Merge remote-tracking branch 'qatar/master'

* qatar/master: (23 commits)
  x86inc: use sse versions of common macros instead of sse2 when applicable
  doc/APIchanges: add missing dates and hashes
  lavf: don't return from void av_update_cur_dts()
  Changelog: add more entries.
  Changelog: update ffmpeg/avconv incompatibility list.
  avconv: remove some redundant temporary variables.
  avconv: fix broken indentation
  avconv: move copy_initial_nonkeyframes to the options context.
  avconv: use file:stream instead of file.stream in log messages.
  doc/avconv: elaborate on basic functionality.
  doc/avconv: -sample_fmts, not -help sample_fmts prints the sample formats
  openssl: Only use CRYPTO_set_id_callback on OpenSSL < 1.0.0
  Call avformat_network_init/deinit in the programs
  Remove leftover includes of strings.h
  avutil: Don't allow using strcasecmp/strncasecmp
  Replace all usage of strcasecmp/strncasecmp
  avstring: Add locale independent implementations of strcasecmp/strncasecmp
  avstring: Add locale independent implementations of toupper/tolower
  cosmetics: insert some spaces in explicit enum value assignments
  move 8SVX audio codecs to the audio codec list part on the next bump
  ...

Conflicts:
	avprobe.c
	doc/APIchanges
	ffplay.c
	ffserver.c
	libavcodec/avcodec.h
	libavdevice/bktr.c
	libavdevice/v4l.c
	libavdevice/v4l2.c
	libavformat/matroskaenc.c
	libavformat/wtv.c
	libavutil/avstring.c
	libavutil/avstring.h
	libavutil/avutil.h
	libswscale/x86/swscale_template.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

3 files changed, 170 insertions, 26 deletions

diff --git a/libswscale/x86/scale.asm b/libswscale/x86/scale.asm
index 2e754862f0..5f831aaebf 100644
--- a/libswscale/x86/scale.asm
+++ b/libswscale/x86/scale.asm
@@ -34,6 +34,12 @@ yuv2yuvX_10_start:  times 4 dd 0x10000
 yuv2yuvX_9_start:   times 4 dd 0x20000
 yuv2yuvX_10_upper:  times 8 dw 0x3ff
 yuv2yuvX_9_upper:   times 8 dw 0x1ff
+pd_4:          times 4 dd 4
+pd_4min0x40000:times 4 dd 4 - (0x40000)
+pw_16:         times 8 dw 16
+pw_32:         times 8 dw 32
+pw_512:        times 8 dw 512
+pw_1024:       times 8 dw 1024
 
 SECTION .text
 
@@ -665,3 +671,139 @@ INIT_AVX
 yuv2planeX_fn avx,   8, 10, 7
 yuv2planeX_fn avx,   9,  7, 5
 yuv2planeX_fn avx,  10,  7, 5
+
+; %1=outout-bpc, %2=alignment (u/a)
+%macro yuv2plane1_mainloop 2
+.loop_%2:
+%if %1 == 8
+    paddsw          m0, m2, [r0+r2*2+mmsize*0]
+    paddsw          m1, m3, [r0+r2*2+mmsize*1]
+    psraw           m0, 7
+    psraw           m1, 7
+    packuswb        m0, m1
+    mov%2      [r1+r2], m0
+%elif %1 == 16
+    paddd           m0, m4, [r0+r2*4+mmsize*0]
+    paddd           m1, m4, [r0+r2*4+mmsize*1]
+    paddd           m2, m4, [r0+r2*4+mmsize*2]
+    paddd           m3, m4, [r0+r2*4+mmsize*3]
+    psrad           m0, 3
+    psrad           m1, 3
+    psrad           m2, 3
+    psrad           m3, 3
+%if cpuflag(sse4) ; avx/sse4
+    packusdw        m0, m1
+    packusdw        m2, m3
+%else ; mmx/sse2
+    packssdw        m0, m1
+    packssdw        m2, m3
+    paddw           m0, m5
+    paddw           m2, m5
+%endif ; mmx/sse2/sse4/avx
+    mov%2    [r1+r2*2], m0
+    mov%2    [r1+r2*2+mmsize], m2
+%else
+    paddsw          m0, m2, [r0+r2*2+mmsize*0]
+    paddsw          m1, m2, [r0+r2*2+mmsize*1]
+    psraw           m0, 15 - %1
+    psraw           m1, 15 - %1
+    pmaxsw          m0, m4
+    pmaxsw          m1, m4
+    pminsw          m0, m3
+    pminsw          m1, m3
+    mov%2    [r1+r2*2], m0
+    mov%2    [r1+r2*2+mmsize], m1
+%endif
+    add             r2, mmsize
+    jl .loop_%2
+%endmacro
+
+%macro yuv2plane1_fn 3
+cglobal yuv2plane1_%1, %3, %3, %2
+%if %1 == 8
+    add             r1, r2
+%else ; %1 != 8
+    lea             r1, [r1+r2*2]
+%endif ; %1 == 8
+%if %1 == 16
+    lea             r0, [r0+r2*4]
+%else ; %1 != 16
+    lea             r0, [r0+r2*2]
+%endif ; %1 == 16
+    neg             r2
+
+%if %1 == 8
+    pxor            m4, m4               ; zero
+
+    ; create registers holding dither
+    movq            m3, [r3]             ; dither
+    test           r4d, r4d
+    jz              .no_rot
+%if mmsize == 16
+    punpcklqdq      m3, m3
+%endif ; mmsize == 16
+    PALIGNR_MMX     m3, m3, 3, m2
+.no_rot:
+%if mmsize == 8
+    mova            m2, m3
+    punpckhbw       m3, m4               ; byte->word
+    punpcklbw       m2, m4               ; byte->word
+%else
+    punpcklbw       m3, m4
+    mova            m2, m3
+%endif
+%elif %1 == 9
+    pxor            m4, m4
+    mova            m3, [pw_512]
+    mova            m2, [pw_32]
+%elif %1 == 10
+    pxor            m4, m4
+    mova            m3, [pw_1024]
+    mova            m2, [pw_16]
+%else ; %1 == 16
+%if cpuflag(sse4) ; sse4/avx
+    mova            m4, [pd_4]
+%else ; mmx/sse2
+    mova            m4, [pd_4min0x40000]
+    mova            m5, [minshort]
+%endif ; mmx/sse2/sse4/avx
+%endif ; %1 == ..
+
+    ; actual pixel scaling
+%if mmsize == 8
+    yuv2plane1_mainloop %1, a
+%else ; mmsize == 16
+    test            r1, 15
+    jnz .unaligned
+    yuv2plane1_mainloop %1, a
+    REP_RET
+.unaligned:
+    yuv2plane1_mainloop %1, u
+%endif ; mmsize == 8/16
+    REP_RET
+%endmacro
+
+%ifdef ARCH_X86_32
+INIT_MMX mmx
+yuv2plane1_fn  8, 0, 5
+yuv2plane1_fn 16, 0, 3
+
+INIT_MMX mmx2
+yuv2plane1_fn  9, 0, 3
+yuv2plane1_fn 10, 0, 3
+%endif
+
+INIT_XMM sse2
+yuv2plane1_fn  8, 5, 5
+yuv2plane1_fn  9, 5, 3
+yuv2plane1_fn 10, 5, 3
+yuv2plane1_fn 16, 6, 3
+
+INIT_XMM sse4
+yuv2plane1_fn 16, 5, 3
+
+INIT_XMM avx
+yuv2plane1_fn  8, 5, 5
+yuv2plane1_fn  9, 5, 3
+yuv2plane1_fn 10, 5, 3
+yuv2plane1_fn 16, 5, 3
diff --git a/libswscale/x86/swscale_mmx.c b/libswscale/x86/swscale_mmx.c
index c16be83f13..8869274dd4 100644
--- a/libswscale/x86/swscale_mmx.c
+++ b/libswscale/x86/swscale_mmx.c
@@ -289,6 +289,22 @@ VSCALEX_FUNCS(sse4, sse4);
 VSCALEX_FUNC(16, sse4);
 VSCALEX_FUNCS(avx,  avx);
 
+#define VSCALE_FUNC(size, opt) \
+extern void ff_yuv2plane1_ ## size ## _ ## opt(const int16_t *src, uint8_t *dst, int dstW, \
+                                               const uint8_t *dither, int offset)
+#define VSCALE_FUNCS(opt1, opt2) \
+    VSCALE_FUNC(8,  opt1); \
+    VSCALE_FUNC(9,  opt2); \
+    VSCALE_FUNC(10, opt2); \
+    VSCALE_FUNC(16, opt1)
+
+#if ARCH_X86_32
+VSCALE_FUNCS(mmx, mmx2);
+#endif
+VSCALE_FUNCS(sse2, sse2);
+VSCALE_FUNC(16, sse4);
+VSCALE_FUNCS(avx, avx);
+
 void ff_sws_init_swScale_mmx(SwsContext *c)
 {
     int cpu_flags = av_get_cpu_flags();
@@ -336,11 +352,19 @@ switch(c->dstBpc){ \
     case 9:  if (!isBE(c->dstFormat) && opt2chk) /*vscalefn = ff_yuv2planeX_9_  ## opt2;*/ break; \
     default:                                     /*vscalefn = ff_yuv2planeX_8_  ## opt1;*/ break; \
     }
+#define ASSIGN_VSCALE_FUNC(vscalefn, opt1, opt2, opt2chk) \
+    switch(c->dstBpc){ \
+    case 16: if (!isBE(c->dstFormat))            vscalefn = ff_yuv2plane1_16_ ## opt1; break; \
+    case 10: if (!isBE(c->dstFormat) && opt2chk) vscalefn = ff_yuv2plane1_10_ ## opt2; break; \
+    case 9:  if (!isBE(c->dstFormat) && opt2chk) vscalefn = ff_yuv2plane1_9_  ## opt2;  break; \
+    default:                                     vscalefn = ff_yuv2plane1_8_  ## opt1;  break; \
+    }
 #if ARCH_X86_32
     if (cpu_flags & AV_CPU_FLAG_MMX) {
         ASSIGN_MMX_SCALE_FUNC(c->hyScale, c->hLumFilterSize, mmx, mmx);
         ASSIGN_MMX_SCALE_FUNC(c->hcScale, c->hChrFilterSize, mmx, mmx);
         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, mmx, mmx2, cpu_flags & AV_CPU_FLAG_MMX2,);
+        ASSIGN_VSCALE_FUNC(c->yuv2plane1, mmx, mmx2, cpu_flags & AV_CPU_FLAG_MMX2);
     }
 #endif
 #define ASSIGN_SSE_SCALE_FUNC(hscalefn, filtersize, opt1, opt2) \
@@ -355,6 +379,7 @@ switch(c->dstBpc){ \
         ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, sse2, sse2);
         ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, sse2, sse2);
         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, sse2, sse2, 1,);
+        ASSIGN_VSCALE_FUNC(c->yuv2plane1, sse2, sse2, 1);
     }
     if (cpu_flags & AV_CPU_FLAG_SSSE3) {
         ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, ssse3, ssse3);
@@ -366,10 +391,13 @@ switch(c->dstBpc){ \
         ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, sse4, ssse3);
         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, sse4, sse4, 1,
                             if (!isBE(c->dstFormat)) c->yuv2planeX = ff_yuv2planeX_16_sse4);
+        if (c->dstBpc == 16 && !isBE(c->dstFormat))
+            c->yuv2plane1 = ff_yuv2plane1_16_sse4;
     }
 
     if (cpu_flags & AV_CPU_FLAG_AVX) {
         ASSIGN_VSCALEX_FUNC(c->yuv2planeX, avx, avx, 1,);
+        ASSIGN_VSCALE_FUNC(c->yuv2plane1, avx, avx, 1);
     }
 #endif
 }
diff --git a/libswscale/x86/swscale_template.c b/libswscale/x86/swscale_template.c
index 0f177176cb..5be6072d3b 100644
--- a/libswscale/x86/swscale_template.c
+++ b/libswscale/x86/swscale_template.c
@@ -109,29 +109,6 @@ static void RENAME(yuv2yuvX)(const int16_t *filter, int filterSize,
     );
 }
 
-static void RENAME(yuv2yuv1_ar)(const int16_t *src, uint8_t *dst, int dstW, const uint8_t *dither, int offset)
-{
-    dither_8to16(dither, offset);
-    __asm__ volatile(
-        "mov %2, %%"REG_a"                    \n\t"
-        ".p2align                4            \n\t" /* FIXME Unroll? */
-        "1:                                   \n\t"
-        "movq  (%0, %%"REG_a", 2), %%mm0      \n\t"
-        "movq 8(%0, %%"REG_a", 2), %%mm1      \n\t"
-        "paddsw             %%mm3, %%mm0      \n\t"
-        "paddsw             %%mm4, %%mm1      \n\t"
-        "psraw                 $7, %%mm0      \n\t"
-        "psraw                 $7, %%mm1      \n\t"
-        "packuswb           %%mm1, %%mm0      \n\t"
-        MOVNTQ(%%mm0, (%1, %%REGa))
-        "add                   $8, %%"REG_a"  \n\t"
-        "jnc                   1b             \n\t"
-        :: "r" (src + dstW), "r" (dst + dstW),
-           "g" ((x86_reg)-dstW)
-        : "%"REG_a
-    );
-}
-
 #define YSCALEYUV2PACKEDX_UV \
     __asm__ volatile(\
         "xor                   %%"REG_a", %%"REG_a"     \n\t"\
@@ -1881,9 +1858,7 @@ static av_cold void RENAME(sws_init_swScale)(SwsContext *c)
     c->use_mmx_vfilter= 0;
     if (!is16BPS(dstFormat) && !is9_OR_10BPS(dstFormat) && dstFormat != PIX_FMT_NV12
         && dstFormat != PIX_FMT_NV21 && !(c->flags & SWS_BITEXACT)) {
-            c->yuv2plane1 = RENAME(yuv2yuv1_ar    );
             if (c->flags & SWS_ACCURATE_RND) {
-                //c->yuv2yuv1 = RENAME(yuv2yuv1_ar    );
                 if (!(c->flags & SWS_FULL_CHR_H_INT)) {
                     switch (c->dstFormat) {
                     case PIX_FMT_RGB32:   c->yuv2packedX = RENAME(yuv2rgb32_X_ar);   break;
@@ -1896,7 +1871,6 @@ static av_cold void RENAME(sws_init_swScale)(SwsContext *c)
                 }
             } else {
                 int should_dither= isNBPS(c->srcFormat) || is16BPS(c->srcFormat);
-                //c->yuv2plane1 = should_dither ? RENAME(yuv2yuv1_ar    ) : RENAME(yuv2yuv1    );
                 c->use_mmx_vfilter= 1;
                 c->yuv2planeX = RENAME(yuv2yuvX    );
                 if (!(c->flags & SWS_FULL_CHR_H_INT)) {