x86: Split pixels{4, 8, 16} functions to a separate file

Message ID 1361899920-12516-1-git-send-email-martin@martin.st
State Superseded
Headers show

Commit Message

Martin Storsjö Feb. 26, 2013, 5:32 p.m.
Since 71155d7b4, these functions were unconditionally referred to
from dsputil_mmx.o regardless of whether h264qpel was enabled or
not.

This moves the shared functions to a separate file that can be
built unconditionally.

This fixes builds with h264qpel disabled.
---
 libavcodec/x86/Makefile           |    1 +
 libavcodec/x86/h264_qpel_8bit.asm |  169 -----------------------------
 libavcodec/x86/pixels4816.asm     |  216 +++++++++++++++++++++++++++++++++++++
 3 files changed, 217 insertions(+), 169 deletions(-)
 create mode 100644 libavcodec/x86/pixels4816.asm

Comments

Diego Biurrun Feb. 26, 2013, 5:48 p.m. | #1
On Tue, Feb 26, 2013 at 07:32:00PM +0200, Martin Storsjö wrote:
> Since 71155d7b4, these functions were unconditionally referred to
> from dsputil_mmx.o regardless of whether h264qpel was enabled or
> not.
> 
> This moves the shared functions to a separate file that can be
> built unconditionally.
> 
> This fixes builds with h264qpel disabled.
> ---
>  libavcodec/x86/Makefile           |    1 +
>  libavcodec/x86/h264_qpel_8bit.asm |  169 -----------------------------
>  libavcodec/x86/pixels4816.asm     |  216 +++++++++++++++++++++++++++++++++++++
>  3 files changed, 217 insertions(+), 169 deletions(-)
>  create mode 100644 libavcodec/x86/pixels4816.asm

How is this different or better than moving those functions to dsputil.asm
as I did in a previous patch?

Diego
Martin Storsjö Feb. 26, 2013, 8:44 p.m. | #2
On Tue, 26 Feb 2013, Diego Biurrun wrote:

> On Tue, Feb 26, 2013 at 07:32:00PM +0200, Martin Storsjö wrote:
>> Since 71155d7b4, these functions were unconditionally referred to
>> from dsputil_mmx.o regardless of whether h264qpel was enabled or
>> not.
>>
>> This moves the shared functions to a separate file that can be
>> built unconditionally.
>>
>> This fixes builds with h264qpel disabled.
>> ---
>>  libavcodec/x86/Makefile           |    1 +
>>  libavcodec/x86/h264_qpel_8bit.asm |  169 -----------------------------
>>  libavcodec/x86/pixels4816.asm     |  216 +++++++++++++++++++++++++++++++++++++
>>  3 files changed, 217 insertions(+), 169 deletions(-)
>>  create mode 100644 libavcodec/x86/pixels4816.asm
>
> How is this different or better than moving those functions to dsputil.asm
> as I did in a previous patch?

It's not. I wasn't aware of such a patch, can you point me to it?

// Martin
Diego Biurrun Feb. 26, 2013, 10:13 p.m. | #3
On Tue, Feb 26, 2013 at 10:44:22PM +0200, Martin Storsjö wrote:
> On Tue, 26 Feb 2013, Diego Biurrun wrote:
> >On Tue, Feb 26, 2013 at 07:32:00PM +0200, Martin Storsjö wrote:
> >>Since 71155d7b4, these functions were unconditionally referred to
> >>from dsputil_mmx.o regardless of whether h264qpel was enabled or
> >>not.
> >>
> >>This moves the shared functions to a separate file that can be
> >>built unconditionally.
> >>
> >>This fixes builds with h264qpel disabled.
> >>---
> >> libavcodec/x86/Makefile           |    1 +
> >> libavcodec/x86/h264_qpel_8bit.asm |  169 -----------------------------
> >> libavcodec/x86/pixels4816.asm     |  216 +++++++++++++++++++++++++++++++++++++
> >> 3 files changed, 217 insertions(+), 169 deletions(-)
> >> create mode 100644 libavcodec/x86/pixels4816.asm
> >
> >How is this different or better than moving those functions to dsputil.asm
> >as I did in a previous patch?
> 
> It's not. I wasn't aware of such a patch, can you point me to it?

  Revert "Move H264/QPEL specific asm from dsputil.asm to h264_qpel_*.asm."

Diego
Martin Storsjö Feb. 26, 2013, 10:19 p.m. | #4
On Tue, 26 Feb 2013, Diego Biurrun wrote:

> On Tue, Feb 26, 2013 at 10:44:22PM +0200, Martin Storsjö wrote:
>> On Tue, 26 Feb 2013, Diego Biurrun wrote:
>>> On Tue, Feb 26, 2013 at 07:32:00PM +0200, Martin Storsjö wrote:
>>>> Since 71155d7b4, these functions were unconditionally referred to
>>>> from dsputil_mmx.o regardless of whether h264qpel was enabled or
>>>> not.
>>>>
>>>> This moves the shared functions to a separate file that can be
>>>> built unconditionally.
>>>>
>>>> This fixes builds with h264qpel disabled.
>>>> ---
>>>> libavcodec/x86/Makefile           |    1 +
>>>> libavcodec/x86/h264_qpel_8bit.asm |  169 -----------------------------
>>>> libavcodec/x86/pixels4816.asm     |  216 +++++++++++++++++++++++++++++++++++++
>>>> 3 files changed, 217 insertions(+), 169 deletions(-)
>>>> create mode 100644 libavcodec/x86/pixels4816.asm
>>>
>>> How is this different or better than moving those functions to dsputil.asm
>>> as I did in a previous patch?
>>
>> It's not. I wasn't aware of such a patch, can you point me to it?
>
>  Revert "Move H264/QPEL specific asm from dsputil.asm to h264_qpel_*.asm."

Oh, sorry - I misread that one.

// Martin

Patch

diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index a759e6e..3ec8cae 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -76,3 +76,4 @@  YASM-OBJS                              += x86/dsputil.o                 \
                                           x86/fmtconvert.o              \
                                           x86/hpeldsp.o                 \
                                           x86/mpeg4qpel.o               \
+                                          x86/pixels4816.o              \
diff --git a/libavcodec/x86/h264_qpel_8bit.asm b/libavcodec/x86/h264_qpel_8bit.asm
index b94d96d..bc6c725 100644
--- a/libavcodec/x86/h264_qpel_8bit.asm
+++ b/libavcodec/x86/h264_qpel_8bit.asm
@@ -860,172 +860,3 @@  INIT_XMM ssse3
 QPEL16_H_LOWPASS_L2_OP put
 QPEL16_H_LOWPASS_L2_OP avg
 %endif
-
-; void pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
-%macro PIXELS4_L2 1
-%define OP op_%1h
-cglobal %1_pixels4_l2, 6,6
-    movsxdifnidn r3, r3d
-    movsxdifnidn r4, r4d
-    test        r5d, 1
-    je        .loop
-    movd         m0, [r1]
-    movd         m1, [r2]
-    add          r1, r4
-    add          r2, 4
-    pavgb        m0, m1
-    OP           m0, [r0], m3
-    add          r0, r3
-    dec         r5d
-.loop:
-    mova         m0, [r1]
-    mova         m1, [r1+r4]
-    lea          r1, [r1+2*r4]
-    pavgb        m0, [r2]
-    pavgb        m1, [r2+4]
-    OP           m0, [r0], m3
-    OP           m1, [r0+r3], m3
-    lea          r0, [r0+2*r3]
-    mova         m0, [r1]
-    mova         m1, [r1+r4]
-    lea          r1, [r1+2*r4]
-    pavgb        m0, [r2+8]
-    pavgb        m1, [r2+12]
-    OP           m0, [r0], m3
-    OP           m1, [r0+r3], m3
-    lea          r0, [r0+2*r3]
-    add          r2, 16
-    sub         r5d, 4
-    jne       .loop
-    REP_RET
-%endmacro
-
-INIT_MMX mmxext
-PIXELS4_L2 put
-PIXELS4_L2 avg
-
-; void pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
-%macro PIXELS8_L2 1
-%define OP op_%1
-cglobal %1_pixels8_l2, 6,6
-    movsxdifnidn r3, r3d
-    movsxdifnidn r4, r4d
-    test        r5d, 1
-    je        .loop
-    mova         m0, [r1]
-    mova         m1, [r2]
-    add          r1, r4
-    add          r2, 8
-    pavgb        m0, m1
-    OP           m0, [r0]
-    add          r0, r3
-    dec         r5d
-.loop:
-    mova         m0, [r1]
-    mova         m1, [r1+r4]
-    lea          r1, [r1+2*r4]
-    pavgb        m0, [r2]
-    pavgb        m1, [r2+8]
-    OP           m0, [r0]
-    OP           m1, [r0+r3]
-    lea          r0, [r0+2*r3]
-    mova         m0, [r1]
-    mova         m1, [r1+r4]
-    lea          r1, [r1+2*r4]
-    pavgb        m0, [r2+16]
-    pavgb        m1, [r2+24]
-    OP           m0, [r0]
-    OP           m1, [r0+r3]
-    lea          r0, [r0+2*r3]
-    add          r2, 32
-    sub         r5d, 4
-    jne       .loop
-    REP_RET
-%endmacro
-
-INIT_MMX mmxext
-PIXELS8_L2 put
-PIXELS8_L2 avg
-
-; void pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
-%macro PIXELS16_L2 1
-%define OP op_%1
-cglobal %1_pixels16_l2, 6,6
-    movsxdifnidn r3, r3d
-    movsxdifnidn r4, r4d
-    test        r5d, 1
-    je        .loop
-    mova         m0, [r1]
-    mova         m1, [r1+8]
-    pavgb        m0, [r2]
-    pavgb        m1, [r2+8]
-    add          r1, r4
-    add          r2, 16
-    OP           m0, [r0]
-    OP           m1, [r0+8]
-    add          r0, r3
-    dec         r5d
-.loop:
-    mova         m0, [r1]
-    mova         m1, [r1+8]
-    add          r1, r4
-    pavgb        m0, [r2]
-    pavgb        m1, [r2+8]
-    OP           m0, [r0]
-    OP           m1, [r0+8]
-    add          r0, r3
-    mova         m0, [r1]
-    mova         m1, [r1+8]
-    add          r1, r4
-    pavgb        m0, [r2+16]
-    pavgb        m1, [r2+24]
-    OP           m0, [r0]
-    OP           m1, [r0+8]
-    add          r0, r3
-    add          r2, 32
-    sub         r5d, 2
-    jne       .loop
-    REP_RET
-%endmacro
-
-INIT_MMX mmxext
-PIXELS16_L2 put
-PIXELS16_L2 avg
-
-INIT_MMX mmxext
-; void pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)
-%macro PIXELS48 2
-%if %2 == 4
-%define OP movh
-%else
-%define OP mova
-%endif
-cglobal %1_pixels%2, 4,5
-    movsxdifnidn r2, r2d
-    lea          r4, [r2*3]
-.loop:
-    OP           m0, [r1]
-    OP           m1, [r1+r2]
-    OP           m2, [r1+r2*2]
-    OP           m3, [r1+r4]
-    lea          r1, [r1+r2*4]
-%ifidn %1, avg
-    pavgb        m0, [r0]
-    pavgb        m1, [r0+r2]
-    pavgb        m2, [r0+r2*2]
-    pavgb        m3, [r0+r4]
-%endif
-    OP         [r0], m0
-    OP      [r0+r2], m1
-    OP    [r0+r2*2], m2
-    OP      [r0+r4], m3
-    sub         r3d, 4
-    lea          r0, [r0+r2*4]
-    jne       .loop
-    RET
-%endmacro
-
-PIXELS48 put, 4
-PIXELS48 avg, 4
-PIXELS48 put, 8
-PIXELS48 avg, 8
diff --git a/libavcodec/x86/pixels4816.asm b/libavcodec/x86/pixels4816.asm
new file mode 100644
index 0000000..55d9d3e
--- /dev/null
+++ b/libavcodec/x86/pixels4816.asm
@@ -0,0 +1,216 @@ 
+;*****************************************************************************
+;* MMX-optimized pixel routines
+;*****************************************************************************
+;* Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
+;* Copyright (C) 2012 Daniel Kang
+;*
+;* Authors: Daniel Kang <daniel.d.kang@gmail.com>
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav 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
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION .text
+
+%macro op_avgh 3
+    movh   %3, %2
+    pavgb  %1, %3
+    movh   %2, %1
+%endmacro
+
+%macro op_avg 2-3
+    pavgb  %1, %2
+    mova   %2, %1
+%endmacro
+
+%macro op_puth 2-3
+    movh   %2, %1
+%endmacro
+
+%macro op_put 2-3
+    mova   %2, %1
+%endmacro
+
+; void pixels4_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+%macro PIXELS4_L2 1
+%define OP op_%1h
+cglobal %1_pixels4_l2, 6,6
+    movsxdifnidn r3, r3d
+    movsxdifnidn r4, r4d
+    test        r5d, 1
+    je        .loop
+    movd         m0, [r1]
+    movd         m1, [r2]
+    add          r1, r4
+    add          r2, 4
+    pavgb        m0, m1
+    OP           m0, [r0], m3
+    add          r0, r3
+    dec         r5d
+.loop:
+    mova         m0, [r1]
+    mova         m1, [r1+r4]
+    lea          r1, [r1+2*r4]
+    pavgb        m0, [r2]
+    pavgb        m1, [r2+4]
+    OP           m0, [r0], m3
+    OP           m1, [r0+r3], m3
+    lea          r0, [r0+2*r3]
+    mova         m0, [r1]
+    mova         m1, [r1+r4]
+    lea          r1, [r1+2*r4]
+    pavgb        m0, [r2+8]
+    pavgb        m1, [r2+12]
+    OP           m0, [r0], m3
+    OP           m1, [r0+r3], m3
+    lea          r0, [r0+2*r3]
+    add          r2, 16
+    sub         r5d, 4
+    jne       .loop
+    REP_RET
+%endmacro
+
+INIT_MMX mmxext
+PIXELS4_L2 put
+PIXELS4_L2 avg
+
+; void pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+%macro PIXELS8_L2 1
+%define OP op_%1
+cglobal %1_pixels8_l2, 6,6
+    movsxdifnidn r3, r3d
+    movsxdifnidn r4, r4d
+    test        r5d, 1
+    je        .loop
+    mova         m0, [r1]
+    mova         m1, [r2]
+    add          r1, r4
+    add          r2, 8
+    pavgb        m0, m1
+    OP           m0, [r0]
+    add          r0, r3
+    dec         r5d
+.loop:
+    mova         m0, [r1]
+    mova         m1, [r1+r4]
+    lea          r1, [r1+2*r4]
+    pavgb        m0, [r2]
+    pavgb        m1, [r2+8]
+    OP           m0, [r0]
+    OP           m1, [r0+r3]
+    lea          r0, [r0+2*r3]
+    mova         m0, [r1]
+    mova         m1, [r1+r4]
+    lea          r1, [r1+2*r4]
+    pavgb        m0, [r2+16]
+    pavgb        m1, [r2+24]
+    OP           m0, [r0]
+    OP           m1, [r0+r3]
+    lea          r0, [r0+2*r3]
+    add          r2, 32
+    sub         r5d, 4
+    jne       .loop
+    REP_RET
+%endmacro
+
+INIT_MMX mmxext
+PIXELS8_L2 put
+PIXELS8_L2 avg
+
+; void pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+%macro PIXELS16_L2 1
+%define OP op_%1
+cglobal %1_pixels16_l2, 6,6
+    movsxdifnidn r3, r3d
+    movsxdifnidn r4, r4d
+    test        r5d, 1
+    je        .loop
+    mova         m0, [r1]
+    mova         m1, [r1+8]
+    pavgb        m0, [r2]
+    pavgb        m1, [r2+8]
+    add          r1, r4
+    add          r2, 16
+    OP           m0, [r0]
+    OP           m1, [r0+8]
+    add          r0, r3
+    dec         r5d
+.loop:
+    mova         m0, [r1]
+    mova         m1, [r1+8]
+    add          r1, r4
+    pavgb        m0, [r2]
+    pavgb        m1, [r2+8]
+    OP           m0, [r0]
+    OP           m1, [r0+8]
+    add          r0, r3
+    mova         m0, [r1]
+    mova         m1, [r1+8]
+    add          r1, r4
+    pavgb        m0, [r2+16]
+    pavgb        m1, [r2+24]
+    OP           m0, [r0]
+    OP           m1, [r0+8]
+    add          r0, r3
+    add          r2, 32
+    sub         r5d, 2
+    jne       .loop
+    REP_RET
+%endmacro
+
+INIT_MMX mmxext
+PIXELS16_L2 put
+PIXELS16_L2 avg
+
+INIT_MMX mmxext
+; void pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)
+%macro PIXELS48 2
+%if %2 == 4
+%define OP movh
+%else
+%define OP mova
+%endif
+cglobal %1_pixels%2, 4,5
+    movsxdifnidn r2, r2d
+    lea          r4, [r2*3]
+.loop:
+    OP           m0, [r1]
+    OP           m1, [r1+r2]
+    OP           m2, [r1+r2*2]
+    OP           m3, [r1+r4]
+    lea          r1, [r1+r2*4]
+%ifidn %1, avg
+    pavgb        m0, [r0]
+    pavgb        m1, [r0+r2]
+    pavgb        m2, [r0+r2*2]
+    pavgb        m3, [r0+r4]
+%endif
+    OP         [r0], m0
+    OP      [r0+r2], m1
+    OP    [r0+r2*2], m2
+    OP      [r0+r4], m3
+    sub         r3d, 4
+    lea          r0, [r0+r2*4]
+    jne       .loop
+    RET
+%endmacro
+
+PIXELS48 put, 4
+PIXELS48 avg, 4
+PIXELS48 put, 8
+PIXELS48 avg, 8