har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'qatar/master'

Browse Source 
* qatar/master:
  dxa: remove useless code
  lavf: don't select an attached picture as default stream for seeking.
  avconv: remove pointless checks.
  avconv: check for get_filtered_frame() failure.
  avconv: remove a pointless check.
  swscale: convert hscale() to use named arguments.
  x86inc: add *mp named argument support to DEFINE_ARGS.
  swscale: convert hscale to cpuflags().

Conflicts:
	ffmpeg.c
	libswscale/x86/scale.asm

Merged-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer 2012-03-16 01:36:49 +01:00
commit 4257ce112c
5 changed files with 143 additions and 136 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ffmpeg.c
View File

@ -2178,13 +2178,13 @@ static int transcode_video(InputStream *ist, AVPacket *pkt, int *got_output, int
} }
if (!ist->filtered_frame && !(ist->filtered_frame = avcodec_alloc_frame())) { if (!ist->filtered_frame && !(ist->filtered_frame = avcodec_alloc_frame())) {
ret = AVERROR(ENOMEM); ret = AVERROR(ENOMEM);
goto end; goto fail;
} }
filtered_frame = ist->filtered_frame; filtered_frame = ist->filtered_frame;
*filtered_frame= *decoded_frame; //for me_threshold *filtered_frame= *decoded_frame; //for me_threshold
avfilter_fill_frame_from_video_buffer_ref(filtered_frame, ost->picref); avfilter_fill_frame_from_video_buffer_ref(filtered_frame, ost->picref);
filtered_frame->pts = av_rescale_q(ost->picref->pts, ist_pts_tb, AV_TIME_BASE_Q); filtered_frame->pts = av_rescale_q(ost->picref->pts, ist_pts_tb, AV_TIME_BASE_Q);
if (ost->picref->video && !ost->frame_aspect_ratio) if (!ost->frame_aspect_ratio)
ost->st->codec->sample_aspect_ratio = ost->picref->video->sample_aspect_ratio; ost->st->codec->sample_aspect_ratio = ost->picref->video->sample_aspect_ratio;
do_video_out(output_files[ost->file_index].ctx, ost, ist, filtered_frame); do_video_out(output_files[ost->file_index].ctx, ost, ist, filtered_frame);
cont: cont:
@ -2195,7 +2195,7 @@ static int transcode_video(InputStream *ist, AVPacket *pkt, int *got_output, int
#endif #endif
} }
end: fail:
av_free(buffer_to_free); av_free(buffer_to_free);
return ret; return ret;
} }

2
libavcodec/dxa.c
View File

@ -36,7 +36,6 @@
* Decoder context * Decoder context
*/ */
typedef struct DxaDecContext { typedef struct DxaDecContext {
AVCodecContext *avctx;
AVFrame pic, prev; AVFrame pic, prev;
int dsize; int dsize;
@ -292,7 +291,6 @@ static av_cold int decode_init(AVCodecContext *avctx)
{ {
DxaDecContext * const c = avctx->priv_data; DxaDecContext * const c = avctx->priv_data;
c->avctx = avctx;
avctx->pix_fmt = PIX_FMT_PAL8; avctx->pix_fmt = PIX_FMT_PAL8;
avcodec_get_frame_defaults(&c->pic); avcodec_get_frame_defaults(&c->pic);

3
libavformat/utils.c
View File

@ -1399,7 +1399,8 @@ int av_find_default_stream_index(AVFormatContext *s)
return -1; return -1;
for(i = 0; i < s->nb_streams; i++) { for(i = 0; i < s->nb_streams; i++) {
st = s->streams[i]; st = s->streams[i];
if (st->codec->codec_type == AVMEDIA_TYPE_VIDEO) { if (st->codec->codec_type == AVMEDIA_TYPE_VIDEO &&
!(st->disposition & AV_DISPOSITION_ATTACHED_PIC)) {
return i; return i;
} }
if (first_audio_index < 0 && st->codec->codec_type == AVMEDIA_TYPE_AUDIO) if (first_audio_index < 0 && st->codec->codec_type == AVMEDIA_TYPE_AUDIO)

2
libavutil/x86/x86inc.asm
View File

@ -246,6 +246,7 @@ DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9
CAT_UNDEF arg_name %+ %%i, w CAT_UNDEF arg_name %+ %%i, w
CAT_UNDEF arg_name %+ %%i, b CAT_UNDEF arg_name %+ %%i, b
CAT_UNDEF arg_name %+ %%i, m CAT_UNDEF arg_name %+ %%i, m
CAT_UNDEF arg_name %+ %%i, mp
CAT_UNDEF arg_name, %%i CAT_UNDEF arg_name, %%i
%assign %%i %%i+1 %assign %%i %%i+1
%endrep %endrep
@ -260,6 +261,7 @@ DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9
%xdefine %1w r %+ %%i %+ w %xdefine %1w r %+ %%i %+ w
%xdefine %1b r %+ %%i %+ b %xdefine %1b r %+ %%i %+ b
%xdefine %1m r %+ %%i %+ m %xdefine %1m r %+ %%i %+ m
%xdefine %1mp r %+ %%i %+ mp
CAT_XDEFINE arg_name, %%i, %1 CAT_XDEFINE arg_name, %%i, %1
%assign %%i %%i+1 %assign %%i %%i+1
%rotate 1 %rotate 1

266
libswscale/x86/scale.asm
View File

@ -48,11 +48,15 @@ SECTION .text
; the first pixel is given in filterPos[nOutputPixel]. ; the first pixel is given in filterPos[nOutputPixel].
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
; SCALE_FUNC source_width, intermediate_nbits, filtersize, filtersuffix, opt, n_args, n_xmm ; SCALE_FUNC source_width, intermediate_nbits, filtersize, filtersuffix, n_args, n_xmm
%macro SCALE_FUNC 7 %macro SCALE_FUNC 6
cglobal hscale%1to%2_%4_%5, %6, 7, %7 %ifnidn %3, X
cglobal hscale%1to%2_%4, %5, 7, %6, pos0, dst, w, src, filter, fltpos, pos1
%else
cglobal hscale%1to%2_%4, %5, 7, %6, pos0, dst, w, srcmem, filter, fltpos, fltsize
%endif
%if ARCH_X86_64 %if ARCH_X86_64
movsxd r2, r2d movsxd wq, wd
%define mov32 movsxd %define mov32 movsxd
%else ; x86-32 %else ; x86-32
%define mov32 mov %define mov32 mov
@ -60,7 +64,7 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
%if %2 == 19 %if %2 == 19
%if mmsize == 8 ; mmx %if mmsize == 8 ; mmx
mova m2, [max_19bit_int] mova m2, [max_19bit_int]
%elifidn %5, sse4 %elif cpuflag(sse4)
mova m2, [max_19bit_int] mova m2, [max_19bit_int]
%else ; ssse3/sse2 %else ; ssse3/sse2
mova m2, [max_19bit_flt] mova m2, [max_19bit_flt]
@ -87,48 +91,48 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
; setup loop ; setup loop
%if %3 == 8 %if %3 == 8
shl r2, 1 ; this allows *16 (i.e. now *8) in lea instructions for the 8-tap filter shl wq, 1 ; this allows *16 (i.e. now *8) in lea instructions for the 8-tap filter
%define r2shr 1 %define wshr 1
%else ; %3 == 4 %else ; %3 == 4
%define r2shr 0 %define wshr 0
%endif ; %3 == 8 %endif ; %3 == 8
lea r4, [r4+r2*8] lea filterq, [filterq+wq*8]
%if %2 == 15 %if %2 == 15
lea r1, [r1+r2*(2>>r2shr)] lea dstq, [dstq+wq*(2>>wshr)]
%else ; %2 == 19 %else ; %2 == 19
lea r1, [r1+r2*(4>>r2shr)] lea dstq, [dstq+wq*(4>>wshr)]
%endif ; %2 == 15/19 %endif ; %2 == 15/19
lea r5, [r5+r2*(4>>r2shr)] lea fltposq, [fltposq+wq*(4>>wshr)]
neg r2 neg wq
.loop: .loop:
%if %3 == 4 ; filterSize == 4 scaling %if %3 == 4 ; filterSize == 4 scaling
; load 2x4 or 4x4 source pixels into m0/m1 ; load 2x4 or 4x4 source pixels into m0/m1
mov32 r0, dword [r5+r2*4+0] ; filterPos[0] mov32 pos0q, dword [fltposq+wq*4+ 0] ; filterPos[0]
mov32 r6, dword [r5+r2*4+4] ; filterPos[1] mov32 pos1q, dword [fltposq+wq*4+ 4] ; filterPos[1]
movlh m0, [r3+r0*srcmul] ; src[filterPos[0] + {0,1,2,3}] movlh m0, [srcq+pos0q*srcmul] ; src[filterPos[0] + {0,1,2,3}]
%if mmsize == 8 %if mmsize == 8
movlh m1, [r3+r6*srcmul] ; src[filterPos[1] + {0,1,2,3}] movlh m1, [srcq+pos1q*srcmul] ; src[filterPos[1] + {0,1,2,3}]
%else ; mmsize == 16 %else ; mmsize == 16
%if %1 > 8 %if %1 > 8
movhps m0, [r3+r6*srcmul] ; src[filterPos[1] + {0,1,2,3}] movhps m0, [srcq+pos1q*srcmul] ; src[filterPos[1] + {0,1,2,3}]
%else ; %1 == 8 %else ; %1 == 8
movd m4, [r3+r6*srcmul] ; src[filterPos[1] + {0,1,2,3}] movd m4, [srcq+pos1q*srcmul] ; src[filterPos[1] + {0,1,2,3}]
%endif %endif
mov32 r0, dword [r5+r2*4+8] ; filterPos[2] mov32 pos0q, dword [fltposq+wq*4+ 8] ; filterPos[2]
mov32 r6, dword [r5+r2*4+12] ; filterPos[3] mov32 pos1q, dword [fltposq+wq*4+12] ; filterPos[3]
movlh m1, [r3+r0*srcmul] ; src[filterPos[2] + {0,1,2,3}] movlh m1, [srcq+pos0q*srcmul] ; src[filterPos[2] + {0,1,2,3}]
%if %1 > 8 %if %1 > 8
movhps m1, [r3+r6*srcmul] ; src[filterPos[3] + {0,1,2,3}] movhps m1, [srcq+pos1q*srcmul] ; src[filterPos[3] + {0,1,2,3}]
%else ; %1 == 8 %else ; %1 == 8
movd m5, [r3+r6*srcmul] ; src[filterPos[3] + {0,1,2,3}] movd m5, [srcq+pos1q*srcmul] ; src[filterPos[3] + {0,1,2,3}]
punpckldq m0, m4 punpckldq m0, m4
punpckldq m1, m5 punpckldq m1, m5
%endif ; %1 == 8 && %5 <= ssse %endif ; %1 == 8
%endif ; mmsize == 8/16 %endif ; mmsize == 8/16
%if %1 == 8 %if %1 == 8
punpcklbw m0, m3 ; byte -> word punpcklbw m0, m3 ; byte -> word
punpcklbw m1, m3 ; byte -> word punpcklbw m1, m3 ; byte -> word
%endif ; %1 == 8 %endif ; %1 == 8
; multiply with filter coefficients ; multiply with filter coefficients
@ -137,8 +141,8 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
psubw m0, m6 psubw m0, m6
psubw m1, m6 psubw m1, m6
%endif ; %1 == 16 %endif ; %1 == 16
pmaddwd m0, [r4+r2*8+mmsize*0] ; *= filter[{0,1,..,6,7}] pmaddwd m0, [filterq+wq*8+mmsize*0] ; *= filter[{0,1,..,6,7}]
pmaddwd m1, [r4+r2*8+mmsize*1] ; *= filter[{8,9,..,14,15}] pmaddwd m1, [filterq+wq*8+mmsize*1] ; *= filter[{8,9,..,14,15}]
; add up horizontally (4 srcpix * 4 coefficients -> 1 dstpix) ; add up horizontally (4 srcpix * 4 coefficients -> 1 dstpix)
%if mmsize == 8 ; mmx %if mmsize == 8 ; mmx
@ -146,38 +150,38 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
punpckldq m0, m1 punpckldq m0, m1
punpckhdq m4, m1 punpckhdq m4, m1
paddd m0, m4 paddd m0, m4
%elifidn %5, sse2 %elif notcpuflag(ssse3) ; sse2
mova m4, m0 mova m4, m0
shufps m0, m1, 10001000b shufps m0, m1, 10001000b
shufps m4, m1, 11011101b shufps m4, m1, 11011101b
paddd m0, m4 paddd m0, m4
%else ; ssse3/sse4 %else ; ssse3/sse4
phaddd m0, m1 ; filter[{ 0, 1, 2, 3}]*src[filterPos[0]+{0,1,2,3}], phaddd m0, m1 ; filter[{ 0, 1, 2, 3}]*src[filterPos[0]+{0,1,2,3}],
; filter[{ 4, 5, 6, 7}]*src[filterPos[1]+{0,1,2,3}], ; filter[{ 4, 5, 6, 7}]*src[filterPos[1]+{0,1,2,3}],
; filter[{ 8, 9,10,11}]*src[filterPos[2]+{0,1,2,3}], ; filter[{ 8, 9,10,11}]*src[filterPos[2]+{0,1,2,3}],
; filter[{12,13,14,15}]*src[filterPos[3]+{0,1,2,3}] ; filter[{12,13,14,15}]*src[filterPos[3]+{0,1,2,3}]
%endif ; mmx/sse2/ssse3/sse4 %endif ; mmx/sse2/ssse3/sse4
%else ; %3 == 8, i.e. filterSize == 8 scaling %else ; %3 == 8, i.e. filterSize == 8 scaling
; load 2x8 or 4x8 source pixels into m0, m1, m4 and m5 ; load 2x8 or 4x8 source pixels into m0, m1, m4 and m5
mov32 r0, dword [r5+r2*2+0] ; filterPos[0] mov32 pos0q, dword [fltposq+wq*2+0] ; filterPos[0]
mov32 r6, dword [r5+r2*2+4] ; filterPos[1] mov32 pos1q, dword [fltposq+wq*2+4] ; filterPos[1]
movbh m0, [r3+ r0 *srcmul] ; src[filterPos[0] + {0,1,2,3,4,5,6,7}] movbh m0, [srcq+ pos0q *srcmul] ; src[filterPos[0] + {0,1,2,3,4,5,6,7}]
%if mmsize == 8 %if mmsize == 8
movbh m1, [r3+(r0+4)*srcmul] ; src[filterPos[0] + {4,5,6,7}] movbh m1, [srcq+(pos0q+4)*srcmul] ; src[filterPos[0] + {4,5,6,7}]
movbh m4, [r3+ r6 *srcmul] ; src[filterPos[1] + {0,1,2,3}] movbh m4, [srcq+ pos1q *srcmul] ; src[filterPos[1] + {0,1,2,3}]
movbh m5, [r3+(r6+4)*srcmul] ; src[filterPos[1] + {4,5,6,7}] movbh m5, [srcq+(pos1q+4)*srcmul] ; src[filterPos[1] + {4,5,6,7}]
%else ; mmsize == 16 %else ; mmsize == 16
movbh m1, [r3+ r6 *srcmul] ; src[filterPos[1] + {0,1,2,3,4,5,6,7}] movbh m1, [srcq+ pos1q *srcmul] ; src[filterPos[1] + {0,1,2,3,4,5,6,7}]
mov32 r0, dword [r5+r2*2+8] ; filterPos[2] mov32 pos0q, dword [fltposq+wq*2+8] ; filterPos[2]
mov32 r6, dword [r5+r2*2+12] ; filterPos[3] mov32 pos1q, dword [fltposq+wq*2+12] ; filterPos[3]
movbh m4, [r3+ r0 *srcmul] ; src[filterPos[2] + {0,1,2,3,4,5,6,7}] movbh m4, [srcq+ pos0q *srcmul] ; src[filterPos[2] + {0,1,2,3,4,5,6,7}]
movbh m5, [r3+ r6 *srcmul] ; src[filterPos[3] + {0,1,2,3,4,5,6,7}] movbh m5, [srcq+ pos1q *srcmul] ; src[filterPos[3] + {0,1,2,3,4,5,6,7}]
%endif ; mmsize == 8/16 %endif ; mmsize == 8/16
%if %1 == 8 %if %1 == 8
punpcklbw m0, m3 ; byte -> word punpcklbw m0, m3 ; byte -> word
punpcklbw m1, m3 ; byte -> word punpcklbw m1, m3 ; byte -> word
punpcklbw m4, m3 ; byte -> word punpcklbw m4, m3 ; byte -> word
punpcklbw m5, m3 ; byte -> word punpcklbw m5, m3 ; byte -> word
%endif ; %1 == 8 %endif ; %1 == 8
; multiply ; multiply
@ -188,10 +192,10 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
psubw m4, m6 psubw m4, m6
psubw m5, m6 psubw m5, m6
%endif ; %1 == 16 %endif ; %1 == 16
pmaddwd m0, [r4+r2*8+mmsize*0] ; *= filter[{0,1,..,6,7}] pmaddwd m0, [filterq+wq*8+mmsize*0] ; *= filter[{0,1,..,6,7}]
pmaddwd m1, [r4+r2*8+mmsize*1] ; *= filter[{8,9,..,14,15}] pmaddwd m1, [filterq+wq*8+mmsize*1] ; *= filter[{8,9,..,14,15}]
pmaddwd m4, [r4+r2*8+mmsize*2] ; *= filter[{16,17,..,22,23}] pmaddwd m4, [filterq+wq*8+mmsize*2] ; *= filter[{16,17,..,22,23}]
pmaddwd m5, [r4+r2*8+mmsize*3] ; *= filter[{24,25,..,30,31}] pmaddwd m5, [filterq+wq*8+mmsize*3] ; *= filter[{24,25,..,30,31}]
; add up horizontally (8 srcpix * 8 coefficients -> 1 dstpix) ; add up horizontally (8 srcpix * 8 coefficients -> 1 dstpix)
%if mmsize == 8 %if mmsize == 8
@ -201,7 +205,7 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
punpckldq m0, m4 punpckldq m0, m4
punpckhdq m1, m4 punpckhdq m1, m4
paddd m0, m1 paddd m0, m1
%elifidn %5, sse2 %elif notcpuflag(ssse3) ; sse2
%if %1 == 8 %if %1 == 8
%define mex m6 %define mex m6
%else %else
@ -226,55 +230,55 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
; of 3 x phaddd here, faster on older cpus ; of 3 x phaddd here, faster on older cpus
phaddd m0, m1 phaddd m0, m1
phaddd m4, m5 phaddd m4, m5
phaddd m0, m4 ; filter[{ 0, 1,..., 6, 7}]*src[filterPos[0]+{0,1,...,6,7}], phaddd m0, m4 ; filter[{ 0, 1,..., 6, 7}]*src[filterPos[0]+{0,1,...,6,7}],
; filter[{ 8, 9,...,14,15}]*src[filterPos[1]+{0,1,...,6,7}], ; filter[{ 8, 9,...,14,15}]*src[filterPos[1]+{0,1,...,6,7}],
; filter[{16,17,...,22,23}]*src[filterPos[2]+{0,1,...,6,7}], ; filter[{16,17,...,22,23}]*src[filterPos[2]+{0,1,...,6,7}],
; filter[{24,25,...,30,31}]*src[filterPos[3]+{0,1,...,6,7}] ; filter[{24,25,...,30,31}]*src[filterPos[3]+{0,1,...,6,7}]
%endif ; mmx/sse2/ssse3/sse4 %endif ; mmx/sse2/ssse3/sse4
%endif ; %3 == 4/8 %endif ; %3 == 4/8
%else ; %3 == X, i.e. any filterSize scaling %else ; %3 == X, i.e. any filterSize scaling
%ifidn %4, X4 %ifidn %4, X4
%define r6sub 4 %define dlt 4
%else ; %4 == X || %4 == X8 %else ; %4 == X || %4 == X8
%define r6sub 0 %define dlt 0
%endif ; %4 ==/!= X4 %endif ; %4 ==/!= X4
%if ARCH_X86_64 %if ARCH_X86_64
push r12 push r12
movsxd r6, r6d ; filterSize %define srcq r11
lea r12, [r3+(r6-r6sub)*srcmul] ; &src[filterSize&~4] %define pos1q r10
%define src_reg r11 %define srcendq r12
%define r1x r10 movsxd fltsizeq, fltsized ; filterSize
%define filter2 r12 lea srcendq, [srcmemq+(fltsizeq-dlt)*srcmul] ; &src[filterSize&~4]
%else ; x86-32 %else ; x86-32
lea r0, [r3+(r6-r6sub)*srcmul] ; &src[filterSize&~4] %define srcq srcmemq
mov r6m, r0 %define pos1q dstq
%define src_reg r3 %define srcendq r6m
%define r1x r1 lea pos0q, [srcmemq+(fltsizeq-dlt)*srcmul] ; &src[filterSize&~4]
%define filter2 r6m mov srcendq, pos0q
%endif ; x86-32/64 %endif ; x86-32/64
lea r5, [r5+r2*4] lea fltposq, [fltposq+wq*4]
%if %2 == 15 %if %2 == 15
lea r1, [r1+r2*2] lea dstq, [dstq+wq*2]
%else ; %2 == 19 %else ; %2 == 19
lea r1, [r1+r2*4] lea dstq, [dstq+wq*4]
%endif ; %2 == 15/19 %endif ; %2 == 15/19
movifnidn r1mp, r1 movifnidn dstmp, dstq
neg r2 neg wq
.loop: .loop:
mov32 r0, dword [r5+r2*4+0] ; filterPos[0] mov32 pos0q, dword [fltposq+wq*4+0] ; filterPos[0]
mov32 r1x, dword [r5+r2*4+4] ; filterPos[1] mov32 pos1q, dword [fltposq+wq*4+4] ; filterPos[1]
; FIXME maybe do 4px/iteration on x86-64 (x86-32 wouldn't have enough regs)? ; FIXME maybe do 4px/iteration on x86-64 (x86-32 wouldn't have enough regs)?
pxor m4, m4 pxor m4, m4
pxor m5, m5 pxor m5, m5
mov src_reg, r3mp mov srcq, srcmemmp
.innerloop: .innerloop:
; load 2x4 (mmx) or 2x8 (sse) source pixels into m0/m1 -> m4/m5 ; load 2x4 (mmx) or 2x8 (sse) source pixels into m0/m1 -> m4/m5
movbh m0, [src_reg+r0 *srcmul] ; src[filterPos[0] + {0,1,2,3(,4,5,6,7)}] movbh m0, [srcq+ pos0q *srcmul] ; src[filterPos[0] + {0,1,2,3(,4,5,6,7)}]
movbh m1, [src_reg+(r1x+r6sub)*srcmul] ; src[filterPos[1] + {0,1,2,3(,4,5,6,7)}] movbh m1, [srcq+(pos1q+dlt)*srcmul] ; src[filterPos[1] + {0,1,2,3(,4,5,6,7)}]
%if %1 == 8 %if %1 == 8
punpcklbw m0, m3 punpcklbw m0, m3
punpcklbw m1, m3 punpcklbw m1, m3
@ -286,25 +290,25 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
psubw m0, m6 psubw m0, m6
psubw m1, m6 psubw m1, m6
%endif ; %1 == 16 %endif ; %1 == 16
pmaddwd m0, [r4 ] ; filter[{0,1,2,3(,4,5,6,7)}] pmaddwd m0, [filterq] ; filter[{0,1,2,3(,4,5,6,7)}]
pmaddwd m1, [r4+(r6+r6sub)*2] ; filter[filtersize+{0,1,2,3(,4,5,6,7)}] pmaddwd m1, [filterq+(fltsizeq+dlt)*2]; filter[filtersize+{0,1,2,3(,4,5,6,7)}]
paddd m4, m0 paddd m4, m0
paddd m5, m1 paddd m5, m1
add r4, mmsize add filterq, mmsize
add src_reg, srcmul*mmsize/2 add srcq, srcmul*mmsize/2
cmp src_reg, filter2 ; while (src += 4) < &src[filterSize] cmp srcq, srcendq ; while (src += 4) < &src[filterSize]
jl .innerloop jl .innerloop
%ifidn %4, X4 %ifidn %4, X4
mov32 r1x, dword [r5+r2*4+4] ; filterPos[1] mov32 pos1q, dword [fltposq+wq*4+4] ; filterPos[1]
movlh m0, [src_reg+r0 *srcmul] ; split last 4 srcpx of dstpx[0] movlh m0, [srcq+ pos0q *srcmul] ; split last 4 srcpx of dstpx[0]
sub r1x, r6 ; and first 4 srcpx of dstpx[1] sub pos1q, fltsizeq ; and first 4 srcpx of dstpx[1]
%if %1 > 8 %if %1 > 8
movhps m0, [src_reg+(r1x+r6sub)*srcmul] movhps m0, [srcq+(pos1q+dlt)*srcmul]
%else ; %1 == 8 %else ; %1 == 8
movd m1, [src_reg+(r1x+r6sub)*srcmul] movd m1, [srcq+(pos1q+dlt)*srcmul]
punpckldq m0, m1 punpckldq m0, m1
%endif ; %1 == 8 && %5 <= ssse %endif ; %1 == 8
%if %1 == 8 %if %1 == 8
punpcklbw m0, m3 punpcklbw m0, m3
%endif ; %1 == 8 %endif ; %1 == 8
@ -312,10 +316,10 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
; add back 0x8000 * sum(coeffs) after the horizontal add ; add back 0x8000 * sum(coeffs) after the horizontal add
psubw m0, m6 psubw m0, m6
%endif ; %1 == 16 %endif ; %1 == 16
pmaddwd m0, [r4] pmaddwd m0, [filterq]
%endif ; %4 == X4 %endif ; %4 == X4
lea r4, [r4+(r6+r6sub)*2] lea filterq, [filterq+(fltsizeq+dlt)*2]
%if mmsize == 8 ; mmx %if mmsize == 8 ; mmx
movq m0, m4 movq m0, m4
@ -323,7 +327,7 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
punpckhdq m0, m5 punpckhdq m0, m5
paddd m0, m4 paddd m0, m4
%else ; mmsize == 16 %else ; mmsize == 16
%ifidn %5, sse2 %if notcpuflag(ssse3) ; sse2
mova m1, m4 mova m1, m4
punpcklqdq m4, m5 punpcklqdq m4, m5
punpckhqdq m1, m5 punpckhqdq m1, m5
@ -334,7 +338,7 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
%ifidn %4, X4 %ifidn %4, X4
paddd m4, m0 paddd m4, m0
%endif ; %3 == X4 %endif ; %3 == X4
%ifidn %5, sse2 %if notcpuflag(ssse3) ; sse2
pshufd m4, m4, 11011000b pshufd m4, m4, 11011000b
movhlps m0, m4 movhlps m0, m4
paddd m0, m4 paddd m0, m4
@ -352,19 +356,19 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
; clip, store ; clip, store
psrad m0, 14 + %1 - %2 psrad m0, 14 + %1 - %2
%ifidn %3, X %ifidn %3, X
movifnidn r1, r1mp movifnidn dstq, dstmp
%endif ; %3 == X %endif ; %3 == X
%if %2 == 15 %if %2 == 15
packssdw m0, m0 packssdw m0, m0
%ifnidn %3, X %ifnidn %3, X
movh [r1+r2*(2>>r2shr)], m0 movh [dstq+wq*(2>>wshr)], m0
%else ; %3 == X %else ; %3 == X
movd [r1+r2*2], m0 movd [dstq+wq*2], m0
%endif ; %3 ==/!= X %endif ; %3 ==/!= X
%else ; %2 == 19 %else ; %2 == 19
%if mmsize == 8 %if mmsize == 8
PMINSD_MMX m0, m2, m4 PMINSD_MMX m0, m2, m4
%elifidn %5, sse4 %elif cpuflag(sse4)
pminsd m0, m2 pminsd m0, m2
%else ; sse2/ssse3 %else ; sse2/ssse3
cvtdq2ps m0, m0 cvtdq2ps m0, m0
@ -372,16 +376,16 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
cvtps2dq m0, m0 cvtps2dq m0, m0
%endif ; mmx/sse2/ssse3/sse4 %endif ; mmx/sse2/ssse3/sse4
%ifnidn %3, X %ifnidn %3, X
mova [r1+r2*(4>>r2shr)], m0 mova [dstq+wq*(4>>wshr)], m0
%else ; %3 == X %else ; %3 == X
movq [r1+r2*4], m0 movq [dstq+wq*4], m0
%endif ; %3 ==/!= X %endif ; %3 ==/!= X
%endif ; %2 == 15/19 %endif ; %2 == 15/19
%ifnidn %3, X %ifnidn %3, X
add r2, (mmsize<<r2shr)/4 ; both 8tap and 4tap really only do 4 pixels (or for mmx: 2 pixels) add wq, (mmsize<<wshr)/4 ; both 8tap and 4tap really only do 4 pixels (or for mmx: 2 pixels)
; per iteration. see "shl r2,1" above as for why we do this ; per iteration. see "shl wq,1" above as for why we do this
%else ; %3 == X %else ; %3 == X
add r2, 2 add wq, 2
%endif ; %3 ==/!= X %endif ; %3 ==/!= X
jl .loop jl .loop
%ifnidn %3, X %ifnidn %3, X
@ -396,39 +400,41 @@ cglobal hscale%1to%2_%4_%5, %6, 7, %7
%endif ; %3 ==/!= X %endif ; %3 ==/!= X
%endmacro %endmacro
; SCALE_FUNCS source_width, intermediate_nbits, opt, n_xmm ; SCALE_FUNCS source_width, intermediate_nbits, n_xmm
%macro SCALE_FUNCS 4 %macro SCALE_FUNCS 3
SCALE_FUNC %1, %2, 4, 4, %3, 6, %4 SCALE_FUNC %1, %2, 4, 4, 6, %3
SCALE_FUNC %1, %2, 8, 8, %3, 6, %4 SCALE_FUNC %1, %2, 8, 8, 6, %3
%if mmsize == 8 %if mmsize == 8
SCALE_FUNC %1, %2, X, X, %3, 7, %4 SCALE_FUNC %1, %2, X, X, 7, %3
%else %else
SCALE_FUNC %1, %2, X, X4, %3, 7, %4 SCALE_FUNC %1, %2, X, X4, 7, %3
SCALE_FUNC %1, %2, X, X8, %3, 7, %4 SCALE_FUNC %1, %2, X, X8, 7, %3
%endif %endif
%endmacro %endmacro
; SCALE_FUNCS2 opt, 8_xmm_args, 9to10_xmm_args, 16_xmm_args ; SCALE_FUNCS2 8_xmm_args, 9to10_xmm_args, 16_xmm_args
%macro SCALE_FUNCS2 4 %macro SCALE_FUNCS2 3
%ifnidn %1, sse4 %if notcpuflag(sse4)
SCALE_FUNCS 8, 15, %1, %2 SCALE_FUNCS 8, 15, %1
SCALE_FUNCS 9, 15, %1, %3 SCALE_FUNCS 9, 15, %2
SCALE_FUNCS 10, 15, %1, %3 SCALE_FUNCS 10, 15, %2
SCALE_FUNCS 14, 15, %1, %3 SCALE_FUNCS 14, 15, %2
SCALE_FUNCS 16, 15, %1, %4 SCALE_FUNCS 16, 15, %3
%endif ; !sse4 %endif ; !sse4
SCALE_FUNCS 8, 19, %1, %2 SCALE_FUNCS 8, 19, %1
SCALE_FUNCS 9, 19, %1, %3 SCALE_FUNCS 9, 19, %2
SCALE_FUNCS 10, 19, %1, %3 SCALE_FUNCS 10, 19, %2
SCALE_FUNCS 14, 19, %1, %3 SCALE_FUNCS 14, 19, %2
SCALE_FUNCS 16, 19, %1, %4 SCALE_FUNCS 16, 19, %3
%endmacro %endmacro
%if ARCH_X86_32 %if ARCH_X86_32
INIT_MMX INIT_MMX mmx
SCALE_FUNCS2 mmx, 0, 0, 0 SCALE_FUNCS2 0, 0, 0
%endif %endif
INIT_XMM INIT_XMM sse2
SCALE_FUNCS2 sse2, 6, 7, 8 SCALE_FUNCS2 6, 7, 8
SCALE_FUNCS2 ssse3, 6, 6, 8 INIT_XMM ssse3
SCALE_FUNCS2 sse4, 6, 6, 8 SCALE_FUNCS2 6, 6, 8
INIT_XMM sse4
SCALE_FUNCS2 6, 6, 8