[x265] [PATCH] asm: 16bpp asm code for intra_pred_ang4_5

[yuvaraj at multicorewareinc.com]

# HG changeset patch
# User Yuvaraj Venkatesh <yuvaraj at multicorewareinc.com>
# Date 1386586389 -19800
#      Mon Dec 09 16:23:09 2013 +0530
# Node ID 9fb6af500fb708b1272da679625e75212fb2ca67
# Parent  721dbb9393998858a29be037afab6f73c882eb1e
asm: 16bpp asm code for intra_pred_ang4_5

diff -r 721dbb939399 -r 9fb6af500fb7 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Dec 09 15:58:51 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Dec 09 16:23:09 2013 +0530
@@ -688,6 +688,8 @@
 
         SETUP_INTRA_ANG4(3, 3, sse4);
         SETUP_INTRA_ANG4(4, 4, sse4);
+        SETUP_INTRA_ANG4(5, 5, sse4);
+        SETUP_INTRA_ANG4(31, 5, sse4);
         SETUP_INTRA_ANG4(32, 4, sse4);
         SETUP_INTRA_ANG4(33, 3, sse4);
     }
diff -r 721dbb939399 -r 9fb6af500fb7 source/common/x86/intrapred16.asm
--- a/source/common/x86/intrapred16.asm	Mon Dec 09 15:58:51 2013 +0530
+++ b/source/common/x86/intrapred16.asm	Mon Dec 09 16:23:09 2013 +0530
@@ -502,3 +502,22 @@
     mova        m6, [r3 + 13 * 16]  ; [31]
     mova        m7, [r3 +  2 * 16]  ; [20]
     jmp         mangle(private_prefix %+ _ %+ intra_pred_ang4_3 %+ SUFFIX %+ .do_filter4x4)
+
+cglobal intra_pred_ang4_5, 3,4,8
+    cmp         r4m, byte 31
+    cmove       r2, r3mp
+    lea         r3, [ang_table + 10 * 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]
+    palignr     m6, m0, 4       ; [x x 8 7 6 5 4 3]
+    punpcklwd   m3, m1, m6      ; [6 5 5 4 4 3 3 2]
+    mova        m4, m3
+    palignr     m7, m0, 6       ; [x x x 8 7 6 5 4]
+    punpcklwd   m5, m6, m7      ; [7 6 6 5 5 4 4 3]
+
+    mova        m0, [r3 +  7 * 16]  ; [17]
+    mova        m1, [r3 -  8 * 16]  ; [ 2]
+    mova        m6, [r3 +  9 * 16]  ; [19]
+    mova        m7, [r3 -  6 * 16]  ; [ 4]
+    jmp         mangle(private_prefix %+ _ %+ intra_pred_ang4_3 %+ SUFFIX %+ .do_filter4x4)