Allocate cached ION buffers.
We're doing lots of CPU access of ION buffers and with no caching this
access is pretty slow. By allocating with ION_FLAG_CACHED CPU mapped
pages will be cached and for example a memcpy of a decoded 1080p video
frame is about 3x faster. When writing to a buffer from CPU ("software
rendering") ION will ensure proper cache management on munmap.
For cases not involving CPU access (e.g. VPU directly to GPU) this is
a no-op.
In the Android world this corresponds to gralloc-ing with
GRALLOC_USAGE_SW_READ | GRALLOC_USAGE_SW_WRITE.
Change-Id: I52ca0ac7cc4ebfe3cb8a76aa38e437ac9a04cf02
(cherry picked from commit a77a0ca4a8c7022cf5ef177e41a9dd5a5c95ee54)
diff --git a/gst-libs/gst/allocators/gstionmemory.c b/gst-libs/gst/allocators/gstionmemory.c
index 4b1e1e4..daa246b 100644
--- a/gst-libs/gst/allocators/gstionmemory.c
+++ b/gst-libs/gst/allocators/gstionmemory.c
@@ -41,7 +41,7 @@
#define gst_ion_allocator_parent_class parent_class
#define DEFAULT_HEAP_ID 0
-#define DEFAULT_FLAG 0
+#define DEFAULT_FLAG ION_FLAG_CACHED
enum
{
