[x265] [PATCH 6 of 7] asm: intra_pred_ang4_29 improved by ~45% over SSE4, 157.78c -> 88.14c

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1427889817 -19800
#      Wed Apr 01 17:33:37 2015 +0530
# Node ID f4147dbe79cdd1f9e677bd953aaa03816100d1ad
# Parent  3369c2892450268d0c4d55adcddba65e5be20887
asm: intra_pred_ang4_29 improved by ~45% over SSE4, 157.78c -> 88.14c

diff -r 3369c2892450 -r f4147dbe79cd source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Wed Apr 01 17:16:45 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Wed Apr 01 17:33:37 2015 +0530
@@ -1608,6 +1608,7 @@
         // intra_pred functions
         p.cu[BLOCK_4x4].intra_pred[27] = x265_intra_pred_ang4_27_avx2;
         p.cu[BLOCK_4x4].intra_pred[28] = x265_intra_pred_ang4_28_avx2;
+        p.cu[BLOCK_4x4].intra_pred[29] = x265_intra_pred_ang4_29_avx2;
         p.cu[BLOCK_8x8].intra_pred[3] = x265_intra_pred_ang8_3_avx2;
         p.cu[BLOCK_8x8].intra_pred[33] = x265_intra_pred_ang8_33_avx2;
         p.cu[BLOCK_8x8].intra_pred[4] = x265_intra_pred_ang8_4_avx2;
diff -r 3369c2892450 -r f4147dbe79cd source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h	Wed Apr 01 17:16:45 2015 +0530
+++ b/source/common/x86/intrapred.h	Wed Apr 01 17:33:37 2015 +0530
@@ -176,6 +176,7 @@
 #undef DECL_ANG
 void x265_intra_pred_ang4_27_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang4_28_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
+void x265_intra_pred_ang4_29_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang8_3_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang8_33_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
 void x265_intra_pred_ang8_4_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
diff -r 3369c2892450 -r f4147dbe79cd source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm	Wed Apr 01 17:16:45 2015 +0530
+++ b/source/common/x86/intrapred8.asm	Wed Apr 01 17:33:37 2015 +0530
@@ -512,9 +512,11 @@
 
 ALIGN 32
 intra_pred_shuff_0_4:    times 4 db 0, 1, 1, 2, 2, 3, 3, 4
+intra_pred4_shuff1:      db 0, 1, 1, 2, 2, 3, 3, 4, 0, 1, 1, 2, 2, 3, 3, 4, 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5
 
 c_ang4_mode_27:          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_28:          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_29:          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
 
 ALIGN 32
 ;; (blkSize - 1 - x)
@@ -15507,3 +15509,19 @@
     movd              [r0], xm0
     pextrd            [r0 + r1], xm0, 1
     RET
+
+INIT_YMM avx2
+cglobal intra_pred_ang4_29, 3, 3, 1
+    vbroadcasti128    m0, [r2 + 1]
+    pshufb            m0, [intra_pred4_shuff1]
+    pmaddubsw         m0, [c_ang4_mode_29]
+    pmulhrsw          m0, [pw_1024]
+    packuswb          m0, m0
+
+    movd              [r0], xm0
+    pextrd            [r0 + r1], xm0, 1
+    vextracti128      xm0, m0, 1
+    lea               r0, [r0 + 2 * r1]
+    movd              [r0], xm0
+    pextrd            [r0 + r1], xm0, 1
+    RET