Merge commit '652f5185945c8405fc57aed353286858df8d066f'
* commit '652f5185945c8405fc57aed353286858df8d066f': x86: mmx2 ---> mmxext in comments and messages Conflicts: libswscale/x86/swscale_template.c Merged-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
commit
78ec407d5a
@ -253,7 +253,7 @@ Optimization guide for ARM11 (used in Nokia N800 Internet Tablet):
|
||||
http://infocenter.arm.com/help/topic/com.arm.doc.ddi0211j/DDI0211J_arm1136_r1p5_trm.pdf
|
||||
Optimization guide for Intel XScale (used in Sharp Zaurus PDA):
|
||||
http://download.intel.com/design/intelxscale/27347302.pdf
|
||||
Intel Wireless MMX2 Coprocessor: Programmers Reference Manual
|
||||
Intel Wireless MMX 2 Coprocessor: Programmers Reference Manual
|
||||
http://download.intel.com/design/intelxscale/31451001.pdf
|
||||
|
||||
PowerPC-specific:
|
||||
|
@ -58,7 +58,7 @@ Input to YUV Converter
|
||||
|
||||
Horizontal scaler
|
||||
There are several horizontal scalers. A special case worth mentioning is
|
||||
the fast bilinear scaler that is made of runtime-generated MMX2 code
|
||||
the fast bilinear scaler that is made of runtime-generated MMXEXT code
|
||||
using specially tuned pshufw instructions.
|
||||
The remaining scalers are specially-tuned for various filter lengths.
|
||||
They scale 8-bit unsigned planar data to 16-bit signed planar data.
|
||||
|
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* DSP utils : average functions are compiled twice for 3dnow/mmx2
|
||||
* DSP utils : average functions are compiled twice for 3dnow/mmxext
|
||||
* Copyright (c) 2000, 2001 Fabrice Bellard
|
||||
* Copyright (c) 2002-2004 Michael Niedermayer
|
||||
*
|
||||
|
@ -206,11 +206,11 @@ DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
|
||||
#undef OP_AVG
|
||||
|
||||
/***********************************/
|
||||
/* MMX2 specific */
|
||||
/* MMXEXT specific */
|
||||
|
||||
#define DEF(x) x ## _mmx2
|
||||
|
||||
/* Introduced only in MMX2 set */
|
||||
/* Introduced only in MMXEXT set */
|
||||
#define PAVGB "pavgb"
|
||||
#define OP_AVG PAVGB
|
||||
|
||||
|
@ -122,7 +122,7 @@ void (*yuyvtoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
|
||||
/*
|
||||
* RGB15->RGB16 original by Strepto/Astral
|
||||
* ported to gcc & bugfixed : A'rpi
|
||||
* MMX2, 3DNOW optimization by Nick Kurshev
|
||||
* MMXEXT, 3DNOW optimization by Nick Kurshev
|
||||
* 32-bit C version, and and&add trick by Michael Niedermayer
|
||||
*/
|
||||
|
||||
|
@ -77,7 +77,7 @@ static inline void rgb32tobgr24_c(const uint8_t *src, uint8_t *dst,
|
||||
/*
|
||||
* original by Strepto/Astral
|
||||
* ported to gcc & bugfixed: A'rpi
|
||||
* MMX2, 3DNOW optimization by Nick Kurshev
|
||||
* MMXEXT, 3DNOW optimization by Nick Kurshev
|
||||
* 32-bit C version, and and&add trick by Michael Niedermayer
|
||||
*/
|
||||
static inline void rgb15to16_c(const uint8_t *src, uint8_t *dst, int src_size)
|
||||
|
@ -312,10 +312,10 @@ typedef struct SwsContext {
|
||||
int vChrFilterSize; ///< Vertical filter size for chroma pixels.
|
||||
//@}
|
||||
|
||||
int lumMmx2FilterCodeSize; ///< Runtime-generated MMX2 horizontal fast bilinear scaler code size for luma/alpha planes.
|
||||
int chrMmx2FilterCodeSize; ///< Runtime-generated MMX2 horizontal fast bilinear scaler code size for chroma planes.
|
||||
uint8_t *lumMmx2FilterCode; ///< Runtime-generated MMX2 horizontal fast bilinear scaler code for luma/alpha planes.
|
||||
uint8_t *chrMmx2FilterCode; ///< Runtime-generated MMX2 horizontal fast bilinear scaler code for chroma planes.
|
||||
int lumMmx2FilterCodeSize; ///< Runtime-generated MMXEXT horizontal fast bilinear scaler code size for luma/alpha planes.
|
||||
int chrMmx2FilterCodeSize; ///< Runtime-generated MMXEXT horizontal fast bilinear scaler code size for chroma planes.
|
||||
uint8_t *lumMmx2FilterCode; ///< Runtime-generated MMXEXT horizontal fast bilinear scaler code for luma/alpha planes.
|
||||
uint8_t *chrMmx2FilterCode; ///< Runtime-generated MMXEXT horizontal fast bilinear scaler code for chroma planes.
|
||||
|
||||
int canMMX2BeUsed;
|
||||
|
||||
|
@ -643,7 +643,7 @@ static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,
|
||||
int xpos, i;
|
||||
|
||||
// create an optimized horizontal scaling routine
|
||||
/* This scaler is made of runtime-generated MMX2 code using specially tuned
|
||||
/* This scaler is made of runtime-generated MMXEXT code using specially tuned
|
||||
* pshufw instructions. For every four output pixels, if four input pixels
|
||||
* are enough for the fast bilinear scaling, then a chunk of fragmentB is
|
||||
* used. If five input pixels are needed, then a chunk of fragmentA is used.
|
||||
@ -1069,7 +1069,7 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
|
||||
&& (flags & SWS_FAST_BILINEAR)) {
|
||||
if (flags & SWS_PRINT_INFO)
|
||||
av_log(c, AV_LOG_INFO,
|
||||
"output width is not a multiple of 32 -> no MMX2 scaler\n");
|
||||
"output width is not a multiple of 32 -> no MMXEXT scaler\n");
|
||||
}
|
||||
if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
|
||||
c->canMMX2BeUsed=0;
|
||||
@ -1313,7 +1313,7 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
|
||||
av_get_pix_fmt_name(dstFormat));
|
||||
|
||||
if (INLINE_MMXEXT(cpu_flags))
|
||||
av_log(c, AV_LOG_INFO, "using MMX2\n");
|
||||
av_log(c, AV_LOG_INFO, "using MMXEXT\n");
|
||||
else if (INLINE_AMD3DNOW(cpu_flags))
|
||||
av_log(c, AV_LOG_INFO, "using 3DNOW\n");
|
||||
else if (INLINE_MMX(cpu_flags))
|
||||
|
@ -218,10 +218,10 @@ cglobal yuv2planeX_%1, %3, 8, %2, filter, fltsize, src, dst, w, dither, offset
|
||||
%else ; %1 == 9/10
|
||||
%if cpuflag(sse4)
|
||||
packusdw m2, m1
|
||||
%else ; mmx2/sse2
|
||||
%else ; mmxext/sse2
|
||||
packssdw m2, m1
|
||||
pmaxsw m2, m6
|
||||
%endif ; mmx2/sse2/sse4/avx
|
||||
%endif ; mmxext/sse2/sse4/avx
|
||||
pminsw m2, [yuv2yuvX_%1_upper]
|
||||
%endif ; %1 == 9/10/16
|
||||
mova [dstq+r5*2], m2
|
||||
|
@ -87,7 +87,7 @@ DECLARE_ASM_CONST(8, uint64_t, mul16_mid) = 0x2080208020802080ULL;
|
||||
#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
|
||||
#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
|
||||
|
||||
//Note: We have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW + MMX2 one.
|
||||
// Note: We have C, MMX, MMXEXT, 3DNOW versions, there is no 3DNOW + MMXEXT one.
|
||||
|
||||
#define COMPILE_TEMPLATE_MMXEXT 0
|
||||
#define COMPILE_TEMPLATE_AMD3DNOW 0
|
||||
@ -98,7 +98,7 @@ DECLARE_ASM_CONST(8, uint64_t, mul16_mid) = 0x2080208020802080ULL;
|
||||
#define RENAME(a) a ## _MMX
|
||||
#include "rgb2rgb_template.c"
|
||||
|
||||
//MMX2 versions
|
||||
// MMXEXT versions
|
||||
#undef RENAME
|
||||
#undef COMPILE_TEMPLATE_MMXEXT
|
||||
#define COMPILE_TEMPLATE_MMXEXT 1
|
||||
@ -126,7 +126,7 @@ DECLARE_ASM_CONST(8, uint64_t, mul16_mid) = 0x2080208020802080ULL;
|
||||
/*
|
||||
RGB15->RGB16 original by Strepto/Astral
|
||||
ported to gcc & bugfixed : A'rpi
|
||||
MMX2, 3DNOW optimization by Nick Kurshev
|
||||
MMXEXT, 3DNOW optimization by Nick Kurshev
|
||||
32-bit C version, and and&add trick by Michael Niedermayer
|
||||
*/
|
||||
|
||||
|
@ -178,7 +178,7 @@ static inline void RENAME(rgb32tobgr24)(const uint8_t *src, uint8_t *dst, int sr
|
||||
/*
|
||||
original by Strepto/Astral
|
||||
ported to gcc & bugfixed: A'rpi
|
||||
MMX2, 3DNOW optimization by Nick Kurshev
|
||||
MMXEXT, 3DNOW optimization by Nick Kurshev
|
||||
32-bit C version, and and&add trick by Michael Niedermayer
|
||||
*/
|
||||
static inline void RENAME(rgb15to16)(const uint8_t *src, uint8_t *dst, int src_size)
|
||||
|
@ -80,7 +80,7 @@ DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
|
||||
#include "swscale_template.c"
|
||||
#endif
|
||||
|
||||
//MMX2 versions
|
||||
// MMXEXT versions
|
||||
#if HAVE_MMXEXT_INLINE
|
||||
#undef RENAME
|
||||
#undef COMPILE_TEMPLATE_MMXEXT
|
||||
|
@ -1690,7 +1690,7 @@ static av_cold void RENAME(sws_init_swScale)(SwsContext *c)
|
||||
}
|
||||
|
||||
if (c->srcBpc == 8 && c->dstBpc <= 14) {
|
||||
// Use the new MMX scaler if the MMX2 one can't be used (it is faster than the x86 ASM one).
|
||||
// Use the new MMX scaler if the MMXEXT one can't be used (it is faster than the x86 ASM one).
|
||||
#if COMPILE_TEMPLATE_MMXEXT
|
||||
if (c->flags & SWS_FAST_BILINEAR && c->canMMX2BeUsed)
|
||||
{
|
||||
|
@ -3,7 +3,7 @@
|
||||
*
|
||||
* Copyright (C) 2009 Konstantin Shishkov
|
||||
*
|
||||
* MMX/MMX2 template stuff (needed for fast movntq support),
|
||||
* MMX/MMXEXT template stuff (needed for fast movntq support),
|
||||
* 1,4,8bpp support and context / deglobalize stuff
|
||||
* by Michael Niedermayer (michaelni@gmx.at)
|
||||
*
|
||||
@ -58,7 +58,7 @@ DECLARE_ASM_CONST(8, uint64_t, pb_07) = 0x0707070707070707ULL;
|
||||
#include "yuv2rgb_template.c"
|
||||
#endif /* HAVE_MMX_INLINE */
|
||||
|
||||
//MMX2 versions
|
||||
// MMXEXT versions
|
||||
#if HAVE_MMXEXT_INLINE
|
||||
#undef RENAME
|
||||
#undef COMPILE_TEMPLATE_MMXEXT
|
||||
|
Loading…
Reference in New Issue
Block a user