commit ffe7ae7cb178ccd8ce91b4b51ec4fb3536d1a702
author Remya <remya.prakasan@ittiam.com> Mon Sep 24 19:59:00 2018 +0530
committer Yunqing Wang <yunqingwang@google.com> Mon Sep 24 19:22:48 2018 +0000
tree de7c5eef6fc33bdb4f74ff27218777869e2be3bb
parent c199fb47ed1d301461f05e3ed1b1050241a7b9c8

Optimize highbd 4x16 and 16x4 inv_txfm

Enabled sse4_1 optimizations for tx_sizes 4x16 and 16x4.

Module level gains:
Tx_size    Gain w.r.t. C
4x16         6.01x
16x4         7.36x

When tested for 20 frames of crowd_run_360p_10 at 1 mbps
for speed=1 preset, observed ~0.5% reduction in encoder time.

Change-Id: I0c5d6530c666150be0de062a7084c7a2bf61410f

av1/common/av1_rtcd_defs.pl

diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index a6d5138..033eeb7 100755
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -135,6 +135,10 @@
 specialize qw/av1_highbd_inv_txfm_add_4x8 sse4_1/;
 add_proto qw/void av1_highbd_inv_txfm_add_8x4/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
 specialize qw/av1_highbd_inv_txfm_add_8x4 sse4_1/;
+add_proto qw/void av1_highbd_inv_txfm_add_4x16/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_highbd_inv_txfm_add_4x16 sse4_1/;
+add_proto qw/void av1_highbd_inv_txfm_add_16x4/,  "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
+specialize qw/av1_highbd_inv_txfm_add_16x4 sse4_1/;
 
 add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
 add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";