid.
+	 */
+	unsigned int fence_invalid : 1;
+
+	/**
+	 * Used for checking the object doesn't appear more than once
+	 * in an execbuffer object list.
+	 */
+	unsigned int in_execbuffer : 1;
+
+	/**
+	 * Advice: are the backing pages purgeable?
+	 */
+	unsigned int madv : 2;
+
+	/**
+	 * Refcount for the pages array. With the current locking scheme, there
+	 * are at most two concurrent users: Binding a bo to the gtt and
+	 * pwrite/pread using physical addresses. So two bits for a maximum
+	 * of two users are enough.
+	 */
+	unsigned int pages_refcount : 2;
+#define DRM_I915_GEM_OBJECT_MAX_PAGES_REFCOUNT 0x3
+
+	/**
+	 * Current tiling mode for the object.
+	 */
+	unsigned int tiling_mode : 2;
+
+	/** How many users have pinned this object in GTT space. The following
+	 * users can each hold at most one reference: pwrite/pread, pin_ioctl
+	 * (via user_pin_count), execbuffer (objects are not allowed multiple
+	 * times for the same batchbuffer), and the framebuffer code. When
+	 * switching/pageflipping, the framebuffer code has at most two buffers
+	 * pinned per crtc.
+	 *
+	 * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
+	 * bits with absolutely no headroom. So use 4 bits. */
+	int pin_count : 4;
+#define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
 
 	/** AGP memory structure for our GTT binding. */
 	DRM_AGP_MEM *agp_mem;
 
 	struct page **pages;
-	int pages_refcount;
 
 	/**
 	 * Current offset of the object in GTT space.
@@ -689,27 +740,10 @@ struct drm_i915_gem_object {
 	 */
 	uint64_t mmap_offset;
 
-	/**
-	 * Fence register bits (if any) for this object.  Will be set
-	 * as needed when mapped into the GTT.
-	 * Protected by dev->struct_mutex.
-	 */
-	int fence_reg;
-
-	/**
-	 * Tiling changes can happen asynchronous to avoid gpu stalls.
-	 * This will be set if the current fencing may be invalid.
-	 */
-	int fence_invalid;
-
-	/** How many users have pinned this object in GTT space */
-	int pin_count;
-
 	/** Breadcrumb of last rendering to the buffer. */
 	uint32_t last_rendering_seqno;
 
-	/** Current tiling mode for the object. */
-	uint32_t tiling_mode;
+	/** Current tiling stride for the object, if it's tiled. */
 	uint32_t stride;
 
 	/** Record of address bit 17 of each page at last unbind. */
@@ -732,17 +766,6 @@ struct drm_i915_gem_object {
 	struct drm_i915_gem_phys_object *phys_obj;
 
 	/**
-	 * Used for checking the object doesn't appear more than once
-	 * in an execbuffer object list.
-	 */
-	int in_execbuffer;
-
-	/**
-	 * Advice: are the backing pages purgeable?
-	 */
-	int madv;
-
-	/**
 	 * Number of crtcs where this object is currently the fb, but
 	 * will be page flipped away on the next vblank.  When it
 	 * reaches 0, dev_priv->pending_flip_queue will be woken up.
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index bd6b73e..99c283f 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2256,6 +2256,9 @@ i915_gem_object_get_pages(struct drm_gem_object *obj,
 	struct inode *inode;
 	struct page *page;
 
+	BUG_ON(obj_priv->pages_refcount
+			== DRM_I915_GEM_OBJECT_MAX_PAGES_REFCOUNT);
+
 	if (obj_priv->pages_refcount++ != 0)
 		return 0;
 
@@ -4235,6 +4238,8 @@ i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment)
 	struct drm_i915_gem_object *obj_priv = to_intel_bo(obj);
 	int ret;
 
+	BUG_ON(obj_priv->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT);
+
 	i915_verify_inactive(dev, __FILE__, __LINE__);
 	if (obj_priv->gtt_space == NULL) {
 		ret = i915_gem_object_bind_to_gtt(obj, alignment);
-- 
1.7.0.4




More information about the Intel-gfx mailing list