[x265] [PATCH] asm: 16bpp asm code for intra_pred_ang4_7

[yuvaraj at multicorewareinc.com]

# HG changeset patch
# User Yuvaraj Venkatesh <yuvaraj at multicorewareinc.com>
# Date 1386589080 -19800
#      Mon Dec 09 17:08:00 2013 +0530
# Node ID 9ceeba76f4714972bd93bda3d1513e8bb68a2384
# Parent  5acffa8a6fe00074bd0d53d29c6fce30dff50743
asm: 16bpp asm code for intra_pred_ang4_7

diff -r 5acffa8a6fe0 -r 9ceeba76f471 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Dec 09 16:40:23 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Dec 09 17:08:00 2013 +0530
@@ -690,6 +690,8 @@
         SETUP_INTRA_ANG4(4, 4, sse4);
         SETUP_INTRA_ANG4(5, 5, sse4);
         SETUP_INTRA_ANG4(6, 6, sse4);
+        SETUP_INTRA_ANG4(7, 7, sse4);
+        SETUP_INTRA_ANG4(29, 7, sse4);
         SETUP_INTRA_ANG4(30, 6, sse4);
         SETUP_INTRA_ANG4(31, 5, sse4);
         SETUP_INTRA_ANG4(32, 4, sse4);
diff -r 5acffa8a6fe0 -r 9ceeba76f471 source/common/x86/intrapred16.asm
--- a/source/common/x86/intrapred16.asm	Mon Dec 09 16:40:23 2013 +0530
+++ b/source/common/x86/intrapred16.asm	Mon Dec 09 17:08:00 2013 +0530
@@ -539,3 +539,21 @@
     mova        m6, [r3 - 12 * 16]  ; [ 7]
     mova        m7, [r3 +  1 * 16]  ; [20]
     jmp         mangle(private_prefix %+ _ %+ intra_pred_ang4_3 %+ SUFFIX %+ .do_filter4x4)
+
+cglobal intra_pred_ang4_7, 3,4,8
+    cmp         r4m, byte 29
+    cmove       r2, r3mp
+    lea         r3, [ang_table + 20 * 16]
+    movu        m0, [r2 + 2]    ; [8 7 6 5 4 3 2 1]
+    palignr     m1, m0, 2       ; [x 8 7 6 5 4 3 2]
+    punpcklwd   m2, m0, m1      ; [5 4 4 3 3 2 2 1]
+    mova        m3, m2
+    mova        m4, m2
+    palignr     m6, m0, 4       ; [x x 8 7 6 5 4 3]
+    punpcklwd   m5, m1, m6      ; [6 5 5 4 4 3 3 2]
+
+    mova        m0, [r3 - 11 * 16]  ; [ 9]
+    mova        m1, [r3 -  2 * 16]  ; [18]
+    mova        m6, [r3 +  7 * 16]  ; [27]
+    mova        m7, [r3 - 16 * 16]  ; [ 4]
+    jmp         mangle(private_prefix %+ _ %+ intra_pred_ang4_3 %+ SUFFIX %+ .do_filter4x4)