Thanks, can you merge with the latest tip?
On Fri, Jun 27, 2014 at 12:13 AM, Aarthi Priya Thirumalai
aar...@multicorewareinc.com wrote:
# HG changeset patch
# User Aarthi Thirumalaiaar...@multicorewareinc.com
# Date 1403808050 -19800
# Fri Jun 27 00:10:50 2014 +0530
# Node ID
# HG changeset patch
# User Sumalatha Polureddysumala...@multicorewareinc.com
# Date 1403858781 -19800
# Node ID a789870889fcc9a31deff7fc6961d143b0db86c1
# Parent 1b669c33ff3a8d8f6c9bd1e18979c009baed2433
psyrd: fix for inconsistent output
maximum buffer size for zeropel is
# HG changeset patch
# User Satoshi Nakagawa nakagawa...@oki.com
# Date 1403877807 -32400
# Fri Jun 27 23:03:27 2014 +0900
# Node ID 77f443fe169ca23969df5d5ee6968543bfa5e794
# Parent 32aa6cc3cf4d108ac92f5d29258b2c38ca888d29
fix emms: move selectLambda() into xRateDistOptQuant() and issue
On 6/27/2014 4:05 PM, chen wrote:
I can't understand what's your means. could you tell me more?
I use some SSSE3 instruction and process 16 pixels every loop.
I meant keep both sse2 and ssse3 variants. Not sure if x86inc.asm macros
help with this or not.
- Derek
On 6/27/2014 6:08 PM, chen wrote:
I use ssse3 instruction PSHUFB to replace 3 SSE2 instructions, the x86inc
macro can't handle it.
After patch, this function is faster ~20% and codeCoeffNxN ~7% speedup, so I
don't worry about old CPU's performance.
I guess SSSE3 is very prevalent nowadays