[x265] [PATCH 4 of 6] asm: intra_pred_ang4_5 improved by ~41% over SSE4

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1428302934 -19800
#      Mon Apr 06 12:18:54 2015 +0530
# Node ID 6a3e3ace9c9d46eed867a78568bebc56c8310791
# Parent  7259464babb177fee1f1d9418017dc143e9a06fe
asm: intra_pred_ang4_5 improved by ~41% over SSE4

AVX2:
intra_ang_4x4[ 5]       9.56x    103.43          989.01

SSE4:
intra_ang_4x4[ 5]       5.99x    176.06          1055.48

diff -r 7259464babb1 -r 6a3e3ace9c9d source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Apr 06 12:05:52 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Apr 06 12:18:54 2015 +0530
@@ -1701,6 +1701,7 @@
         p.weight_sp = x265_weight_sp_avx2;
 
         // intra_pred functions
+        p.cu[BLOCK_4x4].intra_pred[5] = x265_intra_pred_ang4_5_avx2;
         p.cu[BLOCK_4x4].intra_pred[6] = x265_intra_pred_ang4_6_avx2;
         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;
diff -r 7259464babb1 -r 6a3e3ace9c9d source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h	Mon Apr 06 12:05:52 2015 +0530
+++ b/source/common/x86/intrapred.h	Mon Apr 06 12:18:54 2015 +0530
@@ -187,6 +187,7 @@
 DECL_ANG(32, 33, sse4);
 
 #undef DECL_ANG
+void x265_intra_pred_ang4_5_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang4_6_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 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);
diff -r 7259464babb1 -r 6a3e3ace9c9d source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Mon Apr 06 12:05:52 2015 +0530
+++ b/source/common/x86/intrapred8.asm	Mon Apr 06 12:18:54 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_shuff5:      db 9, 10, 10, 11, 11, 12, 12, 13, 10, 11, 11, 12, 12, 13, 13, 14, 10, 11, 11, 12, 12, 13, 13, 14, 11, 12, 12, 13, 13, 14, 14, 15
 intra_pred4_shuff6:      db 9, 10, 10, 11, 11, 12, 12, 13, 9, 10, 10, 11, 11, 12, 12, 13, 10, 11, 11, 12, 12, 13, 13, 14, 10, 11, 11, 12, 12, 13, 13, 14
 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
@@ -537,6 +538,7 @@
 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_5:           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_6:           db 19, 13, 19, 13, 19, 13, 19, 13, 6, 26, 6, 26, 6, 26, 6, 26, 25, 7, 25, 7, 25, 7, 25, 7, 12, 20, 12, 20, 12, 20, 12, 20
 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
@@ -16096,6 +16098,18 @@
     INTRA_PRED_STORE_4x4
     RET
 
+
+INIT_YMM avx2
+cglobal intra_pred_ang4_5, 3, 3, 1
+    vbroadcasti128    m0, [r2]
+    pshufb            m0, [intra_pred4_shuff5]
+    pmaddubsw         m0, [c_ang4_mode_5]
+    pmulhrsw          m0, [pw_1024]
+    packuswb          m0, m0
+
+    INTRA_PRED_TRANS_STORE_4x4
+    RET
+
 INIT_YMM avx2
 cglobal intra_pred_ang4_6, 3, 3, 1
     vbroadcasti128    m0, [r2]