[x265] [PATCH 8 of 9] asm: intra_pred_ang4_9 improved by ~35% over SSE4

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1428053156 -19800
#      Fri Apr 03 14:55:56 2015 +0530
# Node ID 5f6d4c27a01c1b8cacb5cfcc7373ae7fb6dd82f4
# Parent  49eab7332eeba68cecdd55b11b95355cfe2b9557
asm: intra_pred_ang4_9 improved by ~35% over SSE4

AVX2:
intra_ang_4x4[ 9]       10.27x   104.54          1073.82

SSE4:
intra_ang_4x4[ 9]       6.48x    162.27          1051.73

diff -r 49eab7332eeb -r 5f6d4c27a01c source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Fri Apr 03 14:41:25 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Fri Apr 03 14:55:56 2015 +0530
@@ -1633,6 +1633,7 @@
         p.weight_sp = x265_weight_sp_avx2;
 
         // intra_pred functions
+        p.cu[BLOCK_4x4].intra_pred[9] = x265_intra_pred_ang4_9_avx2;
         p.cu[BLOCK_4x4].intra_pred[11] = x265_intra_pred_ang4_11_avx2;
         p.cu[BLOCK_4x4].intra_pred[12] = x265_intra_pred_ang4_12_avx2;
         p.cu[BLOCK_4x4].intra_pred[13] = x265_intra_pred_ang4_13_avx2;
diff -r 49eab7332eeb -r 5f6d4c27a01c source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h	Fri Apr 03 14:41:25 2015 +0530
+++ b/source/common/x86/intrapred.h	Fri Apr 03 14:55:56 2015 +0530
@@ -184,6 +184,7 @@
 DECL_ANG(32, 33, sse4);
 
 #undef DECL_ANG
+void x265_intra_pred_ang4_9_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang4_11_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang4_12_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang4_13_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
diff -r 49eab7332eeb -r 5f6d4c27a01c source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Fri Apr 03 14:41:25 2015 +0530
+++ b/source/common/x86/intrapred8.asm	Fri Apr 03 14:55:56 2015 +0530
@@ -515,6 +515,7 @@
 intra_pred4_shuff2:      db 0, 1, 1, 2, 2, 3, 3, 4, 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5, 1, 2, 2, 3, 3, 4, 4, 5
 intra_pred4_shuff31:     db 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5, 1, 2, 2, 3, 3, 4, 4, 5, 2, 3, 3, 4, 4, 5, 5, 6
 intra_pred4_shuff33:     db 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5, 2, 3, 3, 4, 4, 5, 5, 6, 3, 4, 4, 5, 5, 6, 6, 7
+intra_pred4_shuff9:      db 9, 10, 10, 11, 11, 12, 12, 13, 9, 10, 10, 11, 11, 12, 12, 13, 9, 10, 10, 11, 11, 12, 12, 13, 9, 10, 10, 11, 11, 12, 12, 13
 intra_pred4_shuff12:     db 0, 9, 9, 10, 10, 11, 11, 12, 0, 9, 9, 10, 10, 11, 11, 12, 0, 9, 9, 10, 10, 11, 11, 12,0, 9, 9, 10, 10, 11, 11, 12
 intra_pred4_shuff13:     db 0, 9, 9, 10, 10, 11, 11, 12, 0, 9, 9, 10, 10, 11, 11, 12, 0, 9, 9, 10, 10, 11, 11, 12, 4, 0, 0, 9, 9, 10, 10, 11
 intra_pred4_shuff14:     db 0, 9, 9, 10, 10, 11, 11, 12, 0, 9, 9, 10, 10, 11, 11, 12, 2, 0, 0, 9, 9, 10, 10, 11, 2, 0, 0, 9, 9, 10, 10, 11
@@ -534,6 +535,7 @@
 c_ang4_mode_31:          db 15, 17, 15, 17, 15, 17, 15, 17, 30, 2, 30, 2, 30, 2, 30, 2, 13, 19, 13, 19, 13, 19, 13, 19, 28, 4, 28, 4, 28, 4, 28, 4
 c_ang4_mode_32:          db 11, 21, 11, 21, 11, 21, 11, 21, 22, 10, 22, 10, 22, 10, 22, 10, 1, 31, 1, 31, 1, 31, 1, 31, 12, 20, 12, 20, 12, 20, 12, 20
 c_ang4_mode_33:          db 6, 26, 6, 26, 6, 26, 6, 26, 12, 20, 12, 20, 12, 20, 12, 20, 18, 14, 18, 14, 18, 14, 18, 14, 24, 8, 24, 8, 24, 8, 24, 8
+c_ang4_mode_9:           db 30, 2, 30, 2, 30, 2, 30, 2, 28, 4, 28, 4, 28, 4, 28, 4, 26, 6, 26, 6, 26, 6, 26, 6, 24, 8, 24, 8, 24, 8, 24, 8
 c_ang4_mode_11:          db 2, 30, 2, 30, 2, 30, 2, 30, 4, 28, 4, 28, 4, 28, 4, 28, 6, 26, 6, 26, 6, 26, 6, 26, 8, 24, 8, 24, 8, 24, 8, 24
 c_ang4_mode_12:          db 5, 27, 5, 27, 5, 27, 5, 27, 10, 22, 10, 22, 10, 22, 10, 22, 15, 17, 15, 17, 15, 17, 15, 17, 20, 12, 20, 12, 20, 12, 20, 12
 c_ang4_mode_13:          db 9, 23, 9, 23, 9, 23, 9, 23, 18, 14, 18, 14, 18, 14, 18, 14, 27, 5, 27, 5, 27, 5, 27, 5, 4, 28, 4, 28, 4, 28, 4, 28
@@ -15873,6 +15875,26 @@
     INTRA_PRED_STORE_4x4
     RET
 
+
+INIT_YMM avx2
+cglobal intra_pred_ang4_9, 3, 3, 1
+    vbroadcasti128    m0, [r2]
+    pshufb            m0, [intra_pred4_shuff9]
+    pmaddubsw         m0, [c_ang4_mode_9]
+    pmulhrsw          m0, [pw_1024]
+    packuswb          m0, m0
+
+    vpermq            m0, m0, 00001000b
+    pshufb            m0, [c_trans_4x4]
+
+    ;store
+    movd              [r0], xm0
+    pextrd            [r0 + r1], xm0, 1
+    lea               r0, [r0 + 2 * r1]
+    pextrd            [r0], xm0, 2
+    pextrd            [r0 + r1], xm0, 3
+    RET
+
 INIT_YMM avx2
 cglobal intra_pred_ang4_11, 3, 3, 1
     vbroadcasti128    m0, [r2]