summaryrefslogtreecommitdiffstats
path: root/libjpegtwrp/README-VeNum
diff options
context:
space:
mode:
authorDees_Troy <dees_troy@teamw.in>2012-09-05 21:24:24 +0200
committerDees_Troy <dees_troy@teamw.in>2012-09-05 21:24:31 +0200
commit51a0e82eb29a6dfc79f93479883383fbdbf8bcc2 (patch)
tree52fc18206eb0feba9f50dc3b0ede9fdc5e40f35e /libjpegtwrp/README-VeNum
parentInitial stub of partitions.hpp (diff)
downloadandroid_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.tar
android_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.tar.gz
android_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.tar.bz2
android_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.tar.lz
android_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.tar.xz
android_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.tar.zst
android_bootable_recovery-51a0e82eb29a6dfc79f93479883383fbdbf8bcc2.zip
Diffstat (limited to 'libjpegtwrp/README-VeNum')
-rw-r--r--libjpegtwrp/README-VeNum20
1 files changed, 20 insertions, 0 deletions
diff --git a/libjpegtwrp/README-VeNum b/libjpegtwrp/README-VeNum
new file mode 100644
index 000000000..ee51449c7
--- /dev/null
+++ b/libjpegtwrp/README-VeNum
@@ -0,0 +1,20 @@
+README-VeNum
+Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum
+=================================================================
+
+Consumers are increasingly browsing Web based photo galleries incorporating
+high-resolution JPEG images with their mobile devices. These images are
+decoded by the mobile device, and typically also scaled down to fit the user's
+zoom selection on the device's browser.
+
+In order to improve both decode and download times, Qualcomm Innovation Center
+has optimized the JPEG library found on many common OS platforms such as Web
+OS, Android, and Chrome OS. Our team re-implemented several routines to utilize the
+DSP-like SIMD capabilities of the ARM NEON instruction set. These were then
+tuned and tested on Qualcomm's Snapdragon platform which implements the VeNum
+implementation of these same instructions.
+
+The specific areas of focus cover VeNum/NEON acceleration of Inverse Discrete
+Cosine Transform (iDCT) for 8x8, 4x4, 2x2, and 1x1 block sizes and YUV to RGB
+color space conversion. This resulted in a range of 18-32% improvement in JPEG
+decode and downscale times for images greater than 2Mpixels.