* [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions
@ 2022-11-06 16:27 Andreas Rheinhardt
2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 2/7] avcodec/x86/cavsdsp: Remove unused 3DNow-macro Andreas Rheinhardt
` (6 more replies)
0 siblings, 7 replies; 9+ messages in thread
From: Andreas Rheinhardt @ 2022-11-06 16:27 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Andreas Rheinhardt
postprocess.c currently has C, MMX, MMXEXT, 3DNow as well as
SSE2 versions of its internal functions. But given that only
ancient 32-bit x86 CPUs don't support SSE2, the MMX, MMXEXT
and 3DNow versions are obsolete and are therefore removed by
this commit. This saves about 56KB here.
(The SSE2 version in particular is not really complete,
so that it often falls back to MMXEXT (which means that
there were some identical (apart from the name) MMXEXT
and SSE2 functions; this duplication no longer exists
with this commit.)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
---
The PP_CPU_CAPS_MMX(2)/3DNOW could now be deprecated.
libpostproc/postprocess.c | 69 ++--
libpostproc/postprocess_template.c | 521 ++---------------------------
2 files changed, 57 insertions(+), 533 deletions(-)
diff --git a/libpostproc/postprocess.c b/libpostproc/postprocess.c
index 383c691cb4..0586e458b4 100644
--- a/libpostproc/postprocess.c
+++ b/libpostproc/postprocess.c
@@ -26,28 +26,27 @@
*/
/*
- C MMX MMX2 3DNow AltiVec
-isVertDC Ec Ec Ec
-isVertMinMaxOk Ec Ec Ec
-doVertLowPass E e e Ec
-doVertDefFilter Ec Ec e e Ec
-isHorizDC Ec Ec Ec
-isHorizMinMaxOk a E Ec
-doHorizLowPass E e e Ec
-doHorizDefFilter Ec Ec e e Ec
-do_a_deblock Ec E Ec E
-deRing E e e* Ecp
-Vertical RKAlgo1 E a a
-Horizontal RKAlgo1 a a
-Vertical X1# a E E
-Horizontal X1# a E E
-LinIpolDeinterlace e E E*
-CubicIpolDeinterlace a e e*
-LinBlendDeinterlace e E E*
+ C MMX MMX2 AltiVec
+isVertDC Ec Ec Ec
+isVertMinMaxOk Ec Ec Ec
+doVertLowPass E e Ec
+doVertDefFilter Ec Ec e Ec
+isHorizDC Ec Ec Ec
+isHorizMinMaxOk a E Ec
+doHorizLowPass E e Ec
+doHorizDefFilter Ec Ec e Ec
+do_a_deblock Ec E Ec
+deRing E e Ecp
+Vertical RKAlgo1 E a
+Horizontal RKAlgo1 a
+Vertical X1# a E
+Horizontal X1# a E
+LinIpolDeinterlace e E
+CubicIpolDeinterlace a e
+LinBlendDeinterlace e E
MedianDeinterlace# E Ec Ec
-TempDeNoiser# E e e Ec
+TempDeNoiser# E e Ec
-* I do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work
# more or less selfinvented filters so the exactness is not too meaningful
E = Exact implementation
e = almost exact implementation (slightly different rounding,...)
@@ -83,7 +82,6 @@ try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
#include <stdlib.h>
#include <string.h>
//#undef HAVE_MMXEXT_INLINE
-//#define HAVE_AMD3DNOW_INLINE
//#undef HAVE_MMX_INLINE
//#undef ARCH_X86
//#define DEBUG_BRIGHTNESS
@@ -494,7 +492,7 @@ static av_always_inline void do_a_deblock_C(uint8_t *src, int step,
}
}
-//Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
+//Note: we have C and SSE2 version (which uses MMX(EXT) when advantageous)
//Plain C versions
//we always compile C for testing which needs bitexactness
#define TEMPLATE_PP_C 1
@@ -508,27 +506,12 @@ static av_always_inline void do_a_deblock_C(uint8_t *src, int step,
#if ARCH_X86 && HAVE_INLINE_ASM
# if CONFIG_RUNTIME_CPUDETECT
-# define TEMPLATE_PP_MMX 1
-# include "postprocess_template.c"
-# define TEMPLATE_PP_MMXEXT 1
-# include "postprocess_template.c"
-# define TEMPLATE_PP_3DNOW 1
-# include "postprocess_template.c"
# define TEMPLATE_PP_SSE2 1
# include "postprocess_template.c"
# else
# if HAVE_SSE2_INLINE
# define TEMPLATE_PP_SSE2 1
# include "postprocess_template.c"
-# elif HAVE_MMXEXT_INLINE
-# define TEMPLATE_PP_MMXEXT 1
-# include "postprocess_template.c"
-# elif HAVE_AMD3DNOW_INLINE
-# define TEMPLATE_PP_3DNOW 1
-# include "postprocess_template.c"
-# elif HAVE_MMX_INLINE
-# define TEMPLATE_PP_MMX 1
-# include "postprocess_template.c"
# endif
# endif
#endif
@@ -549,21 +532,12 @@ static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[]
#if ARCH_X86 && HAVE_INLINE_ASM
// ordered per speed fastest first
if (c->cpuCaps & AV_CPU_FLAG_SSE2) pp = postProcess_SSE2;
- else if (c->cpuCaps & AV_CPU_FLAG_MMXEXT) pp = postProcess_MMX2;
- else if (c->cpuCaps & AV_CPU_FLAG_3DNOW) pp = postProcess_3DNow;
- else if (c->cpuCaps & AV_CPU_FLAG_MMX) pp = postProcess_MMX;
#elif HAVE_ALTIVEC
if (c->cpuCaps & AV_CPU_FLAG_ALTIVEC) pp = postProcess_altivec;
#endif
#else /* CONFIG_RUNTIME_CPUDETECT */
#if HAVE_SSE2_INLINE
pp = postProcess_SSE2;
-#elif HAVE_MMXEXT_INLINE
- pp = postProcess_MMX2;
-#elif HAVE_AMD3DNOW_INLINE
- pp = postProcess_3DNow;
-#elif HAVE_MMX_INLINE
- pp = postProcess_MMX;
#elif HAVE_ALTIVEC
pp = postProcess_altivec;
#endif
@@ -877,9 +851,6 @@ av_cold pp_context *pp_get_context(int width, int height, int cpuCaps){
c->cpuCaps = av_get_cpu_flags();
} else {
c->cpuCaps = 0;
- if (cpuCaps & PP_CPU_CAPS_MMX) c->cpuCaps |= AV_CPU_FLAG_MMX;
- if (cpuCaps & PP_CPU_CAPS_MMX2) c->cpuCaps |= AV_CPU_FLAG_MMXEXT;
- if (cpuCaps & PP_CPU_CAPS_3DNOW) c->cpuCaps |= AV_CPU_FLAG_3DNOW;
if (cpuCaps & PP_CPU_CAPS_ALTIVEC) c->cpuCaps |= AV_CPU_FLAG_ALTIVEC;
}
diff --git a/libpostproc/postprocess_template.c b/libpostproc/postprocess_template.c
index 9f76b7c587..bcf7bdad66 100644
--- a/libpostproc/postprocess_template.c
+++ b/libpostproc/postprocess_template.c
@@ -20,7 +20,7 @@
/**
* @file
- * mmx/mmx2/3dnow postprocess code.
+ * mmx/mmx2/sse2 postprocess code.
*/
#include "libavutil/mem_internal.h"
@@ -57,14 +57,6 @@
# define TEMPLATE_PP_MMXEXT 0
#endif
-#ifdef TEMPLATE_PP_3DNOW
-# undef TEMPLATE_PP_MMX
-# define TEMPLATE_PP_MMX 1
-# define RENAME(a) a ## _3DNow
-#else
-# define TEMPLATE_PP_3DNOW 0
-#endif
-
#ifdef TEMPLATE_PP_SSE2
# undef TEMPLATE_PP_MMX
# define TEMPLATE_PP_MMX 1
@@ -82,30 +74,19 @@
#if TEMPLATE_PP_MMXEXT
#define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
-#elif TEMPLATE_PP_3DNOW
-#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
#endif
#define PAVGB(a,b) REAL_PAVGB(a,b)
#if TEMPLATE_PP_MMXEXT
#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
-#elif TEMPLATE_PP_MMX
-#define PMINUB(b,a,t) \
- "movq " #a ", " #t " \n\t"\
- "psubusb " #b ", " #t " \n\t"\
- "psubb " #t ", " #a " \n\t"
#endif
#if TEMPLATE_PP_MMXEXT
#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
-#elif TEMPLATE_PP_MMX
-#define PMAXUB(a,b) \
- "psubusb " #a ", " #b " \n\t"\
- "paddb " #a ", " #b " \n\t"
#endif
//FIXME? |255-0| = 1 (should not be a problem ...)
-#if TEMPLATE_PP_MMX
+#if TEMPLATE_PP_MMXEXT
/**
* Check if the middle 8x8 Block in the given 8x16 block is flat
*/
@@ -185,20 +166,8 @@ static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContex
"psubusb %%mm3, %%mm4 \n\t"
" \n\t"
-#if TEMPLATE_PP_MMXEXT
"pxor %%mm7, %%mm7 \n\t"
"psadbw %%mm7, %%mm0 \n\t"
-#else
- "movq %%mm0, %%mm1 \n\t"
- "psrlw $8, %%mm0 \n\t"
- "paddb %%mm1, %%mm0 \n\t"
- "movq %%mm0, %%mm1 \n\t"
- "psrlq $16, %%mm0 \n\t"
- "paddb %%mm1, %%mm0 \n\t"
- "movq %%mm0, %%mm1 \n\t"
- "psrlq $32, %%mm0 \n\t"
- "paddb %%mm1, %%mm0 \n\t"
-#endif
"movq %4, %%mm7 \n\t" // QP,..., QP
"paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
"psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> 0
@@ -219,7 +188,7 @@ static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContex
return 2;
}
}
-#endif //TEMPLATE_PP_MMX
+#endif //TEMPLATE_PP_MMXEXT
/**
* Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
@@ -228,7 +197,7 @@ static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContex
#if !TEMPLATE_PP_ALTIVEC
static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
{
-#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_MMXEXT
src+= stride*3;
__asm__ volatile( //"movv %0 %1 %2\n\t"
"movq %2, %%mm0 \n\t" // QP,..., QP
@@ -355,7 +324,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
: "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
: "%"FF_REG_a, "%"FF_REG_c
);
-#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#else //TEMPLATE_PP_MMXEXT
const int l1= stride;
const int l2= stride + l1;
const int l3= stride + l2;
@@ -394,7 +363,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
src++;
}
-#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_MMXEXT
}
#endif //TEMPLATE_PP_ALTIVEC
@@ -407,7 +376,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
*/
static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
{
-#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_MMXEXT
src+= stride*3;
__asm__ volatile(
@@ -494,7 +463,7 @@ static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
NAMED_CONSTRAINTS_ADD(b01)
: "%"FF_REG_a, "%"FF_REG_c
);
-#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#else //TEMPLATE_PP_MMXEXT
const int l1= stride;
const int l2= stride + l1;
@@ -528,13 +497,13 @@ static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
}
src++;
}
-#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_MMXEXT
}
#if !TEMPLATE_PP_ALTIVEC
static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
{
-#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_MMXEXT
/*
uint8_t tmp[16];
const int l1= stride;
@@ -816,239 +785,7 @@ static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext
}
}
*/
-#elif TEMPLATE_PP_MMX
- DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars
- src+= stride*4;
- __asm__ volatile(
- "pxor %%mm7, %%mm7 \n\t"
-// 0 1 2 3 4 5 6 7
-// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1
-// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1
-
- "movq (%0), %%mm0 \n\t"
- "movq %%mm0, %%mm1 \n\t"
- "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
- "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
-
- "movq (%0, %1), %%mm2 \n\t"
- "lea (%0, %1, 2), %%"FF_REG_a" \n\t"
- "movq %%mm2, %%mm3 \n\t"
- "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
- "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
-
- "movq (%%"FF_REG_a"), %%mm4 \n\t"
- "movq %%mm4, %%mm5 \n\t"
- "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
- "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
-
- "paddw %%mm0, %%mm0 \n\t" // 2L0
- "paddw %%mm1, %%mm1 \n\t" // 2H0
- "psubw %%mm4, %%mm2 \n\t" // L1 - L2
- "psubw %%mm5, %%mm3 \n\t" // H1 - H2
- "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
- "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
-
- "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
- "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
- "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
- "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
-
- "movq (%%"FF_REG_a", %1), %%mm2 \n\t"
- "movq %%mm2, %%mm3 \n\t"
- "punpcklbw %%mm7, %%mm2 \n\t" // L3
- "punpckhbw %%mm7, %%mm3 \n\t" // H3
-
- "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
- "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
- "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
- "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
- "movq %%mm0, (%3) \n\t" // 2L0 - 5L1 + 5L2 - 2L3
- "movq %%mm1, 8(%3) \n\t" // 2H0 - 5H1 + 5H2 - 2H3
-
- "movq (%%"FF_REG_a", %1, 2), %%mm0 \n\t"
- "movq %%mm0, %%mm1 \n\t"
- "punpcklbw %%mm7, %%mm0 \n\t" // L4
- "punpckhbw %%mm7, %%mm1 \n\t" // H4
-
- "psubw %%mm0, %%mm2 \n\t" // L3 - L4
- "psubw %%mm1, %%mm3 \n\t" // H3 - H4
- "movq %%mm2, 16(%3) \n\t" // L3 - L4
- "movq %%mm3, 24(%3) \n\t" // H3 - H4
- "paddw %%mm4, %%mm4 \n\t" // 2L2
- "paddw %%mm5, %%mm5 \n\t" // 2H2
- "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
- "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
-
- "lea (%%"FF_REG_a", %1), %0 \n\t"
- "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
- "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
- "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
- "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
-//50 opcodes so far
- "movq (%0, %1, 2), %%mm2 \n\t"
- "movq %%mm2, %%mm3 \n\t"
- "punpcklbw %%mm7, %%mm2 \n\t" // L5
- "punpckhbw %%mm7, %%mm3 \n\t" // H5
- "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
- "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
- "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
- "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
-
- "movq (%%"FF_REG_a", %1, 4), %%mm6 \n\t"
- "punpcklbw %%mm7, %%mm6 \n\t" // L6
- "psubw %%mm6, %%mm2 \n\t" // L5 - L6
- "movq (%%"FF_REG_a", %1, 4), %%mm6 \n\t"
- "punpckhbw %%mm7, %%mm6 \n\t" // H6
- "psubw %%mm6, %%mm3 \n\t" // H5 - H6
-
- "paddw %%mm0, %%mm0 \n\t" // 2L4
- "paddw %%mm1, %%mm1 \n\t" // 2H4
- "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
- "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
-
- "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
- "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
- "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
- "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
-
- "movq (%0, %1, 4), %%mm2 \n\t"
- "movq %%mm2, %%mm3 \n\t"
- "punpcklbw %%mm7, %%mm2 \n\t" // L7
- "punpckhbw %%mm7, %%mm3 \n\t" // H7
-
- "paddw %%mm2, %%mm2 \n\t" // 2L7
- "paddw %%mm3, %%mm3 \n\t" // 2H7
- "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
- "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
-
- "movq (%3), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
- "movq 8(%3), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
-
-#if TEMPLATE_PP_MMXEXT
- "movq %%mm7, %%mm6 \n\t" // 0
- "psubw %%mm0, %%mm6 \n\t"
- "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
- "movq %%mm7, %%mm6 \n\t" // 0
- "psubw %%mm1, %%mm6 \n\t"
- "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
- "movq %%mm7, %%mm6 \n\t" // 0
- "psubw %%mm2, %%mm6 \n\t"
- "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
- "movq %%mm7, %%mm6 \n\t" // 0
- "psubw %%mm3, %%mm6 \n\t"
- "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
-#else
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm0, %%mm6 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm1, %%mm6 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm2, %%mm6 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm3, %%mm6 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
-#endif
-
-#if TEMPLATE_PP_MMXEXT
- "pminsw %%mm2, %%mm0 \n\t"
- "pminsw %%mm3, %%mm1 \n\t"
-#else
- "movq %%mm0, %%mm6 \n\t"
- "psubusw %%mm2, %%mm6 \n\t"
- "psubw %%mm6, %%mm0 \n\t"
- "movq %%mm1, %%mm6 \n\t"
- "psubusw %%mm3, %%mm6 \n\t"
- "psubw %%mm6, %%mm1 \n\t"
-#endif
-
- "movd %2, %%mm2 \n\t" // QP
- "punpcklbw %%mm7, %%mm2 \n\t"
-
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
- "pxor %%mm6, %%mm4 \n\t"
- "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
- "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
- "pxor %%mm7, %%mm5 \n\t"
- "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
-// 100 opcodes
- "psllw $3, %%mm2 \n\t" // 8QP
- "movq %%mm2, %%mm3 \n\t" // 8QP
- "pcmpgtw %%mm4, %%mm2 \n\t"
- "pcmpgtw %%mm5, %%mm3 \n\t"
- "pand %%mm2, %%mm4 \n\t"
- "pand %%mm3, %%mm5 \n\t"
-
-
- "psubusw %%mm0, %%mm4 \n\t" // hd
- "psubusw %%mm1, %%mm5 \n\t" // ld
-
-
- "movq "MANGLE(w05)", %%mm2 \n\t" // 5
- "pmullw %%mm2, %%mm4 \n\t"
- "pmullw %%mm2, %%mm5 \n\t"
- "movq "MANGLE(w20)", %%mm2 \n\t" // 32
- "paddw %%mm2, %%mm4 \n\t"
- "paddw %%mm2, %%mm5 \n\t"
- "psrlw $6, %%mm4 \n\t"
- "psrlw $6, %%mm5 \n\t"
-
- "movq 16(%3), %%mm0 \n\t" // L3 - L4
- "movq 24(%3), %%mm1 \n\t" // H3 - H4
-
- "pxor %%mm2, %%mm2 \n\t"
- "pxor %%mm3, %%mm3 \n\t"
-
- "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
- "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
- "pxor %%mm2, %%mm0 \n\t"
- "pxor %%mm3, %%mm1 \n\t"
- "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
- "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
- "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
- "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
-
- "pxor %%mm6, %%mm2 \n\t"
- "pxor %%mm7, %%mm3 \n\t"
- "pand %%mm2, %%mm4 \n\t"
- "pand %%mm3, %%mm5 \n\t"
-
-#if TEMPLATE_PP_MMXEXT
- "pminsw %%mm0, %%mm4 \n\t"
- "pminsw %%mm1, %%mm5 \n\t"
-#else
- "movq %%mm4, %%mm2 \n\t"
- "psubusw %%mm0, %%mm2 \n\t"
- "psubw %%mm2, %%mm4 \n\t"
- "movq %%mm5, %%mm2 \n\t"
- "psubusw %%mm1, %%mm2 \n\t"
- "psubw %%mm2, %%mm5 \n\t"
-#endif
- "pxor %%mm6, %%mm4 \n\t"
- "pxor %%mm7, %%mm5 \n\t"
- "psubw %%mm6, %%mm4 \n\t"
- "psubw %%mm7, %%mm5 \n\t"
- "packsswb %%mm5, %%mm4 \n\t"
- "movq (%0), %%mm0 \n\t"
- "paddb %%mm4, %%mm0 \n\t"
- "movq %%mm0, (%0) \n\t"
- "movq (%0, %1), %%mm0 \n\t"
- "psubb %%mm4, %%mm0 \n\t"
- "movq %%mm0, (%0, %1) \n\t"
-
- : "+r" (src)
- : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp)
- NAMED_CONSTRAINTS_ADD(w05,w20)
- : "%"FF_REG_a
- );
-#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#else //TEMPLATE_PP_MMXEXT
const int l1= stride;
const int l2= stride + l1;
const int l3= stride + l2;
@@ -1086,14 +823,14 @@ static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext
}
src++;
}
-#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_MMXEXT
}
#endif //TEMPLATE_PP_ALTIVEC
#if !TEMPLATE_PP_ALTIVEC
static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
{
-#if HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW)
+#if HAVE_7REGS && TEMPLATE_PP_MMXEXT
DECLARE_ALIGNED(8, uint64_t, tmp)[3];
__asm__ volatile(
"pxor %%mm6, %%mm6 \n\t"
@@ -1113,20 +850,10 @@ static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
#undef REAL_FIND_MIN_MAX
#undef FIND_MIN_MAX
-#if TEMPLATE_PP_MMXEXT
#define REAL_FIND_MIN_MAX(addr)\
"movq " #addr ", %%mm0 \n\t"\
"pminub %%mm0, %%mm7 \n\t"\
"pmaxub %%mm0, %%mm6 \n\t"
-#else
-#define REAL_FIND_MIN_MAX(addr)\
- "movq " #addr ", %%mm0 \n\t"\
- "movq %%mm7, %%mm1 \n\t"\
- "psubusb %%mm0, %%mm6 \n\t"\
- "paddb %%mm0, %%mm6 \n\t"\
- "psubusb %%mm0, %%mm1 \n\t"\
- "psubb %%mm1, %%mm7 \n\t"
-#endif
#define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr)
FIND_MIN_MAX((%%FF_REGa))
@@ -1140,49 +867,20 @@ FIND_MIN_MAX((%0, %1, 8))
"movq %%mm7, %%mm4 \n\t"
"psrlq $8, %%mm7 \n\t"
-#if TEMPLATE_PP_MMXEXT
"pminub %%mm4, %%mm7 \n\t" // min of pixels
"pshufw $0xF9, %%mm7, %%mm4 \n\t"
"pminub %%mm4, %%mm7 \n\t" // min of pixels
"pshufw $0xFE, %%mm7, %%mm4 \n\t"
"pminub %%mm4, %%mm7 \n\t"
-#else
- "movq %%mm7, %%mm1 \n\t"
- "psubusb %%mm4, %%mm1 \n\t"
- "psubb %%mm1, %%mm7 \n\t"
- "movq %%mm7, %%mm4 \n\t"
- "psrlq $16, %%mm7 \n\t"
- "movq %%mm7, %%mm1 \n\t"
- "psubusb %%mm4, %%mm1 \n\t"
- "psubb %%mm1, %%mm7 \n\t"
- "movq %%mm7, %%mm4 \n\t"
- "psrlq $32, %%mm7 \n\t"
- "movq %%mm7, %%mm1 \n\t"
- "psubusb %%mm4, %%mm1 \n\t"
- "psubb %%mm1, %%mm7 \n\t"
-#endif
"movq %%mm6, %%mm4 \n\t"
"psrlq $8, %%mm6 \n\t"
-#if TEMPLATE_PP_MMXEXT
"pmaxub %%mm4, %%mm6 \n\t" // max of pixels
"pshufw $0xF9, %%mm6, %%mm4 \n\t"
"pmaxub %%mm4, %%mm6 \n\t"
"pshufw $0xFE, %%mm6, %%mm4 \n\t"
"pmaxub %%mm4, %%mm6 \n\t"
-#else
- "psubusb %%mm4, %%mm6 \n\t"
- "paddb %%mm4, %%mm6 \n\t"
- "movq %%mm6, %%mm4 \n\t"
- "psrlq $16, %%mm6 \n\t"
- "psubusb %%mm4, %%mm6 \n\t"
- "paddb %%mm4, %%mm6 \n\t"
- "movq %%mm6, %%mm4 \n\t"
- "psrlq $32, %%mm6 \n\t"
- "psubusb %%mm4, %%mm6 \n\t"
- "paddb %%mm4, %%mm6 \n\t"
-#endif
"movq %%mm6, %%mm0 \n\t" // max
"psubb %%mm7, %%mm6 \n\t" // max - min
"push %%"FF_REG_a" \n\t"
@@ -1320,7 +1018,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,
NAMED_CONSTRAINTS_ADD(deringThreshold,b00,b02,b08)
: "%"FF_REG_a, "%"FF_REG_d
);
-#else // HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW)
+#else // HAVE_7REGS && TEMPLATE_PP_MMXEXT
int y;
int min=255;
int max=0;
@@ -1438,7 +1136,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,
// src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
}
#endif
-#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_MMXEXT
}
#endif //TEMPLATE_PP_ALTIVEC
@@ -1450,7 +1148,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,
*/
static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
{
-#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_MMXEXT
src+= 4*stride;
__asm__ volatile(
"lea (%0, %1), %%"FF_REG_a" \n\t"
@@ -1503,14 +1201,13 @@ static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int strid
*/
static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
{
-#if TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_SSE2
src+= stride*3;
__asm__ volatile(
"lea (%0, %1), %%"FF_REG_a" \n\t"
"lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t"
"lea (%%"FF_REG_d", %1, 4), %%"FF_REG_c"\n\t"
"add %1, %%"FF_REG_c" \n\t"
-#if TEMPLATE_PP_SSE2
"pxor %%xmm7, %%xmm7 \n\t"
#define REAL_DEINT_CUBIC(a,b,c,d,e)\
"movq " #a ", %%xmm0 \n\t"\
@@ -1526,33 +1223,6 @@ static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride
"psubw %%xmm0, %%xmm1 \n\t"\
"packuswb %%xmm1, %%xmm1 \n\t"\
"movlps %%xmm1, " #c " \n\t"
-#else //TEMPLATE_PP_SSE2
- "pxor %%mm7, %%mm7 \n\t"
-// 0 1 2 3 4 5 6 7 8 9 10
-// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
-
-#define REAL_DEINT_CUBIC(a,b,c,d,e)\
- "movq " #a ", %%mm0 \n\t"\
- "movq " #b ", %%mm1 \n\t"\
- "movq " #d ", %%mm2 \n\t"\
- "movq " #e ", %%mm3 \n\t"\
- PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
- PAVGB(%%mm3, %%mm0) /* (a+e) /2 */\
- "movq %%mm0, %%mm2 \n\t"\
- "punpcklbw %%mm7, %%mm0 \n\t"\
- "punpckhbw %%mm7, %%mm2 \n\t"\
- "movq %%mm1, %%mm3 \n\t"\
- "punpcklbw %%mm7, %%mm1 \n\t"\
- "punpckhbw %%mm7, %%mm3 \n\t"\
- "psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
- "psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
- "psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
- "psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
- "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
- "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
- "packuswb %%mm3, %%mm1 \n\t"\
- "movq %%mm1, " #c " \n\t"
-#endif //TEMPLATE_PP_SSE2
#define DEINT_CUBIC(a,b,c,d,e) REAL_DEINT_CUBIC(a,b,c,d,e)
DEINT_CUBIC((%0) , (%%FF_REGa, %1), (%%FF_REGa, %1, 2), (%0, %1, 4) , (%%FF_REGd, %1))
@@ -1562,13 +1232,11 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , (%%FF_REGd, %1, 4), (%%FF_REGc)
: : "r" (src), "r" ((x86_reg)stride)
:
-#if TEMPLATE_PP_SSE2
XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm7",)
-#endif
"%"FF_REG_a, "%"FF_REG_d, "%"FF_REG_c
);
#undef REAL_DEINT_CUBIC
-#else //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#else //TEMPLATE_PP_SSE2
int x;
src+= stride*3;
for(x=0; x<8; x++){
@@ -1578,7 +1246,7 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , (%%FF_REGd, %1, 4), (%%FF_REGc)
src[stride*9] = av_clip_uint8((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
src++;
}
-#endif //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_SSE2
}
/**
@@ -1590,7 +1258,7 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , (%%FF_REGd, %1, 4), (%%FF_REGc)
*/
static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
{
-#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_MMXEXT
src+= stride*4;
__asm__ volatile(
"lea (%0, %1), %%"FF_REG_a" \n\t"
@@ -1639,7 +1307,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4
: : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
: "%"FF_REG_a, "%"FF_REG_d
);
-#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#else //TEMPLATE_PP_MMXEXT
int x;
src+= stride*4;
for(x=0; x<8; x++){
@@ -1657,7 +1325,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4
src++;
}
-#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_MMXEXT
}
/**
@@ -1669,7 +1337,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4
*/
static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
{
-#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
+#if TEMPLATE_PP_MMXEXT && HAVE_6REGS
src+= stride*4;
__asm__ volatile(
"lea (%0, %1), %%"FF_REG_a" \n\t"
@@ -1729,7 +1397,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4
: : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
: "%"FF_REG_a, "%"FF_REG_d
);
-#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
+#else //TEMPLATE_PP_MMXEXT && HAVE_6REGS
int x;
src+= stride*4;
for(x=0; x<8; x++){
@@ -1758,7 +1426,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4
src++;
}
-#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
+#endif // TEMPLATE_PP_MMXEXT && HAVE_6REGS
}
/**
@@ -1770,7 +1438,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4
*/
static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
{
-#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#if TEMPLATE_PP_MMXEXT
src+= 4*stride;
__asm__ volatile(
"lea (%0, %1), %%"FF_REG_a" \n\t"
@@ -1817,7 +1485,7 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin
: : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
: "%"FF_REG_a, "%"FF_REG_d
);
-#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#else //TEMPLATE_PP_MMXEXT
int a, b, c, x;
src+= 4*stride;
@@ -1860,7 +1528,7 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin
src += 4;
tmp += 4;
}
-#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW
+#endif //TEMPLATE_PP_MMXEXT
}
/**
@@ -1871,9 +1539,8 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin
*/
static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
{
-#if TEMPLATE_PP_MMX
- src+= 4*stride;
#if TEMPLATE_PP_MMXEXT
+ src+= 4*stride;
__asm__ volatile(
"lea (%0, %1), %%"FF_REG_a" \n\t"
"lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t"
@@ -1922,48 +1589,6 @@ static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
: "%"FF_REG_a, "%"FF_REG_d
);
-#else // MMX without MMX2
- __asm__ volatile(
- "lea (%0, %1), %%"FF_REG_a" \n\t"
- "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t"
-// 0 1 2 3 4 5 6 7 8 9
-// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
- "pxor %%mm7, %%mm7 \n\t"
-
-#define REAL_MEDIAN(a,b,c)\
- "movq " #a ", %%mm0 \n\t"\
- "movq " #b ", %%mm2 \n\t"\
- "movq " #c ", %%mm1 \n\t"\
- "movq %%mm0, %%mm3 \n\t"\
- "movq %%mm1, %%mm4 \n\t"\
- "movq %%mm2, %%mm5 \n\t"\
- "psubusb %%mm1, %%mm3 \n\t"\
- "psubusb %%mm2, %%mm4 \n\t"\
- "psubusb %%mm0, %%mm5 \n\t"\
- "pcmpeqb %%mm7, %%mm3 \n\t"\
- "pcmpeqb %%mm7, %%mm4 \n\t"\
- "pcmpeqb %%mm7, %%mm5 \n\t"\
- "movq %%mm3, %%mm6 \n\t"\
- "pxor %%mm4, %%mm3 \n\t"\
- "pxor %%mm5, %%mm4 \n\t"\
- "pxor %%mm6, %%mm5 \n\t"\
- "por %%mm3, %%mm1 \n\t"\
- "por %%mm4, %%mm2 \n\t"\
- "por %%mm5, %%mm0 \n\t"\
- "pand %%mm2, %%mm0 \n\t"\
- "pand %%mm1, %%mm0 \n\t"\
- "movq %%mm0, " #b " \n\t"
-#define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c)
-
-MEDIAN((%0) , (%%FF_REGa) , (%%FF_REGa, %1))
-MEDIAN((%%FF_REGa, %1), (%%FF_REGa, %1, 2), (%0, %1, 4))
-MEDIAN((%0, %1, 4) , (%%FF_REGd) , (%%FF_REGd, %1))
-MEDIAN((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8))
-
- : : "r" (src), "r" ((x86_reg)stride)
- : "%"FF_REG_a, "%"FF_REG_d
- );
-#endif //TEMPLATE_PP_MMXEXT
#else //TEMPLATE_PP_MMX
int x, y;
src+= 4*stride;
@@ -2165,7 +1790,7 @@ static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
#define FAST_L2_DIFF
//#define L1_DIFF //u should change the thresholds too if u try that one
-#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
+#if TEMPLATE_PP_MMXEXT && HAVE_6REGS
__asm__ volatile(
"lea (%2, %2, 2), %%"FF_REG_a" \n\t" // 3*stride
"lea (%2, %2, 4), %%"FF_REG_d" \n\t" // 5*stride
@@ -2454,7 +2079,7 @@ L2_DIFF_CORE((%0, %%FF_REGc) , (%1, %%FF_REGc))
NAMED_CONSTRAINTS_ADD(b80)
: "%"FF_REG_a, "%"FF_REG_d, "%"FF_REG_c, "memory"
);
-#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
+#else //TEMPLATE_PP_MMXEXT && HAVE_6REGS
{
int y;
int d=0;
@@ -2537,11 +2162,11 @@ Switch between
}
}
}
-#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS
+#endif //TEMPLATE_PP_MMXEXT && HAVE_6REGS
}
#endif //TEMPLATE_PP_ALTIVEC
-#if TEMPLATE_PP_MMX
+#if TEMPLATE_PP_MMXEXT
/**
* accurate deblock filter
*/
@@ -2945,7 +2570,6 @@ static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int st
"movq (%4), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
"movq 8(%4), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
-#if TEMPLATE_PP_MMXEXT
"movq %%mm7, %%mm6 \n\t" // 0
"psubw %%mm0, %%mm6 \n\t"
"pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
@@ -2958,36 +2582,9 @@ static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int st
"movq %%mm7, %%mm6 \n\t" // 0
"psubw %%mm3, %%mm6 \n\t"
"pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
-#else
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm0, %%mm6 \n\t"
- "pxor %%mm6, %%mm0 \n\t"
- "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm1, %%mm6 \n\t"
- "pxor %%mm6, %%mm1 \n\t"
- "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm2, %%mm6 \n\t"
- "pxor %%mm6, %%mm2 \n\t"
- "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
- "movq %%mm7, %%mm6 \n\t" // 0
- "pcmpgtw %%mm3, %%mm6 \n\t"
- "pxor %%mm6, %%mm3 \n\t"
- "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
-#endif
-#if TEMPLATE_PP_MMXEXT
"pminsw %%mm2, %%mm0 \n\t"
"pminsw %%mm3, %%mm1 \n\t"
-#else
- "movq %%mm0, %%mm6 \n\t"
- "psubusw %%mm2, %%mm6 \n\t"
- "psubw %%mm6, %%mm0 \n\t"
- "movq %%mm1, %%mm6 \n\t"
- "psubusw %%mm3, %%mm6 \n\t"
- "psubw %%mm6, %%mm1 \n\t"
-#endif
"movd %2, %%mm2 \n\t" // QP
"punpcklbw %%mm7, %%mm2 \n\t"
@@ -3041,17 +2638,8 @@ static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int st
"pand %%mm2, %%mm4 \n\t"
"pand %%mm3, %%mm5 \n\t"
-#if TEMPLATE_PP_MMXEXT
"pminsw %%mm0, %%mm4 \n\t"
"pminsw %%mm1, %%mm5 \n\t"
-#else
- "movq %%mm4, %%mm2 \n\t"
- "psubusw %%mm0, %%mm2 \n\t"
- "psubw %%mm2, %%mm4 \n\t"
- "movq %%mm5, %%mm2 \n\t"
- "psubusw %%mm1, %%mm2 \n\t"
- "psubw %%mm2, %%mm5 \n\t"
-#endif
"pxor %%mm6, %%mm4 \n\t"
"pxor %%mm7, %%mm5 \n\t"
"psubw %%mm6, %%mm4 \n\t"
@@ -3088,18 +2676,14 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
int levelFix, int64_t *packedOffsetAndScale)
{
-#if !TEMPLATE_PP_MMX || !HAVE_6REGS
- int i;
-#endif
if(levelFix){
-#if TEMPLATE_PP_MMX && HAVE_6REGS
+#if TEMPLATE_PP_MMXEXT && HAVE_6REGS
__asm__ volatile(
"movq (%%"FF_REG_a"), %%mm2 \n\t" // packedYOffset
"movq 8(%%"FF_REG_a"), %%mm3 \n\t" // packedYScale
"lea (%2,%4), %%"FF_REG_a" \n\t"
"lea (%3,%5), %%"FF_REG_d" \n\t"
"pxor %%mm4, %%mm4 \n\t"
-#if TEMPLATE_PP_MMXEXT
#define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
"movq " #src1 ", %%mm0 \n\t"\
"movq " #src1 ", %%mm5 \n\t"\
@@ -3122,34 +2706,6 @@ static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t
"movq %%mm0, " #dst1 " \n\t"\
"movq %%mm1, " #dst2 " \n\t"\
-#else //TEMPLATE_PP_MMXEXT
-#define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
- "movq " #src1 ", %%mm0 \n\t"\
- "movq " #src1 ", %%mm5 \n\t"\
- "punpcklbw %%mm4, %%mm0 \n\t"\
- "punpckhbw %%mm4, %%mm5 \n\t"\
- "psubw %%mm2, %%mm0 \n\t"\
- "psubw %%mm2, %%mm5 \n\t"\
- "movq " #src2 ", %%mm1 \n\t"\
- "psllw $6, %%mm0 \n\t"\
- "psllw $6, %%mm5 \n\t"\
- "pmulhw %%mm3, %%mm0 \n\t"\
- "movq " #src2 ", %%mm6 \n\t"\
- "pmulhw %%mm3, %%mm5 \n\t"\
- "punpcklbw %%mm4, %%mm1 \n\t"\
- "punpckhbw %%mm4, %%mm6 \n\t"\
- "psubw %%mm2, %%mm1 \n\t"\
- "psubw %%mm2, %%mm6 \n\t"\
- "psllw $6, %%mm1 \n\t"\
- "psllw $6, %%mm6 \n\t"\
- "pmulhw %%mm3, %%mm1 \n\t"\
- "pmulhw %%mm3, %%mm6 \n\t"\
- "packuswb %%mm5, %%mm0 \n\t"\
- "packuswb %%mm6, %%mm1 \n\t"\
- "movq %%mm0, " #dst1 " \n\t"\
- "movq %%mm1, " #dst2 " \n\t"\
-
-#endif //TEMPLATE_PP_MMXEXT
#define SCALED_CPY(src1, src2, dst1, dst2)\
REAL_SCALED_CPY(src1, src2, dst1, dst2)
@@ -3170,7 +2726,7 @@ SCALED_CPY((%%FF_REGa, %4), (%%FF_REGa, %4, 2), (%%FF_REGd, %5), (%%FF_REGd, %5,
: "%"FF_REG_d
);
#else //TEMPLATE_PP_MMX && HAVE_6REGS
- for(i=0; i<8; i++)
+ for (int i = 0; i < 8; i++)
memcpy( &(dst[dstStride*i]),
&(src[srcStride*i]), BLOCK_SIZE);
#endif //TEMPLATE_PP_MMX && HAVE_6REGS
@@ -3203,7 +2759,7 @@ SIMPLE_CPY((%%FF_REGa, %2), (%%FF_REGa, %2, 2), (%%FF_REGd, %3), (%%FF_REGd, %3,
: "%"FF_REG_a, "%"FF_REG_d
);
#else //TEMPLATE_PP_MMX && HAVE_6REGS
- for(i=0; i<8; i++)
+ for (int i = 0; i < 8; i++)
memcpy( &(dst[dstStride*i]),
&(src[srcStride*i]), BLOCK_SIZE);
#endif //TEMPLATE_PP_MMX && HAVE_6REGS
@@ -3696,9 +3252,7 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
}
}
}
-#if TEMPLATE_PP_3DNOW
- __asm__ volatile("femms");
-#elif TEMPLATE_PP_MMX
+#if TEMPLATE_PP_MMX
__asm__ volatile("emms");
#endif
@@ -3734,5 +3288,4 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
#undef TEMPLATE_PP_ALTIVEC
#undef TEMPLATE_PP_MMX
#undef TEMPLATE_PP_MMXEXT
-#undef TEMPLATE_PP_3DNOW
#undef TEMPLATE_PP_SSE2
--
2.34.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 9+ messages in thread* [FFmpeg-devel] [PATCH 2/7] avcodec/x86/cavsdsp: Remove unused 3DNow-macro 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt @ 2022-11-06 16:30 ` Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 3/7] swscale/utils: Remove obsolete 3DNow reference Andreas Rheinhardt ` (5 subsequent siblings) 6 siblings, 0 replies; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-06 16:30 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Andreas Rheinhardt Forgotten in 3221aba87989742ea22b639a7bb4af69f4eaa0e7. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libavcodec/x86/cavsdsp.c | 4 ---- 1 file changed, 4 deletions(-) diff --git a/libavcodec/x86/cavsdsp.c b/libavcodec/x86/cavsdsp.c index 4ad977a034..96680836a1 100644 --- a/libavcodec/x86/cavsdsp.c +++ b/libavcodec/x86/cavsdsp.c @@ -308,10 +308,6 @@ static void OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uin }\ #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t" -#define AVG_3DNOW_OP(a,b,temp, size) \ -"mov" #size " " #b ", " #temp " \n\t"\ -"pavgusb " #temp ", " #a " \n\t"\ -"mov" #size " " #a ", " #b " \n\t" #define AVG_MMXEXT_OP(a, b, temp, size) \ "mov" #size " " #b ", " #temp " \n\t"\ "pavgb " #temp ", " #a " \n\t"\ -- 2.34.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* [FFmpeg-devel] [PATCH 3/7] swscale/utils: Remove obsolete 3DNow reference 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 2/7] avcodec/x86/cavsdsp: Remove unused 3DNow-macro Andreas Rheinhardt @ 2022-11-06 16:30 ` Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 4/7] postproc/postprocess_template: Fix alignment Andreas Rheinhardt ` (4 subsequent siblings) 6 siblings, 0 replies; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-06 16:30 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Andreas Rheinhardt swscale does not use 3DNow any more since commit 608319a311a31f7d85333a7b08286c00be38eab6. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libswscale/utils.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/libswscale/utils.c b/libswscale/utils.c index 45baa22b23..85640a143f 100644 --- a/libswscale/utils.c +++ b/libswscale/utils.c @@ -1966,8 +1966,6 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter, if (INLINE_MMXEXT(cpu_flags)) cpucaps = "MMXEXT"; - else if (INLINE_AMD3DNOW(cpu_flags)) - cpucaps = "3DNOW"; else if (INLINE_MMX(cpu_flags)) cpucaps = "MMX"; else if (PPC_ALTIVEC(cpu_flags)) -- 2.34.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* [FFmpeg-devel] [PATCH 4/7] postproc/postprocess_template: Fix alignment 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 2/7] avcodec/x86/cavsdsp: Remove unused 3DNow-macro Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 3/7] swscale/utils: Remove obsolete 3DNow reference Andreas Rheinhardt @ 2022-11-06 16:30 ` Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 5/7] avfilter/vf_qp: Don't store NAN in int Andreas Rheinhardt ` (3 subsequent siblings) 6 siblings, 0 replies; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-06 16:30 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Andreas Rheinhardt postProcess in postprocess_template.c copies a PPContext to the stack, works with this copy and then copies it back again. Said local copy uses a hardcoded alignment of eight, although PPContext has alignment 32 since cbe27006cee0099076d1d68af646f3ef914167d8 (this commit was in anticipation of AVX2 code that never landed). This leads to misalignment in the filter-(pp|pp1|pp2|pp3|qp) FATE-tests which UBSan complains about. So avoid the local copy. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libpostproc/postprocess_template.c | 156 ++++++++++++++--------------- 1 file changed, 77 insertions(+), 79 deletions(-) diff --git a/libpostproc/postprocess_template.c b/libpostproc/postprocess_template.c index bcf7bdad66..ade1d6ce2b 100644 --- a/libpostproc/postprocess_template.c +++ b/libpostproc/postprocess_template.c @@ -2860,14 +2860,13 @@ static inline void RENAME(prefetcht2)(const void *p) * Filter array of bytes (Y or U or V values) */ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height, - const int8_t QPs[], int QPStride, int isColor, PPContext *c2) + const int8_t QPs[], int QPStride, int isColor, PPContext *c) { - DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access int x,y; #ifdef TEMPLATE_PP_TIME_MODE const int mode= TEMPLATE_PP_TIME_MODE; #else - const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode; + const int mode = isColor ? c->ppMode.chromMode : c->ppMode.lumMode; #endif int black=0, white=255; // blackest black and whitest white in the picture int QPCorrecture= 256*256; @@ -2877,29 +2876,29 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ int i; #endif - const int qpHShift= isColor ? 4-c.hChromaSubSample : 4; - const int qpVShift= isColor ? 4-c.vChromaSubSample : 4; + const int qpHShift = isColor ? 4 - c->hChromaSubSample : 4; + const int qpVShift = isColor ? 4 - c->vChromaSubSample : 4; //FIXME remove - uint64_t * const yHistogram= c.yHistogram; - uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride; - uint8_t * const tempDst= (dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride) + 32; + uint64_t * const yHistogram= c->yHistogram; + uint8_t * const tempSrc = srcStride > 0 ? c->tempSrc : c->tempSrc - 23*srcStride; + uint8_t * const tempDst = (dstStride > 0 ? c->tempDst : c->tempDst - 23*dstStride) + 32; //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4; if (mode & VISUALIZE){ if(!(mode & (V_A_DEBLOCK | H_A_DEBLOCK)) || TEMPLATE_PP_MMX) { - av_log(c2, AV_LOG_WARNING, "Visualization is currently only supported with the accurate deblock filter without SIMD\n"); + av_log(c, AV_LOG_WARNING, "Visualization is currently only supported with the accurate deblock filter without SIMD\n"); } } #if TEMPLATE_PP_MMX for(i=0; i<57; i++){ - int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1; + int offset = ((i * c->ppMode.baseDcDiff) >> 8) + 1; int threshold= offset*2 + 1; - c.mmxDcOffset[i]= 0x7F - offset; - c.mmxDcThreshold[i]= 0x7F - threshold; - c.mmxDcOffset[i]*= 0x0101010101010101LL; - c.mmxDcThreshold[i]*= 0x0101010101010101LL; + c->mmxDcOffset[i] = 0x7F - offset; + c->mmxDcThreshold[i] = 0x7F - threshold; + c->mmxDcOffset[i] *= 0x0101010101010101LL; + c->mmxDcThreshold[i] *= 0x0101010101010101LL; } #endif @@ -2925,16 +2924,18 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ uint64_t clipped; AVRational scale; - c.frameNum++; + c->frameNum++; // first frame is fscked so we ignore it - if(c.frameNum == 1) yHistogram[0]= width*(uint64_t)height/64*15/256; + if (c->frameNum == 1) + yHistogram[0] = width * (uint64_t)height/64*15/256; for(i=0; i<256; i++){ sum+= yHistogram[i]; } /* We always get a completely black picture first. */ - maxClipped= av_rescale(sum, c.ppMode.maxClippedThreshold.num, c.ppMode.maxClippedThreshold.den); + maxClipped = av_rescale(sum, c->ppMode.maxClippedThreshold.num, + c->ppMode.maxClippedThreshold.den); clipped= sum; for(black=255; black>0; black--){ @@ -2948,27 +2949,27 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ clipped-= yHistogram[white]; } - scale = (AVRational){c.ppMode.maxAllowedY - c.ppMode.minAllowedY, white - black}; + scale = (AVRational){c->ppMode.maxAllowedY - c->ppMode.minAllowedY, white - black}; #if TEMPLATE_PP_MMXEXT - c.packedYScale = (uint16_t)av_rescale(scale.num, 256, scale.den); - c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF; + c->packedYScale = (uint16_t)av_rescale(scale.num, 256, scale.den); + c->packedYOffset = (((black*c->packedYScale)>>8) - c->ppMode.minAllowedY) & 0xFFFF; #else - c.packedYScale = (uint16_t)av_rescale(scale.num, 1024, scale.den); - c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF; + c->packedYScale = (uint16_t)av_rescale(scale.num, 1024, scale.den); + c->packedYOffset = (black - c->ppMode.minAllowedY) & 0xFFFF; #endif - c.packedYOffset|= c.packedYOffset<<32; - c.packedYOffset|= c.packedYOffset<<16; + c->packedYOffset |= c->packedYOffset<<32; + c->packedYOffset |= c->packedYOffset<<16; - c.packedYScale|= c.packedYScale<<32; - c.packedYScale|= c.packedYScale<<16; + c->packedYScale |= c->packedYScale<<32; + c->packedYScale |= c->packedYScale<<16; if(mode & LEVEL_FIX) QPCorrecture= (int)av_rescale(scale.num, 256*256, scale.den); else QPCorrecture= 256*256; }else{ - c.packedYScale= 0x0100010001000100LL; - c.packedYOffset= 0; + c->packedYScale = 0x0100010001000100LL; + c->packedYOffset = 0; QPCorrecture= 256*256; } @@ -2988,22 +2989,22 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead+1)*dstStride + 32); RENAME(blockCopy)(dstBlock + dstStride*8, dstStride, - srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset); + srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c->packedYOffset); RENAME(duplicate)(dstBlock + dstStride*8, dstStride); if(mode & LINEAR_IPOL_DEINT_FILTER) RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride); else if(mode & LINEAR_BLEND_DEINT_FILTER) - RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x); + RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c->deintTemp + x); else if(mode & MEDIAN_DEINT_FILTER) RENAME(deInterlaceMedian)(dstBlock, dstStride); else if(mode & CUBIC_IPOL_DEINT_FILTER) RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride); else if(mode & FFMPEG_DEINT_FILTER) - RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x); + RENAME(deInterlaceFF)(dstBlock, dstStride, c->deintTemp + x); else if(mode & LOWPASS5_DEINT_FILTER) - RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x); + RENAME(deInterlaceL5)(dstBlock, dstStride, c->deintTemp + x, c->deintTemp + width + x); /* else if(mode & CUBIC_BLEND_DEINT_FILTER) RENAME(deInterlaceBlendCubic)(dstBlock, dstStride); */ @@ -3025,11 +3026,11 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ const uint8_t *srcBlock= &(src[y*srcStride]); uint8_t *dstBlock= &(dst[y*dstStride]); #if TEMPLATE_PP_MMX - uint8_t *tempBlock1= c.tempBlocks; - uint8_t *tempBlock2= c.tempBlocks + 8; + uint8_t *tempBlock1 = c->tempBlocks; + uint8_t *tempBlock2 = c->tempBlocks + 8; #endif const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride]; - int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)]; + int8_t *nonBQPptr = &c->nonBQPTable[(y>>qpVShift)*FFABS(QPStride)]; int QP=0, nonBQP=0; /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards if not than use a temporary buffer */ @@ -3072,8 +3073,8 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ nonBQP= (nonBQP* QPCorrecture + 256*128)>>16; yHistogram[(srcBlock+qp_index*8)[srcStride*12 + 4]]++; } - c.QP_block[qp_index] = QP; - c.nonBQP_block[qp_index] = nonBQP; + c->QP_block[qp_index] = QP; + c->nonBQP_block[qp_index] = nonBQP; #if TEMPLATE_PP_MMX __asm__ volatile( "movd %1, %%mm7 \n\t" @@ -3081,7 +3082,7 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ "packuswb %%mm7, %%mm7 \n\t" // 0,QP, 0, QP, 0,QP, 0, QP "packuswb %%mm7, %%mm7 \n\t" // QP,..., QP "movq %%mm7, %0 \n\t" - : "=m" (c.pQPb_block[qp_index]) + : "=m" (c->pQPb_block[qp_index]) : "r" (QP) ); #endif @@ -3093,20 +3094,20 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ RENAME(prefetcht0)(dstBlock + (((x>>2)&6) + copyAhead+1)*dstStride + 32); RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride, - srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset); + srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c->packedYOffset); if(mode & LINEAR_IPOL_DEINT_FILTER) RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride); else if(mode & LINEAR_BLEND_DEINT_FILTER) - RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x); + RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c->deintTemp + x); else if(mode & MEDIAN_DEINT_FILTER) RENAME(deInterlaceMedian)(dstBlock, dstStride); else if(mode & CUBIC_IPOL_DEINT_FILTER) RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride); else if(mode & FFMPEG_DEINT_FILTER) - RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x); + RENAME(deInterlaceFF)(dstBlock, dstStride, c->deintTemp + x); else if(mode & LOWPASS5_DEINT_FILTER) - RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x); + RENAME(deInterlaceL5)(dstBlock, dstStride, c->deintTemp + x, c->deintTemp + width + x); /* else if(mode & CUBIC_BLEND_DEINT_FILTER) RENAME(deInterlaceBlendCubic)(dstBlock, dstStride); */ @@ -3121,24 +3122,24 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ const int stride= dstStride; //temporary while changing QP stuff to make things continue to work //eventually QP,nonBQP,etc will be arrays and this will be unnecessary - c.QP = c.QP_block[qp_index]; - c.nonBQP = c.nonBQP_block[qp_index]; - c.pQPb = c.pQPb_block[qp_index]; - c.pQPb2 = c.pQPb2_block[qp_index]; + c->QP = c->QP_block[qp_index]; + c->nonBQP = c->nonBQP_block[qp_index]; + c->pQPb = c->pQPb_block[qp_index]; + c->pQPb2 = c->pQPb2_block[qp_index]; /* only deblock if we have 2 blocks */ if(y + 8 < height){ if(mode & V_X1_FILTER) - RENAME(vertX1Filter)(dstBlock, stride, &c); + RENAME(vertX1Filter)(dstBlock, stride, c); else if(mode & V_DEBLOCK){ - const int t= RENAME(vertClassify)(dstBlock, stride, &c); + const int t = RENAME(vertClassify)(dstBlock, stride, c); if(t==1) - RENAME(doVertLowPass)(dstBlock, stride, &c); + RENAME(doVertLowPass)(dstBlock, stride, c); else if(t==2) - RENAME(doVertDefFilter)(dstBlock, stride, &c); + RENAME(doVertDefFilter)(dstBlock, stride, c); }else if(mode & V_A_DEBLOCK){ - RENAME(do_a_deblock)(dstBlock, stride, 1, &c, mode); + RENAME(do_a_deblock)(dstBlock, stride, 1, c, mode); } } @@ -3151,10 +3152,10 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ for(x = startx, qp_index=0; x < endx; x+=BLOCK_SIZE, qp_index++){ const int stride= dstStride; - c.QP = c.QP_block[qp_index]; - c.nonBQP = c.nonBQP_block[qp_index]; - c.pQPb = c.pQPb_block[qp_index]; - c.pQPb2 = c.pQPb2_block[qp_index]; + c->QP = c->QP_block[qp_index]; + c->nonBQP = c->nonBQP_block[qp_index]; + c->pQPb = c->pQPb_block[qp_index]; + c->pQPb2 = c->pQPb2_block[qp_index]; #if TEMPLATE_PP_MMX RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride); #endif @@ -3162,60 +3163,60 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ if(x - 8 >= 0){ #if TEMPLATE_PP_MMX if(mode & H_X1_FILTER) - RENAME(vertX1Filter)(tempBlock1, 16, &c); + RENAME(vertX1Filter)(tempBlock1, 16, c); else if(mode & H_DEBLOCK){ - const int t= RENAME(vertClassify)(tempBlock1, 16, &c); + const int t= RENAME(vertClassify)(tempBlock1, 16, c); if(t==1) - RENAME(doVertLowPass)(tempBlock1, 16, &c); + RENAME(doVertLowPass)(tempBlock1, 16, c); else if(t==2) - RENAME(doVertDefFilter)(tempBlock1, 16, &c); + RENAME(doVertDefFilter)(tempBlock1, 16, c); }else if(mode & H_A_DEBLOCK){ - RENAME(do_a_deblock)(tempBlock1, 16, 1, &c, mode); + RENAME(do_a_deblock)(tempBlock1, 16, 1, c, mode); } RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16); #else if(mode & H_X1_FILTER) - horizX1Filter(dstBlock-4, stride, c.QP); + horizX1Filter(dstBlock-4, stride, c->QP); else if(mode & H_DEBLOCK){ #if TEMPLATE_PP_ALTIVEC DECLARE_ALIGNED(16, unsigned char, tempBlock)[272]; int t; transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride); - t = vertClassify_altivec(tempBlock-48, 16, &c); + t = vertClassify_altivec(tempBlock-48, 16, c); if(t==1) { - doVertLowPass_altivec(tempBlock-48, 16, &c); + doVertLowPass_altivec(tempBlock-48, 16, c); transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride); } else if(t==2) { - doVertDefFilter_altivec(tempBlock-48, 16, &c); + doVertDefFilter_altivec(tempBlock-48, 16, c); transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride); } #else - const int t= RENAME(horizClassify)(dstBlock-4, stride, &c); + const int t= RENAME(horizClassify)(dstBlock-4, stride, c); if(t==1) - RENAME(doHorizLowPass)(dstBlock-4, stride, &c); + RENAME(doHorizLowPass)(dstBlock-4, stride, c); else if(t==2) - RENAME(doHorizDefFilter)(dstBlock-4, stride, &c); + RENAME(doHorizDefFilter)(dstBlock-4, stride, c); #endif }else if(mode & H_A_DEBLOCK){ - RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c, mode); + RENAME(do_a_deblock)(dstBlock-8, 1, stride, c, mode); } #endif //TEMPLATE_PP_MMX if(mode & DERING){ //FIXME filter first line - if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c); + if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, c); } if(mode & TEMP_NOISE_FILTER) { RENAME(tempNoiseReducer)(dstBlock-8, stride, - c.tempBlurred[isColor] + y*dstStride + x, - c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256, - c.ppMode.maxTmpNoise); + c->tempBlurred[isColor] + y*dstStride + x, + c->tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256, + c->ppMode.maxTmpNoise); } } @@ -3229,14 +3230,14 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ } if(mode & DERING){ - if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c); + if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, c); } if((mode & TEMP_NOISE_FILTER)){ RENAME(tempNoiseReducer)(dstBlock-8, dstStride, - c.tempBlurred[isColor] + y*dstStride + x, - c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256, - c.ppMode.maxTmpNoise); + c->tempBlurred[isColor] + y*dstStride + x, + c->tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256, + c->ppMode.maxTmpNoise); } /* did we use a tmp buffer for the last lines*/ @@ -3278,9 +3279,6 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ } } #endif - - *c2= c; //copy local context back - } #undef RENAME -- 2.34.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* [FFmpeg-devel] [PATCH 5/7] avfilter/vf_qp: Don't store NAN in int 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt ` (2 preceding siblings ...) 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 4/7] postproc/postprocess_template: Fix alignment Andreas Rheinhardt @ 2022-11-06 16:30 ` Andreas Rheinhardt 2022-11-06 16:31 ` [FFmpeg-devel] [PATCH 6/7] avfilter/phase_template: Fix left-shift of negative numbers Andreas Rheinhardt ` (2 subsequent siblings) 6 siblings, 0 replies; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-06 16:30 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Andreas Rheinhardt Affected the filter-pp2 FATE-test. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libavfilter/vf_qp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavfilter/vf_qp.c b/libavfilter/vf_qp.c index 7c9d60fe21..a771b51ae1 100644 --- a/libavfilter/vf_qp.c +++ b/libavfilter/vf_qp.c @@ -143,7 +143,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in) for (x = 0; x < s->qstride; x++) { unsigned int block_idx = y * s->qstride + x; AVVideoBlockParams *b = av_video_enc_params_block(par_out, block_idx); - int qp = sd_in ? in_qp_global + BLOCK_QP_DELTA(block_idx) : NAN; + double qp = sd_in ? in_qp_global + BLOCK_QP_DELTA(block_idx) : NAN; double var_values[] = { !!sd_in, qp, x, y, s->qstride, s->h, 0}; double temp_val; -- 2.34.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* [FFmpeg-devel] [PATCH 6/7] avfilter/phase_template: Fix left-shift of negative numbers 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt ` (3 preceding siblings ...) 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 5/7] avfilter/vf_qp: Don't store NAN in int Andreas Rheinhardt @ 2022-11-06 16:31 ` Andreas Rheinhardt 2022-11-06 16:31 ` [FFmpeg-devel] [PATCH 7/7] avfilter/vf_hqdn3d: " Andreas Rheinhardt 2022-11-08 15:41 ` [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt 6 siblings, 0 replies; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-06 16:31 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Andreas Rheinhardt Affected the filter-phase FATE-test. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libavfilter/phase_template.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavfilter/phase_template.c b/libavfilter/phase_template.c index 491612b826..8450670234 100644 --- a/libavfilter/phase_template.c +++ b/libavfilter/phase_template.c @@ -46,7 +46,7 @@ * * (The result is actually multiplied by 25) */ -#define DIFF(a, as, b, bs) ((t) = ((*(a) - (b)[bs]) << 2) + (a)[(as) << 1] - (b)[-(bs)], (t) * (t)) +#define DIFF(a, as, b, bs) ((t) = ((*(a) - (b)[bs]) * 4) + (a)[(as) * 2] - (b)[-(bs)], (t) * (t)) /* * Find which field combination has the smallest average squared difference -- 2.34.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* [FFmpeg-devel] [PATCH 7/7] avfilter/vf_hqdn3d: Fix left-shift of negative numbers 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt ` (4 preceding siblings ...) 2022-11-06 16:31 ` [FFmpeg-devel] [PATCH 6/7] avfilter/phase_template: Fix left-shift of negative numbers Andreas Rheinhardt @ 2022-11-06 16:31 ` Andreas Rheinhardt 2022-11-08 15:41 ` [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt 6 siblings, 0 replies; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-06 16:31 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Andreas Rheinhardt Affected filter-hqdn3d and filter-hqdn3d-sample FATE-tests. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libavfilter/vf_hqdn3d.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavfilter/vf_hqdn3d.c b/libavfilter/vf_hqdn3d.c index ce2b9a2974..c796ea9ab4 100644 --- a/libavfilter/vf_hqdn3d.c +++ b/libavfilter/vf_hqdn3d.c @@ -180,7 +180,7 @@ static void precalc_coefs(double dist25, int depth, int16_t *ct) gamma = log(0.25) / log(1.0 - FFMIN(dist25,252.0)/255.0 - 0.00001); for (i = -(256<<LUT_BITS); i < 256<<LUT_BITS; i++) { - double f = ((i<<(9-LUT_BITS)) + (1<<(8-LUT_BITS)) - 1) / 512.0; // midpoint of the bin + double f = (i * (1 << (9-LUT_BITS)) + (1<<(8-LUT_BITS)) - 1) / 512.0; // midpoint of the bin simil = FFMAX(0, 1.0 - fabs(f) / 255.0); C = pow(simil, gamma) * 256.0 * f; ct[(256<<LUT_BITS)+i] = lrint(C); -- 2.34.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt ` (5 preceding siblings ...) 2022-11-06 16:31 ` [FFmpeg-devel] [PATCH 7/7] avfilter/vf_hqdn3d: " Andreas Rheinhardt @ 2022-11-08 15:41 ` Andreas Rheinhardt 2022-11-08 15:43 ` Paul B Mahol 6 siblings, 1 reply; 9+ messages in thread From: Andreas Rheinhardt @ 2022-11-08 15:41 UTC (permalink / raw) To: ffmpeg-devel Andreas Rheinhardt: > postprocess.c currently has C, MMX, MMXEXT, 3DNow as well as > SSE2 versions of its internal functions. But given that only > ancient 32-bit x86 CPUs don't support SSE2, the MMX, MMXEXT > and 3DNow versions are obsolete and are therefore removed by > this commit. This saves about 56KB here. > > (The SSE2 version in particular is not really complete, > so that it often falls back to MMXEXT (which means that > there were some identical (apart from the name) MMXEXT > and SSE2 functions; this duplication no longer exists > with this commit.) > > Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> > --- > The PP_CPU_CAPS_MMX(2)/3DNOW could now be deprecated. > > libpostproc/postprocess.c | 69 ++-- > libpostproc/postprocess_template.c | 521 ++--------------------------- > 2 files changed, 57 insertions(+), 533 deletions(-) > > diff --git a/libpostproc/postprocess.c b/libpostproc/postprocess.c > index 383c691cb4..0586e458b4 100644 > --- a/libpostproc/postprocess.c > +++ b/libpostproc/postprocess.c > @@ -26,28 +26,27 @@ > */ > > /* > - C MMX MMX2 3DNow AltiVec > -isVertDC Ec Ec Ec > -isVertMinMaxOk Ec Ec Ec > -doVertLowPass E e e Ec > -doVertDefFilter Ec Ec e e Ec > -isHorizDC Ec Ec Ec > -isHorizMinMaxOk a E Ec > -doHorizLowPass E e e Ec > -doHorizDefFilter Ec Ec e e Ec > -do_a_deblock Ec E Ec E > -deRing E e e* Ecp > -Vertical RKAlgo1 E a a > -Horizontal RKAlgo1 a a > -Vertical X1# a E E > -Horizontal X1# a E E > -LinIpolDeinterlace e E E* > -CubicIpolDeinterlace a e e* > -LinBlendDeinterlace e E E* > + C MMX MMX2 AltiVec > +isVertDC Ec Ec Ec > +isVertMinMaxOk Ec Ec Ec > +doVertLowPass E e Ec > +doVertDefFilter Ec Ec e Ec > +isHorizDC Ec Ec Ec > +isHorizMinMaxOk a E Ec > +doHorizLowPass E e Ec > +doHorizDefFilter Ec Ec e Ec > +do_a_deblock Ec E Ec > +deRing E e Ecp > +Vertical RKAlgo1 E a > +Horizontal RKAlgo1 a > +Vertical X1# a E > +Horizontal X1# a E > +LinIpolDeinterlace e E > +CubicIpolDeinterlace a e > +LinBlendDeinterlace e E > MedianDeinterlace# E Ec Ec > -TempDeNoiser# E e e Ec > +TempDeNoiser# E e Ec > > -* I do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work > # more or less selfinvented filters so the exactness is not too meaningful > E = Exact implementation > e = almost exact implementation (slightly different rounding,...) > @@ -83,7 +82,6 @@ try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks > #include <stdlib.h> > #include <string.h> > //#undef HAVE_MMXEXT_INLINE > -//#define HAVE_AMD3DNOW_INLINE > //#undef HAVE_MMX_INLINE > //#undef ARCH_X86 > //#define DEBUG_BRIGHTNESS > @@ -494,7 +492,7 @@ static av_always_inline void do_a_deblock_C(uint8_t *src, int step, > } > } > > -//Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one > +//Note: we have C and SSE2 version (which uses MMX(EXT) when advantageous) > //Plain C versions > //we always compile C for testing which needs bitexactness > #define TEMPLATE_PP_C 1 > @@ -508,27 +506,12 @@ static av_always_inline void do_a_deblock_C(uint8_t *src, int step, > > #if ARCH_X86 && HAVE_INLINE_ASM > # if CONFIG_RUNTIME_CPUDETECT > -# define TEMPLATE_PP_MMX 1 > -# include "postprocess_template.c" > -# define TEMPLATE_PP_MMXEXT 1 > -# include "postprocess_template.c" > -# define TEMPLATE_PP_3DNOW 1 > -# include "postprocess_template.c" > # define TEMPLATE_PP_SSE2 1 > # include "postprocess_template.c" > # else > # if HAVE_SSE2_INLINE > # define TEMPLATE_PP_SSE2 1 > # include "postprocess_template.c" > -# elif HAVE_MMXEXT_INLINE > -# define TEMPLATE_PP_MMXEXT 1 > -# include "postprocess_template.c" > -# elif HAVE_AMD3DNOW_INLINE > -# define TEMPLATE_PP_3DNOW 1 > -# include "postprocess_template.c" > -# elif HAVE_MMX_INLINE > -# define TEMPLATE_PP_MMX 1 > -# include "postprocess_template.c" > # endif > # endif > #endif > @@ -549,21 +532,12 @@ static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[] > #if ARCH_X86 && HAVE_INLINE_ASM > // ordered per speed fastest first > if (c->cpuCaps & AV_CPU_FLAG_SSE2) pp = postProcess_SSE2; > - else if (c->cpuCaps & AV_CPU_FLAG_MMXEXT) pp = postProcess_MMX2; > - else if (c->cpuCaps & AV_CPU_FLAG_3DNOW) pp = postProcess_3DNow; > - else if (c->cpuCaps & AV_CPU_FLAG_MMX) pp = postProcess_MMX; > #elif HAVE_ALTIVEC > if (c->cpuCaps & AV_CPU_FLAG_ALTIVEC) pp = postProcess_altivec; > #endif > #else /* CONFIG_RUNTIME_CPUDETECT */ > #if HAVE_SSE2_INLINE > pp = postProcess_SSE2; > -#elif HAVE_MMXEXT_INLINE > - pp = postProcess_MMX2; > -#elif HAVE_AMD3DNOW_INLINE > - pp = postProcess_3DNow; > -#elif HAVE_MMX_INLINE > - pp = postProcess_MMX; > #elif HAVE_ALTIVEC > pp = postProcess_altivec; > #endif > @@ -877,9 +851,6 @@ av_cold pp_context *pp_get_context(int width, int height, int cpuCaps){ > c->cpuCaps = av_get_cpu_flags(); > } else { > c->cpuCaps = 0; > - if (cpuCaps & PP_CPU_CAPS_MMX) c->cpuCaps |= AV_CPU_FLAG_MMX; > - if (cpuCaps & PP_CPU_CAPS_MMX2) c->cpuCaps |= AV_CPU_FLAG_MMXEXT; > - if (cpuCaps & PP_CPU_CAPS_3DNOW) c->cpuCaps |= AV_CPU_FLAG_3DNOW; > if (cpuCaps & PP_CPU_CAPS_ALTIVEC) c->cpuCaps |= AV_CPU_FLAG_ALTIVEC; > } > > diff --git a/libpostproc/postprocess_template.c b/libpostproc/postprocess_template.c > index 9f76b7c587..bcf7bdad66 100644 > --- a/libpostproc/postprocess_template.c > +++ b/libpostproc/postprocess_template.c > @@ -20,7 +20,7 @@ > > /** > * @file > - * mmx/mmx2/3dnow postprocess code. > + * mmx/mmx2/sse2 postprocess code. > */ > > #include "libavutil/mem_internal.h" > @@ -57,14 +57,6 @@ > # define TEMPLATE_PP_MMXEXT 0 > #endif > > -#ifdef TEMPLATE_PP_3DNOW > -# undef TEMPLATE_PP_MMX > -# define TEMPLATE_PP_MMX 1 > -# define RENAME(a) a ## _3DNow > -#else > -# define TEMPLATE_PP_3DNOW 0 > -#endif > - > #ifdef TEMPLATE_PP_SSE2 > # undef TEMPLATE_PP_MMX > # define TEMPLATE_PP_MMX 1 > @@ -82,30 +74,19 @@ > > #if TEMPLATE_PP_MMXEXT > #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t" > -#elif TEMPLATE_PP_3DNOW > -#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t" > #endif > #define PAVGB(a,b) REAL_PAVGB(a,b) > > #if TEMPLATE_PP_MMXEXT > #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t" > -#elif TEMPLATE_PP_MMX > -#define PMINUB(b,a,t) \ > - "movq " #a ", " #t " \n\t"\ > - "psubusb " #b ", " #t " \n\t"\ > - "psubb " #t ", " #a " \n\t" > #endif > > #if TEMPLATE_PP_MMXEXT > #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t" > -#elif TEMPLATE_PP_MMX > -#define PMAXUB(a,b) \ > - "psubusb " #a ", " #b " \n\t"\ > - "paddb " #a ", " #b " \n\t" > #endif > > //FIXME? |255-0| = 1 (should not be a problem ...) > -#if TEMPLATE_PP_MMX > +#if TEMPLATE_PP_MMXEXT > /** > * Check if the middle 8x8 Block in the given 8x16 block is flat > */ > @@ -185,20 +166,8 @@ static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContex > "psubusb %%mm3, %%mm4 \n\t" > > " \n\t" > -#if TEMPLATE_PP_MMXEXT > "pxor %%mm7, %%mm7 \n\t" > "psadbw %%mm7, %%mm0 \n\t" > -#else > - "movq %%mm0, %%mm1 \n\t" > - "psrlw $8, %%mm0 \n\t" > - "paddb %%mm1, %%mm0 \n\t" > - "movq %%mm0, %%mm1 \n\t" > - "psrlq $16, %%mm0 \n\t" > - "paddb %%mm1, %%mm0 \n\t" > - "movq %%mm0, %%mm1 \n\t" > - "psrlq $32, %%mm0 \n\t" > - "paddb %%mm1, %%mm0 \n\t" > -#endif > "movq %4, %%mm7 \n\t" // QP,..., QP > "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP > "psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> 0 > @@ -219,7 +188,7 @@ static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContex > return 2; > } > } > -#endif //TEMPLATE_PP_MMX > +#endif //TEMPLATE_PP_MMXEXT > > /** > * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle) > @@ -228,7 +197,7 @@ static inline int RENAME(vertClassify)(const uint8_t src[], int stride, PPContex > #if !TEMPLATE_PP_ALTIVEC > static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c) > { > -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_MMXEXT > src+= stride*3; > __asm__ volatile( //"movv %0 %1 %2\n\t" > "movq %2, %%mm0 \n\t" // QP,..., QP > @@ -355,7 +324,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c) > : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb) > : "%"FF_REG_a, "%"FF_REG_c > ); > -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#else //TEMPLATE_PP_MMXEXT > const int l1= stride; > const int l2= stride + l1; > const int l3= stride + l2; > @@ -394,7 +363,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c) > > src++; > } > -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_MMXEXT > } > #endif //TEMPLATE_PP_ALTIVEC > > @@ -407,7 +376,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c) > */ > static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co) > { > -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_MMXEXT > src+= stride*3; > > __asm__ volatile( > @@ -494,7 +463,7 @@ static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co) > NAMED_CONSTRAINTS_ADD(b01) > : "%"FF_REG_a, "%"FF_REG_c > ); > -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#else //TEMPLATE_PP_MMXEXT > > const int l1= stride; > const int l2= stride + l1; > @@ -528,13 +497,13 @@ static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co) > } > src++; > } > -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_MMXEXT > } > > #if !TEMPLATE_PP_ALTIVEC > static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c) > { > -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_MMXEXT > /* > uint8_t tmp[16]; > const int l1= stride; > @@ -816,239 +785,7 @@ static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext > } > } > */ > -#elif TEMPLATE_PP_MMX > - DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars > - src+= stride*4; > - __asm__ volatile( > - "pxor %%mm7, %%mm7 \n\t" > -// 0 1 2 3 4 5 6 7 > -// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1 > -// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 > - > - "movq (%0), %%mm0 \n\t" > - "movq %%mm0, %%mm1 \n\t" > - "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0 > - "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0 > - > - "movq (%0, %1), %%mm2 \n\t" > - "lea (%0, %1, 2), %%"FF_REG_a" \n\t" > - "movq %%mm2, %%mm3 \n\t" > - "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1 > - "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1 > - > - "movq (%%"FF_REG_a"), %%mm4 \n\t" > - "movq %%mm4, %%mm5 \n\t" > - "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2 > - "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2 > - > - "paddw %%mm0, %%mm0 \n\t" // 2L0 > - "paddw %%mm1, %%mm1 \n\t" // 2H0 > - "psubw %%mm4, %%mm2 \n\t" // L1 - L2 > - "psubw %%mm5, %%mm3 \n\t" // H1 - H2 > - "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2 > - "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2 > - > - "psllw $2, %%mm2 \n\t" // 4L1 - 4L2 > - "psllw $2, %%mm3 \n\t" // 4H1 - 4H2 > - "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 > - "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 > - > - "movq (%%"FF_REG_a", %1), %%mm2 \n\t" > - "movq %%mm2, %%mm3 \n\t" > - "punpcklbw %%mm7, %%mm2 \n\t" // L3 > - "punpckhbw %%mm7, %%mm3 \n\t" // H3 > - > - "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3 > - "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3 > - "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3 > - "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3 > - "movq %%mm0, (%3) \n\t" // 2L0 - 5L1 + 5L2 - 2L3 > - "movq %%mm1, 8(%3) \n\t" // 2H0 - 5H1 + 5H2 - 2H3 > - > - "movq (%%"FF_REG_a", %1, 2), %%mm0 \n\t" > - "movq %%mm0, %%mm1 \n\t" > - "punpcklbw %%mm7, %%mm0 \n\t" // L4 > - "punpckhbw %%mm7, %%mm1 \n\t" // H4 > - > - "psubw %%mm0, %%mm2 \n\t" // L3 - L4 > - "psubw %%mm1, %%mm3 \n\t" // H3 - H4 > - "movq %%mm2, 16(%3) \n\t" // L3 - L4 > - "movq %%mm3, 24(%3) \n\t" // H3 - H4 > - "paddw %%mm4, %%mm4 \n\t" // 2L2 > - "paddw %%mm5, %%mm5 \n\t" // 2H2 > - "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4 > - "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4 > - > - "lea (%%"FF_REG_a", %1), %0 \n\t" > - "psllw $2, %%mm2 \n\t" // 4L3 - 4L4 > - "psllw $2, %%mm3 \n\t" // 4H3 - 4H4 > - "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 > - "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 > -//50 opcodes so far > - "movq (%0, %1, 2), %%mm2 \n\t" > - "movq %%mm2, %%mm3 \n\t" > - "punpcklbw %%mm7, %%mm2 \n\t" // L5 > - "punpckhbw %%mm7, %%mm3 \n\t" // H5 > - "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5 > - "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5 > - "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5 > - "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5 > - > - "movq (%%"FF_REG_a", %1, 4), %%mm6 \n\t" > - "punpcklbw %%mm7, %%mm6 \n\t" // L6 > - "psubw %%mm6, %%mm2 \n\t" // L5 - L6 > - "movq (%%"FF_REG_a", %1, 4), %%mm6 \n\t" > - "punpckhbw %%mm7, %%mm6 \n\t" // H6 > - "psubw %%mm6, %%mm3 \n\t" // H5 - H6 > - > - "paddw %%mm0, %%mm0 \n\t" // 2L4 > - "paddw %%mm1, %%mm1 \n\t" // 2H4 > - "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6 > - "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6 > - > - "psllw $2, %%mm2 \n\t" // 4L5 - 4L6 > - "psllw $2, %%mm3 \n\t" // 4H5 - 4H6 > - "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 > - "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 > - > - "movq (%0, %1, 4), %%mm2 \n\t" > - "movq %%mm2, %%mm3 \n\t" > - "punpcklbw %%mm7, %%mm2 \n\t" // L7 > - "punpckhbw %%mm7, %%mm3 \n\t" // H7 > - > - "paddw %%mm2, %%mm2 \n\t" // 2L7 > - "paddw %%mm3, %%mm3 \n\t" // 2H7 > - "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7 > - "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7 > - > - "movq (%3), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3 > - "movq 8(%3), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3 > - > -#if TEMPLATE_PP_MMXEXT > - "movq %%mm7, %%mm6 \n\t" // 0 > - "psubw %%mm0, %%mm6 \n\t" > - "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "psubw %%mm1, %%mm6 \n\t" > - "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "psubw %%mm2, %%mm6 \n\t" > - "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "psubw %%mm3, %%mm6 \n\t" > - "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3| > -#else > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm0, %%mm6 \n\t" > - "pxor %%mm6, %%mm0 \n\t" > - "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm1, %%mm6 \n\t" > - "pxor %%mm6, %%mm1 \n\t" > - "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm2, %%mm6 \n\t" > - "pxor %%mm6, %%mm2 \n\t" > - "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm3, %%mm6 \n\t" > - "pxor %%mm6, %%mm3 \n\t" > - "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3| > -#endif > - > -#if TEMPLATE_PP_MMXEXT > - "pminsw %%mm2, %%mm0 \n\t" > - "pminsw %%mm3, %%mm1 \n\t" > -#else > - "movq %%mm0, %%mm6 \n\t" > - "psubusw %%mm2, %%mm6 \n\t" > - "psubw %%mm6, %%mm0 \n\t" > - "movq %%mm1, %%mm6 \n\t" > - "psubusw %%mm3, %%mm6 \n\t" > - "psubw %%mm6, %%mm1 \n\t" > -#endif > - > - "movd %2, %%mm2 \n\t" // QP > - "punpcklbw %%mm7, %%mm2 \n\t" > - > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5) > - "pxor %%mm6, %%mm4 \n\t" > - "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5| > - "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5) > - "pxor %%mm7, %%mm5 \n\t" > - "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5| > -// 100 opcodes > - "psllw $3, %%mm2 \n\t" // 8QP > - "movq %%mm2, %%mm3 \n\t" // 8QP > - "pcmpgtw %%mm4, %%mm2 \n\t" > - "pcmpgtw %%mm5, %%mm3 \n\t" > - "pand %%mm2, %%mm4 \n\t" > - "pand %%mm3, %%mm5 \n\t" > - > - > - "psubusw %%mm0, %%mm4 \n\t" // hd > - "psubusw %%mm1, %%mm5 \n\t" // ld > - > - > - "movq "MANGLE(w05)", %%mm2 \n\t" // 5 > - "pmullw %%mm2, %%mm4 \n\t" > - "pmullw %%mm2, %%mm5 \n\t" > - "movq "MANGLE(w20)", %%mm2 \n\t" // 32 > - "paddw %%mm2, %%mm4 \n\t" > - "paddw %%mm2, %%mm5 \n\t" > - "psrlw $6, %%mm4 \n\t" > - "psrlw $6, %%mm5 \n\t" > - > - "movq 16(%3), %%mm0 \n\t" // L3 - L4 > - "movq 24(%3), %%mm1 \n\t" // H3 - H4 > - > - "pxor %%mm2, %%mm2 \n\t" > - "pxor %%mm3, %%mm3 \n\t" > - > - "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4) > - "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4) > - "pxor %%mm2, %%mm0 \n\t" > - "pxor %%mm3, %%mm1 \n\t" > - "psubw %%mm2, %%mm0 \n\t" // |L3-L4| > - "psubw %%mm3, %%mm1 \n\t" // |H3-H4| > - "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2 > - "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2 > - > - "pxor %%mm6, %%mm2 \n\t" > - "pxor %%mm7, %%mm3 \n\t" > - "pand %%mm2, %%mm4 \n\t" > - "pand %%mm3, %%mm5 \n\t" > - > -#if TEMPLATE_PP_MMXEXT > - "pminsw %%mm0, %%mm4 \n\t" > - "pminsw %%mm1, %%mm5 \n\t" > -#else > - "movq %%mm4, %%mm2 \n\t" > - "psubusw %%mm0, %%mm2 \n\t" > - "psubw %%mm2, %%mm4 \n\t" > - "movq %%mm5, %%mm2 \n\t" > - "psubusw %%mm1, %%mm2 \n\t" > - "psubw %%mm2, %%mm5 \n\t" > -#endif > - "pxor %%mm6, %%mm4 \n\t" > - "pxor %%mm7, %%mm5 \n\t" > - "psubw %%mm6, %%mm4 \n\t" > - "psubw %%mm7, %%mm5 \n\t" > - "packsswb %%mm5, %%mm4 \n\t" > - "movq (%0), %%mm0 \n\t" > - "paddb %%mm4, %%mm0 \n\t" > - "movq %%mm0, (%0) \n\t" > - "movq (%0, %1), %%mm0 \n\t" > - "psubb %%mm4, %%mm0 \n\t" > - "movq %%mm0, (%0, %1) \n\t" > - > - : "+r" (src) > - : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp) > - NAMED_CONSTRAINTS_ADD(w05,w20) > - : "%"FF_REG_a > - ); > -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#else //TEMPLATE_PP_MMXEXT > const int l1= stride; > const int l2= stride + l1; > const int l3= stride + l2; > @@ -1086,14 +823,14 @@ static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext > } > src++; > } > -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_MMXEXT > } > #endif //TEMPLATE_PP_ALTIVEC > > #if !TEMPLATE_PP_ALTIVEC > static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c) > { > -#if HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) > +#if HAVE_7REGS && TEMPLATE_PP_MMXEXT > DECLARE_ALIGNED(8, uint64_t, tmp)[3]; > __asm__ volatile( > "pxor %%mm6, %%mm6 \n\t" > @@ -1113,20 +850,10 @@ static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c) > > #undef REAL_FIND_MIN_MAX > #undef FIND_MIN_MAX > -#if TEMPLATE_PP_MMXEXT > #define REAL_FIND_MIN_MAX(addr)\ > "movq " #addr ", %%mm0 \n\t"\ > "pminub %%mm0, %%mm7 \n\t"\ > "pmaxub %%mm0, %%mm6 \n\t" > -#else > -#define REAL_FIND_MIN_MAX(addr)\ > - "movq " #addr ", %%mm0 \n\t"\ > - "movq %%mm7, %%mm1 \n\t"\ > - "psubusb %%mm0, %%mm6 \n\t"\ > - "paddb %%mm0, %%mm6 \n\t"\ > - "psubusb %%mm0, %%mm1 \n\t"\ > - "psubb %%mm1, %%mm7 \n\t" > -#endif > #define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr) > > FIND_MIN_MAX((%%FF_REGa)) > @@ -1140,49 +867,20 @@ FIND_MIN_MAX((%0, %1, 8)) > > "movq %%mm7, %%mm4 \n\t" > "psrlq $8, %%mm7 \n\t" > -#if TEMPLATE_PP_MMXEXT > "pminub %%mm4, %%mm7 \n\t" // min of pixels > "pshufw $0xF9, %%mm7, %%mm4 \n\t" > "pminub %%mm4, %%mm7 \n\t" // min of pixels > "pshufw $0xFE, %%mm7, %%mm4 \n\t" > "pminub %%mm4, %%mm7 \n\t" > -#else > - "movq %%mm7, %%mm1 \n\t" > - "psubusb %%mm4, %%mm1 \n\t" > - "psubb %%mm1, %%mm7 \n\t" > - "movq %%mm7, %%mm4 \n\t" > - "psrlq $16, %%mm7 \n\t" > - "movq %%mm7, %%mm1 \n\t" > - "psubusb %%mm4, %%mm1 \n\t" > - "psubb %%mm1, %%mm7 \n\t" > - "movq %%mm7, %%mm4 \n\t" > - "psrlq $32, %%mm7 \n\t" > - "movq %%mm7, %%mm1 \n\t" > - "psubusb %%mm4, %%mm1 \n\t" > - "psubb %%mm1, %%mm7 \n\t" > -#endif > > > "movq %%mm6, %%mm4 \n\t" > "psrlq $8, %%mm6 \n\t" > -#if TEMPLATE_PP_MMXEXT > "pmaxub %%mm4, %%mm6 \n\t" // max of pixels > "pshufw $0xF9, %%mm6, %%mm4 \n\t" > "pmaxub %%mm4, %%mm6 \n\t" > "pshufw $0xFE, %%mm6, %%mm4 \n\t" > "pmaxub %%mm4, %%mm6 \n\t" > -#else > - "psubusb %%mm4, %%mm6 \n\t" > - "paddb %%mm4, %%mm6 \n\t" > - "movq %%mm6, %%mm4 \n\t" > - "psrlq $16, %%mm6 \n\t" > - "psubusb %%mm4, %%mm6 \n\t" > - "paddb %%mm4, %%mm6 \n\t" > - "movq %%mm6, %%mm4 \n\t" > - "psrlq $32, %%mm6 \n\t" > - "psubusb %%mm4, %%mm6 \n\t" > - "paddb %%mm4, %%mm6 \n\t" > -#endif > "movq %%mm6, %%mm0 \n\t" // max > "psubb %%mm7, %%mm6 \n\t" // max - min > "push %%"FF_REG_a" \n\t" > @@ -1320,7 +1018,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5, > NAMED_CONSTRAINTS_ADD(deringThreshold,b00,b02,b08) > : "%"FF_REG_a, "%"FF_REG_d > ); > -#else // HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) > +#else // HAVE_7REGS && TEMPLATE_PP_MMXEXT > int y; > int min=255; > int max=0; > @@ -1438,7 +1136,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5, > // src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255; > } > #endif > -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_MMXEXT > } > #endif //TEMPLATE_PP_ALTIVEC > > @@ -1450,7 +1148,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5, > */ > static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride) > { > -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_MMXEXT > src+= 4*stride; > __asm__ volatile( > "lea (%0, %1), %%"FF_REG_a" \n\t" > @@ -1503,14 +1201,13 @@ static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int strid > */ > static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride) > { > -#if TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_SSE2 > src+= stride*3; > __asm__ volatile( > "lea (%0, %1), %%"FF_REG_a" \n\t" > "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t" > "lea (%%"FF_REG_d", %1, 4), %%"FF_REG_c"\n\t" > "add %1, %%"FF_REG_c" \n\t" > -#if TEMPLATE_PP_SSE2 > "pxor %%xmm7, %%xmm7 \n\t" > #define REAL_DEINT_CUBIC(a,b,c,d,e)\ > "movq " #a ", %%xmm0 \n\t"\ > @@ -1526,33 +1223,6 @@ static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride > "psubw %%xmm0, %%xmm1 \n\t"\ > "packuswb %%xmm1, %%xmm1 \n\t"\ > "movlps %%xmm1, " #c " \n\t" > -#else //TEMPLATE_PP_SSE2 > - "pxor %%mm7, %%mm7 \n\t" > -// 0 1 2 3 4 5 6 7 8 9 10 > -// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx > - > -#define REAL_DEINT_CUBIC(a,b,c,d,e)\ > - "movq " #a ", %%mm0 \n\t"\ > - "movq " #b ", %%mm1 \n\t"\ > - "movq " #d ", %%mm2 \n\t"\ > - "movq " #e ", %%mm3 \n\t"\ > - PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\ > - PAVGB(%%mm3, %%mm0) /* (a+e) /2 */\ > - "movq %%mm0, %%mm2 \n\t"\ > - "punpcklbw %%mm7, %%mm0 \n\t"\ > - "punpckhbw %%mm7, %%mm2 \n\t"\ > - "movq %%mm1, %%mm3 \n\t"\ > - "punpcklbw %%mm7, %%mm1 \n\t"\ > - "punpckhbw %%mm7, %%mm3 \n\t"\ > - "psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\ > - "psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\ > - "psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\ > - "psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\ > - "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\ > - "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\ > - "packuswb %%mm3, %%mm1 \n\t"\ > - "movq %%mm1, " #c " \n\t" > -#endif //TEMPLATE_PP_SSE2 > #define DEINT_CUBIC(a,b,c,d,e) REAL_DEINT_CUBIC(a,b,c,d,e) > > DEINT_CUBIC((%0) , (%%FF_REGa, %1), (%%FF_REGa, %1, 2), (%0, %1, 4) , (%%FF_REGd, %1)) > @@ -1562,13 +1232,11 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , (%%FF_REGd, %1, 4), (%%FF_REGc) > > : : "r" (src), "r" ((x86_reg)stride) > : > -#if TEMPLATE_PP_SSE2 > XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm7",) > -#endif > "%"FF_REG_a, "%"FF_REG_d, "%"FF_REG_c > ); > #undef REAL_DEINT_CUBIC > -#else //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#else //TEMPLATE_PP_SSE2 > int x; > src+= stride*3; > for(x=0; x<8; x++){ > @@ -1578,7 +1246,7 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , (%%FF_REGd, %1, 4), (%%FF_REGc) > src[stride*9] = av_clip_uint8((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4); > src++; > } > -#endif //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_SSE2 > } > > /** > @@ -1590,7 +1258,7 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , (%%FF_REGd, %1, 4), (%%FF_REGc) > */ > static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp) > { > -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_MMXEXT > src+= stride*4; > __asm__ volatile( > "lea (%0, %1), %%"FF_REG_a" \n\t" > @@ -1639,7 +1307,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4 > : : "r" (src), "r" ((x86_reg)stride), "r"(tmp) > : "%"FF_REG_a, "%"FF_REG_d > ); > -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#else //TEMPLATE_PP_MMXEXT > int x; > src+= stride*4; > for(x=0; x<8; x++){ > @@ -1657,7 +1325,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4 > > src++; > } > -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_MMXEXT > } > > /** > @@ -1669,7 +1337,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4 > */ > static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2) > { > -#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS > +#if TEMPLATE_PP_MMXEXT && HAVE_6REGS > src+= stride*4; > __asm__ volatile( > "lea (%0, %1), %%"FF_REG_a" \n\t" > @@ -1729,7 +1397,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4 > : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2) > : "%"FF_REG_a, "%"FF_REG_d > ); > -#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS > +#else //TEMPLATE_PP_MMXEXT && HAVE_6REGS > int x; > src+= stride*4; > for(x=0; x<8; x++){ > @@ -1758,7 +1426,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4 > > src++; > } > -#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS > +#endif // TEMPLATE_PP_MMXEXT && HAVE_6REGS > } > > /** > @@ -1770,7 +1438,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, 8) , (%%FF_REGd, %1, 4 > */ > static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp) > { > -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#if TEMPLATE_PP_MMXEXT > src+= 4*stride; > __asm__ volatile( > "lea (%0, %1), %%"FF_REG_a" \n\t" > @@ -1817,7 +1485,7 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin > : : "r" (src), "r" ((x86_reg)stride), "r" (tmp) > : "%"FF_REG_a, "%"FF_REG_d > ); > -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#else //TEMPLATE_PP_MMXEXT > int a, b, c, x; > src+= 4*stride; > > @@ -1860,7 +1528,7 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin > src += 4; > tmp += 4; > } > -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW > +#endif //TEMPLATE_PP_MMXEXT > } > > /** > @@ -1871,9 +1539,8 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin > */ > static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride) > { > -#if TEMPLATE_PP_MMX > - src+= 4*stride; > #if TEMPLATE_PP_MMXEXT > + src+= 4*stride; > __asm__ volatile( > "lea (%0, %1), %%"FF_REG_a" \n\t" > "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t" > @@ -1922,48 +1589,6 @@ static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride) > : "%"FF_REG_a, "%"FF_REG_d > ); > > -#else // MMX without MMX2 > - __asm__ volatile( > - "lea (%0, %1), %%"FF_REG_a" \n\t" > - "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t" > -// 0 1 2 3 4 5 6 7 8 9 > -// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 > - "pxor %%mm7, %%mm7 \n\t" > - > -#define REAL_MEDIAN(a,b,c)\ > - "movq " #a ", %%mm0 \n\t"\ > - "movq " #b ", %%mm2 \n\t"\ > - "movq " #c ", %%mm1 \n\t"\ > - "movq %%mm0, %%mm3 \n\t"\ > - "movq %%mm1, %%mm4 \n\t"\ > - "movq %%mm2, %%mm5 \n\t"\ > - "psubusb %%mm1, %%mm3 \n\t"\ > - "psubusb %%mm2, %%mm4 \n\t"\ > - "psubusb %%mm0, %%mm5 \n\t"\ > - "pcmpeqb %%mm7, %%mm3 \n\t"\ > - "pcmpeqb %%mm7, %%mm4 \n\t"\ > - "pcmpeqb %%mm7, %%mm5 \n\t"\ > - "movq %%mm3, %%mm6 \n\t"\ > - "pxor %%mm4, %%mm3 \n\t"\ > - "pxor %%mm5, %%mm4 \n\t"\ > - "pxor %%mm6, %%mm5 \n\t"\ > - "por %%mm3, %%mm1 \n\t"\ > - "por %%mm4, %%mm2 \n\t"\ > - "por %%mm5, %%mm0 \n\t"\ > - "pand %%mm2, %%mm0 \n\t"\ > - "pand %%mm1, %%mm0 \n\t"\ > - "movq %%mm0, " #b " \n\t" > -#define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c) > - > -MEDIAN((%0) , (%%FF_REGa) , (%%FF_REGa, %1)) > -MEDIAN((%%FF_REGa, %1), (%%FF_REGa, %1, 2), (%0, %1, 4)) > -MEDIAN((%0, %1, 4) , (%%FF_REGd) , (%%FF_REGd, %1)) > -MEDIAN((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8)) > - > - : : "r" (src), "r" ((x86_reg)stride) > - : "%"FF_REG_a, "%"FF_REG_d > - ); > -#endif //TEMPLATE_PP_MMXEXT > #else //TEMPLATE_PP_MMX > int x, y; > src+= 4*stride; > @@ -2165,7 +1790,7 @@ static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride, > > #define FAST_L2_DIFF > //#define L1_DIFF //u should change the thresholds too if u try that one > -#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS > +#if TEMPLATE_PP_MMXEXT && HAVE_6REGS > __asm__ volatile( > "lea (%2, %2, 2), %%"FF_REG_a" \n\t" // 3*stride > "lea (%2, %2, 4), %%"FF_REG_d" \n\t" // 5*stride > @@ -2454,7 +2079,7 @@ L2_DIFF_CORE((%0, %%FF_REGc) , (%1, %%FF_REGc)) > NAMED_CONSTRAINTS_ADD(b80) > : "%"FF_REG_a, "%"FF_REG_d, "%"FF_REG_c, "memory" > ); > -#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS > +#else //TEMPLATE_PP_MMXEXT && HAVE_6REGS > { > int y; > int d=0; > @@ -2537,11 +2162,11 @@ Switch between > } > } > } > -#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS > +#endif //TEMPLATE_PP_MMXEXT && HAVE_6REGS > } > #endif //TEMPLATE_PP_ALTIVEC > > -#if TEMPLATE_PP_MMX > +#if TEMPLATE_PP_MMXEXT > /** > * accurate deblock filter > */ > @@ -2945,7 +2570,6 @@ static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int st > "movq (%4), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3 > "movq 8(%4), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3 > > -#if TEMPLATE_PP_MMXEXT > "movq %%mm7, %%mm6 \n\t" // 0 > "psubw %%mm0, %%mm6 \n\t" > "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7| > @@ -2958,36 +2582,9 @@ static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int st > "movq %%mm7, %%mm6 \n\t" // 0 > "psubw %%mm3, %%mm6 \n\t" > "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3| > -#else > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm0, %%mm6 \n\t" > - "pxor %%mm6, %%mm0 \n\t" > - "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm1, %%mm6 \n\t" > - "pxor %%mm6, %%mm1 \n\t" > - "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm2, %%mm6 \n\t" > - "pxor %%mm6, %%mm2 \n\t" > - "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3| > - "movq %%mm7, %%mm6 \n\t" // 0 > - "pcmpgtw %%mm3, %%mm6 \n\t" > - "pxor %%mm6, %%mm3 \n\t" > - "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3| > -#endif > > -#if TEMPLATE_PP_MMXEXT > "pminsw %%mm2, %%mm0 \n\t" > "pminsw %%mm3, %%mm1 \n\t" > -#else > - "movq %%mm0, %%mm6 \n\t" > - "psubusw %%mm2, %%mm6 \n\t" > - "psubw %%mm6, %%mm0 \n\t" > - "movq %%mm1, %%mm6 \n\t" > - "psubusw %%mm3, %%mm6 \n\t" > - "psubw %%mm6, %%mm1 \n\t" > -#endif > > "movd %2, %%mm2 \n\t" // QP > "punpcklbw %%mm7, %%mm2 \n\t" > @@ -3041,17 +2638,8 @@ static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int st > "pand %%mm2, %%mm4 \n\t" > "pand %%mm3, %%mm5 \n\t" > > -#if TEMPLATE_PP_MMXEXT > "pminsw %%mm0, %%mm4 \n\t" > "pminsw %%mm1, %%mm5 \n\t" > -#else > - "movq %%mm4, %%mm2 \n\t" > - "psubusw %%mm0, %%mm2 \n\t" > - "psubw %%mm2, %%mm4 \n\t" > - "movq %%mm5, %%mm2 \n\t" > - "psubusw %%mm1, %%mm2 \n\t" > - "psubw %%mm2, %%mm5 \n\t" > -#endif > "pxor %%mm6, %%mm4 \n\t" > "pxor %%mm7, %%mm5 \n\t" > "psubw %%mm6, %%mm4 \n\t" > @@ -3088,18 +2676,14 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ > static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride, > int levelFix, int64_t *packedOffsetAndScale) > { > -#if !TEMPLATE_PP_MMX || !HAVE_6REGS > - int i; > -#endif > if(levelFix){ > -#if TEMPLATE_PP_MMX && HAVE_6REGS > +#if TEMPLATE_PP_MMXEXT && HAVE_6REGS > __asm__ volatile( > "movq (%%"FF_REG_a"), %%mm2 \n\t" // packedYOffset > "movq 8(%%"FF_REG_a"), %%mm3 \n\t" // packedYScale > "lea (%2,%4), %%"FF_REG_a" \n\t" > "lea (%3,%5), %%"FF_REG_d" \n\t" > "pxor %%mm4, %%mm4 \n\t" > -#if TEMPLATE_PP_MMXEXT > #define REAL_SCALED_CPY(src1, src2, dst1, dst2) \ > "movq " #src1 ", %%mm0 \n\t"\ > "movq " #src1 ", %%mm5 \n\t"\ > @@ -3122,34 +2706,6 @@ static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t > "movq %%mm0, " #dst1 " \n\t"\ > "movq %%mm1, " #dst2 " \n\t"\ > > -#else //TEMPLATE_PP_MMXEXT > -#define REAL_SCALED_CPY(src1, src2, dst1, dst2) \ > - "movq " #src1 ", %%mm0 \n\t"\ > - "movq " #src1 ", %%mm5 \n\t"\ > - "punpcklbw %%mm4, %%mm0 \n\t"\ > - "punpckhbw %%mm4, %%mm5 \n\t"\ > - "psubw %%mm2, %%mm0 \n\t"\ > - "psubw %%mm2, %%mm5 \n\t"\ > - "movq " #src2 ", %%mm1 \n\t"\ > - "psllw $6, %%mm0 \n\t"\ > - "psllw $6, %%mm5 \n\t"\ > - "pmulhw %%mm3, %%mm0 \n\t"\ > - "movq " #src2 ", %%mm6 \n\t"\ > - "pmulhw %%mm3, %%mm5 \n\t"\ > - "punpcklbw %%mm4, %%mm1 \n\t"\ > - "punpckhbw %%mm4, %%mm6 \n\t"\ > - "psubw %%mm2, %%mm1 \n\t"\ > - "psubw %%mm2, %%mm6 \n\t"\ > - "psllw $6, %%mm1 \n\t"\ > - "psllw $6, %%mm6 \n\t"\ > - "pmulhw %%mm3, %%mm1 \n\t"\ > - "pmulhw %%mm3, %%mm6 \n\t"\ > - "packuswb %%mm5, %%mm0 \n\t"\ > - "packuswb %%mm6, %%mm1 \n\t"\ > - "movq %%mm0, " #dst1 " \n\t"\ > - "movq %%mm1, " #dst2 " \n\t"\ > - > -#endif //TEMPLATE_PP_MMXEXT > #define SCALED_CPY(src1, src2, dst1, dst2)\ > REAL_SCALED_CPY(src1, src2, dst1, dst2) > > @@ -3170,7 +2726,7 @@ SCALED_CPY((%%FF_REGa, %4), (%%FF_REGa, %4, 2), (%%FF_REGd, %5), (%%FF_REGd, %5, > : "%"FF_REG_d > ); > #else //TEMPLATE_PP_MMX && HAVE_6REGS > - for(i=0; i<8; i++) > + for (int i = 0; i < 8; i++) > memcpy( &(dst[dstStride*i]), > &(src[srcStride*i]), BLOCK_SIZE); > #endif //TEMPLATE_PP_MMX && HAVE_6REGS > @@ -3203,7 +2759,7 @@ SIMPLE_CPY((%%FF_REGa, %2), (%%FF_REGa, %2, 2), (%%FF_REGd, %3), (%%FF_REGd, %3, > : "%"FF_REG_a, "%"FF_REG_d > ); > #else //TEMPLATE_PP_MMX && HAVE_6REGS > - for(i=0; i<8; i++) > + for (int i = 0; i < 8; i++) > memcpy( &(dst[dstStride*i]), > &(src[srcStride*i]), BLOCK_SIZE); > #endif //TEMPLATE_PP_MMX && HAVE_6REGS > @@ -3696,9 +3252,7 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ > } > } > } > -#if TEMPLATE_PP_3DNOW > - __asm__ volatile("femms"); > -#elif TEMPLATE_PP_MMX > +#if TEMPLATE_PP_MMX > __asm__ volatile("emms"); > #endif > > @@ -3734,5 +3288,4 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[ > #undef TEMPLATE_PP_ALTIVEC > #undef TEMPLATE_PP_MMX > #undef TEMPLATE_PP_MMXEXT > -#undef TEMPLATE_PP_3DNOW > #undef TEMPLATE_PP_SSE2 Will apply this patchset tomorrow unless there are objections. - Andreas _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions 2022-11-08 15:41 ` [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt @ 2022-11-08 15:43 ` Paul B Mahol 0 siblings, 0 replies; 9+ messages in thread From: Paul B Mahol @ 2022-11-08 15:43 UTC (permalink / raw) To: FFmpeg development discussions and patches On 11/8/22, Andreas Rheinhardt <andreas.rheinhardt@outlook.com> wrote: > Andreas Rheinhardt: >> postprocess.c currently has C, MMX, MMXEXT, 3DNow as well as >> SSE2 versions of its internal functions. But given that only >> ancient 32-bit x86 CPUs don't support SSE2, the MMX, MMXEXT >> and 3DNow versions are obsolete and are therefore removed by >> this commit. This saves about 56KB here. >> >> (The SSE2 version in particular is not really complete, >> so that it often falls back to MMXEXT (which means that >> there were some identical (apart from the name) MMXEXT >> and SSE2 functions; this duplication no longer exists >> with this commit.) >> >> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> >> --- >> The PP_CPU_CAPS_MMX(2)/3DNOW could now be deprecated. >> >> libpostproc/postprocess.c | 69 ++-- >> libpostproc/postprocess_template.c | 521 ++--------------------------- >> 2 files changed, 57 insertions(+), 533 deletions(-) >> >> diff --git a/libpostproc/postprocess.c b/libpostproc/postprocess.c >> index 383c691cb4..0586e458b4 100644 >> --- a/libpostproc/postprocess.c >> +++ b/libpostproc/postprocess.c >> @@ -26,28 +26,27 @@ >> */ >> >> /* >> - C MMX MMX2 3DNow AltiVec >> -isVertDC Ec Ec Ec >> -isVertMinMaxOk Ec Ec Ec >> -doVertLowPass E e e Ec >> -doVertDefFilter Ec Ec e e Ec >> -isHorizDC Ec Ec Ec >> -isHorizMinMaxOk a E Ec >> -doHorizLowPass E e e Ec >> -doHorizDefFilter Ec Ec e e Ec >> -do_a_deblock Ec E Ec E >> -deRing E e e* Ecp >> -Vertical RKAlgo1 E a a >> -Horizontal RKAlgo1 a a >> -Vertical X1# a E E >> -Horizontal X1# a E E >> -LinIpolDeinterlace e E E* >> -CubicIpolDeinterlace a e e* >> -LinBlendDeinterlace e E E* >> + C MMX MMX2 AltiVec >> +isVertDC Ec Ec Ec >> +isVertMinMaxOk Ec Ec Ec >> +doVertLowPass E e Ec >> +doVertDefFilter Ec Ec e Ec >> +isHorizDC Ec Ec Ec >> +isHorizMinMaxOk a E Ec >> +doHorizLowPass E e Ec >> +doHorizDefFilter Ec Ec e Ec >> +do_a_deblock Ec E Ec >> +deRing E e Ecp >> +Vertical RKAlgo1 E a >> +Horizontal RKAlgo1 a >> +Vertical X1# a E >> +Horizontal X1# a E >> +LinIpolDeinterlace e E >> +CubicIpolDeinterlace a e >> +LinBlendDeinterlace e E >> MedianDeinterlace# E Ec Ec >> -TempDeNoiser# E e e Ec >> +TempDeNoiser# E e Ec >> >> -* I do not have a 3DNow! CPU -> it is untested, but no one said it does >> not work so it seems to work >> # more or less selfinvented filters so the exactness is not too >> meaningful >> E = Exact implementation >> e = almost exact implementation (slightly different rounding,...) >> @@ -83,7 +82,6 @@ try to unroll inner for(x=0 ... loop to avoid these damn >> if(x ... checks >> #include <stdlib.h> >> #include <string.h> >> //#undef HAVE_MMXEXT_INLINE >> -//#define HAVE_AMD3DNOW_INLINE >> //#undef HAVE_MMX_INLINE >> //#undef ARCH_X86 >> //#define DEBUG_BRIGHTNESS >> @@ -494,7 +492,7 @@ static av_always_inline void do_a_deblock_C(uint8_t >> *src, int step, >> } >> } >> >> -//Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one >> +//Note: we have C and SSE2 version (which uses MMX(EXT) when >> advantageous) >> //Plain C versions >> //we always compile C for testing which needs bitexactness >> #define TEMPLATE_PP_C 1 >> @@ -508,27 +506,12 @@ static av_always_inline void do_a_deblock_C(uint8_t >> *src, int step, >> >> #if ARCH_X86 && HAVE_INLINE_ASM >> # if CONFIG_RUNTIME_CPUDETECT >> -# define TEMPLATE_PP_MMX 1 >> -# include "postprocess_template.c" >> -# define TEMPLATE_PP_MMXEXT 1 >> -# include "postprocess_template.c" >> -# define TEMPLATE_PP_3DNOW 1 >> -# include "postprocess_template.c" >> # define TEMPLATE_PP_SSE2 1 >> # include "postprocess_template.c" >> # else >> # if HAVE_SSE2_INLINE >> # define TEMPLATE_PP_SSE2 1 >> # include "postprocess_template.c" >> -# elif HAVE_MMXEXT_INLINE >> -# define TEMPLATE_PP_MMXEXT 1 >> -# include "postprocess_template.c" >> -# elif HAVE_AMD3DNOW_INLINE >> -# define TEMPLATE_PP_3DNOW 1 >> -# include "postprocess_template.c" >> -# elif HAVE_MMX_INLINE >> -# define TEMPLATE_PP_MMX 1 >> -# include "postprocess_template.c" >> # endif >> # endif >> #endif >> @@ -549,21 +532,12 @@ static inline void postProcess(const uint8_t src[], >> int srcStride, uint8_t dst[] >> #if ARCH_X86 && HAVE_INLINE_ASM >> // ordered per speed fastest first >> if (c->cpuCaps & AV_CPU_FLAG_SSE2) pp = >> postProcess_SSE2; >> - else if (c->cpuCaps & AV_CPU_FLAG_MMXEXT) pp = >> postProcess_MMX2; >> - else if (c->cpuCaps & AV_CPU_FLAG_3DNOW) pp = >> postProcess_3DNow; >> - else if (c->cpuCaps & AV_CPU_FLAG_MMX) pp = >> postProcess_MMX; >> #elif HAVE_ALTIVEC >> if (c->cpuCaps & AV_CPU_FLAG_ALTIVEC) pp = >> postProcess_altivec; >> #endif >> #else /* CONFIG_RUNTIME_CPUDETECT */ >> #if HAVE_SSE2_INLINE >> pp = postProcess_SSE2; >> -#elif HAVE_MMXEXT_INLINE >> - pp = postProcess_MMX2; >> -#elif HAVE_AMD3DNOW_INLINE >> - pp = postProcess_3DNow; >> -#elif HAVE_MMX_INLINE >> - pp = postProcess_MMX; >> #elif HAVE_ALTIVEC >> pp = postProcess_altivec; >> #endif >> @@ -877,9 +851,6 @@ av_cold pp_context *pp_get_context(int width, int >> height, int cpuCaps){ >> c->cpuCaps = av_get_cpu_flags(); >> } else { >> c->cpuCaps = 0; >> - if (cpuCaps & PP_CPU_CAPS_MMX) c->cpuCaps |= >> AV_CPU_FLAG_MMX; >> - if (cpuCaps & PP_CPU_CAPS_MMX2) c->cpuCaps |= >> AV_CPU_FLAG_MMXEXT; >> - if (cpuCaps & PP_CPU_CAPS_3DNOW) c->cpuCaps |= >> AV_CPU_FLAG_3DNOW; >> if (cpuCaps & PP_CPU_CAPS_ALTIVEC) c->cpuCaps |= >> AV_CPU_FLAG_ALTIVEC; >> } >> >> diff --git a/libpostproc/postprocess_template.c >> b/libpostproc/postprocess_template.c >> index 9f76b7c587..bcf7bdad66 100644 >> --- a/libpostproc/postprocess_template.c >> +++ b/libpostproc/postprocess_template.c >> @@ -20,7 +20,7 @@ >> >> /** >> * @file >> - * mmx/mmx2/3dnow postprocess code. >> + * mmx/mmx2/sse2 postprocess code. >> */ >> >> #include "libavutil/mem_internal.h" >> @@ -57,14 +57,6 @@ >> # define TEMPLATE_PP_MMXEXT 0 >> #endif >> >> -#ifdef TEMPLATE_PP_3DNOW >> -# undef TEMPLATE_PP_MMX >> -# define TEMPLATE_PP_MMX 1 >> -# define RENAME(a) a ## _3DNow >> -#else >> -# define TEMPLATE_PP_3DNOW 0 >> -#endif >> - >> #ifdef TEMPLATE_PP_SSE2 >> # undef TEMPLATE_PP_MMX >> # define TEMPLATE_PP_MMX 1 >> @@ -82,30 +74,19 @@ >> >> #if TEMPLATE_PP_MMXEXT >> #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t" >> -#elif TEMPLATE_PP_3DNOW >> -#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t" >> #endif >> #define PAVGB(a,b) REAL_PAVGB(a,b) >> >> #if TEMPLATE_PP_MMXEXT >> #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t" >> -#elif TEMPLATE_PP_MMX >> -#define PMINUB(b,a,t) \ >> - "movq " #a ", " #t " \n\t"\ >> - "psubusb " #b ", " #t " \n\t"\ >> - "psubb " #t ", " #a " \n\t" >> #endif >> >> #if TEMPLATE_PP_MMXEXT >> #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t" >> -#elif TEMPLATE_PP_MMX >> -#define PMAXUB(a,b) \ >> - "psubusb " #a ", " #b " \n\t"\ >> - "paddb " #a ", " #b " \n\t" >> #endif >> >> //FIXME? |255-0| = 1 (should not be a problem ...) >> -#if TEMPLATE_PP_MMX >> +#if TEMPLATE_PP_MMXEXT >> /** >> * Check if the middle 8x8 Block in the given 8x16 block is flat >> */ >> @@ -185,20 +166,8 @@ static inline int RENAME(vertClassify)(const uint8_t >> src[], int stride, PPContex >> "psubusb %%mm3, %%mm4 \n\t" >> >> " \n\t" >> -#if TEMPLATE_PP_MMXEXT >> "pxor %%mm7, %%mm7 \n\t" >> "psadbw %%mm7, %%mm0 \n\t" >> -#else >> - "movq %%mm0, %%mm1 \n\t" >> - "psrlw $8, %%mm0 \n\t" >> - "paddb %%mm1, %%mm0 \n\t" >> - "movq %%mm0, %%mm1 \n\t" >> - "psrlq $16, %%mm0 \n\t" >> - "paddb %%mm1, %%mm0 \n\t" >> - "movq %%mm0, %%mm1 \n\t" >> - "psrlq $32, %%mm0 \n\t" >> - "paddb %%mm1, %%mm0 \n\t" >> -#endif >> "movq %4, %%mm7 \n\t" // QP,..., QP >> "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP >> "psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> >> 0 >> @@ -219,7 +188,7 @@ static inline int RENAME(vertClassify)(const uint8_t >> src[], int stride, PPContex >> return 2; >> } >> } >> -#endif //TEMPLATE_PP_MMX >> +#endif //TEMPLATE_PP_MMXEXT >> >> /** >> * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 >> block in the middle) >> @@ -228,7 +197,7 @@ static inline int RENAME(vertClassify)(const uint8_t >> src[], int stride, PPContex >> #if !TEMPLATE_PP_ALTIVEC >> static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, >> PPContext *c) >> { >> -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_MMXEXT >> src+= stride*3; >> __asm__ volatile( //"movv %0 %1 %2\n\t" >> "movq %2, %%mm0 \n\t" // QP,..., QP >> @@ -355,7 +324,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, >> int stride, PPContext *c) >> : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb) >> : "%"FF_REG_a, "%"FF_REG_c >> ); >> -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#else //TEMPLATE_PP_MMXEXT >> const int l1= stride; >> const int l2= stride + l1; >> const int l3= stride + l2; >> @@ -394,7 +363,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, >> int stride, PPContext *c) >> >> src++; >> } >> -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_MMXEXT >> } >> #endif //TEMPLATE_PP_ALTIVEC >> >> @@ -407,7 +376,7 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, >> int stride, PPContext *c) >> */ >> static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, >> PPContext *co) >> { >> -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_MMXEXT >> src+= stride*3; >> >> __asm__ volatile( >> @@ -494,7 +463,7 @@ static inline void RENAME(vertX1Filter)(uint8_t *src, >> int stride, PPContext *co) >> NAMED_CONSTRAINTS_ADD(b01) >> : "%"FF_REG_a, "%"FF_REG_c >> ); >> -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#else //TEMPLATE_PP_MMXEXT >> >> const int l1= stride; >> const int l2= stride + l1; >> @@ -528,13 +497,13 @@ static inline void RENAME(vertX1Filter)(uint8_t >> *src, int stride, PPContext *co) >> } >> src++; >> } >> -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_MMXEXT >> } >> >> #if !TEMPLATE_PP_ALTIVEC >> static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, >> PPContext *c) >> { >> -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_MMXEXT >> /* >> uint8_t tmp[16]; >> const int l1= stride; >> @@ -816,239 +785,7 @@ static inline void RENAME(doVertDefFilter)(uint8_t >> src[], int stride, PPContext >> } >> } >> */ >> -#elif TEMPLATE_PP_MMX >> - DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte >> vars >> - src+= stride*4; >> - __asm__ volatile( >> - "pxor %%mm7, %%mm7 \n\t" >> -// 0 1 2 3 4 5 6 7 >> -// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1 >> -// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 >> - >> - "movq (%0), %%mm0 \n\t" >> - "movq %%mm0, %%mm1 \n\t" >> - "punpcklbw %%mm7, %%mm0 \n\t" // low part of line >> 0 >> - "punpckhbw %%mm7, %%mm1 \n\t" // high part of >> line 0 >> - >> - "movq (%0, %1), %%mm2 \n\t" >> - "lea (%0, %1, 2), %%"FF_REG_a" \n\t" >> - "movq %%mm2, %%mm3 \n\t" >> - "punpcklbw %%mm7, %%mm2 \n\t" // low part of line >> 1 >> - "punpckhbw %%mm7, %%mm3 \n\t" // high part of >> line 1 >> - >> - "movq (%%"FF_REG_a"), %%mm4 \n\t" >> - "movq %%mm4, %%mm5 \n\t" >> - "punpcklbw %%mm7, %%mm4 \n\t" // low part of line >> 2 >> - "punpckhbw %%mm7, %%mm5 \n\t" // high part of >> line 2 >> - >> - "paddw %%mm0, %%mm0 \n\t" // 2L0 >> - "paddw %%mm1, %%mm1 \n\t" // 2H0 >> - "psubw %%mm4, %%mm2 \n\t" // L1 - L2 >> - "psubw %%mm5, %%mm3 \n\t" // H1 - H2 >> - "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2 >> - "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2 >> - >> - "psllw $2, %%mm2 \n\t" // 4L1 - 4L2 >> - "psllw $2, %%mm3 \n\t" // 4H1 - 4H2 >> - "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 >> - "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 >> - >> - "movq (%%"FF_REG_a", %1), %%mm2 \n\t" >> - "movq %%mm2, %%mm3 \n\t" >> - "punpcklbw %%mm7, %%mm2 \n\t" // L3 >> - "punpckhbw %%mm7, %%mm3 \n\t" // H3 >> - >> - "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 >> - L3 >> - "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 >> - H3 >> - "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 >> - 2L3 >> - "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 >> - 2H3 >> - "movq %%mm0, (%3) \n\t" // 2L0 - 5L1 + 5L2 >> - 2L3 >> - "movq %%mm1, 8(%3) \n\t" // 2H0 - 5H1 + 5H2 >> - 2H3 >> - >> - "movq (%%"FF_REG_a", %1, 2), %%mm0 \n\t" >> - "movq %%mm0, %%mm1 \n\t" >> - "punpcklbw %%mm7, %%mm0 \n\t" // L4 >> - "punpckhbw %%mm7, %%mm1 \n\t" // H4 >> - >> - "psubw %%mm0, %%mm2 \n\t" // L3 - L4 >> - "psubw %%mm1, %%mm3 \n\t" // H3 - H4 >> - "movq %%mm2, 16(%3) \n\t" // L3 - L4 >> - "movq %%mm3, 24(%3) \n\t" // H3 - H4 >> - "paddw %%mm4, %%mm4 \n\t" // 2L2 >> - "paddw %%mm5, %%mm5 \n\t" // 2H2 >> - "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4 >> - "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4 >> - >> - "lea (%%"FF_REG_a", %1), %0 \n\t" >> - "psllw $2, %%mm2 \n\t" // 4L3 - 4L4 >> - "psllw $2, %%mm3 \n\t" // 4H3 - 4H4 >> - "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 >> - "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 >> -//50 opcodes so far >> - "movq (%0, %1, 2), %%mm2 \n\t" >> - "movq %%mm2, %%mm3 \n\t" >> - "punpcklbw %%mm7, %%mm2 \n\t" // L5 >> - "punpckhbw %%mm7, %%mm3 \n\t" // H5 >> - "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 >> - L5 >> - "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 >> - H5 >> - "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 >> - 2L5 >> - "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 >> - 2H5 >> - >> - "movq (%%"FF_REG_a", %1, 4), %%mm6 \n\t" >> - "punpcklbw %%mm7, %%mm6 \n\t" // L6 >> - "psubw %%mm6, %%mm2 \n\t" // L5 - L6 >> - "movq (%%"FF_REG_a", %1, 4), %%mm6 \n\t" >> - "punpckhbw %%mm7, %%mm6 \n\t" // H6 >> - "psubw %%mm6, %%mm3 \n\t" // H5 - H6 >> - >> - "paddw %%mm0, %%mm0 \n\t" // 2L4 >> - "paddw %%mm1, %%mm1 \n\t" // 2H4 >> - "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6 >> - "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6 >> - >> - "psllw $2, %%mm2 \n\t" // 4L5 - 4L6 >> - "psllw $2, %%mm3 \n\t" // 4H5 - 4H6 >> - "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 >> - "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 >> - >> - "movq (%0, %1, 4), %%mm2 \n\t" >> - "movq %%mm2, %%mm3 \n\t" >> - "punpcklbw %%mm7, %%mm2 \n\t" // L7 >> - "punpckhbw %%mm7, %%mm3 \n\t" // H7 >> - >> - "paddw %%mm2, %%mm2 \n\t" // 2L7 >> - "paddw %%mm3, %%mm3 \n\t" // 2H7 >> - "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 >> - 2L7 >> - "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 >> - 2H7 >> - >> - "movq (%3), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 >> - 2L3 >> - "movq 8(%3), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 >> - 2H3 >> - >> -#if TEMPLATE_PP_MMXEXT >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "psubw %%mm0, %%mm6 \n\t" >> - "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 >> - 2L7| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "psubw %%mm1, %%mm6 \n\t" >> - "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 >> - 2H7| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "psubw %%mm2, %%mm6 \n\t" >> - "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 >> - 2L3| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "psubw %%mm3, %%mm6 \n\t" >> - "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 >> - 2H3| >> -#else >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm0, %%mm6 \n\t" >> - "pxor %%mm6, %%mm0 \n\t" >> - "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 >> - 2L7| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm1, %%mm6 \n\t" >> - "pxor %%mm6, %%mm1 \n\t" >> - "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 >> - 2H7| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm2, %%mm6 \n\t" >> - "pxor %%mm6, %%mm2 \n\t" >> - "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 >> - 2L3| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm3, %%mm6 \n\t" >> - "pxor %%mm6, %%mm3 \n\t" >> - "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 >> - 2H3| >> -#endif >> - >> -#if TEMPLATE_PP_MMXEXT >> - "pminsw %%mm2, %%mm0 \n\t" >> - "pminsw %%mm3, %%mm1 \n\t" >> -#else >> - "movq %%mm0, %%mm6 \n\t" >> - "psubusw %%mm2, %%mm6 \n\t" >> - "psubw %%mm6, %%mm0 \n\t" >> - "movq %%mm1, %%mm6 \n\t" >> - "psubusw %%mm3, %%mm6 \n\t" >> - "psubw %%mm6, %%mm1 \n\t" >> -#endif >> - >> - "movd %2, %%mm2 \n\t" // QP >> - "punpcklbw %%mm7, %%mm2 \n\t" >> - >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + >> 5L4 - 2L5) >> - "pxor %%mm6, %%mm4 \n\t" >> - "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 >> - 2L5| >> - "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + >> 5H4 - 2H5) >> - "pxor %%mm7, %%mm5 \n\t" >> - "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 >> - 2H5| >> -// 100 opcodes >> - "psllw $3, %%mm2 \n\t" // 8QP >> - "movq %%mm2, %%mm3 \n\t" // 8QP >> - "pcmpgtw %%mm4, %%mm2 \n\t" >> - "pcmpgtw %%mm5, %%mm3 \n\t" >> - "pand %%mm2, %%mm4 \n\t" >> - "pand %%mm3, %%mm5 \n\t" >> - >> - >> - "psubusw %%mm0, %%mm4 \n\t" // hd >> - "psubusw %%mm1, %%mm5 \n\t" // ld >> - >> - >> - "movq "MANGLE(w05)", %%mm2 \n\t" // 5 >> - "pmullw %%mm2, %%mm4 \n\t" >> - "pmullw %%mm2, %%mm5 \n\t" >> - "movq "MANGLE(w20)", %%mm2 \n\t" // 32 >> - "paddw %%mm2, %%mm4 \n\t" >> - "paddw %%mm2, %%mm5 \n\t" >> - "psrlw $6, %%mm4 \n\t" >> - "psrlw $6, %%mm5 \n\t" >> - >> - "movq 16(%3), %%mm0 \n\t" // L3 - L4 >> - "movq 24(%3), %%mm1 \n\t" // H3 - H4 >> - >> - "pxor %%mm2, %%mm2 \n\t" >> - "pxor %%mm3, %%mm3 \n\t" >> - >> - "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4) >> - "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4) >> - "pxor %%mm2, %%mm0 \n\t" >> - "pxor %%mm3, %%mm1 \n\t" >> - "psubw %%mm2, %%mm0 \n\t" // |L3-L4| >> - "psubw %%mm3, %%mm1 \n\t" // |H3-H4| >> - "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2 >> - "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2 >> - >> - "pxor %%mm6, %%mm2 \n\t" >> - "pxor %%mm7, %%mm3 \n\t" >> - "pand %%mm2, %%mm4 \n\t" >> - "pand %%mm3, %%mm5 \n\t" >> - >> -#if TEMPLATE_PP_MMXEXT >> - "pminsw %%mm0, %%mm4 \n\t" >> - "pminsw %%mm1, %%mm5 \n\t" >> -#else >> - "movq %%mm4, %%mm2 \n\t" >> - "psubusw %%mm0, %%mm2 \n\t" >> - "psubw %%mm2, %%mm4 \n\t" >> - "movq %%mm5, %%mm2 \n\t" >> - "psubusw %%mm1, %%mm2 \n\t" >> - "psubw %%mm2, %%mm5 \n\t" >> -#endif >> - "pxor %%mm6, %%mm4 \n\t" >> - "pxor %%mm7, %%mm5 \n\t" >> - "psubw %%mm6, %%mm4 \n\t" >> - "psubw %%mm7, %%mm5 \n\t" >> - "packsswb %%mm5, %%mm4 \n\t" >> - "movq (%0), %%mm0 \n\t" >> - "paddb %%mm4, %%mm0 \n\t" >> - "movq %%mm0, (%0) \n\t" >> - "movq (%0, %1), %%mm0 \n\t" >> - "psubb %%mm4, %%mm0 \n\t" >> - "movq %%mm0, (%0, %1) \n\t" >> - >> - : "+r" (src) >> - : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp) >> - NAMED_CONSTRAINTS_ADD(w05,w20) >> - : "%"FF_REG_a >> - ); >> -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#else //TEMPLATE_PP_MMXEXT >> const int l1= stride; >> const int l2= stride + l1; >> const int l3= stride + l2; >> @@ -1086,14 +823,14 @@ static inline void RENAME(doVertDefFilter)(uint8_t >> src[], int stride, PPContext >> } >> src++; >> } >> -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_MMXEXT >> } >> #endif //TEMPLATE_PP_ALTIVEC >> >> #if !TEMPLATE_PP_ALTIVEC >> static inline void RENAME(dering)(uint8_t src[], int stride, PPContext >> *c) >> { >> -#if HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) >> +#if HAVE_7REGS && TEMPLATE_PP_MMXEXT >> DECLARE_ALIGNED(8, uint64_t, tmp)[3]; >> __asm__ volatile( >> "pxor %%mm6, %%mm6 \n\t" >> @@ -1113,20 +850,10 @@ static inline void RENAME(dering)(uint8_t src[], >> int stride, PPContext *c) >> >> #undef REAL_FIND_MIN_MAX >> #undef FIND_MIN_MAX >> -#if TEMPLATE_PP_MMXEXT >> #define REAL_FIND_MIN_MAX(addr)\ >> "movq " #addr ", %%mm0 \n\t"\ >> "pminub %%mm0, %%mm7 \n\t"\ >> "pmaxub %%mm0, %%mm6 \n\t" >> -#else >> -#define REAL_FIND_MIN_MAX(addr)\ >> - "movq " #addr ", %%mm0 \n\t"\ >> - "movq %%mm7, %%mm1 \n\t"\ >> - "psubusb %%mm0, %%mm6 \n\t"\ >> - "paddb %%mm0, %%mm6 \n\t"\ >> - "psubusb %%mm0, %%mm1 \n\t"\ >> - "psubb %%mm1, %%mm7 \n\t" >> -#endif >> #define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr) >> >> FIND_MIN_MAX((%%FF_REGa)) >> @@ -1140,49 +867,20 @@ FIND_MIN_MAX((%0, %1, 8)) >> >> "movq %%mm7, %%mm4 \n\t" >> "psrlq $8, %%mm7 \n\t" >> -#if TEMPLATE_PP_MMXEXT >> "pminub %%mm4, %%mm7 \n\t" // min of pixels >> "pshufw $0xF9, %%mm7, %%mm4 \n\t" >> "pminub %%mm4, %%mm7 \n\t" // min of pixels >> "pshufw $0xFE, %%mm7, %%mm4 \n\t" >> "pminub %%mm4, %%mm7 \n\t" >> -#else >> - "movq %%mm7, %%mm1 \n\t" >> - "psubusb %%mm4, %%mm1 \n\t" >> - "psubb %%mm1, %%mm7 \n\t" >> - "movq %%mm7, %%mm4 \n\t" >> - "psrlq $16, %%mm7 \n\t" >> - "movq %%mm7, %%mm1 \n\t" >> - "psubusb %%mm4, %%mm1 \n\t" >> - "psubb %%mm1, %%mm7 \n\t" >> - "movq %%mm7, %%mm4 \n\t" >> - "psrlq $32, %%mm7 \n\t" >> - "movq %%mm7, %%mm1 \n\t" >> - "psubusb %%mm4, %%mm1 \n\t" >> - "psubb %%mm1, %%mm7 \n\t" >> -#endif >> >> >> "movq %%mm6, %%mm4 \n\t" >> "psrlq $8, %%mm6 \n\t" >> -#if TEMPLATE_PP_MMXEXT >> "pmaxub %%mm4, %%mm6 \n\t" // max of pixels >> "pshufw $0xF9, %%mm6, %%mm4 \n\t" >> "pmaxub %%mm4, %%mm6 \n\t" >> "pshufw $0xFE, %%mm6, %%mm4 \n\t" >> "pmaxub %%mm4, %%mm6 \n\t" >> -#else >> - "psubusb %%mm4, %%mm6 \n\t" >> - "paddb %%mm4, %%mm6 \n\t" >> - "movq %%mm6, %%mm4 \n\t" >> - "psrlq $16, %%mm6 \n\t" >> - "psubusb %%mm4, %%mm6 \n\t" >> - "paddb %%mm4, %%mm6 \n\t" >> - "movq %%mm6, %%mm4 \n\t" >> - "psrlq $32, %%mm6 \n\t" >> - "psubusb %%mm4, %%mm6 \n\t" >> - "paddb %%mm4, %%mm6 \n\t" >> -#endif >> "movq %%mm6, %%mm0 \n\t" // max >> "psubb %%mm7, %%mm6 \n\t" // max - min >> "push %%"FF_REG_a" \n\t" >> @@ -1320,7 +1018,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, >> 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5, >> NAMED_CONSTRAINTS_ADD(deringThreshold,b00,b02,b08) >> : "%"FF_REG_a, "%"FF_REG_d >> ); >> -#else // HAVE_7REGS && (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) >> +#else // HAVE_7REGS && TEMPLATE_PP_MMXEXT >> int y; >> int min=255; >> int max=0; >> @@ -1438,7 +1136,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, >> 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5, >> // src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255; >> } >> #endif >> -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_MMXEXT >> } >> #endif //TEMPLATE_PP_ALTIVEC >> >> @@ -1450,7 +1148,7 @@ DERING_CORE((%0, %1, 8) ,(%%FF_REGd, %1, >> 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5, >> */ >> static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], >> int stride) >> { >> -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_MMXEXT >> src+= 4*stride; >> __asm__ volatile( >> "lea (%0, %1), %%"FF_REG_a" \n\t" >> @@ -1503,14 +1201,13 @@ static inline void >> RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int strid >> */ >> static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int >> stride) >> { >> -#if TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_SSE2 >> src+= stride*3; >> __asm__ volatile( >> "lea (%0, %1), %%"FF_REG_a" \n\t" >> "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t" >> "lea (%%"FF_REG_d", %1, 4), %%"FF_REG_c"\n\t" >> "add %1, %%"FF_REG_c" \n\t" >> -#if TEMPLATE_PP_SSE2 >> "pxor %%xmm7, %%xmm7 \n\t" >> #define REAL_DEINT_CUBIC(a,b,c,d,e)\ >> "movq " #a ", %%xmm0 \n\t"\ >> @@ -1526,33 +1223,6 @@ static inline void >> RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride >> "psubw %%xmm0, %%xmm1 \n\t"\ >> "packuswb %%xmm1, %%xmm1 \n\t"\ >> "movlps %%xmm1, " #c " \n\t" >> -#else //TEMPLATE_PP_SSE2 >> - "pxor %%mm7, %%mm7 \n\t" >> -// 0 1 2 3 4 5 6 7 8 >> 9 10 >> -// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 >> %0+8%1 edx+4%1 ecx >> - >> -#define REAL_DEINT_CUBIC(a,b,c,d,e)\ >> - "movq " #a ", %%mm0 \n\t"\ >> - "movq " #b ", %%mm1 \n\t"\ >> - "movq " #d ", %%mm2 \n\t"\ >> - "movq " #e ", %%mm3 \n\t"\ >> - PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\ >> - PAVGB(%%mm3, %%mm0) /* (a+e) /2 */\ >> - "movq %%mm0, %%mm2 \n\t"\ >> - "punpcklbw %%mm7, %%mm0 \n\t"\ >> - "punpckhbw %%mm7, %%mm2 \n\t"\ >> - "movq %%mm1, %%mm3 \n\t"\ >> - "punpcklbw %%mm7, %%mm1 \n\t"\ >> - "punpckhbw %%mm7, %%mm3 \n\t"\ >> - "psubw %%mm1, %%mm0 \n\t" /* L(a+e - >> (b+d))/2 */\ >> - "psubw %%mm3, %%mm2 \n\t" /* H(a+e - >> (b+d))/2 */\ >> - "psraw $3, %%mm0 \n\t" /* L(a+e - >> (b+d))/16 */\ >> - "psraw $3, %%mm2 \n\t" /* H(a+e - >> (b+d))/16 */\ >> - "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a >> - e)/16 */\ >> - "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a >> - e)/16 */\ >> - "packuswb %%mm3, %%mm1 \n\t"\ >> - "movq %%mm1, " #c " \n\t" >> -#endif //TEMPLATE_PP_SSE2 >> #define DEINT_CUBIC(a,b,c,d,e) REAL_DEINT_CUBIC(a,b,c,d,e) >> >> DEINT_CUBIC((%0) , (%%FF_REGa, %1), (%%FF_REGa, %1, 2), (%0, >> %1, 4) , (%%FF_REGd, %1)) >> @@ -1562,13 +1232,11 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , >> (%%FF_REGd, %1, 4), (%%FF_REGc) >> >> : : "r" (src), "r" ((x86_reg)stride) >> : >> -#if TEMPLATE_PP_SSE2 >> XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm7",) >> -#endif >> "%"FF_REG_a, "%"FF_REG_d, "%"FF_REG_c >> ); >> #undef REAL_DEINT_CUBIC >> -#else //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#else //TEMPLATE_PP_SSE2 >> int x; >> src+= stride*3; >> for(x=0; x<8; x++){ >> @@ -1578,7 +1246,7 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , >> (%%FF_REGd, %1, 4), (%%FF_REGc) >> src[stride*9] = av_clip_uint8((-src[stride*6] + 9*src[stride*8] + >> 9*src[stride*10] - src[stride*12])>>4); >> src++; >> } >> -#endif //TEMPLATE_PP_SSE2 || TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_SSE2 >> } >> >> /** >> @@ -1590,7 +1258,7 @@ DEINT_CUBIC((%%FF_REGd, %1), (%0, %1, 8) , >> (%%FF_REGd, %1, 4), (%%FF_REGc) >> */ >> static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, >> uint8_t *tmp) >> { >> -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_MMXEXT >> src+= stride*4; >> __asm__ volatile( >> "lea (%0, %1), %%"FF_REG_a" \n\t" >> @@ -1639,7 +1307,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, >> %1, 8) , (%%FF_REGd, %1, 4 >> : : "r" (src), "r" ((x86_reg)stride), "r"(tmp) >> : "%"FF_REG_a, "%"FF_REG_d >> ); >> -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#else //TEMPLATE_PP_MMXEXT >> int x; >> src+= stride*4; >> for(x=0; x<8; x++){ >> @@ -1657,7 +1325,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, >> %1, 8) , (%%FF_REGd, %1, 4 >> >> src++; >> } >> -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_MMXEXT >> } >> >> /** >> @@ -1669,7 +1337,7 @@ DEINT_FF((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, >> %1, 8) , (%%FF_REGd, %1, 4 >> */ >> static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, >> uint8_t *tmp, uint8_t *tmp2) >> { >> -#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS >> +#if TEMPLATE_PP_MMXEXT && HAVE_6REGS >> src+= stride*4; >> __asm__ volatile( >> "lea (%0, %1), %%"FF_REG_a" \n\t" >> @@ -1729,7 +1397,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, >> 8) , (%%FF_REGd, %1, 4 >> : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2) >> : "%"FF_REG_a, "%"FF_REG_d >> ); >> -#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS >> +#else //TEMPLATE_PP_MMXEXT && HAVE_6REGS >> int x; >> src+= stride*4; >> for(x=0; x<8; x++){ >> @@ -1758,7 +1426,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, >> 8) , (%%FF_REGd, %1, 4 >> >> src++; >> } >> -#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS >> +#endif // TEMPLATE_PP_MMXEXT && HAVE_6REGS >> } >> >> /** >> @@ -1770,7 +1438,7 @@ DEINT_L5(%%mm1, %%mm0, (%%FF_REGd, %1, 2), (%0, %1, >> 8) , (%%FF_REGd, %1, 4 >> */ >> static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int >> stride, uint8_t *tmp) >> { >> -#if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#if TEMPLATE_PP_MMXEXT >> src+= 4*stride; >> __asm__ volatile( >> "lea (%0, %1), %%"FF_REG_a" \n\t" >> @@ -1817,7 +1485,7 @@ static inline void >> RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin >> : : "r" (src), "r" ((x86_reg)stride), "r" (tmp) >> : "%"FF_REG_a, "%"FF_REG_d >> ); >> -#else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#else //TEMPLATE_PP_MMXEXT >> int a, b, c, x; >> src+= 4*stride; >> >> @@ -1860,7 +1528,7 @@ static inline void >> RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin >> src += 4; >> tmp += 4; >> } >> -#endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW >> +#endif //TEMPLATE_PP_MMXEXT >> } >> >> /** >> @@ -1871,9 +1539,8 @@ static inline void >> RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin >> */ >> static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride) >> { >> -#if TEMPLATE_PP_MMX >> - src+= 4*stride; >> #if TEMPLATE_PP_MMXEXT >> + src+= 4*stride; >> __asm__ volatile( >> "lea (%0, %1), %%"FF_REG_a" \n\t" >> "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t" >> @@ -1922,48 +1589,6 @@ static inline void >> RENAME(deInterlaceMedian)(uint8_t src[], int stride) >> : "%"FF_REG_a, "%"FF_REG_d >> ); >> >> -#else // MMX without MMX2 >> - __asm__ volatile( >> - "lea (%0, %1), %%"FF_REG_a" \n\t" >> - "lea (%%"FF_REG_a", %1, 4), %%"FF_REG_d"\n\t" >> -// 0 1 2 3 4 5 6 7 8 >> 9 >> -// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 >> %0+8%1 edx+4%1 >> - "pxor %%mm7, %%mm7 \n\t" >> - >> -#define REAL_MEDIAN(a,b,c)\ >> - "movq " #a ", %%mm0 \n\t"\ >> - "movq " #b ", %%mm2 \n\t"\ >> - "movq " #c ", %%mm1 \n\t"\ >> - "movq %%mm0, %%mm3 \n\t"\ >> - "movq %%mm1, %%mm4 \n\t"\ >> - "movq %%mm2, %%mm5 \n\t"\ >> - "psubusb %%mm1, %%mm3 \n\t"\ >> - "psubusb %%mm2, %%mm4 \n\t"\ >> - "psubusb %%mm0, %%mm5 \n\t"\ >> - "pcmpeqb %%mm7, %%mm3 \n\t"\ >> - "pcmpeqb %%mm7, %%mm4 \n\t"\ >> - "pcmpeqb %%mm7, %%mm5 \n\t"\ >> - "movq %%mm3, %%mm6 \n\t"\ >> - "pxor %%mm4, %%mm3 \n\t"\ >> - "pxor %%mm5, %%mm4 \n\t"\ >> - "pxor %%mm6, %%mm5 \n\t"\ >> - "por %%mm3, %%mm1 \n\t"\ >> - "por %%mm4, %%mm2 \n\t"\ >> - "por %%mm5, %%mm0 \n\t"\ >> - "pand %%mm2, %%mm0 \n\t"\ >> - "pand %%mm1, %%mm0 \n\t"\ >> - "movq %%mm0, " #b " \n\t" >> -#define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c) >> - >> -MEDIAN((%0) , (%%FF_REGa) , (%%FF_REGa, %1)) >> -MEDIAN((%%FF_REGa, %1), (%%FF_REGa, %1, 2), (%0, %1, 4)) >> -MEDIAN((%0, %1, 4) , (%%FF_REGd) , (%%FF_REGd, %1)) >> -MEDIAN((%%FF_REGd, %1), (%%FF_REGd, %1, 2), (%0, %1, 8)) >> - >> - : : "r" (src), "r" ((x86_reg)stride) >> - : "%"FF_REG_a, "%"FF_REG_d >> - ); >> -#endif //TEMPLATE_PP_MMXEXT >> #else //TEMPLATE_PP_MMX >> int x, y; >> src+= 4*stride; >> @@ -2165,7 +1790,7 @@ static inline void RENAME(tempNoiseReducer)(uint8_t >> *src, int stride, >> >> #define FAST_L2_DIFF >> //#define L1_DIFF //u should change the thresholds too if u try that one >> -#if (TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS >> +#if TEMPLATE_PP_MMXEXT && HAVE_6REGS >> __asm__ volatile( >> "lea (%2, %2, 2), %%"FF_REG_a" \n\t" // 3*stride >> "lea (%2, %2, 4), %%"FF_REG_d" \n\t" // 5*stride >> @@ -2454,7 +2079,7 @@ L2_DIFF_CORE((%0, %%FF_REGc) , (%1, %%FF_REGc)) >> NAMED_CONSTRAINTS_ADD(b80) >> : "%"FF_REG_a, "%"FF_REG_d, "%"FF_REG_c, "memory" >> ); >> -#else //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS >> +#else //TEMPLATE_PP_MMXEXT && HAVE_6REGS >> { >> int y; >> int d=0; >> @@ -2537,11 +2162,11 @@ Switch between >> } >> } >> } >> -#endif //(TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW) && HAVE_6REGS >> +#endif //TEMPLATE_PP_MMXEXT && HAVE_6REGS >> } >> #endif //TEMPLATE_PP_ALTIVEC >> >> -#if TEMPLATE_PP_MMX >> +#if TEMPLATE_PP_MMXEXT >> /** >> * accurate deblock filter >> */ >> @@ -2945,7 +2570,6 @@ static av_always_inline void >> RENAME(do_a_deblock)(uint8_t *src, int step, int st >> "movq (%4), %%mm2 \n\t" // 2L0 - 5L1 + >> 5L2 - 2L3 >> "movq 8(%4), %%mm3 \n\t" // 2H0 - 5H1 + >> 5H2 - 2H3 >> >> -#if TEMPLATE_PP_MMXEXT >> "movq %%mm7, %%mm6 \n\t" // 0 >> "psubw %%mm0, %%mm6 \n\t" >> "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + >> 5L6 - 2L7| >> @@ -2958,36 +2582,9 @@ static av_always_inline void >> RENAME(do_a_deblock)(uint8_t *src, int step, int st >> "movq %%mm7, %%mm6 \n\t" // 0 >> "psubw %%mm3, %%mm6 \n\t" >> "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + >> 5H2 - 2H3| >> -#else >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm0, %%mm6 \n\t" >> - "pxor %%mm6, %%mm0 \n\t" >> - "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + >> 5L6 - 2L7| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm1, %%mm6 \n\t" >> - "pxor %%mm6, %%mm1 \n\t" >> - "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + >> 5H6 - 2H7| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm2, %%mm6 \n\t" >> - "pxor %%mm6, %%mm2 \n\t" >> - "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + >> 5L2 - 2L3| >> - "movq %%mm7, %%mm6 \n\t" // 0 >> - "pcmpgtw %%mm3, %%mm6 \n\t" >> - "pxor %%mm6, %%mm3 \n\t" >> - "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + >> 5H2 - 2H3| >> -#endif >> >> -#if TEMPLATE_PP_MMXEXT >> "pminsw %%mm2, %%mm0 \n\t" >> "pminsw %%mm3, %%mm1 \n\t" >> -#else >> - "movq %%mm0, %%mm6 \n\t" >> - "psubusw %%mm2, %%mm6 \n\t" >> - "psubw %%mm6, %%mm0 \n\t" >> - "movq %%mm1, %%mm6 \n\t" >> - "psubusw %%mm3, %%mm6 \n\t" >> - "psubw %%mm6, %%mm1 \n\t" >> -#endif >> >> "movd %2, %%mm2 \n\t" // QP >> "punpcklbw %%mm7, %%mm2 \n\t" >> @@ -3041,17 +2638,8 @@ static av_always_inline void >> RENAME(do_a_deblock)(uint8_t *src, int step, int st >> "pand %%mm2, %%mm4 \n\t" >> "pand %%mm3, %%mm5 \n\t" >> >> -#if TEMPLATE_PP_MMXEXT >> "pminsw %%mm0, %%mm4 \n\t" >> "pminsw %%mm1, %%mm5 \n\t" >> -#else >> - "movq %%mm4, %%mm2 \n\t" >> - "psubusw %%mm0, %%mm2 \n\t" >> - "psubw %%mm2, %%mm4 \n\t" >> - "movq %%mm5, %%mm2 \n\t" >> - "psubusw %%mm1, %%mm2 \n\t" >> - "psubw %%mm2, %%mm5 \n\t" >> -#endif >> "pxor %%mm6, %%mm4 \n\t" >> "pxor %%mm7, %%mm5 \n\t" >> "psubw %%mm6, %%mm4 \n\t" >> @@ -3088,18 +2676,14 @@ static void RENAME(postProcess)(const uint8_t >> src[], int srcStride, uint8_t dst[ >> static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const >> uint8_t src[], int srcStride, >> int levelFix, int64_t >> *packedOffsetAndScale) >> { >> -#if !TEMPLATE_PP_MMX || !HAVE_6REGS >> - int i; >> -#endif >> if(levelFix){ >> -#if TEMPLATE_PP_MMX && HAVE_6REGS >> +#if TEMPLATE_PP_MMXEXT && HAVE_6REGS >> __asm__ volatile( >> "movq (%%"FF_REG_a"), %%mm2 \n\t" // packedYOffset >> "movq 8(%%"FF_REG_a"), %%mm3 \n\t" // packedYScale >> "lea (%2,%4), %%"FF_REG_a" \n\t" >> "lea (%3,%5), %%"FF_REG_d" \n\t" >> "pxor %%mm4, %%mm4 \n\t" >> -#if TEMPLATE_PP_MMXEXT >> #define REAL_SCALED_CPY(src1, src2, dst1, dst2) >> \ >> "movq " #src1 ", %%mm0 \n\t"\ >> "movq " #src1 ", %%mm5 \n\t"\ >> @@ -3122,34 +2706,6 @@ static inline void RENAME(blockCopy)(uint8_t dst[], >> int dstStride, const uint8_t >> "movq %%mm0, " #dst1 " \n\t"\ >> "movq %%mm1, " #dst2 " \n\t"\ >> >> -#else //TEMPLATE_PP_MMXEXT >> -#define REAL_SCALED_CPY(src1, src2, dst1, dst2) >> \ >> - "movq " #src1 ", %%mm0 \n\t"\ >> - "movq " #src1 ", %%mm5 \n\t"\ >> - "punpcklbw %%mm4, %%mm0 \n\t"\ >> - "punpckhbw %%mm4, %%mm5 \n\t"\ >> - "psubw %%mm2, %%mm0 \n\t"\ >> - "psubw %%mm2, %%mm5 \n\t"\ >> - "movq " #src2 ", %%mm1 \n\t"\ >> - "psllw $6, %%mm0 \n\t"\ >> - "psllw $6, %%mm5 \n\t"\ >> - "pmulhw %%mm3, %%mm0 \n\t"\ >> - "movq " #src2 ", %%mm6 \n\t"\ >> - "pmulhw %%mm3, %%mm5 \n\t"\ >> - "punpcklbw %%mm4, %%mm1 \n\t"\ >> - "punpckhbw %%mm4, %%mm6 \n\t"\ >> - "psubw %%mm2, %%mm1 \n\t"\ >> - "psubw %%mm2, %%mm6 \n\t"\ >> - "psllw $6, %%mm1 \n\t"\ >> - "psllw $6, %%mm6 \n\t"\ >> - "pmulhw %%mm3, %%mm1 \n\t"\ >> - "pmulhw %%mm3, %%mm6 \n\t"\ >> - "packuswb %%mm5, %%mm0 \n\t"\ >> - "packuswb %%mm6, %%mm1 \n\t"\ >> - "movq %%mm0, " #dst1 " \n\t"\ >> - "movq %%mm1, " #dst2 " \n\t"\ >> - >> -#endif //TEMPLATE_PP_MMXEXT >> #define SCALED_CPY(src1, src2, dst1, dst2)\ >> REAL_SCALED_CPY(src1, src2, dst1, dst2) >> >> @@ -3170,7 +2726,7 @@ SCALED_CPY((%%FF_REGa, %4), (%%FF_REGa, %4, 2), >> (%%FF_REGd, %5), (%%FF_REGd, %5, >> : "%"FF_REG_d >> ); >> #else //TEMPLATE_PP_MMX && HAVE_6REGS >> - for(i=0; i<8; i++) >> + for (int i = 0; i < 8; i++) >> memcpy( &(dst[dstStride*i]), >> &(src[srcStride*i]), BLOCK_SIZE); >> #endif //TEMPLATE_PP_MMX && HAVE_6REGS >> @@ -3203,7 +2759,7 @@ SIMPLE_CPY((%%FF_REGa, %2), (%%FF_REGa, %2, 2), >> (%%FF_REGd, %3), (%%FF_REGd, %3, >> : "%"FF_REG_a, "%"FF_REG_d >> ); >> #else //TEMPLATE_PP_MMX && HAVE_6REGS >> - for(i=0; i<8; i++) >> + for (int i = 0; i < 8; i++) >> memcpy( &(dst[dstStride*i]), >> &(src[srcStride*i]), BLOCK_SIZE); >> #endif //TEMPLATE_PP_MMX && HAVE_6REGS >> @@ -3696,9 +3252,7 @@ static void RENAME(postProcess)(const uint8_t src[], >> int srcStride, uint8_t dst[ >> } >> } >> } >> -#if TEMPLATE_PP_3DNOW >> - __asm__ volatile("femms"); >> -#elif TEMPLATE_PP_MMX >> +#if TEMPLATE_PP_MMX >> __asm__ volatile("emms"); >> #endif >> >> @@ -3734,5 +3288,4 @@ static void RENAME(postProcess)(const uint8_t src[], >> int srcStride, uint8_t dst[ >> #undef TEMPLATE_PP_ALTIVEC >> #undef TEMPLATE_PP_MMX >> #undef TEMPLATE_PP_MMXEXT >> -#undef TEMPLATE_PP_3DNOW >> #undef TEMPLATE_PP_SSE2 > > Will apply this patchset tomorrow unless there are objections. > LGTM > - Andreas > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2022-11-08 15:43 UTC | newest] Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2022-11-06 16:27 [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 2/7] avcodec/x86/cavsdsp: Remove unused 3DNow-macro Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 3/7] swscale/utils: Remove obsolete 3DNow reference Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 4/7] postproc/postprocess_template: Fix alignment Andreas Rheinhardt 2022-11-06 16:30 ` [FFmpeg-devel] [PATCH 5/7] avfilter/vf_qp: Don't store NAN in int Andreas Rheinhardt 2022-11-06 16:31 ` [FFmpeg-devel] [PATCH 6/7] avfilter/phase_template: Fix left-shift of negative numbers Andreas Rheinhardt 2022-11-06 16:31 ` [FFmpeg-devel] [PATCH 7/7] avfilter/vf_hqdn3d: " Andreas Rheinhardt 2022-11-08 15:41 ` [FFmpeg-devel] [PATCH 1/7] postproc/postprocess: Remove obsolete MMX(EXT)/3Dnow functions Andreas Rheinhardt 2022-11-08 15:43 ` Paul B Mahol
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel This inbox may be cloned and mirrored by anyone: git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git # If you have public-inbox 1.1+ installed, you may # initialize and index your mirror using the following commands: public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \ ffmpegdev@gitmailbox.com public-inbox-index ffmpegdev Example config snippet for mirrors. AGPL code for this site: git clone https://public-inbox.org/public-inbox.git