This is new algorithm for intra_ang16x16.
1075 cycles - current AVX2 asm
827 cycles - new AVX2 asm (improved 23% over current avx2 asm)
On Thu, Aug 6, 2015 at 10:41 AM, Deepthi Nandakumar
deep...@multicorewareinc.com wrote:
Please be sure to mention what is the baseline - for instance, what is
ok. I will change the commit message and resend the patch
2015-08-06 11:19 GMT+05:30 Dnyaneshwar Gorade
dnyanesh...@multicorewareinc.com:
This is new algorithm for intra_ang16x16.
1075 cycles - current AVX2 asm
827 cycles - new AVX2 asm (improved 23% over current avx2 asm)
On Thu, Aug 6,
Please be sure to mention what is the baseline - for instance, what is 1075
cycles?
On Wed, Aug 5, 2015 at 6:06 PM, raj...@multicorewareinc.com wrote:
# HG changeset patch
# User Rajesh Paulrajraj...@multicorewareinc.com
# Date 1438766294 -19800
# Wed Aug 05 14:48:14 2015 +0530
# Node
# HG changeset patch
# User Rajesh Paulrajraj...@multicorewareinc.com
# Date 1438766294 -19800
# Wed Aug 05 14:48:14 2015 +0530
# Node ID 4a71c4261e5a7955a7ecdda61db1f20744254b0e
# Parent 3fa7f6838098854de79d3800b2d775dabaf45705
asm: avx2 code for intra_ang_16 modes 3 33
# HG changeset patch
# User Rajesh Paulrajraj...@multicorewareinc.com
# Date 1438162042 -19800
# Wed Jul 29 14:57:22 2015 +0530
# Node ID 90bf2117cff7e5727c43e65f926870e8874b9ed2
# Parent e08a245054434090ccb1fc1b985f955a66711157
asm: avx2 code for intra_ang_16 modes 3 33