[x265] [PATCH 1 of 2] asm: assembly code for calcrecon[]
chen
chenm003 at 163.com
Thu Nov 14 15:25:47 CET 2013
Sure
在 2013-11-14 22:19:07，"Steve Borho" <steve at borho.org> 写道：
On Thu, Nov 14, 2013 at 2:47 AM, Min Chen <chenm003 at 163.com> wrote:
# HG changeset patch
# User Min Chen <chenm003 at 163.com>
# Date 1384418703 -28800
# Node ID 8e22129119d6d8049996ed5f487625e4801b0a50
# Parent d80ab2913b31e678334fb0941066c313dcb2d3b7
asm: assembly code for calcrecon[]
the EOLN changes to pixel-util.asm are not making it through your email client properly, so this patch is not importing. Can you send this to me as a patch file?
diff -r d80ab2913b31 -r 8e22129119d6 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Wed Nov 13 14:30:22 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp Thu Nov 14 16:45:03 2013 +0800
@@ -450,6 +450,8 @@
p.cvt32to16_shr = x265_cvt32to16_shr_sse2;
p.ipfilter_ss[FILTER_V_S_S_8] = x265_interp_8tap_v_ss_sse2;
+ p.calcrecon[BLOCK_4x4] = x265_calcRecons4_sse2;
+ p.calcrecon[BLOCK_8x8] = x265_calcRecons8_sse2;
}
if (cpuMask & X265_CPU_SSSE3)
{
@@ -525,6 +527,9 @@
p.chroma_copy_sp[CHROMA_2x4] = x265_blockcopy_sp_2x4_sse4;
p.chroma_copy_sp[CHROMA_2x8] = x265_blockcopy_sp_2x8_sse4;
p.chroma_copy_sp[CHROMA_6x8] = x265_blockcopy_sp_6x8_sse4;
+ p.calcrecon[BLOCK_16x16] = x265_calcRecons16_sse4;
+ p.calcrecon[BLOCK_32x32] = x265_calcRecons32_sse4;
+ p.calcrecon[BLOCK_64x64] = x265_calcRecons64_sse4;
}
if (cpuMask & X265_CPU_AVX)
{
diff -r d80ab2913b31 -r 8e22129119d6 source/common/x86/pixel-util.asm
--- a/source/common/x86/pixel-util.asm Wed Nov 13 14:30:22 2013 +0530
+++ b/source/common/x86/pixel-util.asm Thu Nov 14 16:45:03 2013 +0800
@@ -1,103 +1,475 @@
-;*****************************************************************************
-;* Copyright (C) 2013 x265 project
-;*
-;* Authors: Min Chen <chenm003 at 163.com> <min.chen at multicorewareinc.com>
-;*
-;* This program is free software; you can redistribute it and/or modify
-;* it under the terms of the GNU General Public License as published by
-;* the Free Software Foundation; either version 2 of the License, or
-;* (at your option) any later version.
-;*
-;* This program is distributed in the hope that it will be useful,
-;* but WITHOUT ANY WARRANTY; without even the implied warranty of
-;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-;* GNU General Public License for more details.
-;*
-;* You should have received a copy of the GNU General Public License
-;* along with this program; if not, write to the Free Software
-;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
-;*
-;* This program is also available under a commercial proprietary license.
-;* For more information, contact us at licensing at multicorewareinc.com.
-;*****************************************************************************/
-
-%include "x86inc.asm"
-%include "x86util.asm"
-
-SECTION_RODATA 32
-
-SECTION .text
-
-
-;-----------------------------------------------------------------------------
-; void cvt32to16_shr(short *dst, int *src, intptr_t stride, int shift, int size)
-;-----------------------------------------------------------------------------
-INIT_XMM sse2
-cglobal cvt32to16_shr, 5, 7, 1, dst, src, stride
-%define rnd m7
-%define shift m6
-
- ; make shift
- mov r5d, r3m
- movd shift, r5d
-
- ; make round
- dec r5
- xor r6, r6
- bts r6, r5
-
- movd rnd, r6d
- pshufd rnd, rnd, 0
-
- ; register alloc
- ; r0 - dst
- ; r1 - src
- ; r2 - stride * 2 (short*)
- ; r3 - lx
- ; r4 - size
- ; r5 - ly
- ; r6 - diff
- lea r2, [r2 * 2]
-
- mov r4d, r4m
- mov r5, r4
- mov r6, r2
- sub r6, r4
- lea r6, [r6 * 2]
-
- shr r5, 1
-.loop_row:
-
- mov r3, r4
- shr r3, 2
-.loop_col:
- ; row 0
- movu m0, [r1]
- paddd m0, rnd
- psrad m0, shift
- packssdw m0, m0
- movh [r0], m0
-
- ; row 1
- movu m0, [r1 + r4 * 4]
- paddd m0, rnd
- psrad m0, shift
- packssdw m0, m0
- movh [r0 + r2], m0
-
- ; move col pointer
- add r1, 16
- add r0, 8
-
- dec r3
- jg .loop_col
-
- ; update pointer
- lea r1, [r1 + r4 * 4]
- add r0, r6
-
- ; end of loop_row
- dec r5
- jg .loop_row
-
- RET
+;*****************************************************************************
+;* Copyright (C) 2013 x265 project
+;*
+;* Authors: Min Chen <chenm003 at 163.com> <min.chen at multicorewareinc.com>
+;*
+;* This program is free software; you can redistribute it and/or modify
+;* it under the terms of the GNU General Public License as published by
+;* the Free Software Foundation; either version 2 of the License, or
+;* (at your option) any later version.
+;*
+;* This program is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+;* GNU General Public License for more details.
+;*
+;* You should have received a copy of the GNU General Public License
+;* along with this program; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at licensing at multicorewareinc.com.
+;*****************************************************************************/
+
+%include "x86inc.asm"
+%include "x86util.asm"
+
+SECTION_RODATA 32
+
+SECTION .text
+
+
+;-----------------------------------------------------------------------------
+; void cvt32to16_shr(short *dst, int *src, intptr_t stride, int shift, int size)
+;-----------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal cvt32to16_shr, 5, 7, 1, dst, src, stride
+%define rnd m7
+%define shift m6
+
+ ; make shift
+ mov r5d, r3m
+ movd shift, r5d
+
+ ; make round
+ dec r5
+ xor r6, r6
+ bts r6, r5
+
+ movd rnd, r6d
+ pshufd rnd, rnd, 0
+
+ ; register alloc
+ ; r0 - dst
+ ; r1 - src
+ ; r2 - stride * 2 (short*)
+ ; r3 - lx
+ ; r4 - size
+ ; r5 - ly
+ ; r6 - diff
+ lea r2, [r2 * 2]
+
+ mov r4d, r4m
+ mov r5, r4
+ mov r6, r2
+ sub r6, r4
+ lea r6, [r6 * 2]
+
+ shr r5, 1
+.loop_row:
+
+ mov r3, r4
+ shr r3, 2
+.loop_col:
+ ; row 0
+ movu m0, [r1]
+ paddd m0, rnd
+ psrad m0, shift
+ packssdw m0, m0
+ movh [r0], m0
+
+ ; row 1
+ movu m0, [r1 + r4 * 4]
+ paddd m0, rnd
+ psrad m0, shift
+ packssdw m0, m0
+ movh [r0 + r2], m0
+
+ ; move col pointer
+ add r1, 16
+ add r0, 8
+
+ dec r3
+ jg .loop_col
+
+ ; update pointer
+ lea r1, [r1 + r4 * 4]
+ add r0, r6
+
+ ; end of loop_row
+ dec r5
+ jg .loop_row
+
+ RET
+
+
+;-----------------------------------------------------------------------------
+; void calcrecon(pixel* pred, int16_t* residual, pixel* recon, int16_t* reconqt, pixel *reconipred, int stride, int strideqt, int strideipred)
+;-----------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal calcRecons4
+%if ARCH_X86_64 == 1
+ DECLARE_REG_TMP 0,1,2,3,4,5,6,7,8
+ PROLOGUE 6,9,4
+%else
+ DECLARE_REG_TMP 0,1,2,3,4,5
+ PROLOGUE 6,7,4
+ %define t6 r6m
+ %define t6d r6d
+ %define t7 r7m
+ %define t8d r6d
+%endif
+
+ mov t6d, r6m
+%if ARCH_X86_64 == 0
+ add t6d, t6d
+ mov r6m, t6d
+%else
+ mov r5d, r5m
+ mov r7d, r7m
+ add t6d, t6d
+%endif
+
+ pxor m0, m0
+ mov t8d, 4/2
+.loop:
+ movd m1, [t0]
+ movd m2, [t0 + t5]
+ punpckldq m1, m2
+ punpcklbw m1, m0
+ movh m2, [t1]
+ movh m3, [t1 + t5 * 2]
+ punpcklqdq m2, m3
+ paddw m1, m2
+ packuswb m1, m1
+
+ ; store recon[] and recipred[]
+ movd [t2], m1
+ movd [t4], m1
+ add t4, t7
+ pshufd m2, m1, 1
+ movd [t2 + t5], m2
+ movd [t4], m2
+ add t4, t7
+
+ ; store recqt[]
+ punpcklbw m1, m0
+ movlps [t3], m1
+ add t3, t6
+ movhps [t3], m1
+ add t3, t6
+
+ lea t0, [t0 + t5 * 2]
+ lea t1, [t1 + t5 * 4]
+ lea t2, [t2 + t5 * 2]
+
+ dec t8d
+ jnz .loop
+ RET
+
+
+INIT_XMM sse2
+cglobal calcRecons8
+%if ARCH_X86_64 == 1
+ DECLARE_REG_TMP 0,1,2,3,4,5,6,7,8
+ PROLOGUE 6,9,5
+%else
+ DECLARE_REG_TMP 0,1,2,3,4,5
+ PROLOGUE 6,7,5
+ %define t6 r6m
+ %define t6d r6d
+ %define t7 r7m
+ %define t8d r6d
+%endif
+
+ mov t6d, r6m
+%if ARCH_X86_64 == 0
+ add t6d, t6d
+ mov r6m, t6d
+%else
+ mov r5d, r5m
+ mov r7d, r7m
+ add t6d, t6d
+%endif
+
+ pxor m0, m0
+ mov t8d, 8/2
+.loop:
+ movh m1, [t0]
+ movh m2, [t0 + t5]
+ punpcklbw m1, m0
+ punpcklbw m2, m0
+ movu m3, [t1]
+ movu m4, [t1 + t5 * 2]
+ paddw m1, m3
+ paddw m2, m4
+ packuswb m1, m2
+
+ ; store recon[] and recipred[]
+ movlps [t2], m1
+ movhps [t2 + t5], m1
+ movlps [t4], m1
+%if ARCH_X86_64 == 0
+ add t4, t7
+ movhps [t4], m1
+ add t4, t7
+%else
+ movhps [t4 + t7], m1
+ lea t4, [t4 + t7 * 2]
+%endif
+
+ ; store recqt[]
+ punpcklbw m2, m1, m0
+ punpckhbw m1, m0
+ movu [t3], m2
+ add t3, t6
+ movu [t3], m1
+ add t3, t6
+
+ lea t0, [t0 + t5 * 2]
+ lea t1, [t1 + t5 * 4]
+ lea t2, [t2 + t5 * 2]
+
+ dec t8d
+ jnz .loop
+ RET
+
+
+INIT_XMM sse4
+cglobal calcRecons16
+%if ARCH_X86_64 == 1
+ DECLARE_REG_TMP 0,1,2,3,4,5,6,7,8
+ PROLOGUE 6,9,5
+%else
+ DECLARE_REG_TMP 0,1,2,3,4,5
+ PROLOGUE 6,7,5
+ %define t6 r6m
+ %define t6d r6d
+ %define t7 r7m
+ %define t8d r6d
+%endif
+
+ mov t6d, r6m
+%if ARCH_X86_64 == 0
+ add t6d, t6d
+ mov r6m, t6d
+%else
+ mov r5d, r5m
+ mov r7d, r7m
+ add t6d, t6d
+%endif
+
+ pxor m0, m0
+ mov t8d, 16
+.loop:
+ movu m2, [t0]
+ pmovzxbw m1, m2
+ punpckhbw m2, m0
+ movu m3, [t1]
+ movu m4, [t1 + 16]
+ paddw m1, m3
+ paddw m2, m4
+ packuswb m1, m2
+
+ ; store recon[] and recipred[]
+ movu [t2], m1
+ movu [t4], m1
+
+ ; store recqt[]
+ pmovzxbw m2, m1
+ punpckhbw m1, m0
+ movu [t3], m2
+ movu [t3 + 16], m1
+
+ add t3, t6
+ add t4, t7
+ add t0, t5
+ lea t1, [t1 + t5 * 2]
+ add t2, t5
+
+ dec t8d
+ jnz .loop
+ RET
+
+
+INIT_XMM sse4
+cglobal calcRecons32
+%if ARCH_X86_64 == 1
+ DECLARE_REG_TMP 0,1,2,3,4,5,6,7,8
+ PROLOGUE 6,9,7
+%else
+ DECLARE_REG_TMP 0,1,2,3,4,5
+ PROLOGUE 6,7,7
+ %define t6 r6m
+ %define t6d r6d
+ %define t7 r7m
+ %define t8d r6d
+%endif
+
+ mov t6d, r6m
+%if ARCH_X86_64 == 0
+ add t6d, t6d
+ mov r6m, t6d
+%else
+ mov r5d, r5m
+ mov r7d, r7m
+ add t6d, t6d
+%endif
+
+ pxor m0, m0
+ mov t8d, 32
+.loop:
+ movu m2, [t0]
+ movu m4, [t0 + 16]
+ pmovzxbw m1, m2
+ punpckhbw m2, m0
+ pmovzxbw m3, m4
+ punpckhbw m4, m0
+
+ movu m5, [t1 + 0 * 16]
+ movu m6, [t1 + 1 * 16]
+ paddw m1, m5
+ paddw m2, m6
+ packuswb m1, m2
+
+ movu m5, [t1 + 2 * 16]
+ movu m6, [t1 + 3 * 16]
+ paddw m3, m5
+ paddw m4, m6
+ packuswb m3, m4
+
+ ; store recon[] and recipred[]
+ movu [t2], m1
+ movu [t2 + 16], m3
+ movu [t4], m1
+ movu [t4 + 16], m3
+
+ ; store recqt[]
+ pmovzxbw m2, m1
+ punpckhbw m1, m0
+ movu [t3 + 0 * 16], m2
+ movu [t3 + 1 * 16], m1
+ pmovzxbw m4, m3
+ punpckhbw m3, m0
+ movu [t3 + 2 * 16], m4
+ movu [t3 + 3 * 16], m3
+
+ add t3, t6
+ add t4, t7
+ add t0, t5
+ lea t1, [t1 + t5 * 2]
+ add t2, t5
+
+ dec t8d
+ jnz .loop
+ RET
+
+
+INIT_XMM sse4
+cglobal calcRecons64
+%if ARCH_X86_64 == 1
+ DECLARE_REG_TMP 0,1,2,3,4,5,6,7,8
+ PROLOGUE 6,9,7
+%else
+ DECLARE_REG_TMP 0,1,2,3,4,5
+ PROLOGUE 6,7,7
+ %define t6 r6m
+ %define t6d r6d
+ %define t7 r7m
+ %define t8d r6d
+%endif
+
+ mov t6d, r6m
+%if ARCH_X86_64 == 0
+ add t6d, t6d
+ mov r6m, t6d
+%else
+ mov r5d, r5m
+ mov r7d, r7m
+ add t6d, t6d
+%endif
+
+ pxor m0, m0
+ mov t8d, 64
+.loop:
+ ; left 32 pixel
+ movu m2, [t0 + 0 * 16]
+ movu m4, [t0 + 1 * 16]
+ pmovzxbw m1, m2
+ punpckhbw m2, m0
+ pmovzxbw m3, m4
+ punpckhbw m4, m0
+
+ movu m5, [t1 + 0 * 16]
+ movu m6, [t1 + 1 * 16]
+ paddw m1, m5
+ paddw m2, m6
+ packuswb m1, m2
+
+ movu m5, [t1 + 2 * 16]
+ movu m6, [t1 + 3 * 16]
+ paddw m3, m5
+ paddw m4, m6
+ packuswb m3, m4
+
+ ; store recon[] and recipred[]
+ movu [t2 + 0 * 16], m1
+ movu [t2 + 1 * 16], m3
+ movu [t4 + 0 * 16], m1
+ movu [t4 + 1 * 16], m3
+
+ ; store recqt[]
+ pmovzxbw m2, m1
+ punpckhbw m1, m0
+ movu [t3 + 0 * 16], m2
+ movu [t3 + 1 * 16], m1
+ pmovzxbw m4, m3
+ punpckhbw m3, m0
+ movu [t3 + 2 * 16], m4
+ movu [t3 + 3 * 16], m3
+
+ ; right 32 pixel
+ movu m2, [t0 + 2 * 16]
+ movu m4, [t0 + 3 * 16]
+ pmovzxbw m1, m2
+ punpckhbw m2, m0
+ pmovzxbw m3, m4
+ punpckhbw m4, m0
+
+ movu m5, [t1 + 4 * 16]
+ movu m6, [t1 + 5 * 16]
+ paddw m1, m5
+ paddw m2, m6
+ packuswb m1, m2
+
+ movu m5, [t1 + 6 * 16]
+ movu m6, [t1 + 7 * 16]
+ paddw m3, m5
+ paddw m4, m6
+ packuswb m3, m4
+
+ ; store recon[] and recipred[]
+ movu [t2 + 2 * 16], m1
+ movu [t2 + 3 * 16], m3
+ movu [t4 + 2 * 16], m1
+ movu [t4 + 3 * 16], m3
+
+ ; store recqt[]
+ pmovzxbw m2, m1
+ punpckhbw m1, m0
+ movu [t3 + 4 * 16], m2
+ movu [t3 + 5 * 16], m1
+ pmovzxbw m4, m3
+ punpckhbw m3, m0
+ movu [t3 + 6 * 16], m4
+ movu [t3 + 7 * 16], m3
+
+ add t3, t6
+ add t4, t7
+ add t0, t5
+ lea t1, [t1 + t5 * 2]
+ add t2, t5
+
+ dec t8d
+ jnz .loop
+ RET
diff -r d80ab2913b31 -r 8e22129119d6 source/common/x86/pixel.h
--- a/source/common/x86/pixel.h Wed Nov 13 14:30:22 2013 +0530
+++ b/source/common/x86/pixel.h Thu Nov 14 16:45:03 2013 +0800
@@ -355,4 +355,10 @@
#undef CHROMA_PIXELSUB_DEF
#undef LUMA_PIXELSUB_DEF
+void x265_calcRecons4_sse2(pixel* pred, int16_t* residual, pixel* recon, int16_t* reconqt, pixel *reconipred, int stride, int strideqt, int strideipred);
+void x265_calcRecons8_sse2(pixel* pred, int16_t* residual, pixel* recon, int16_t* reconqt, pixel *reconipred, int stride, int strideqt, int strideipred);
+void x265_calcRecons16_sse4(pixel* pred, int16_t* residual, pixel* recon, int16_t* reconqt, pixel *reconipred, int stride, int strideqt, int strideipred);
+void x265_calcRecons32_sse4(pixel* pred, int16_t* residual, pixel* recon, int16_t* reconqt, pixel *reconipred, int stride, int strideqt, int strideipred);
+void x265_calcRecons64_sse4(pixel* pred, int16_t* residual, pixel* recon, int16_t* reconqt, pixel *reconipred, int stride, int strideqt, int strideipred);
+
#endif // ifndef X265_I386_PIXEL_H
diff -r d80ab2913b31 -r 8e22129119d6 source/test/pixelharness.cpp
--- a/source/test/pixelharness.cpp Wed Nov 13 14:30:22 2013 +0530
+++ b/source/test/pixelharness.cpp Thu Nov 14 16:45:03 2013 +0800
@@ -286,8 +286,8 @@
for (int i = 0; i < ITERS; i++)
{
int stride = STRIDE;
+ ref(pbuf1 + j, sbuf1 + j, ref_reco, ref_recq, ref_pred, stride, stride, stride);
opt(pbuf1 + j, sbuf1 + j, opt_reco, opt_recq, opt_pred, stride, stride, stride);
- ref(pbuf1 + j, sbuf1 + j, ref_reco, ref_recq, ref_pred, stride, stride, stride);
if (memcmp(ref_recq, opt_recq, 64 * 64 * sizeof(int16_t)))
return false;
_______________________________________________
x265-devel mailing list
x265-devel at videolan.org
https://mailman.videolan.org/listinfo/x265-devel
--
Steve Borho
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20131114/4c1177db/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-asm-assembly-code-for-calcrecon.patch
Type: application/octet-stream
Size: 17229 bytes
Desc: not available
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20131114/4c1177db/attachment-0001.obj>
More information about the x265-devel
mailing list