We try to avoid writing the relocations through the uncached GTT, if the
buffer is currently in the CPU write domain and so will be flushed out to
main memory afterwards anyway. Also on SandyBridge we can safely write
to the pages in cacheable memory, so long as the buffer is LLC mapped.
In either of these caches, we therefore do not need to force the
reallocation of the buffer into the mappable region of the GTT, reducing
the aperture pressure.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/i915_gem_execbuffer.c | 24 +++++++++++++++++-------
1 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index f71f26e..9b8d543 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -265,6 +265,12 @@ eb_destroy(struct eb_objects *eb)
kfree(eb);
}
+static inline int use_cpu_reloc(struct drm_i915_gem_object *obj)
+{
+ return (obj->base.write_domain == I915_GEM_DOMAIN_CPU ||
+ obj->cache_level != I915_CACHE_NONE);
+}
+
static int
i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
struct eb_objects *eb,
@@ -351,7 +357,7 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
}
reloc->delta += target_offset;
- if (obj->base.write_domain == I915_GEM_DOMAIN_CPU) {
+ if (use_cpu_reloc(obj)) {
uint32_t page_offset = reloc->offset & ~PAGE_MASK;
char *vaddr;
@@ -463,6 +469,13 @@ i915_gem_execbuffer_relocate(struct drm_device *dev,
#define __EXEC_OBJECT_HAS_FENCE (1<<31)
static int
+need_reloc_mappable(struct drm_i915_gem_object *obj)
+{
+ struct drm_i915_gem_exec_object2 *entry = obj->exec_entry;
+ return entry->relocation_count && !use_cpu_reloc(obj);
+}
+
+static int
pin_and_fence_object(struct drm_i915_gem_object *obj,
struct intel_ring_buffer *ring)
{
@@ -475,8 +488,7 @@ pin_and_fence_object(struct drm_i915_gem_object *obj,
has_fenced_gpu_access &&
entry->flags & EXEC_OBJECT_NEEDS_FENCE &&
obj->tiling_mode != I915_TILING_NONE;
- need_mappable =
- entry->relocation_count ? true : need_fence;
+ need_mappable = need_fence || need_reloc_mappable(obj);
ret = i915_gem_object_pin(obj, entry->alignment, need_mappable);
if (ret)
@@ -532,8 +544,7 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,
has_fenced_gpu_access &&
entry->flags & EXEC_OBJECT_NEEDS_FENCE &&
obj->tiling_mode != I915_TILING_NONE;
- need_mappable =
- entry->relocation_count ? true : need_fence;
+ need_mappable = need_fence || need_reloc_mappable(obj);
if (need_mappable)
list_move(&obj->exec_list, &ordered_objects);
@@ -573,8 +584,7 @@ i915_gem_execbuffer_reserve(struct intel_ring_buffer *ring,
has_fenced_gpu_access &&
entry->flags & EXEC_OBJECT_NEEDS_FENCE &&
obj->tiling_mode != I915_TILING_NONE;
- need_mappable =
- entry->relocation_count ? true : need_fence;
+ need_mappable = need_fence || need_reloc_mappable(obj);
if ((entry->alignment && obj->gtt_offset & (entry->alignment - 1)) ||
(need_mappable && !obj->map_and_fenceable))
--
1.7.7.3
