[x265] [PATCH 2 of 6] asm: intra_pred_ang4_7 improved by ~42% over SSE4

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1428300062 -19800
#      Mon Apr 06 11:31:02 2015 +0530
# Node ID a69d1b623145e68a11e6e2eba3802096a2f18fcc
# Parent  05f6ee179cbd45995c7bbea43ad2cf8a824928d7
asm: intra_pred_ang4_7 improved by ~42% over SSE4

AVX2:
intra_ang_4x4[ 7]       10.24x   98.65           1009.92

SSE4:
intra_ang_4x4[ 7]       6.25x    169.98          1061.89

diff -r 05f6ee179cbd -r a69d1b623145 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Apr 06 10:58:30 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Apr 06 11:31:02 2015 +0530
@@ -1701,6 +1701,7 @@
         p.weight_sp = x265_weight_sp_avx2;
 
         // intra_pred functions
+        p.cu[BLOCK_4x4].intra_pred[7] = x265_intra_pred_ang4_7_avx2;
         p.cu[BLOCK_4x4].intra_pred[8] = x265_intra_pred_ang4_8_avx2;
         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;
diff -r 05f6ee179cbd -r a69d1b623145 source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h	Mon Apr 06 10:58:30 2015 +0530
+++ b/source/common/x86/intrapred.h	Mon Apr 06 11:31:02 2015 +0530
@@ -187,6 +187,7 @@
 DECL_ANG(32, 33, sse4);
 
 #undef DECL_ANG
+void x265_intra_pred_ang4_7_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang4_8_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 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);
diff -r 05f6ee179cbd -r a69d1b623145 source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Mon Apr 06 10:58:30 2015 +0530
+++ b/source/common/x86/intrapred8.asm	Mon Apr 06 11:31:02 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_shuff7:      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, 10, 11, 11, 12, 12, 13, 13, 14
 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
@@ -535,6 +536,7 @@
 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_7:           db 23, 9, 23, 9, 23, 9, 23, 9, 14, 18, 14, 18, 14, 18, 14, 18, 5, 27, 5, 27, 5, 27, 5, 27, 28, 4, 28, 4, 28, 4, 28, 4
 c_ang4_mode_8:           db 27, 5, 27, 5, 27, 5, 27, 5, 22, 10, 22, 10, 22, 10, 22, 10, 17, 15, 17, 15, 17, 15, 17, 15, 12, 20, 12, 20, 12, 20, 12, 20
 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
@@ -16092,6 +16094,18 @@
     INTRA_PRED_STORE_4x4
     RET
 
+
+INIT_YMM avx2
+cglobal intra_pred_ang4_7, 3, 3, 1
+    vbroadcasti128    m0, [r2]
+    pshufb            m0, [intra_pred4_shuff7]
+    pmaddubsw         m0, [c_ang4_mode_7]
+    pmulhrsw          m0, [pw_1024]
+    packuswb          m0, m0
+
+    INTRA_PRED_TRANS_STORE_4x4
+    RET
+
 INIT_YMM avx2
 cglobal intra_pred_ang4_8, 3, 3, 1
     vbroadcasti128    m0, [r2]