[x265] [PATCH] asm: intrapred dc16 sse2 high bit

dtyx265 at gmail.com dtyx265 at gmail.com
Mon Mar 2 23:48:38 CET 2015


# HG changeset patch
# User David T Yuen <dtyx265 at gmail.com>
# Date 1425336406 28800
# Node ID 4641827f98c935603f608425de8c76785aef1114
# Parent  30df8e0058c7ca1fe1a59a7e8bce344071348641
asm: intrapred dc16 sse2 high bit

This replaces c code for systems using ssse3 to sse2 processors
The code is backported from intrapred dc16 sse4 high bit

./test/TestBench --testbench intrapred | grep 16x16
intra_dc_16x16[f=0]	2.53x 	 1030.01  	 2607.97
intra_dc_16x16[f=1]	2.57x 	 1507.50  	 3877.95

diff -r 30df8e0058c7 -r 4641827f98c9 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Feb 26 21:03:33 2015 -0800
+++ b/source/common/x86/asm-primitives.cpp	Mon Mar 02 14:46:46 2015 -0800
@@ -870,6 +870,7 @@
 
         p.cu[BLOCK_4x4].intra_pred[DC_IDX] = x265_intra_pred_dc4_sse2;
         p.cu[BLOCK_8x8].intra_pred[DC_IDX] = x265_intra_pred_dc8_sse2;
+        p.cu[BLOCK_16x16].intra_pred[DC_IDX] = x265_intra_pred_dc16_sse2;
 
         p.cu[BLOCK_4x4].sse_ss = x265_pixel_ssd_ss_4x4_mmx2;
         ALL_LUMA_CU(sse_ss, pixel_ssd_ss, sse2);
diff -r 30df8e0058c7 -r 4641827f98c9 source/common/x86/intrapred16.asm
--- a/source/common/x86/intrapred16.asm	Thu Feb 26 21:03:33 2015 -0800
+++ b/source/common/x86/intrapred16.asm	Mon Mar 02 14:46:46 2015 -0800
@@ -246,6 +246,146 @@
 .end:
     RET
 
+;-------------------------------------------------------------------------------------------------------
+; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel* left, pixel* above, int dirMode, int filter)
+;-------------------------------------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal intra_pred_dc16, 5, 10, 4
+    lea             r3,                  [r2 + 66]
+    add             r2,                  2
+    add             r1,                  r1
+    movu            m0,                  [r3]
+    movu            m1,                  [r3 + 16]
+    movu            m2,                  [r2]
+    movu            m3,                  [r2 + 16]
+
+    paddw           m0,                  m1
+    paddw           m2,                  m3
+    paddw           m0,                  m2
+    movhlps         m1,                  m0
+    paddw           m0,                  m1
+    pshuflw         m1,                  m0, 0x6E
+    paddw           m0,                  m1
+    pmaddwd         m0,                  [pw_1]
+
+    paddw           m0,                  [pw_16]
+    psraw           m0,                  5
+    movd            r5d,                 m0
+    pshuflw         m0,                  m0, 0 ; m0 = word [dc_val ...]
+    pshufd          m0,                  m0, 0
+
+    test            r4d,                 r4d
+
+    ; store DC 16x16
+    lea             r6,            [r1 + r1 * 2]        ;index 3
+    lea             r7,            [r1 + r1 * 4]        ;index 5
+    lea             r8,            [r6 + r1 * 4]        ;index 7
+    lea             r9,            [r0 + r8]            ;base + 7
+    movu            [r0],                m0
+    movu            [r0 + 16],           m0
+    movu            [r0 + r1],           m0
+    movu            [r0 + 16 + r1],      m0
+    movu            [r0 + r1 * 2],       m0
+    movu            [r0 + r1 * 2 + 16],  m0
+    movu            [r0 + r6],           m0
+    movu            [r0 + r6 + 16],      m0
+    movu            [r0 + r1 * 4],       m0
+    movu            [r0 + r1 * 4 + 16],  m0
+    movu            [r0 + r7],           m0
+    movu            [r0 + r7 + 16],      m0
+    movu            [r0 + r6 * 2],       m0
+    movu            [r0 + r6 * 2 + 16],  m0
+    movu            [r9],                m0
+    movu            [r9 + 16],           m0
+    movu            [r0 + r1 * 8],       m0
+    movu            [r0 + r1 * 8 + 16],  m0
+    movu            [r9 + r1 * 2],       m0
+    movu            [r9 + r1 * 2 + 16],  m0
+    movu            [r0 + r7 * 2],       m0
+    movu            [r0 + r7 * 2 + 16],  m0
+    movu            [r9 + r1 * 4],       m0
+    movu            [r9 + r1 * 4 + 16],  m0
+    movu            [r0 + r6 * 4],       m0
+    movu            [r0 + r6 * 4 + 16],  m0
+    movu            [r9 + r6 * 2],       m0
+    movu            [r9 + r6 * 2 + 16],  m0
+    movu            [r9 + r8],           m0
+    movu            [r9 + r8 + 16],      m0
+    movu            [r9 + r1 * 8],       m0
+    movu            [r9 + r1 * 8 + 16],  m0
+
+    ; Do DC Filter
+    jz              .end
+    mova            m1,                 [pw_2]
+    pmullw          m1,                 m0
+    paddw           m1,                  [pw_2]
+    movd            r4d,                 m1
+    paddw           m1,                  m0
+
+    ; filter top
+    movu            m2,                  [r2]
+    paddw           m2,                  m1
+    psraw           m2,                  2
+    movu            [r0],                m2
+    movu            m3,                  [r2 + 16]
+    paddw           m3,                  m1
+    psraw           m3,                  2
+    movu            [r0 + 16],           m3
+
+    ; filter top-left
+    movzx           r4d,                 r4w
+    movzx           r5d, word            [r3]
+    add             r4d,                 r5d
+    movzx           r5d, word            [r2]
+    add             r5d,                 r4d
+    shr             r5d,                 2
+    mov             [r0],                r5w
+
+    ; filter left
+    movu            m2,                  [r3 + 2]
+    paddw           m2,                  m1
+    psraw           m2,                  2
+
+    movq            r2,                  m2
+    pshufd          m2,                  m2, 0xEE
+    mov             [r0 + r1],           r2w
+    shr             r2,                  16
+    mov             [r0 + r1 * 2],       r2w
+    shr             r2,                  16
+    mov             [r0 + r6],           r2w
+    shr             r2,                  16
+    mov             [r0 + r1 * 4],       r2w
+    movq            r2,                  m2
+    mov             [r0 + r7],           r2w
+    shr             r2,                  16
+    mov             [r0 + r6 * 2],       r2w
+    shr             r2,                  16
+    mov             [r9],                r2w
+    shr             r2,                  16
+    mov             [r0 + r1 * 8],       r2w
+
+    movu            m3,                  [r3 + 18]
+    paddw           m3,                  m1
+    psraw           m3,                  2
+
+    movq            r2,                  m3
+    pshufd          m3,                  m3, 0xEE
+    mov             [r9 + r1 * 2],       r2w
+    shr             r2,                  16
+    mov             [r0 + r7 * 2],       r2w
+    shr             r2,                  16
+    mov             [r9 + r1 * 4],       r2w
+    shr             r2,                  16
+    mov             [r0 + r6 * 4],           r2w
+    movq            r2,                  m3
+    mov             [r9 + r6 * 2],       r2w
+    shr             r2,                  16
+    mov             [r9 + r8],           r2w
+    shr             r2,                  16
+    mov             [r9 + r1 * 8],       r2w
+.end:
+    RET
+
 ;-----------------------------------------------------------------------------------
 ; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel* above, int, int filter)
 ;-----------------------------------------------------------------------------------


More information about the x265-devel mailing list