September 2025 - ffmpeg-cvslog

[ffmpeg] branch master updated. 55e71768cf avformat/dhav: Fix off by length of read element error
by ffmpeg-git＠ffmpeg.org 21 Sep '25

21 Sep '25

The branch, master has been updated via 55e71768cf7d9cb395187be0cf38e97b9c454ea1 (commit) from 8d65da767b2b727a80672969f3120286ee478290 (commit) - Log ----------------------------------------------------------------- commit 55e71768cf7d9cb395187be0cf38e97b9c454ea1 Author: Michael Niedermayer <michael(a)niedermayer.cc> AuthorDate: Sat Sep 20 01:25:51 2025 +0200 Commit: michaelni <michael(a)niedermayer.cc> CommitDate: Sun Sep 21 00:49:44 2025 +0000 avformat/dhav: Fix off by length of read element error Fixes: out of array read Fixes: BIGSLEEP-445383648 Fixes: crash_input_1/2/3.dhav Found-by: Google Big Sleep Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> diff --git a/libavformat/dhav.c b/libavformat/dhav.c index 21c9c530d6..b3fa612ab3 100644 --- a/libavformat/dhav.c +++ b/libavformat/dhav.c @@ -251,7 +251,7 @@ static int64_t get_duration(AVFormatContext *s) unsigned date; int64_t size = avio_size(s->pb); - if (start_pos + 16 > size) + if (start_pos + 20 > size) return 0; avio_skip(s->pb, 16); @@ -279,7 +279,7 @@ static int64_t get_duration(AVFormatContext *s) } } - if (pos < buffer_pos || pos + 16 > buffer_pos + buffer_size) + if (pos < buffer_pos || pos + 20 > buffer_pos + buffer_size) goto fail; date = AV_RL32(buffer + (pos - buffer_pos) + 16); ----------------------------------------------------------------------- Summary of changes: libavformat/dhav.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. 8d65da767b lavf: fix demuxing of FLAC files with id3v2 tags
by ffmpeg-git＠ffmpeg.org 20 Sep '25

20 Sep '25

The branch, master has been updated via 8d65da767b2b727a80672969f3120286ee478290 (commit) via 99285cba210399447bc9893dcc30e4882b3b7c30 (commit) from 82b5a0faba9ce558682de728586929bc19d106e2 (commit) - Log ----------------------------------------------------------------- commit 8d65da767b2b727a80672969f3120286ee478290 Author: Lynne <dev(a)lynne.ee> AuthorDate: Wed Sep 17 00:11:11 2025 +0900 Commit: Lynne <dev(a)lynne.ee> CommitDate: Sat Sep 20 22:46:39 2025 +0900 lavf: fix demuxing of FLAC files with id3v2 tags Due to the recent id3v2 refactor, FLAC was left out due to earlier code not checking for id3v2 presence on FLAC. Without the id3v2 data parsed, detection of FLAC and therefore demuxing fails. Fixes 9d037c54f209958d47ac376d2a9561608f98dfae diff --git a/libavformat/flacdec.c b/libavformat/flacdec.c index fc90c01ddf..347b66d6d7 100644 --- a/libavformat/flacdec.c +++ b/libavformat/flacdec.c @@ -385,4 +385,5 @@ const FFInputFormat ff_flac_demuxer = { .read_timestamp = flac_read_timestamp, .raw_codec_id = AV_CODEC_ID_FLAC, .priv_data_size = sizeof(FLACDecContext), + .flags_internal = FF_INFMT_FLAG_ID3V2_AUTO, }; commit 99285cba210399447bc9893dcc30e4882b3b7c30 Author: Lynne <dev(a)lynne.ee> AuthorDate: Wed Aug 27 17:39:16 2025 +0900 Commit: Lynne <dev(a)lynne.ee> CommitDate: Sat Sep 20 22:46:35 2025 +0900 lavf/id3v2: report incorrect BOM value on error diff --git a/libavformat/id3v2.c b/libavformat/id3v2.c index 037d1a2e23..849f22c60d 100644 --- a/libavformat/id3v2.c +++ b/libavformat/id3v2.c @@ -273,13 +273,14 @@ static int decode_str(AVFormatContext *s, AVIOContext *pb, int encoding, *dst = NULL; return AVERROR_INVALIDDATA; } - switch (avio_rb16(pb)) { + uint16_t bom = avio_rb16(pb); + switch (bom) { case 0xfffe: get = avio_rl16; case 0xfeff: break; default: - av_log(s, AV_LOG_ERROR, "Incorrect BOM value\n"); + av_log(s, AV_LOG_ERROR, "Incorrect BOM value: 0x%x\n", bom); ffio_free_dyn_buf(&dynbuf); *dst = NULL; *maxread = left; ----------------------------------------------------------------------- Summary of changes: libavformat/flacdec.c | 1 + libavformat/id3v2.c | 5 +++-- 2 files changed, 4 insertions(+), 2 deletions(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. 82b5a0faba vp9: Remove 8bpc AVX asm for inverse transforms
by ffmpeg-git＠ffmpeg.org 19 Sep '25

19 Sep '25

The branch, master has been updated via 82b5a0faba9ce558682de728586929bc19d106e2 (commit) via 0b5d46ee1c568d616385abaa79708250d3d49c77 (commit) from 49b5c17e0924df1de72db44c5f49a1b65422d202 (commit) - Log ----------------------------------------------------------------- commit 82b5a0faba9ce558682de728586929bc19d106e2 Author: Henrik Gramner <gramner(a)twoorioles.com> AuthorDate: Mon Sep 15 14:11:58 2025 +0200 Commit: Henrik Gramner <henrik(a)gramner.com> CommitDate: Fri Sep 19 23:12:59 2025 +0000 vp9: Remove 8bpc AVX asm for inverse transforms There's very little performance difference vs SSE2/SSSE3 and most systems will use the AVX2 implementations anyway. This reduces code size and compilation time by a significant amount. diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c index a1e47445a8..72edf6bb45 100644 --- a/libavcodec/x86/vp9dsp_init.c +++ b/libavcodec/x86/vp9dsp_init.c @@ -111,14 +111,11 @@ itxfm_funcs(4, ssse3); itxfm_funcs(4, avx2); itxfm_funcs(8, sse2); itxfm_funcs(8, ssse3); -itxfm_funcs(8, avx); itxfm_funcs(8, avx2); itxfm_funcs(16, sse2); itxfm_funcs(16, ssse3); -itxfm_funcs(16, avx); itxfm_func(idct, idct, 32, sse2); itxfm_func(idct, idct, 32, ssse3); -itxfm_func(idct, idct, 32, avx); itxfm_func(iwht, iwht, 4, mmx); itxfm_func(iwht, iwht, 4, avx2); itxfm_funcs(16, avx2); @@ -368,18 +365,6 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact) } if (EXTERNAL_AVX(cpu_flags)) { - dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx; - dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx; - dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx; - dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx; - dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx; - dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx; - dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx; - dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx; - dsp->itxfm_add[TX_32X32][ADST_ADST] = - dsp->itxfm_add[TX_32X32][ADST_DCT] = - dsp->itxfm_add[TX_32X32][DCT_ADST] = - dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx; init_lpf(avx); init_dir_tm_h_ipred(8, avx); init_dir_tm_h_ipred(16, avx); diff --git a/libavcodec/x86/vp9itxfm.asm b/libavcodec/x86/vp9itxfm.asm index a616c90c48..fe650d519c 100644 --- a/libavcodec/x86/vp9itxfm.asm +++ b/libavcodec/x86/vp9itxfm.asm @@ -463,7 +463,6 @@ IADST4_FN iadst, IADST4, iadst, IADST4, ssse3 pmulhrsw m7, W_11585x2_REG ; m7=t5 pmulhrsw m5, W_11585x2_REG ; m5=t6 SWAP 5, 1 - ; merged VP9_IDCT8_1D_FINALIZE to make register-sharing w/ avx easier psubw m6, m0, m3 ; m6=t0-t7 paddw m3, m0 ; m3=t0+t7 psubw m2, m0, m1 ; m2=t1-t6 @@ -711,7 +710,6 @@ cglobal vp9_idct_idct_8x8_add, 4, 4, %2, dst, stride, block, eob VP9_IDCT_IDCT_8x8_ADD_XMM sse2, 12 VP9_IDCT_IDCT_8x8_ADD_XMM ssse3, 13 -VP9_IDCT_IDCT_8x8_ADD_XMM avx, 13 ;--------------------------------------------------------------------------------------------- ; void vp9_iadst_iadst_8x8_add_<opt>(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob); @@ -885,11 +883,8 @@ IADST8_FN idct, IDCT8, iadst, IADST8, sse2, 15 IADST8_FN iadst, IADST8, idct, IDCT8, sse2, 15 IADST8_FN iadst, IADST8, iadst, IADST8, sse2, 15 IADST8_FN idct, IDCT8, iadst, IADST8, ssse3, 16 -IADST8_FN idct, IDCT8, iadst, IADST8, avx, 16 IADST8_FN iadst, IADST8, idct, IDCT8, ssse3, 16 -IADST8_FN iadst, IADST8, idct, IDCT8, avx, 16 IADST8_FN iadst, IADST8, iadst, IADST8, ssse3, 16 -IADST8_FN iadst, IADST8, iadst, IADST8, avx, 16 ;--------------------------------------------------------------------------------------------- ; void vp9_idct_idct_16x16_add_<opt>(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob); @@ -1427,7 +1422,6 @@ cglobal vp9_idct_idct_16x16_add, 4, 6, 16, 512, dst, stride, block, eob VP9_IDCT_IDCT_16x16_ADD_XMM sse2 VP9_IDCT_IDCT_16x16_ADD_XMM ssse3 -VP9_IDCT_IDCT_16x16_ADD_XMM avx %macro VP9_IDCT16_YMM_1D 0 VP9_UNPACK_MULSUB_2W_4X 1, 15, 16305, 1606, [pd_8192], 0, 4 ; t8, t15 @@ -1894,9 +1888,6 @@ IADST16_FN iadst, IADST16, iadst, IADST16, sse2 IADST16_FN idct, IDCT16, iadst, IADST16, ssse3 IADST16_FN iadst, IADST16, idct, IDCT16, ssse3 IADST16_FN iadst, IADST16, iadst, IADST16, ssse3 -IADST16_FN idct, IDCT16, iadst, IADST16, avx -IADST16_FN iadst, IADST16, idct, IDCT16, avx -IADST16_FN iadst, IADST16, iadst, IADST16, avx ; in: data in m[0-15] except m0/m4, which are in [blockq+0] and [blockq+128] ; out: m[0-15] except m6, which is in [blockq+192] @@ -2836,4 +2827,3 @@ cglobal vp9_idct_idct_32x32_add, 0, 6 + ARCH_X86_64 * 3, 16, 2048, dst, stride, VP9_IDCT_IDCT_32x32_ADD_XMM sse2 VP9_IDCT_IDCT_32x32_ADD_XMM ssse3 -VP9_IDCT_IDCT_32x32_ADD_XMM avx commit 0b5d46ee1c568d616385abaa79708250d3d49c77 Author: Henrik Gramner <gramner(a)twoorioles.com> AuthorDate: Mon Sep 15 14:11:57 2025 +0200 Commit: Henrik Gramner <henrik(a)gramner.com> CommitDate: Fri Sep 19 23:12:59 2025 +0000 vp9: Add 8bpc AVX2 asm for inverse transforms diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index ebd2bdb310..461753c2fe 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -186,6 +186,7 @@ X86ASM-OBJS-$(CONFIG_VP6_DECODER) += x86/vp6dsp.o X86ASM-OBJS-$(CONFIG_VP9_DECODER) += x86/vp9intrapred.o \ x86/vp9intrapred_16bpp.o \ x86/vp9itxfm.o \ + x86/vp9itxfm_avx2.o \ x86/vp9itxfm_avx512.o \ x86/vp9itxfm_16bpp.o \ x86/vp9itxfm_16bpp_avx512.o \ diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c index bbabcf38c3..a1e47445a8 100644 --- a/libavcodec/x86/vp9dsp_init.c +++ b/libavcodec/x86/vp9dsp_init.c @@ -108,9 +108,11 @@ itxfm_func(idct, iadst, 4, sse2); itxfm_func(iadst, idct, 4, sse2); itxfm_func(iadst, iadst, 4, sse2); itxfm_funcs(4, ssse3); +itxfm_funcs(4, avx2); itxfm_funcs(8, sse2); itxfm_funcs(8, ssse3); itxfm_funcs(8, avx); +itxfm_funcs(8, avx2); itxfm_funcs(16, sse2); itxfm_funcs(16, ssse3); itxfm_funcs(16, avx); @@ -118,6 +120,7 @@ itxfm_func(idct, idct, 32, sse2); itxfm_func(idct, idct, 32, ssse3); itxfm_func(idct, idct, 32, avx); itxfm_func(iwht, iwht, 4, mmx); +itxfm_func(iwht, iwht, 4, avx2); itxfm_funcs(16, avx2); itxfm_funcs(16, avx512icl); itxfm_func(idct, idct, 32, avx2); @@ -392,6 +395,18 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact) init_fpel_func(0, 1, 64, avg, _8, avx2); if (ARCH_X86_64) { #if ARCH_X86_64 && HAVE_AVX2_EXTERNAL + dsp->itxfm_add[4 /* lossless */][DCT_DCT] = + dsp->itxfm_add[4 /* lossless */][ADST_DCT] = + dsp->itxfm_add[4 /* lossless */][DCT_ADST] = + dsp->itxfm_add[4 /* lossless */][ADST_ADST] = ff_vp9_iwht_iwht_4x4_add_avx2; + dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_avx2; + dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_avx2; + dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_avx2; + dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_avx2; + dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx2; + dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx2; + dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx2; + dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx2; dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx2; dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx2; dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx2; diff --git a/libavcodec/x86/vp9itxfm.asm b/libavcodec/x86/vp9itxfm.asm index 2f290f2f88..a616c90c48 100644 --- a/libavcodec/x86/vp9itxfm.asm +++ b/libavcodec/x86/vp9itxfm.asm @@ -24,36 +24,36 @@ %include "libavutil/x86/x86util.asm" %include "vp9itxfm_template.asm" -SECTION_RODATA 32 +SECTION_RODATA 16 %macro VP9_IDCT_COEFFS 2-3 0 const pw_m%1_%2 -times 8 dw -%1, %2 +times 4 dw -%1, %2 const pw_%2_%1 -times 8 dw %2, %1 +times 4 dw %2, %1 %if %3 == 1 const pw_m%2_m%1 -times 8 dw -%2, -%1 +times 4 dw -%2, -%1 %if %1 != %2 const pw_m%2_%1 -times 8 dw -%2, %1 +times 4 dw -%2, %1 const pw_%1_%2 -times 8 dw %1, %2 +times 4 dw %1, %2 %endif %endif %if %1 < 11585 -pw_m%1x2: times 16 dw -%1*2 +pw_m%1x2: times 8 dw -%1*2 %elif %1 > 11585 -pw_%1x2: times 16 dw %1*2 +pw_%1x2: times 8 dw %1*2 %else const pw_%1x2 -times 16 dw %1*2 +times 8 dw %1*2 %endif %if %2 != %1 -pw_%2x2: times 16 dw %2*2 +pw_%2x2: times 8 dw %2*2 %endif %endmacro @@ -127,24 +127,13 @@ SECTION .text %endmacro %macro VP9_STORE_2X 5-6 dstq ; reg1, reg2, tmp1, tmp2, zero, dst -%if mmsize == 32 - pmovzxbw m%3, [%6] - pmovzxbw m%4, [%6+strideq] -%else movh m%3, [%6] movh m%4, [%6+strideq] punpcklbw m%3, m%5 punpcklbw m%4, m%5 -%endif paddw m%3, m%1 paddw m%4, m%2 -%if mmsize == 32 - packuswb m%3, m%4 - ; Intel... - vpermq m%3, m%3, q3120 - mova [%6], xm%3 - vextracti128 [%6+strideq], m%3, 1 -%elif mmsize == 16 +%if mmsize == 16 packuswb m%3, m%4 movh [%6], m%3 movhps [%6+strideq], m%3 @@ -1534,83 +1523,6 @@ VP9_IDCT_IDCT_16x16_ADD_XMM avx vextracti128 [dstq+stride3q], m%4, 1 %endmacro -%if ARCH_X86_64 && HAVE_AVX2_EXTERNAL -INIT_YMM avx2 -cglobal vp9_idct_idct_16x16_add, 4, 4, 16, dst, stride, block, eob - cmp eobd, 1 ; faster path for when only DC is set - jg .idctfull - - ; dc-only - mova m1, [pw_11585x2] - vpbroadcastw m0, [blockq] - pmulhrsw m0, m1 - pmulhrsw m0, m1 - pxor m5, m5 - pmulhrsw m0, [pw_512] - movd [blockq], xm5 - - DEFINE_ARGS dst, stride, stride3, cnt - mov cntd, 4 - lea stride3q, [strideq*3] -.loop_dc: - VP9_STORE_YMM_DC_4X 0, 1, 2, 3, 4, 5 - lea dstq, [dstq+4*strideq] - dec cntd - jg .loop_dc - RET - - DEFINE_ARGS dst, stride, block, eob -.idctfull: - mova m1, [blockq+ 32] - mova m2, [blockq+ 64] - mova m3, [blockq+ 96] - mova m5, [blockq+160] - mova m6, [blockq+192] - mova m7, [blockq+224] - mova m8, [blockq+256] - mova m9, [blockq+288] - mova m10, [blockq+320] - mova m11, [blockq+352] - mova m12, [blockq+384] - mova m13, [blockq+416] - mova m14, [blockq+448] - mova m15, [blockq+480] - - VP9_IDCT16_YMM_1D - TRANSPOSE16x16W 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, \ - [blockq+192], [blockq+128], 1 - mova [blockq+ 0], m0 - VP9_IDCT16_YMM_1D - - mova [blockq+224], m7 - - ; store - VP9_IDCT8_WRITEx2 0, 1, 6, 7, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 2, 3, 6, 7, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 4, 5, 6, 7, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - mova m6, [blockq+192] - mova m7, [blockq+224] - VP9_IDCT8_WRITEx2 6, 7, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 8, 9, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 10, 11, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 12, 13, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 14, 15, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - - ; at the end of the loop, m0 should still be zero - ; use that to zero out block coefficients - pxor m0, m0 - ZERO_BLOCK blockq, 32, 16, m0 - RET -%endif - ;--------------------------------------------------------------------------------------------- ; void vp9_iadst_iadst_16x16_add_<opt>(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob); ;--------------------------------------------------------------------------------------------- @@ -2094,65 +2006,6 @@ IADST16_FN iadst, IADST16, iadst, IADST16, avx SWAP 5, 9, 15 %endmacro -%if ARCH_X86_64 && HAVE_AVX2_EXTERNAL -%macro IADST16_YMM_FN 4 -INIT_YMM avx2 -cglobal vp9_%1_%3_16x16_add, 4, 4, 16, dst, stride, block, eob - mova m1, [blockq+ 32] - mova m2, [blockq+ 64] - mova m3, [blockq+ 96] - mova m5, [blockq+160] - mova m6, [blockq+192] - mova m7, [blockq+224] - mova m8, [blockq+256] - mova m9, [blockq+288] - mova m10, [blockq+320] - mova m11, [blockq+352] - mova m12, [blockq+384] - mova m13, [blockq+416] - mova m14, [blockq+448] - mova m15, [blockq+480] - - VP9_%2_YMM_1D - TRANSPOSE16x16W 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, \ - [blockq+192], [blockq+128], 1 - mova [blockq+ 0], m0 - VP9_%4_YMM_1D - - mova [blockq+224], m7 - - ; store - VP9_IDCT8_WRITEx2 0, 1, 6, 7, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 2, 3, 6, 7, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 4, 5, 6, 7, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - mova m6, [blockq+192] - mova m7, [blockq+224] - VP9_IDCT8_WRITEx2 6, 7, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 8, 9, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 10, 11, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 12, 13, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - VP9_IDCT8_WRITEx2 14, 15, 1, 2, unused, [pw_512], 6 - lea dstq, [dstq+2*strideq] - - ; at the end of the loop, m0 should still be zero - ; use that to zero out block coefficients - pxor m0, m0 - ZERO_BLOCK blockq, 32, 16, m0 - RET -%endmacro - -IADST16_YMM_FN idct, IDCT16, iadst, IADST16 -IADST16_YMM_FN iadst, IADST16, idct, IDCT16 -IADST16_YMM_FN iadst, IADST16, iadst, IADST16 -%endif - ;--------------------------------------------------------------------------------------------- ; void vp9_idct_idct_32x32_add_<opt>(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob); ;--------------------------------------------------------------------------------------------- @@ -2468,115 +2321,6 @@ IADST16_YMM_FN iadst, IADST16, iadst, IADST16 SUMSUB_BA w, 5, 13, 8 mova m10, [tmpq+28*%%str] ; t7 SUMSUB_BA w, 4, 10, 8 -%if cpuflag(avx2) - ; the "shitty" about this idct is that the final pass does the outermost - ; interleave sumsubs (t0/31, t1/30, etc) but the tN for the 16x16 need - ; to be sequential, which means I need to load/store half of the sumsub - ; intermediates back to/from memory to get a 16x16 transpose going... - ; This would be easier if we had more (e.g. 32) YMM regs here. - mova [tmpq+ 7*%%str], m9 - mova [tmpq+11*%%str], m12 - mova [tmpq+15*%%str], m11 - mova [tmpq+19*%%str], m2 - mova [tmpq+23*%%str], m3 - mova [tmpq+27*%%str], m13 - mova [tmpq+31*%%str], m10 - mova [tmpq+12*%%str], m5 - - mova m13, [tmpq+30*%%str] ; t8 - mova m12, [tmpq+26*%%str] ; t9 - mova m11, [tmpq+22*%%str] ; t10 - mova m10, [tmpq+18*%%str] ; t11 - mova m9, [tmpq+17*%%str] ; t20 - mova m8, [tmpq+ 1*%%str] ; t21 - mova m3, [tmpq+25*%%str] ; t22 - mova m2, [tmpq+ 5*%%str] ; t23 - - SUMSUB_BA w, 9, 10, 5 - SUMSUB_BA w, 8, 11, 5 - SUMSUB_BA w, 3, 12, 5 - SUMSUB_BA w, 2, 13, 5 - mova [tmpq+ 1*%%str], m10 - mova [tmpq+ 5*%%str], m11 - mova [tmpq+17*%%str], m12 - mova [tmpq+25*%%str], m13 - - mova m13, [tmpq+14*%%str] ; t12 - mova m12, [tmpq+10*%%str] ; t13 - mova m11, [tmpq+ 9*%%str] ; t18 - mova m10, [tmpq+13*%%str] ; t19 - - SUMSUB_BA w, 11, 12, 5 - SUMSUB_BA w, 10, 13, 5 - mova [tmpq+ 9*%%str], m13 - mova [tmpq+13*%%str], m12 - mova [tmpq+10*%%str], m10 - mova [tmpq+14*%%str], m11 - - mova m13, [tmpq+ 6*%%str] ; t14 - mova m12, [tmpq+ 2*%%str] ; t15 - mova m11, [tmpq+21*%%str] ; t16 - mova m10, [tmpq+29*%%str] ; t17 - SUMSUB_BA w, 11, 12, 5 - SUMSUB_BA w, 10, 13, 5 - mova [tmpq+21*%%str], m12 - mova [tmpq+29*%%str], m13 - mova m12, [tmpq+10*%%str] - mova m13, [tmpq+14*%%str] - - TRANSPOSE16x16W 6, 0, 15, 14, 1, 7, 5, 4, \ - 2, 3, 8, 9, 12, 13, 10, 11, \ - [tmpq+12*%%str], [tmpq+ 8*%%str], 1 - mova [tmpq+ 0*%%str], m6 - mova [tmpq+ 2*%%str], m0 - mova [tmpq+ 4*%%str], m15 - mova [tmpq+ 6*%%str], m14 - mova [tmpq+10*%%str], m7 - mova [tmpq+12*%%str], m5 - mova [tmpq+14*%%str], m4 - mova [tmpq+16*%%str], m2 - mova [tmpq+18*%%str], m3 - mova [tmpq+20*%%str], m8 - mova [tmpq+22*%%str], m9 - mova [tmpq+24*%%str], m12 - mova [tmpq+26*%%str], m13 - mova [tmpq+28*%%str], m10 - mova [tmpq+30*%%str], m11 - - mova m0, [tmpq+21*%%str] - mova m1, [tmpq+29*%%str] - mova m2, [tmpq+13*%%str] - mova m3, [tmpq+ 9*%%str] - mova m4, [tmpq+ 1*%%str] - mova m5, [tmpq+ 5*%%str] - mova m7, [tmpq+25*%%str] - mova m8, [tmpq+31*%%str] - mova m9, [tmpq+27*%%str] - mova m10, [tmpq+23*%%str] - mova m11, [tmpq+19*%%str] - mova m12, [tmpq+15*%%str] - mova m13, [tmpq+11*%%str] - mova m14, [tmpq+ 7*%%str] - mova m15, [tmpq+ 3*%%str] - TRANSPOSE16x16W 0, 1, 2, 3, 4, 5, 6, 7, \ - 8, 9, 10, 11, 12, 13, 14, 15, \ - [tmpq+17*%%str], [tmpq+ 9*%%str], 1 - mova [tmpq+ 1*%%str], m0 - mova [tmpq+ 3*%%str], m1 - mova [tmpq+ 5*%%str], m2 - mova [tmpq+ 7*%%str], m3 - mova [tmpq+11*%%str], m5 - mova [tmpq+13*%%str], m6 - mova [tmpq+15*%%str], m7 - mova [tmpq+17*%%str], m8 - mova [tmpq+19*%%str], m9 - mova [tmpq+21*%%str], m10 - mova [tmpq+23*%%str], m11 - mova [tmpq+25*%%str], m12 - mova [tmpq+27*%%str], m13 - mova [tmpq+29*%%str], m14 - mova [tmpq+31*%%str], m15 -%else ; !avx2 TRANSPOSE8x8W 6, 0, 15, 14, 1, 7, 5, 4, 8 mova [tmpq+ 0*%%str], m6 mova [tmpq+ 4*%%str], m0 @@ -2645,7 +2389,6 @@ IADST16_YMM_FN iadst, IADST16, iadst, IADST16 mova [tmpq+22*%%str], m13 mova [tmpq+26*%%str], m14 mova [tmpq+30*%%str], m15 -%endif ; avx2 %else mova m2, [tmpq+24*%%str] ; t6 mova m3, [tmpq+28*%%str] ; t7 @@ -3094,106 +2837,3 @@ cglobal vp9_idct_idct_32x32_add, 0, 6 + ARCH_X86_64 * 3, 16, 2048, dst, stride, VP9_IDCT_IDCT_32x32_ADD_XMM sse2 VP9_IDCT_IDCT_32x32_ADD_XMM ssse3 VP9_IDCT_IDCT_32x32_ADD_XMM avx - -; this is almost identical to VP9_STORE_2X, but it does two rows -; for slightly improved interleaving, and it omits vpermq since the -; input is DC so all values are identical -%macro VP9_STORE_YMM_DC_2X2 6 ; reg, tmp1, tmp2, tmp3, tmp4, zero - mova m%2, [dstq] - mova m%4, [dstq+strideq] - punpckhbw m%3, m%2, m%6 - punpcklbw m%2, m%6 - punpckhbw m%5, m%4, m%6 - punpcklbw m%4, m%6 - paddw m%3, m%1 - paddw m%2, m%1 - paddw m%5, m%1 - paddw m%4, m%1 - packuswb m%2, m%3 - packuswb m%4, m%5 - mova [dstq+strideq*0], m%2 - mova [dstq+strideq*1], m%4 -%endmacro - -%if ARCH_X86_64 && HAVE_AVX2_EXTERNAL -INIT_YMM avx2 -cglobal vp9_idct_idct_32x32_add, 4, 9, 16, 2048, dst, stride, block, eob - cmp eobd, 135 - jg .idctfull - cmp eobd, 1 - jg .idct16x16 - - ; dc-only case - mova m1, [pw_11585x2] - vpbroadcastw m0, [blockq] - pmulhrsw m0, m1 - pmulhrsw m0, m1 - pxor m5, m5 - pmulhrsw m0, [pw_512] - movd [blockq], xm5 - - DEFINE_ARGS dst, stride, cnt - mov cntd, 16 -.loop_dc: - VP9_STORE_YMM_DC_2X2 0, 1, 2, 3, 4, 5 - lea dstq, [dstq+2*strideq] - dec cntd - jg .loop_dc - RET - - DEFINE_ARGS dst_bak, stride, block, cnt, dst, stride30, dst_end, stride2, tmp -.idct16x16: - mov tmpq, rsp - VP9_IDCT32_1D blockq, 1, 16 - - mov stride30q, strideq ; stride - lea stride2q, [strideq*2] ; stride*2 - shl stride30q, 5 ; stride*32 - mov cntd, 2 - sub stride30q, stride2q ; stride*30 -.loop2_16x16: - mov dstq, dst_bakq - lea dst_endq, [dstq+stride30q] - VP9_IDCT32_1D tmpq, 2, 16 - add dst_bakq, 16 - add tmpq, 32 - dec cntd - jg .loop2_16x16 - - ; at the end of the loop, m1 should still be zero - ; use that to zero out block coefficients - ZERO_BLOCK blockq, 64, 16, m1 - RET - -.idctfull: - mov cntd, 2 - mov tmpq, rsp -.loop1_full: - VP9_IDCT32_1D blockq, 1 - add blockq, 32 - add tmpq, 1024 - dec cntd - jg .loop1_full - - sub blockq, 64 - - mov stride30q, strideq ; stride - lea stride2q, [strideq*2] ; stride*2 - shl stride30q, 5 ; stride*32 - mov cntd, 2 - mov tmpq, rsp - sub stride30q, stride2q ; stride*30 -.loop2_full: - mov dstq, dst_bakq - lea dst_endq, [dstq+stride30q] - VP9_IDCT32_1D tmpq, 2 - add dst_bakq, 16 - add tmpq, 32 - dec cntd - jg .loop2_full - - ; at the end of the loop, m1 should still be zero - ; use that to zero out block coefficients - ZERO_BLOCK blockq, 64, 32, m1 - RET -%endif diff --git a/libavcodec/x86/vp9itxfm_avx2.asm b/libavcodec/x86/vp9itxfm_avx2.asm new file mode 100644 index 0000000000..c5ee2426e2 --- /dev/null +++ b/libavcodec/x86/vp9itxfm_avx2.asm @@ -0,0 +1,1640 @@ +;****************************************************************************** +;* VP9 IDCT SIMD optimizations +;* +;* Copyright (C) 2025 Two Orioles, LLC +;* +;* This file is part of FFmpeg. +;* +;* FFmpeg is free software; you can redistribute it and/or +;* modify it under the terms of the GNU Lesser General Public +;* License as published by the Free Software Foundation; either +;* version 2.1 of the License, or (at your option) any later version. +;* +;* FFmpeg is distributed in the hope that it will be useful, +;* but WITHOUT ANY WARRANTY; without even the implied warranty of +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +;* Lesser General Public License for more details. +;* +;* You should have received a copy of the GNU Lesser General Public +;* License along with FFmpeg; if not, write to the Free Software +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA +;****************************************************************************** + +%include "libavutil/x86/x86util.asm" + +%if ARCH_X86_64 && HAVE_AVX2_EXTERNAL + +SECTION_RODATA 16 + +deint_shuf: db 0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15 + +pw_512: times 2 dw 512 ; 16-byte aligned +pd_8192: dd 8192 +pw_m512: times 2 dw -512 +pw_2048: times 2 dw 2048 +pw_1024: times 2 dw 1024 ; 16-byte aligned + +pw_804x2: times 2 dw 804*2 +pw_1606x2: times 2 dw 1606*2 +pw_3196x2: times 2 dw 3196*2 +pw_3981x2: times 2 dw 3981*2 +pw_6270x2: times 2 dw 6270*2 +pw_7005x2: times 2 dw 7005*2 +pw_7723x2: times 2 dw 7723*2 +pw_9760x2: times 2 dw 9760*2 +pw_11585x2: times 2 dw 11585*2 +pw_12140x2: times 2 dw 12140*2 +pw_12665x2: times 2 dw 12665*2 +pw_13160x2: times 2 dw 13160*2 +pw_13623x2: times 2 dw 13623*2 +pw_14053x2: times 2 dw 14053*2 +pw_14449x2: times 2 dw 14449*2 +pw_14811x2: times 2 dw 14811*2 +pw_15137x2: times 2 dw 15137*2 +pw_15426x2: times 2 dw 15426*2 +pw_15679x2: times 2 dw 15679*2 +pw_15893x2: times 2 dw 15893*2 +pw_16069x2: times 2 dw 16069*2 +pw_16207x2: times 2 dw 16207*2 +pw_16305x2: times 2 dw 16305*2 +pw_16364x2: times 2 dw 16364*2 +pw_m2404x2: times 2 dw -2404*2 +pw_m4756x2: times 2 dw -4756*2 +pw_m5520x2: times 2 dw -5520*2 +pw_m8423x2: times 2 dw -8423*2 +pw_m9102x2: times 2 dw -9102*2 +pw_m10394x2: times 2 dw -10394*2 +pw_m11003x2: times 2 dw -11003*2 +pw_m11585x2: times 2 dw -11585*2 + +%macro COEF_PAIR 2-3 +pw_%1_%2: dw %1, %2 +pw_m%2_%1: dw -%2, %1 +%if %0 == 3 +pw_m%1_m%2: dw -%1, -%2 +%endif +%endmacro + +COEF_PAIR 804, 16364 +COEF_PAIR 1606, 16305 +COEF_PAIR 3196, 16069, 1 +COEF_PAIR 3981, 15893 +COEF_PAIR 6270, 15137, 1 +COEF_PAIR 7005, 14811 +COEF_PAIR 7723, 14449 +COEF_PAIR 9102, 13623 +COEF_PAIR 9760, 13160 +COEF_PAIR 11585, 11585 +COEF_PAIR 12140, 11003 +COEF_PAIR 12665, 10394 +COEF_PAIR 13623, 9102, 1 +COEF_PAIR 14053, 8423 +COEF_PAIR 15137, 6270 +COEF_PAIR 15426, 5520 +COEF_PAIR 15679, 4756 +COEF_PAIR 16069, 3196 +COEF_PAIR 16207, 2404 + +; ADST4-only: +pw_0_13377: dw 0, 13377 +pw_13377_m13377: dw 13377, -13377 +pw_m13377_m5283: dw -13377, -5283 +pw_13377_m15212: dw 13377, -15212 +pw_9929_m5283: dw 9929, -5283 +pw_5283_15212: dw 5283, 15212 +pw_13377_9929: dw 13377, 9929 + +; ADST16-only: +pw_8423_3981: dw 8423, 3981 +pw_m8423_3981: dw -8423, 3981 +pw_14053_m15893: dw 14053, -15893 +pw_m14053_m15893: dw -14053, -15893 +pw_2404_9760: dw 2404, 9760 +pw_m2404_9760: dw -2404, 9760 +pw_16207_m13160: dw 16207, -13160 +pw_m16207_m13160: dw -16207, -13160 +pw_11003_804: dw 11003, 804 +pw_m11003_804: dw -11003, 804 +pw_12140_m16364: dw 12140, -16364 +pw_m12140_m16364: dw -12140, -16364 +pw_5520_7005: dw 5520, 7005 +pw_m5520_7005: dw -5520, 7005 +pw_15426_m14811: dw 15426, -14811 +pw_m15426_m14811: dw -15426, -14811 + +SECTION .text + +%define o_base pw_512 + 128 +%define o(x) (r6 - (o_base) + (x)) +%define m(x) mangle(private_prefix %+ _ %+ x %+ SUFFIX) + +%macro IWHT4_1D_PACKED 0 + psubw m2, m0, m3 + paddw m0, m3 + punpckhqdq m2, m2 ; t2 t2 + punpcklqdq m0, m0 ; t0 t0 + psubw m1, m0, m2 + psraw m1, 1 + psubw m1, m3 ; t1 t3 + psubw m0, m1 ; ____ out0 + paddw m2, m1 ; out3 ____ +%endmacro + +INIT_XMM avx2 +cglobal vp9_iwht_iwht_4x4_add, 3, 3, 6, dst, stride, c + mova m0, [cq+16*0] + mova m1, [cq+16*1] + pxor m2, m2 + mova [cq+16*0], m2 + mova [cq+16*1], m2 + lea r2, [dstq+strideq*2] + punpckhqdq m3, m0, m1 ; in1 in3 + punpcklqdq m0, m1 ; in0 in2 + movd m4, [r2 +strideq*0] + pinsrd m4, [dstq+strideq*1], 1 + movd m5, [r2 +strideq*1] + pinsrd m5, [dstq+strideq*0], 1 + psraw m3, 2 + psraw m0, 2 + IWHT4_1D_PACKED + punpckhwd m0, m1 + punpcklwd m1, m2 + punpckhdq m2, m0, m1 + punpckldq m0, m1 + punpckhqdq m3, m0, m2 + punpcklqdq m0, m2 + IWHT4_1D_PACKED + pmovzxbw m4, m4 + pmovzxbw m5, m5 + vpblendd m0, m2, 0x03 + paddw m1, m4 + paddw m0, m5 + packuswb m0, m1 + pextrd [dstq+strideq*0], m0, 1 + pextrd [dstq+strideq*1], m0, 3 + pextrd [r2 +strideq*0], m0, 2 + movd [r2 +strideq*1], m0 + RET + +%macro ITX_MUL2X_PACK 6-7 0 ; dst/src, tmp[1-2], rnd, coef[1-2], swap +%if %7 + vpbroadcastd m%2, [o(pw_%5_%6)] + vpbroadcastd m%3, [o(pw_m%6_%5)] +%else + vpbroadcastd m%2, [o(pw_m%6_%5)] + vpbroadcastd m%3, [o(pw_%5_%6)] +%endif + pmaddwd m%2, m%1 + pmaddwd m%1, m%3 + paddd m%2, m%4 + paddd m%1, m%4 + psrad m%2, 14 + psrad m%1, 14 + packssdw m%1, m%2 +%endmacro + +; dst1 = (src1 * coef1 - src2 * coef2 + rnd) >> 12 +; dst2 = (src1 * coef2 + src2 * coef1 + rnd) >> 12 +%macro ITX_MULSUB_2W 7 ; dst/src[1-2], tmp[1-2], rnd, coef[1-2] + punpckhwd m%3, m%2, m%1 + punpcklwd m%2, m%1 +%if %7 < 32 + pmaddwd m%1, m%7, m%2 + pmaddwd m%4, m%7, m%3 +%else + vpbroadcastd m%1, [o(pw_m%7_%6)] + pmaddwd m%4, m%3, m%1 + pmaddwd m%1, m%2 +%endif + paddd m%4, m%5 + paddd m%1, m%5 + psrad m%4, 14 + psrad m%1, 14 + packssdw m%1, m%4 +%if %7 < 32 + pmaddwd m%3, m%6 + pmaddwd m%2, m%6 +%else + vpbroadcastd m%4, [o(pw_%6_%7)] + pmaddwd m%3, m%4 + pmaddwd m%2, m%4 +%endif + paddd m%3, m%5 + paddd m%2, m%5 + psrad m%3, 14 + psrad m%2, 14 + packssdw m%2, m%3 +%endmacro + +%macro ADST_MULSUB_2W 8-10 ; dst/src[1-2], dst[3-4], tmp, rnd, coef[1-4] + vpbroadcastd m%3, [o(pw_m%8_%7)] + vpbroadcastd m%4, [o(pw_%7_%8)] + pmaddwd m%3, m%1 + pmaddwd m%1, m%4 +%if %0 == 8 + vpbroadcastd m%5, [o(pw_%8_%7)] + vpbroadcastd m%4, [o(pw_m%7_%8)] +%else + vpbroadcastd m%5, [o(pw_m%10_%9)] + vpbroadcastd m%4, [o(pw_%9_%10)] +%endif + pmaddwd m%5, m%2 + pmaddwd m%4, m%2 + paddd m%3, m%6 + paddd m%1, m%6 + psubd m%2, m%1, m%4 + paddd m%1, m%4 + psubd m%4, m%3, m%5 + paddd m%3, m%5 +%endmacro + +%macro ADST_MULSUB_4W 12-14 ; dst/src[1-4], tmp[1-5], rnd, coef[1-4] + vpbroadcastd m%8, [o(pw_%11_%12)] + vpbroadcastd m%7, [o(pw_m%12_%11)] + punpckhwd m%5, m%2, m%1 + punpcklwd m%2, m%1 +%if %0 == 12 + vpbroadcastd m%1, [o(pw_m%11_%12)] + vpbroadcastd m%9, [o(pw_%12_%11)] +%else + vpbroadcastd m%1, [o(pw_%13_%14)] + vpbroadcastd m%9, [o(pw_m%14_%13)] +%endif + pmaddwd m%6, m%5, m%8 + pmaddwd m%8, m%2 + pmaddwd m%5, m%7 + pmaddwd m%2, m%7 + punpckhwd m%7, m%4, m%3 + punpcklwd m%4, m%3 + pmaddwd m%3, m%7, m%1 + pmaddwd m%1, m%4 + pmaddwd m%7, m%9 + pmaddwd m%9, m%4 + REPX {paddd x, m%10}, m%6, m%8, m%5, m%2 + psubd m%4, m%6, m%3 + paddd m%6, m%3 + psubd m%3, m%8, m%1 + paddd m%1, m%8 + REPX {psrad x, 14}, m%4, m%3, m%6, m%1 + psubd m%8, m%5, m%7 + paddd m%5, m%7 + packssdw m%3, m%4 + psubd m%4, m%2, m%9 + paddd m%2, m%9 + packssdw m%1, m%6 + REPX {psrad x, 14}, m%8, m%4, m%5, m%2 + packssdw m%4, m%8 + packssdw m%2, m%5 +%endmacro + +%macro INV_TXFM_FN 3-4 0 ; type1, type2, size, eob_offset +cglobal vp9_i%1_i%2_%3_add, 4, 5, 0, dst, stride, c, eob, tx2 + %undef cmp + %define %%p1 m(vp9_i%1_%3_internal) + lea r6, [o_base] + ; Jump to the 1st txfm function if we're not taking the fast path, which + ; in turn performs an indirect jump to the 2nd txfm function. + lea tx2q, [m(vp9_i%2_%3_internal).pass2] +%ifidn %1_%2, dct_dct + cmp eobd, 1 + jne %%p1 +%else +%if %4 + add eobd, %4 +%endif + ; jump to the 1st txfm function unless it's located directly after this + times ((%%end - %%p1) >> 31) & 1 jmp %%p1 +ALIGN function_align +%%end: +%endif +%endmacro + +%macro INV_TXFM_4X4_FN 2 ; type1, type2 + INV_TXFM_FN %1, %2, 4x4 +%ifidn %1_%2, dct_dct + vpbroadcastw m0, [cq] + vpbroadcastd m1, [o(pw_11585x2)] + pmulhrsw m0, m1 + pmulhrsw m0, m1 + mova m1, m0 + jmp m(vp9_idct_4x4_internal).pass2_end +%endif +%endmacro + +INV_TXFM_4X4_FN dct, dct +INV_TXFM_4X4_FN dct, adst + +cglobal vp9_idct_4x4_internal, 0, 5, 6, dst, stride, c, eob, tx2 + mova m0, [cq+16*0] + mova m1, [cq+16*1] + call .main +.pass1_end: + mova m2, [o(deint_shuf)] + shufps m3, m0, m1, q1331 + shufps m0, m1, q0220 + pshufb m0, m2 + pshufb m1, m3, m2 + jmp tx2q +.pass2: + call .main +.pass2_end: + vpbroadcastd m2, [o(pw_2048)] + pmulhrsw m0, m2 + pmulhrsw m1, m2 + lea r3, [dstq+strideq*2] + movd m2, [dstq+strideq*0] + pinsrd m2, [dstq+strideq*1], 1 + movd m3, [r3 +strideq*1] + pinsrd m3, [r3 +strideq*0], 1 + pxor m4, m4 + pmovzxbw m2, m2 + mova [cq+16*0], m4 + pmovzxbw m3, m3 + mova [cq+16*1], m4 + paddw m0, m2 + paddw m1, m3 + packuswb m0, m1 + movd [dstq+strideq*0], m0 + pextrd [dstq+strideq*1], m0, 1 + pextrd [r3 +strideq*0], m0, 3 + pextrd [r3 +strideq*1], m0, 2 + RET +ALIGN function_align +.main: + vpbroadcastd m4, [o(pd_8192)] + punpckhwd m2, m1, m0 + psubw m3, m0, m1 + paddw m0, m1 + punpcklqdq m0, m3 + ITX_MUL2X_PACK 2, 1, 3, 4, 6270, 15137 + vpbroadcastd m4, [o(pw_11585x2)] + pmulhrsw m0, m4 ; t0 t1 + psubw m1, m0, m2 ; out3 out2 + paddw m0, m2 ; out0 out1 + ret + +INV_TXFM_4X4_FN adst, dct +INV_TXFM_4X4_FN adst, adst + +cglobal vp9_iadst_4x4_internal, 0, 5, 6, dst, stride, c, eob, tx2 + mova m0, [cq+16*0] + mova m1, [cq+16*1] + call .main + jmp m(vp9_idct_4x4_internal).pass1_end +.pass2: + call .main + jmp m(vp9_idct_4x4_internal).pass2_end +ALIGN function_align +.main: + vpbroadcastd m4, [o(pw_0_13377)] + punpckhwd m2, m0, m1 + vpbroadcastd m5, [o(pw_13377_m13377)] + punpcklwd m0, m1 + vpbroadcastd m1, [o(pw_m13377_m5283)] + vpbroadcastd m3, [o(pw_13377_m15212)] + pmaddwd m4, m2 + pmaddwd m5, m0 + pmaddwd m1, m2 + pmaddwd m3, m2 + paddd m4, m5 ; 2 + vpbroadcastd m5, [o(pw_9929_m5283)] + pmaddwd m5, m0 + paddd m1, m5 + paddd m3, m5 ; 1 + vpbroadcastd m5, [o(pw_5283_15212)] + pmaddwd m0, m5 + vpbroadcastd m5, [o(pw_13377_9929)] + pmaddwd m2, m5 + vpbroadcastd m5, [o(pd_8192)] + paddd m4, m5 + paddd m0, m5 + paddd m3, m5 + paddd m1, m0 ; 3 + paddd m0, m2 ; 0 + REPX {psrad x, 14}, m4, m1, m3, m0 + packssdw m1, m4 ; out3 out2 + packssdw m0, m3 ; out0 out1 + ret + +%macro WRITE_8X4 4-7 strideq*1, strideq*2, r3 ; coefs[1-2], tmp[1-2], off[1-3] + movq xm%3, [dstq ] + movhps xm%3, [dstq+%5] + movq xm%4, [dstq+%6] + movhps xm%4, [dstq+%7] + pmovzxbw m%3, xm%3 + pmovzxbw m%4, xm%4 + paddw m%3, m%1 + paddw m%4, m%2 + packuswb m%3, m%4 + vextracti128 xm%4, m%3, 1 + movq [dstq ], xm%3 + movhps [dstq+%6], xm%3 + movq [dstq+%5], xm%4 + movhps [dstq+%7], xm%4 +%endmacro + +%macro INV_TXFM_8X8_FN 2 ; type1, type2 + INV_TXFM_FN %1, %2, 8x8 +%ifidn %1_%2, dct_dct + vpbroadcastw xm2, [cq] + vpbroadcastd xm1, [o(pw_11585x2)] + vpbroadcastd xm0, [o(pw_1024)] + mov word [cq], 0 + pmulhrsw xm2, xm1 + add r3d, 3 + pmulhrsw xm2, xm1 + pmulhrsw xm2, xm0 +.dconly_loop: + pmovzxbw xm0, [dstq+strideq*0] + pmovzxbw xm1, [dstq+strideq*1] + paddw xm0, xm2 + paddw xm1, xm2 + packuswb xm0, xm1 + movq [dstq+strideq*0], xm0 + movhps [dstq+strideq*1], xm0 + lea dstq, [dstq+strideq*2] + dec r3d + jg .dconly_loop + RET +%endif +%endmacro + +INIT_YMM avx2 +INV_TXFM_8X8_FN dct, dct +INV_TXFM_8X8_FN dct, adst + +cglobal vp9_idct_8x8_internal, 0, 5, 8, dst, stride, c, eob, tx2 + vpermq m0, [cq+32*0], q3120 ; 0 1 + vpermq m3, [cq+32*3], q3120 ; 6 7 + vpermq m2, [cq+32*2], q3120 ; 4 5 + vpermq m1, [cq+32*1], q3120 ; 2 3 + call .main + shufps m4, m0, m1, q0220 + shufps m5, m0, m1, q1331 + vbroadcasti128 m0, [o(deint_shuf)] + shufps m1, m2, m3, q0220 + shufps m3, m2, m3, q1331 + REPX {pshufb x, m0}, m4, m5, m1, m3 + vinserti128 m0, m4, xm1, 1 + vperm2i128 m2, m4, m1, 0x31 + vinserti128 m1, m5, xm3, 1 + vperm2i128 m3, m5, m3, 0x31 + jmp tx2q +.pass2: + call .main + vpbroadcastd m4, [o(pw_1024)] + vpermq m1, m1, q2031 + vpermq m3, m3, q2031 +.end: + vpermq m0, m0, q3120 + vpermq m2, m2, q3120 + REPX {pmulhrsw x, m4}, m0, m1, m2, m3 + pxor m4, m4 + REPX {mova [cq+32*x], m4}, 0, 1, 2, 3 + lea r3, [strideq*3] + WRITE_8X4 0, 1, 4, 5 + lea dstq, [dstq+strideq*4] + WRITE_8X4 2, 3, 4, 5 + RET +ALIGN function_align +.main: + vpbroadcastd m6, [o(pd_8192)] + punpckhwd m5, m3, m0 ; in7 in1 + punpckhwd m4, m1, m2 ; in3 in5 + punpcklwd m3, m1 ; in2 in6 + psubw m1, m0, m2 + paddw m0, m2 + punpcklqdq m0, m1 ; in0+in4 in0-in4 + ITX_MUL2X_PACK 5, 1, 2, 6, 3196, 16069, 1 ; t4a t7a + ITX_MUL2X_PACK 4, 1, 2, 6, 13623, 9102, 1 ; t5a t6a + ITX_MUL2X_PACK 3, 1, 2, 6, 6270, 15137 ; t3 t2 + vpbroadcastd m6, [o(pw_11585x2)] + psubw m2, m5, m4 ; t4 t7 + paddw m5, m4 ; t5a t6a + pshufd m4, m2, q1032 + psubw m1, m2, m4 + paddw m4, m2 + vpblendd m4, m1, 0xcc + pmulhrsw m0, m6 ; t0 t1 + pmulhrsw m4, m6 ; t6 t5 + psubw m1, m0, m3 ; tmp3 tmp2 + paddw m0, m3 ; tmp0 tmp1 + shufps m2, m5, m4, q1032 + vpblendd m5, m4, 0xcc + psubw m3, m0, m2 ; out7 out6 + paddw m0, m2 ; out0 out1 + psubw m2, m1, m5 ; out4 out5 + paddw m1, m5 ; out3 out2 + ret + +INV_TXFM_8X8_FN adst, dct +INV_TXFM_8X8_FN adst, adst + +cglobal vp9_iadst_8x8_internal, 0, 5, 8, dst, stride, c, eob, tx2 + vpermq m4, [cq+32*0], q1302 ; 1 0 + vpermq m3, [cq+32*3], q3120 ; 6 7 + vpermq m5, [cq+32*1], q1302 ; 3 2 + vpermq m2, [cq+32*2], q3120 ; 4 5 + call .main + punpcklwd m4, m0, m1 + punpckhwd m0, m1 + punpcklwd m1, m2, m3 + punpckhwd m2, m3 + pxor m3, m3 + psubw m0, m3, m0 + psubw m2, m3, m2 + punpcklwd m3, m4, m0 + punpckhwd m4, m0 + punpcklwd m0, m1, m2 + punpckhwd m1, m2 + vperm2i128 m2, m3, m0, 0x31 + vinserti128 m0, m3, xm0, 1 + vperm2i128 m3, m4, m1, 0x31 + vinserti128 m1, m4, xm1, 1 + jmp tx2q +.pass2: + pshufd m4, m0, q1032 + pshufd m5, m1, q1032 + call .main + vpbroadcastd m5, [o(pw_1024)] + vpbroadcastd xm4, [o(pw_2048)] + psubw m4, m5 ; lower half = 1024, upper half = -1024 + REPX {vpermq x, x, q3120}, m1, m3 + jmp m(vp9_idct_8x8_internal).end +ALIGN function_align +.main: + vpbroadcastd m7, [o(pd_8192)] + punpckhwd m0, m4, m3 ; 0 7 + punpckhwd m1, m5, m2 ; 2 5 + punpcklwd m2, m5 ; 4 3 + punpcklwd m3, m4 ; 6 1 + ADST_MULSUB_2W 0, 2, 4, 5, 6, 7, 1606, 16305, 12665, 10394 ; t0, t4, t1, t5 + pslld m2, 2 + REPX {psrad x, 14}, m0, m5, m4 + pblendw m2, m5, 0x55 ; t5 t4 + packssdw m0, m4 ; t0 t1 + ADST_MULSUB_2W 1, 3, 4, 5, 6, 7, 7723, 14449, 15679, 4756 ; t2, t6, t3, t7 + pslld m5, 2 + REPX {psrad x, 14}, m3, m1, m4 + pblendw m3, m5, 0xaa ; t6 t7 + packssdw m1, m4 ; t2 t3 + ADST_MULSUB_2W 2, 3, 4, 5, 6, 7, 6270, 15137 ; t4, t6, t5, t7 + REPX {psrad x, 14}, m3, m2, m5, m4 + packssdw m3, m5 ; t6 t7 + packssdw m2, m4 ; -out1 out6 + vpbroadcastd m5, [o(pw_11585x2)] + psubw m4, m0, m1 ; t2 t3 + paddw m0, m1 ; out0 -out7 + punpckhqdq m1, m4, m3 ; t3 t7 + punpcklqdq m4, m3 ; t2 t6 + punpckhqdq m3, m2, m0 ; out6 -out7 + punpcklqdq m0, m2 ; out0 -out1 + psubw m2, m4, m1 + paddw m1, m4 + pshufd m1, m1, q1032 + pmulhrsw m2, m5 ; out4 -out5 + pmulhrsw m1, m5 ; out2 -out3 + ret + +%macro WRITE_16X2 6 ; coefs[1-2], tmp[1-2], offset[1-2] + pmovzxbw m%3, [dstq+%5] +%ifnum %1 + paddw m%3, m%1 +%else + paddw m%3, %1 +%endif + pmovzxbw m%4, [dstq+%6] +%ifnum %2 + paddw m%4, m%2 +%else + paddw m%4, %2 +%endif + packuswb m%3, m%4 + vpermq m%3, m%3, q3120 + mova [dstq+%5], xm%3 + vextracti128 [dstq+%6], m%3, 1 +%endmacro + +%macro INV_TXFM_16X16_FN 2-3 0 ; type1, type2, eob_offset + INV_TXFM_FN %1, %2, 16x16, %3 +%ifidn %1_%2, dct_dct + movd xm0, [o(pw_11585x2)] + pmulhrsw xm3, xm0, [cq] + pxor m2, m2 + pmulhrsw xm3, xm0 + pmulhrsw xm3, [o(pw_512)] + movd [cq], xm2 + add r3d, 7 + vpbroadcastw m3, xm3 +.dconly_loop: + mova xm1, [dstq+strideq*0] + vinserti128 m1, [dstq+strideq*1], 1 + punpcklbw m0, m1, m2 + punpckhbw m1, m2 + paddw m0, m3 + paddw m1, m3 + packuswb m0, m1 + mova [dstq+strideq*0], xm0 + vextracti128 [dstq+strideq*1], m0, 1 + lea dstq, [dstq+strideq*2] + dec r3d + jg .dconly_loop + RET +%endif +%endmacro + +%macro IDCT16_MAIN 2 ; name, idct32 (uses 32-bit intermediates for 11585 multiplies) +%1_fast: + vpbroadcastd m10, [o(pw_m9102x2)] + vpbroadcastd m8, [o(pw_13623x2)] + vpbroadcastd m14, [o(pw_16069x2)] + vpbroadcastd m15, [o(pw_3196x2)] + pmulhrsw m10, m6 + vpbroadcastd m12, [o(pw_15137x2)] + pmulhrsw m6, m8 + vpbroadcastd m8, [o(pw_6270x2)] + pmulhrsw m14, m2 + vpbroadcastd m9, [o(pw_11585x2)] + pmulhrsw m2, m15 + vpbroadcastd m15, [o(pd_8192)] + pmulhrsw m12, m4 + pmulhrsw m4, m8 + pmulhrsw m0, m9 + mova m8, m0 + jmp %%main2 +ALIGN function_align +%1: + mova [rsp+gprsize+32*1], m13 + mova [rsp+gprsize+32*2], m9 + vpbroadcastd m15, [o(pd_8192)] + ITX_MULSUB_2W 10, 6, 9, 13, 15, 13623, 9102 ; t5a, t6a + ITX_MULSUB_2W 2, 14, 9, 13, 15, 3196, 16069 ; t4a, t7a + ITX_MULSUB_2W 4, 12, 9, 13, 15, 6270, 15137 ; t2, t3 + ITX_MULSUB_2W 0, 8, 9, 13, 15, 11585, 11585 ; t1, t0 +%%main2: + paddw m13, m14, m6 ; t7 + psubw m14, m6 ; t6a + paddw m6, m2, m10 ; t4 + psubw m2, m10 ; t5a +%if %2 + ITX_MULSUB_2W 14, 2, 9, 10, 15, 11585, 11585 ; t5, t6 + psubw m10, m0, m4 ; t2 + paddw m4, m0 ; t1 + paddw m0, m8, m12 ; t0 + psubw m8, m12 ; t3 + psubw m9, m4, m2 ; t6 + paddw m2, m4 ; t1 + psubw m4, m8, m6 ; t4 + paddw m8, m6 ; t3 + psubw m6, m10, m14 ; t5 + paddw m10, m14 ; t2 +%else + vpbroadcastd m9, [o(pw_11585x2)] + psubw m10, m14, m2 + paddw m2, m14 + pmulhrsw m10, m9 ; t5 + pmulhrsw m2, m9 ; t6 + psubw m14, m0, m4 ; t2 + paddw m4, m0 ; t1 + paddw m0, m8, m12 ; t0 + psubw m8, m12 ; t3 + psubw m9, m4, m2 ; t6 + paddw m2, m4 ; t1 + psubw m4, m8, m6 ; t4 + paddw m8, m6 ; t3 + psubw m6, m14, m10 ; t5 + paddw m10, m14 ; t2 +%endif + psubw m14, m0, m13 ; t7 + paddw m0, m13 ; t0 + test eobd, eobd + jl %%main3_fast + mova m12, [rsp+gprsize+32*2] ; in9 + mova [rsp+gprsize+32*2], m14 + mova m13, [rsp+gprsize+32*1] ; in13 + mova [rsp+gprsize+32*1], m2 + mova m14, [rsp+gprsize+32*0] ; in15 + mova [rsp+gprsize+32*0], m8 + ITX_MULSUB_2W 1, 14, 2, 8, 15, 1606, 16305 ; t8a, t15a + ITX_MULSUB_2W 12, 7, 2, 8, 15, 12665, 10394 ; t9a, t14a + ITX_MULSUB_2W 5, 11, 2, 8, 15, 7723, 14449 ; t10a, t13a + ITX_MULSUB_2W 13, 3, 2, 8, 15, 15679, 4756 ; t11a, t12a + jmp %%main3 +%%main3_fast: + mova [rsp+gprsize+32*2], m14 + mova [rsp+gprsize+32*1], m2 + mova [rsp+gprsize+32*0], m8 + vpbroadcastd m14, [o(pw_16305x2)] + vpbroadcastd m2, [o(pw_1606x2)] + vpbroadcastd m12, [o(pw_m10394x2)] + vpbroadcastd m8, [o(pw_12665x2)] + pmulhrsw m14, m1 + vpbroadcastd m11, [o(pw_14449x2)] + pmulhrsw m1, m2 + vpbroadcastd m2, [o(pw_7723x2)] + pmulhrsw m12, m7 + vpbroadcastd m13, [o(pw_m4756x2)] + pmulhrsw m7, m8 + vpbroadcastd m8, [o(pw_15679x2)] + pmulhrsw m11, m5 + pmulhrsw m5, m2 + pmulhrsw m13, m3 + pmulhrsw m3, m8 +%%main3: + paddw m2, m11, m3 ; t12 + psubw m3, m11 ; t13 + psubw m11, m14, m7 ; t14 + paddw m14, m7 ; t15 + psubw m7, m13, m5 ; t10 + paddw m5, m13 ; t11 + psubw m13, m1, m12 ; t9 + paddw m12, m1 ; t8 + ITX_MULSUB_2W 11, 13, 1, 8, 15, 6270, 15137 ; t9a, t14a + ITX_MULSUB_2W 3, 7, 1, 8, 15, m15137, 6270 ; t10a, t13a +%if %2 + psubw m1, m12, m5 ; t11a + paddw m12, m5 ; t8a + psubw m5, m13, m7 ; t13 + paddw m13, m7 ; t14 + psubw m7, m14, m2 ; t12a + paddw m14, m2 ; t15a + psubw m2, m11, m3 ; t10 + paddw m3, m11 ; t9 + ITX_MULSUB_2W 5, 2, 8, 11, 15, 11585, 11585 ; t10a, t13a + ITX_MULSUB_2W 7, 1, 8, 11, 15, 11585, 11585 ; t11, t12 +%else + vpbroadcastd m15, [o(pw_11585x2)] + psubw m8, m12, m5 ; t11a + paddw m12, m5 ; t8a + psubw m5, m13, m7 ; t13 + paddw m13, m7 ; t14 + psubw m7, m14, m2 ; t12a + paddw m14, m2 ; t15a + psubw m1, m11, m3 ; t10 + paddw m3, m11 ; t9 + paddw m2, m5, m1 ; t13a + psubw m5, m1 ; t10a + paddw m1, m7, m8 ; t12 + psubw m7, m8 ; t11 + REPX {pmulhrsw x, m15}, m2, m5, m1, m7 +%endif + mova m8, [rsp+gprsize+32*1] ; t1 + psubw m15, m0, m14 ; out15 + paddw m0, m14 ; out0 + psubw m14, m8, m13 ; out14 + paddw m8, m13 ; out1 + psubw m13, m10, m2 ; out13 + paddw m2, m10 ; out2 + psubw m11, m4, m7 ; out11 + paddw m4, m7 ; out4 + mova m7, [rsp+gprsize+32*2] ; t7 + psubw m10, m6, m5 ; out10 + paddw m5, m6 ; out5 + paddw m6, m9, m3 ; out6 + psubw m9, m3 ; out9 + mova m3, [rsp+gprsize+32*0] ; t3 + mova [rsp+gprsize+32*1], m8 + psubw m8, m7, m12 ; out8 + paddw m7, m12 ; out7 + psubw m12, m3, m1 ; out12 + paddw m3, m1 ; out3 +%endmacro + +INV_TXFM_16X16_FN dct, dct +INV_TXFM_16X16_FN dct, adst, 39-23 + +cglobal vp9_idct_16x16_internal, 0, 5, 16, 32*6, dst, stride, c, eob, tx2 + mova m0, [cq+32*0] + mova m1, [cq+32*1] + mova m2, [cq+32*2] + mova m3, [cq+32*3] + mova m4, [cq+32*4] + mova m5, [cq+32*5] + mova m6, [cq+32*6] + mova m7, [cq+32*7] + sub eobd, 39 + jl .pass1_fast + add cq, 32*12 + mova m8, [cq-32*4] + mova m9, [cq-32*3] + mova m10, [cq-32*2] + mova m11, [cq-32*1] + mova m12, [cq+32*0] + mova m13, [cq+32*1] + mova m14, [cq+32*2] + mova m15, [cq+32*3] + mova [rsp], m15 + call .main + vextracti128 [rsp+16*4], m0, 1 + mova [rsp+16*0], xm0 +.pass1_end: + vextracti128 [rsp+16*5], m8, 1 + mova [rsp+16*1], xm8 + mova xm1, [rsp+32*1+16*0] + vinserti128 m8, m9, [rsp+32*1+16*1], 0 + vinserti128 m1, xm9, 1 + vperm2i128 m9, m2, m10, 0x31 + vinserti128 m2, xm10, 1 + vperm2i128 m10, m3, m11, 0x31 + vinserti128 m3, xm11, 1 + vperm2i128 m11, m4, m12, 0x31 + vinserti128 m4, xm12, 1 + vperm2i128 m12, m5, m13, 0x31 + vinserti128 m5, xm13, 1 + vperm2i128 m13, m6, m14, 0x31 + vinserti128 m6, xm14, 1 + vperm2i128 m14, m7, m15, 0x31 + vinserti128 m7, xm15, 1 + mova m15, [rsp+32*2] + pxor m0, m0 + mov r3, -32*12 +.zero_loop: + mova [cq+r3+32*0], m0 + mova [cq+r3+32*1], m0 + mova [cq+r3+32*2], m0 + mova [cq+r3+32*3], m0 + add r3, 32*4 + jle .zero_loop + punpcklwd m0, m9, m10 + punpckhwd m9, m10 + punpcklwd m10, m15, m8 + punpckhwd m15, m8 + punpckhwd m8, m11, m12 + punpcklwd m11, m12 + punpckhwd m12, m13, m14 + punpcklwd m13, m14 + punpckhdq m14, m11, m13 + punpckldq m11, m13 + punpckldq m13, m15, m9 + punpckhdq m15, m9 + punpckldq m9, m10, m0 + punpckhdq m10, m0 + punpckhdq m0, m8, m12 + punpckldq m8, m12 + punpcklqdq m12, m13, m8 + punpckhqdq m13, m8 + punpcklqdq m8, m9, m11 + punpckhqdq m9, m11 + punpckhqdq m11, m10, m14 + punpcklqdq m10, m14 + punpcklqdq m14, m15, m0 + punpckhqdq m15, m0 + mova m0, [rsp] + mova [rsp], m15 + punpckhwd m15, m4, m5 + punpcklwd m4, m5 + punpckhwd m5, m0, m1 + punpcklwd m0, m1 + punpckhwd m1, m6, m7 + punpcklwd m6, m7 + punpckhwd m7, m2, m3 + punpcklwd m2, m3 + punpckhdq m3, m0, m2 + punpckldq m0, m2 + punpckldq m2, m4, m6 + punpckhdq m4, m6 + punpckhdq m6, m5, m7 + punpckldq m5, m7 + punpckldq m7, m15, m1 + punpckhdq m15, m1 + punpckhqdq m1, m0, m2 + punpcklqdq m0, m2 + punpcklqdq m2, m3, m4 + punpckhqdq m3, m4 + punpcklqdq m4, m5, m7 + punpckhqdq m5, m7 + punpckhqdq m7, m6, m15 + punpcklqdq m6, m15 + jmp tx2q +.pass1_fast: + call .main_fast + mova xm1, [rsp+32*1] +.pass1_fast_end: + vinserti128 m0, xm8, 1 + vinserti128 m1, xm9, 1 + vinserti128 m2, xm10, 1 + vinserti128 m3, xm11, 1 + vinserti128 m4, xm12, 1 + vinserti128 m5, xm13, 1 + vinserti128 m6, xm14, 1 + vinserti128 m7, xm15, 1 + pxor m8, m8 + REPX {mova [cq+32*x], m8}, 0, 1, 2, 3, 4, 5, 6, 7 + call .transpose_8x8 + jmp tx2q +.pass2: + test eobd, eobd + jl .pass2_fast + call .main + jmp .pass2_end +.pass2_fast: + call .main_fast +.pass2_end: + vpbroadcastd m1, [o(pw_512)] + REPX {pmulhrsw x, m1}, m0, m2, m4, m5, m6, m7, m8, m9, m10, m11, m12, m14 +.end: + REPX {pmulhrsw x, m1}, m3, m13, m15 + pmulhrsw m1, [rsp+32*1] + mova [rsp], m6 + lea r3, [strideq*3] + WRITE_16X2 0, 1, 6, 0, strideq*0, strideq*1 + WRITE_16X2 2, 3, 0, 1, strideq*2, r3 + lea dstq, [dstq+strideq*4] + WRITE_16X2 4, 5, 0, 1, strideq*0, strideq*1 + WRITE_16X2 [rsp], 7, 0, 1, strideq*2, r3 + lea dstq, [dstq+strideq*4] + WRITE_16X2 8, 9, 0, 1, strideq*0, strideq*1 + WRITE_16X2 10, 11, 0, 1, strideq*2, r3 + lea dstq, [dstq+strideq*4] + WRITE_16X2 12, 13, 0, 1, strideq*0, strideq*1 + WRITE_16X2 14, 15, 0, 1, strideq*2, r3 + RET +ALIGN function_align + IDCT16_MAIN .main, 0 + ret +ALIGN function_align +.transpose_8x8: + punpckhwd m8, m4, m5 + punpcklwd m4, m5 + punpckhwd m5, m0, m1 + punpcklwd m0, m1 + punpckhwd m1, m6, m7 + punpcklwd m6, m7 + punpckhwd m7, m2, m3 + punpcklwd m2, m3 + punpckhdq m3, m0, m2 + punpckldq m0, m2 + punpckldq m2, m4, m6 + punpckhdq m4, m6 + punpckhdq m6, m5, m7 + punpckldq m5, m7 + punpckldq m7, m8, m1 + punpckhdq m8, m1 + punpckhqdq m1, m0, m2 + punpcklqdq m0, m2 + punpcklqdq m2, m3, m4 + punpckhqdq m3, m4 + punpcklqdq m4, m5, m7 + punpckhqdq m5, m7 + punpckhqdq m7, m6, m8 + punpcklqdq m6, m8 + ret + +%macro ADST_MULSUB_4W_FAST 12 ; dst/src[1-4], tmp[1-3], rnd, coef[1-4] + vpbroadcastd m%5, [o(pw_%11_m%10)] + vpbroadcastd m%6, [o(pw_m%11_m%10)] + punpckhwd m%7, m%3, m%2 + punpcklwd m%3, m%2 + pmaddwd m%2, m%3, m%5 + pmaddwd m%5, m%7 + pmaddwd m%4, m%3, m%6 + pmaddwd m%6, m%7 + REPX {paddd x, m%8}, m%2, m%5, m%4, m%6 + REPX {psrad x, 14 }, m%2, m%5, m%4, m%6 + packssdw m%2, m%5 + vpbroadcastd m%5, [o(pw_%12_%9)] + packssdw m%4, m%6 + vpbroadcastd m%6, [o(pw_m%12_%9)] + pmaddwd m%1, m%3, m%5 + pmaddwd m%5, m%7 + pmaddwd m%3, m%6 + pmaddwd m%6, m%7 + REPX {paddd x, m%8}, m%1, m%5, m%3, m%6 + REPX {psrad x, 14 }, m%1, m%5, m%3, m%6 + packssdw m%1, m%5 + packssdw m%3, m%6 +%endmacro + +INV_TXFM_16X16_FN adst, dct, 39-18 +INV_TXFM_16X16_FN adst, adst + +cglobal vp9_iadst_16x16_internal, 0, 5, 16, 32*6, dst, stride, c, eob, tx2 + mova m0, [cq+32*0] + mova m1, [cq+32*1] + mova m2, [cq+32*2] + mova m3, [cq+32*3] + mova m4, [cq+32*4] + mova m5, [cq+32*5] + mova m6, [cq+32*6] + mova m7, [cq+32*7] + sub eobd, 39 + jl .pass1_fast + add cq, 32*12 + mova m8, [cq-32*4] + mova m9, [cq-32*3] + mova m10, [cq-32*2] + mova m11, [cq-32*1] + mova m12, [cq+32*0] + mova m13, [cq+32*1] + mova m14, [cq+32*2] + mova m15, [cq+32*3] + mova [rsp+32*0], m15 + call .main + call .pass1_main_part2 + mova [rsp+32*1], m1 + jmp m(vp9_idct_16x16_internal).pass1_end +.pass1_fast: + call .main_fast + call .pass1_main_part2 + mova xm0, [rsp+32*0] + jmp m(vp9_idct_16x16_internal).pass1_fast_end +.pass2: + test eobd, eobd + jl .pass2_fast + call .main + jmp .pass2_end +.pass2_fast: + call .main_fast +.pass2_end: + ; In pass 2 we're going to clip to pixels afterwards anyway, so clipping to + ; 16-bit here will produce the same result as using 32-bit intermediates. + paddsw m5, m10, m11 ; -out5 + psubsw m10, m11 ; out10 + psubsw m11, m8, m4 ; out11 + paddsw m4, m8 ; out4 + psubsw m8, m7, m9 ; out8 + paddsw m7, m9 ; -out7 + psubsw m9, m6, m1 ; out9 + paddsw m6, m1 ; out6 + vpbroadcastd m1, [o(pw_11585x2)] + REPX {pmulhrsw x, m1}, m4, m6, m8, m9, m10, m11 + vpbroadcastd m1, [o(pw_m11585x2)] + pmulhrsw m5, m1 + pmulhrsw m7, m1 + vpbroadcastd m1, [o(pw_512)] + REPX {pmulhrsw x, m1}, m0, m2, m4, m5, m6, m7, m8, m9, m10, m11, m12, m14 + vpbroadcastd m1, [o(pw_m512)] + jmp m(vp9_idct_16x16_internal).end +ALIGN function_align +.main_fast: + mova [rsp+gprsize+32*1], m0 + mova [rsp+gprsize+32*2], m3 + mova [rsp+gprsize+32*3], m4 + vpbroadcastd m15, [o(pd_8192)] + ADST_MULSUB_4W_FAST 13, 2, 5, 10, 0, 3, 4, 15, 3981, 15893, 14053, 8423 + ADST_MULSUB_4W_FAST 9, 6, 1, 14, 0, 3, 4, 15, 9760, 13160, 16207, 2404 + jmp .main2 +ALIGN function_align +.main: + mova [rsp+gprsize+32*1], m0 + mova [rsp+gprsize+32*2], m3 + mova [rsp+gprsize+32*3], m4 + mova [rsp+gprsize+32*4], m12 + mova [rsp+gprsize+32*5], m8 + vpbroadcastd m15, [o(pd_8192)] + ADST_MULSUB_4W 13, 2, 5, 10, 0, 3, 4, 8, 12, 15, 3981, 15893, 14053, 8423 ; t2a, t3a, t10a, t11a + ADST_MULSUB_4W 9, 6, 1, 14, 0, 3, 4, 8, 12, 15, 9760, 13160, 16207, 2404 ; t6a, t7a, t14a, t15a +.main2: + ADST_MULSUB_4W 5, 10, 14, 1, 0, 3, 4, 8, 12, 15, 13623, 9102 ; t10a, t11a, t14a, t15a + psubw m4, m2, m6 ; t7 + paddw m2, m6 ; t3 + psubw m8, m13, m9 ; t6 + paddw m13, m9 ; t2 + mova m0, [rsp+gprsize+32*0] ; in15 + mova [rsp+gprsize+32*0], m10 + mova m10, [rsp+gprsize+32*1] ; in0 + mova [rsp+gprsize+32*1], m13 + mova m13, [rsp+gprsize+32*2] ; in3 + mova [rsp+gprsize+32*2], m1 + mova m6, [rsp+gprsize+32*3] ; in4 + mova [rsp+gprsize+32*3], m2 + mova m2, [rsp+gprsize+32*4] ; in12 + mova [rsp+gprsize+32*4], m5 + mova m5, [rsp+gprsize+32*5] ; in8 + mova [rsp+gprsize+32*5], m14 + test eobd, eobd + jl .main3_fast + ADST_MULSUB_4W 0, 10, 7, 5, 1, 3, 9, 12, 14, 15, 804, 16364, 12140, 11003 ; t0a, t1a, t8a, t9a + ADST_MULSUB_4W 11, 6, 13, 2, 1, 3, 9, 12, 14, 15, 7005, 14811, 15426, 5520 ; t4a, t5a, t12a, t13a + jmp .main3 +.main3_fast: + ADST_MULSUB_4W_FAST 0, 10, 7, 5, 1, 3, 9, 15, 804, 16364, 12140, 11003 + ADST_MULSUB_4W_FAST 11, 6, 13, 2, 1, 3, 9, 15, 7005, 14811, 15426, 5520 +.main3: + ADST_MULSUB_4W 7, 5, 2, 13, 1, 3, 9, 12, 14, 15, 3196, 16069 ; t8a, t9a, t12a, t13a + psubw m3, m0, m11 ; t4 + paddw m0, m11 ; t0 + psubw m12, m10, m6 ; t5 + paddw m10, m6 ; t1 + mova m11, [rsp+gprsize+32*5] ; t14a + mova [rsp+gprsize+32*5], m10 + mova m10, [rsp+gprsize+32*2] ; t15a + mova [rsp+gprsize+32*2], m0 + ADST_MULSUB_4W 2, 13, 10, 11, 0, 1, 6, 9, 14, 15, 6270, 15137 ; out2, -out13, t14a, t15a + ADST_MULSUB_4W 3, 12, 4, 8, 0, 1, 6, 9, 14, 15, 6270, 15137 ; -out3, out12, t6, t7 + mova m6, [rsp+gprsize+32*4] ; t10a + mova m14, [rsp+gprsize+32*0] ; t11a + mova m9, [rsp+gprsize+32*1] ; t2 + mova m0, [rsp+gprsize+32*2] ; t0 + mova m15, [rsp+gprsize+32*5] ; t1 + psubw m1, m7, m6 ; t10 + paddw m7, m6 ; -out1 + psubw m6, m5, m14 ; t11 + paddw m14, m5 ; out14 + mova m5, [rsp+gprsize+32*3] ; t3 + mova [rsp+gprsize+32*1], m7 + psubw m7, m0, m9 ; t2a + paddw m0, m9 ; out0 + psubw m9, m15, m5 ; t3a + paddw m15, m5 ; -out15 + ret +ALIGN function_align +.pass1_main_part2: + mova [rsp+gprsize+16*0], xm0 + vextracti128 [rsp+gprsize+16*4], m0, 1 + mova [rsp+gprsize+32*3], m15 + mova [rsp+gprsize+32*4], m13 + mova [rsp+gprsize+32*5], m3 + vpbroadcastd m15, [o(pw_m11585_11585)] + vpbroadcastd m13, [o(pw_11585_11585)] + vpbroadcastd m3, [o(pd_8192)] + punpcklwd m5, m11, m10 + punpckhwd m11, m10 + pmaddwd m10, m15, m5 + pmaddwd m0, m15, m11 + pmaddwd m5, m13 + pmaddwd m11, m13 + paddd m10, m3 + paddd m0, m3 + psubd m5, m3, m5 + psubd m11, m3, m11 + REPX {psrad x, 14}, m10, m0, m5, m11 + packssdw m10, m0 ; out10 + packssdw m5, m11 ; out5 + punpcklwd m11, m8, m4 + punpckhwd m8, m4 + pmaddwd m4, m13, m11 + pmaddwd m0, m13, m8 + pmaddwd m11, m15 + pmaddwd m8, m15 + paddd m4, m3 + paddd m0, m3 + psubd m11, m3, m11 + psubd m8, m3, m8 + REPX {psrad x, 14}, m4, m0, m11, m8 + packssdw m4, m0 ; out4 + packssdw m11, m8 ; out11 + punpcklwd m8, m9, m7 + punpckhwd m9, m7 + pmaddwd m7, m13, m8 + pmaddwd m0, m13, m9 + pmaddwd m8, m15 + pmaddwd m9, m15 + psubd m7, m3, m7 + psubd m0, m3, m0 + paddd m8, m3 + paddd m9, m3 + REPX {psrad x, 14}, m7, m0, m8, m9 + packssdw m7, m0 ; out7 + packssdw m8, m9 ; out8 + punpckhwd m0, m6, m1 + punpcklwd m6, m1 + pmaddwd m1, m15, m0 + pmaddwd m9, m15, m6 + pmaddwd m0, m13 + pmaddwd m6, m13 + psubd m1, m3, m1 + psubd m9, m3, m9 + paddd m0, m3 + paddd m6, m3 + pxor m3, m3 + psubw m15, m3, [rsp+gprsize+32*3] ; out15 + REPX {psrad x, 14}, m1, m9, m0, m6 + psubw m13, m3, [rsp+gprsize+32*4] ; out13 + packssdw m9, m1 ; out7 + psubw m1, m3, [rsp+gprsize+32*1] ; out1 + packssdw m6, m0 ; out8 + psubw m3, [rsp+gprsize+32*5] ; out3 + ret + +%macro LOAD_8ROWS 2 ; src, stride + mova m0, [%1+%2*0] + mova m1, [%1+%2*1] + mova m2, [%1+%2*2] + mova m3, [%1+%2*3] + mova m4, [%1+%2*4] + mova m5, [%1+%2*5] + mova m6, [%1+%2*6] + mova m7, [%1+%2*7] +%endmacro + +%macro LOAD_8ROWS_H 2 ; src, stride + mova m8, [%1+%2*0] + mova m9, [%1+%2*1] + mova m10, [%1+%2*2] + mova m11, [%1+%2*3] + mova m12, [%1+%2*4] + mova m13, [%1+%2*5] + mova m14, [%1+%2*6] + mova m15, [%1+%2*7] +%endmacro + +; Perform the final sumsub step and YMM lane shuffling +%macro IDCT32_PASS1_END 4 ; row[1-2], tmp[1-2] + mova m%3, [tmp2q+32*( 3-%1)] + psubw m%4, m%1, m%3 + paddw m%1, m%3 + mova m%3, [tmp1q+32*(11-%2)] + mova [tmp1q+32*(11-%2)+16], xm%4 + vextracti128 [tmp2q+32*( 3-%1)+16], m%4, 1 + paddw m%4, m%2, m%3 + psubw m%2, m%3 + mova [tmp1q+32*(11-%2)], xm%2 + vextracti128 [tmp2q+32*( 3-%1)], m%2, 1 + vperm2i128 m%2, m%1, m%4, 0x31 + vinserti128 m%1, xm%4, 1 +%endmacro + +%macro IDCT32_PASS2_END 7 ; coefs[1-2], tmp[1-2], rnd, offset[1-2] + mova m%4, [%2] + paddw m%3, m%1, m%4 + psubw m%1, m%4 + pmovzxbw m%4, [dstq+%6] + pmulhrsw m%3, m%5 + pmulhrsw m%1, m%5 + paddw m%3, m%4 + pmovzxbw m%4, [r2+%7] + paddw m%1, m%4 + packuswb m%3, m%1 + vpermq m%3, m%3, q3120 + mova [dstq+%6], xm%3 + vextracti128 [r2+%7], m%3, 1 +%endmacro + +cglobal vp9_idct_idct_32x32_add, 4, 4, 0, dst, stride, c, eob + lea r6, [o_base] + sub eobd, 1 + jnz .pass1 + movd xm0, [o(pw_11585x2)] + pmulhrsw xm5, xm0, [cq] + pxor m4, m4 + pmulhrsw xm5, xm0 + pmulhrsw xm5, [o(pw_512)] + movd [cq], xm4 + or r3d, 16 + vpbroadcastw m5, xm5 +.dconly_loop: + mova m2, [dstq+strideq*0] + mova m3, [dstq+strideq*1] + punpcklbw m0, m2, m4 + punpckhbw m2, m4 + punpcklbw m1, m3, m4 + punpckhbw m3, m4 + REPX {paddw x, m5}, m0, m2, m1, m3 + packuswb m0, m2 + packuswb m1, m3 + mova [dstq+strideq*0], m0 + mova [dstq+strideq*1], m1 + lea dstq, [dstq+strideq*2] + dec r3d + jg .dconly_loop + RET +.pass1: + PROLOGUE 0, 9, 16, 32*67, dst, stride, c, eob, tmp1, \ + tmp2, base, tmp3, tmp4 + %undef cmp + lea tmp1q, [rsp+32*7] + sub eobd, 135 + lea tmp2q, [tmp1q+32*8] + mov tmp4d, eobd +.pass1_loop: + LOAD_8ROWS cq+64*1, 64*2 + test eobd, eobd + jl .pass1_fast + LOAD_8ROWS_H cq+64*17, 64*2 + call .main + LOAD_8ROWS_H cq+64*16, 64*2 + mova [rsp], m15 + LOAD_8ROWS cq+64*0, 64*2 + call .idct16 + mov tmp3d, 64*30 + jmp .pass1_loop_end +.pass1_fast: + call .main_fast + LOAD_8ROWS cq+64*0, 64*2 + call .idct16_fast + mov tmp3d, 64*14 +.pass1_loop_end: + pxor m1, m1 +.zero_loop: + mova [cq+tmp3q+64*1], m1 + mova [cq+tmp3q+64*0], m1 + mova [cq+tmp3q-64*1], m1 + mova [cq+tmp3q-64*2], m1 + sub tmp3d, 64*4 + jg .zero_loop + mova [rsp+32*0], m9 + IDCT32_PASS1_END 0, 8, 1, 9 + IDCT32_PASS1_END 2, 10, 1, 9 + IDCT32_PASS1_END 3, 11, 1, 9 + IDCT32_PASS1_END 4, 12, 1, 9 + IDCT32_PASS1_END 5, 13, 1, 9 + IDCT32_PASS1_END 6, 14, 1, 9 + IDCT32_PASS1_END 7, 15, 1, 9 + mova m1, [rsp+32*1] + mova m9, [rsp+32*0] + mova [rsp+32*0], m6 + mova [rsp+32*1], m7 + IDCT32_PASS1_END 1, 9, 6, 7 + mova m7, [rsp+32*1] + punpckhwd m6, m12, m13 + punpcklwd m12, m13 + punpckhwd m13, m8, m9 + punpcklwd m8, m9 + punpckhwd m9, m14, m15 + punpcklwd m14, m15 + punpckhwd m15, m10, m11 + punpcklwd m10, m11 + punpckhdq m11, m8, m10 + punpckldq m8, m10 + punpckldq m10, m12, m14 + punpckhdq m12, m14 + punpckhdq m14, m13, m15 + punpckldq m13, m15 + punpckldq m15, m6, m9 + punpckhdq m6, m9 + punpckhqdq m9, m8, m10 + punpcklqdq m8, m10 + punpcklqdq m10, m11, m12 + punpckhqdq m11, m12 + punpcklqdq m12, m13, m15 + punpckhqdq m13, m15 + punpckhqdq m15, m14, m6 + punpcklqdq m14, m6 + mova m6, [rsp+32*0] + mova [rsp+32*0], m8 + call m(vp9_idct_16x16_internal).transpose_8x8 + lea tmp3q, [tmp1q+32*32] + mova m8, [rsp] + mova [tmp3q-32*4], m0 + mova [tmp3q-32*3], m2 + mova [tmp3q-32*2], m4 + mova [tmp3q-32*1], m6 + mova [tmp3q+32*0], m8 + mova [tmp3q+32*1], m10 + mova [tmp3q+32*2], m12 + mova [tmp3q+32*3], m14 + add tmp3q, 32*8 + mova [tmp3q-32*4], m1 + mova [tmp3q-32*3], m3 + mova [tmp3q-32*2], m5 + mova [tmp3q-32*1], m7 + mova [tmp3q+32*0], m9 + mova [tmp3q+32*1], m11 + mova [tmp3q+32*2], m13 + mova [tmp3q+32*3], m15 + mova m0, [tmp1q-32*4] + mova m1, [tmp1q-32*3] + mova m2, [tmp1q-32*2] + mova m3, [tmp1q-32*1] + mova m4, [tmp1q+32*0] + mova m5, [tmp1q+32*1] + mova m6, [tmp1q+32*2] + mova m7, [tmp1q+32*3] + call m(vp9_idct_16x16_internal).transpose_8x8 + mova [tmp1q-32*4], m0 + mova m0, [tmp2q-32*4] + mova [tmp2q-32*4], m1 + mova m1, [tmp2q-32*3] + mova [tmp1q-32*3], m2 + mova m2, [tmp2q-32*2] + mova [tmp2q-32*3], m3 + mova m3, [tmp2q-32*1] + mova [tmp1q-32*2], m4 + mova m4, [tmp2q+32*0] + mova [tmp2q-32*2], m5 + mova m5, [tmp2q+32*1] + mova [tmp1q-32*1], m6 + mova m6, [tmp2q+32*2] + mova [tmp2q-32*1], m7 + mova m7, [tmp2q+32*3] + call m(vp9_idct_16x16_internal).transpose_8x8 + mova [tmp1q+32*0], m0 + mova [tmp2q+32*0], m1 + mova [tmp1q+32*1], m2 + mova [tmp2q+32*1], m3 + mova [tmp1q+32*2], m4 + mova [tmp2q+32*2], m5 + mova [tmp1q+32*3], m6 + mova [tmp2q+32*3], m7 + add cq, 32 + add tmp1q, 32*16 + add tmp2q, 32*16 + add tmp4d, 0x80000000 + jnc .pass1_loop + add tmp1q, 32*24 + imul r2, strideq, 19 + lea tmp4q, [strideq*3] + add r2, dstq + test eobd, eobd + jge .pass2_loop + add tmp1q, 32*16 + add tmp2q, 32*16 + add tmp3q, 32*16 +.pass2_loop: + LOAD_8ROWS tmp2q-32*4, 32 + test eobd, eobd + jl .pass2_fast + LOAD_8ROWS_H tmp3q-32*4, 32 + call .main + sub tmp3q, 32*8 + LOAD_8ROWS_H tmp3q-32*4, 32 + sub tmp3q, 32*16 + LOAD_8ROWS tmp3q-32*4, 32 + mova [rsp], m15 + call .idct16 + jmp .pass2_loop_end +.pass2_fast: + call .main_fast + sub tmp3q, 32*24 + LOAD_8ROWS tmp3q-32*4, 32 + call .idct16_fast +.pass2_loop_end: + mova [rsp+32*0], m7 + mova [rsp+32*2], m15 + vpbroadcastd m15, [o(pw_512)] + IDCT32_PASS2_END 0, tmp2q+32*3, 1, 7, 15, strideq*0, tmp4q*4 + IDCT32_PASS2_END 4, tmp2q-32*1, 0, 7, 15, strideq*4, strideq*8 + IDCT32_PASS2_END 8, tmp1q+32*3, 0, 4, 15, strideq*8, strideq*4 + IDCT32_PASS2_END 12, tmp1q-32*1, 0, 4, 15, tmp4q*4, strideq*0 + add dstq, strideq + sub r2, strideq + mova m1, [rsp+32*1] + IDCT32_PASS2_END 1, tmp2q+32*2, 0, 4, 15, strideq*0, tmp4q*4 + IDCT32_PASS2_END 5, tmp2q-32*2, 0, 4, 15, strideq*4, strideq*8 + IDCT32_PASS2_END 9, tmp1q+32*2, 0, 4, 15, strideq*8, strideq*4 + IDCT32_PASS2_END 13, tmp1q-32*2, 0, 4, 15, tmp4q*4, strideq*0 + add dstq, strideq + sub r2, strideq + IDCT32_PASS2_END 2, tmp2q+32*1, 0, 4, 15, strideq*0, tmp4q*4 + IDCT32_PASS2_END 6, tmp2q-32*3, 0, 4, 15, strideq*4, strideq*8 + IDCT32_PASS2_END 10, tmp1q+32*1, 0, 4, 15, strideq*8, strideq*4 + IDCT32_PASS2_END 14, tmp1q-32*3, 0, 4, 15, tmp4q*4, strideq*0 + add dstq, strideq + sub r2, strideq + mova m7, [rsp+32*0] + mova m1, [rsp+32*2] + IDCT32_PASS2_END 3, tmp2q+32*0, 0, 4, 15, strideq*0, tmp4q*4 + IDCT32_PASS2_END 7, tmp2q-32*4, 0, 4, 15, strideq*4, strideq*8 + IDCT32_PASS2_END 11, tmp1q+32*0, 0, 4, 15, strideq*8, strideq*4 + IDCT32_PASS2_END 1, tmp1q-32*4, 0, 4, 15, tmp4q*4, strideq*0 + lea tmp3q, [tmp1q-32*32] + cmp tmp2q, tmp3q + jb .ret + sub tmp2q, 32*32 + sub dstq, tmp4q + lea r2, [r2+tmp4q+16] + add dstq, 16 + jmp .pass2_loop +.ret: + RET +ALIGN function_align + IDCT16_MAIN .idct16, 1 + ret +ALIGN function_align +.main_fast: + mova [tmp1q+32*0], m7 + vpbroadcastd m11, [o(pw_14811x2)] + vpbroadcastd m7, [o(pw_7005x2)] + vpbroadcastd m12, [o(pw_m5520x2)] + vpbroadcastd m8, [o(pw_15426x2)] + vpbroadcastd m13, [o(pw_15893x2)] + vpbroadcastd m15, [o(pw_3981x2)] + pmulhrsw m11, m4 ; t29a + vpbroadcastd m10, [o(pw_m8423x2)] + pmulhrsw m4, m7 ; t18a + vpbroadcastd m7, [o(pw_14053x2)] + pmulhrsw m12, m3 ; t19a + vpbroadcastd m9, [o(pw_13160x2)] + pmulhrsw m3, m8 ; t28a + vpbroadcastd m8, [o(pw_9760x2)] + pmulhrsw m13, m2 ; t27a + vpbroadcastd m14, [o(pw_m2404x2)] + pmulhrsw m2, m15 ; t20a + vpbroadcastd m15, [o(pw_16207x2)] + pmulhrsw m10, m5 ; t21a + pmulhrsw m5, m7 ; t26a + pmulhrsw m9, m6 ; t25a + pmulhrsw m6, m8 ; t22a + pmulhrsw m14, m1 ; t23a + pmulhrsw m1, m15 ; t24a + vpbroadcastd m15, [o(pd_8192)] + jmp .main2 +ALIGN function_align +.main: + mova [tmp1q+32*0], m7 + mova [tmp1q-32*1], m15 + mova [tmp1q-32*2], m8 + vpbroadcastd m15, [o(pd_8192)] + ITX_MULSUB_2W 4, 11, 7, 8, 15, 7005, 14811 ; t18a, t29a + ITX_MULSUB_2W 12, 3, 7, 8, 15, 15426, 5520 ; t19a, t28a + ITX_MULSUB_2W 2, 13, 7, 8, 15, 3981, 15893 ; t20a, t27a + ITX_MULSUB_2W 10, 5, 7, 8, 15, 14053, 8423 ; t21a, t26a + ITX_MULSUB_2W 6, 9, 7, 8, 15, 9760, 13160 ; t22a, t25a + ITX_MULSUB_2W 14, 1, 7, 8, 15, 16207, 2404 ; t23a, t24a +.main2: + psubw m7, m12, m4 ; t18 + paddw m12, m4 ; t19 + psubw m4, m2, m10 ; t21 + paddw m2, m10 ; t20 + psubw m10, m14, m6 ; t22 + paddw m14, m6 ; t23 + psubw m6, m1, m9 ; t25 + paddw m1, m9 ; t24 + psubw m9, m13, m5 ; t26 + paddw m13, m5 ; t27 + psubw m5, m3, m11 ; t29 + paddw m3, m11 ; t28 + ITX_MULSUB_2W 5, 7, 8, 11, 15, m16069, 3196 ; t18a, t29a + ITX_MULSUB_2W 9, 4, 8, 11, 15, 13623, 9102 ; t21a, t26a + ITX_MULSUB_2W 6, 10, 8, 11, 15, m9102, 13623 ; t22a, t25a + psubw m8, m14, m2 ; t20a + paddw m14, m2 ; t23a + psubw m2, m1, m13 ; t27a + paddw m1, m13 ; t24a + psubw m13, m6, m9 ; t21 + paddw m6, m9 ; t22 + psubw m9, m10, m4 ; t26 + paddw m10, m4 ; t25 + ITX_MULSUB_2W 2, 8, 4, 11, 15, m15137, 6270 ; t20, t27 + ITX_MULSUB_2W 9, 13, 4, 11, 15, m15137, 6270 ; t21a, t26a + mova [tmp1q+32*1], m6 + mova [tmp1q+32*2], m14 + mova [tmp1q+32*3], m1 + mova m4, [tmp1q+32*0] ; in15 + test eobd, eobd + jl .main3_fast + mova m6, [tmp1q-32*1] ; in31 + mova m14, [tmp1q-32*2] ; in17 + ITX_MULSUB_2W 0, 6, 1, 11, 15, 804, 16364 ; t16a, t31a + ITX_MULSUB_2W 14, 4, 1, 11, 15, 12140, 11003 ; t17a, t30a + jmp .main3 +.main3_fast: + vpbroadcastd m6, [o(pw_16364x2)] + vpbroadcastd m1, [o(pw_804x2)] + vpbroadcastd m14, [o(pw_m11003x2)] + vpbroadcastd m11, [o(pw_12140x2)] + pmulhrsw m6, m0 ; t31a + pmulhrsw m0, m1 ; t16a + pmulhrsw m14, m4 ; t17a + pmulhrsw m4, m11 ; t30a +.main3: + psubw m1, m0, m14 ; t17 + paddw m0, m14 ; t16 + psubw m14, m6, m4 ; t30 + paddw m4, m6 ; t31 + ITX_MULSUB_2W 14, 1, 6, 11, 15, 3196, 16069 ; t17a, t30a + psubw m6, m0, m12 ; t19a + paddw m0, m12 ; t16a + psubw m12, m4, m3 ; t28a + paddw m4, m3 ; t31a + psubw m3, m14, m5 ; t18 + paddw m14, m5 ; t17 + psubw m5, m1, m7 ; t29 + paddw m1, m7 ; t30 + ITX_MULSUB_2W 5, 3, 7, 11, 15, 6270, 15137 ; t18a, t29a + ITX_MULSUB_2W 12, 6, 7, 11, 15, 6270, 15137 ; t19, t28 + psubw m7, m1, m10 ; t25a + paddw m1, m10 ; t30a + psubw m10, m5, m9 ; t21 + paddw m5, m9 ; t18 + psubw m9, m12, m2 ; t20a + paddw m12, m2 ; t19a + psubw m2, m3, m13 ; t26 + paddw m3, m13 ; t29 + psubw m13, m6, m8 ; t27a + paddw m6, m8 ; t28a + mova [tmp1q-32*2], m5 + mova [tmp1q-32*1], m12 + mova [tmp2q+32*0], m6 + mova [tmp2q+32*1], m3 + mova [tmp2q+32*2], m1 + mova m5, [tmp1q+32*1] ; t22 + mova m6, [tmp1q+32*2] ; t23 + mova m3, [tmp1q+32*3] ; t24a + psubw m1, m14, m5 ; t22a + paddw m14, m5 ; t17a + psubw m5, m0, m6 ; t23 + paddw m0, m6 ; t16 + psubw m6, m4, m3 ; t24 + paddw m4, m3 ; t31 + vpbroadcastd m8, [o(pw_m11585_11585)] + vpbroadcastd m3, [o(pw_11585_11585)] + mova [tmp1q-32*4], m0 + mova [tmp1q-32*3], m14 + mova [tmp2q+32*3], m4 + ITX_MULSUB_2W 13, 9, 0, 4, 15, 3, 8 ; t20, t27 + ITX_MULSUB_2W 2, 10, 0, 4, 15, 3, 8 ; t21a, t26a + ITX_MULSUB_2W 7, 1, 0, 4, 15, 3, 8 ; t22, t25 + ITX_MULSUB_2W 6, 5, 0, 4, 15, 3, 8 ; t23a, t24a + mova [tmp1q+32*0], m13 + mova [tmp1q+32*1], m2 + mova [tmp1q+32*2], m7 + mova [tmp1q+32*3], m6 + mova [tmp2q-32*4], m5 + mova [tmp2q-32*3], m1 + mova [tmp2q-32*2], m10 + mova [tmp2q-32*1], m9 + ret + +%endif ----------------------------------------------------------------------- Summary of changes: libavcodec/x86/Makefile | 1 + libavcodec/x86/vp9dsp_init.c | 30 +- libavcodec/x86/vp9itxfm.asm | 392 +-------- libavcodec/x86/vp9itxfm_avx2.asm | 1640 ++++++++++++++++++++++++++++++++++++++ 4 files changed, 1667 insertions(+), 396 deletions(-) create mode 100644 libavcodec/x86/vp9itxfm_avx2.asm hooks/post-receive --

1 0

[ffmpeg] branch master updated. 49b5c17e09 avformat/mccdec: Constify data
by ffmpeg-git＠ffmpeg.org 19 Sep '25

19 Sep '25

The branch, master has been updated via 49b5c17e0924df1de72db44c5f49a1b65422d202 (commit) via cce1af54132ff7044a89388064490f53e6e80579 (commit) via f65875c0825ec9900836fec37c75e6f106c43f7f (commit) via e965fe9057ff4436fe0778c8c5c396793f3f331c (commit) via 28f0b7a237f24b16d09d292d39aa190fed39077f (commit) via 3c79219e93a37d671271d934e206ad5e8f63405b (commit) via 159330016c26d0a56ac3bf7b1a6b3aca3287511f (commit) via d81a971b3503ec636fcb15655975e2ceb0e9a1ef (commit) via 62dd3e1e7db374400cf7d9f8be1cac11d95916c5 (commit) via 0cd78bddc40f95a770700d2bec098cabdcbd51f1 (commit) via a34a85ed54089682fdab4ee4ff93999d4a7b3776 (commit) from 82a09aef05e5b738d6dec16a0e0080543ff30c87 (commit) - Log ----------------------------------------------------------------- commit 49b5c17e0924df1de72db44c5f49a1b65422d202 Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 20:56:36 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:31:10 2025 +0200 avformat/mccdec: Constify data Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccdec.c b/libavformat/mccdec.c index 71882d13a9..e9b6fa14b4 100644 --- a/libavformat/mccdec.c +++ b/libavformat/mccdec.c @@ -158,7 +158,7 @@ struct ValidTimeCodeRate { char str[5]; }; -static struct ValidTimeCodeRate valid_time_code_rates[] = { +static const struct ValidTimeCodeRate valid_time_code_rates[] = { { .rate = { .num = 24, .den = 1 }, .str = "24" }, { .rate = { .num = 25, .den = 1 }, .str = "25" }, { .rate = { .num = 30000, .den = 1001 }, .str = "30DF" }, commit cce1af54132ff7044a89388064490f53e6e80579 Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 20:56:10 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:31:05 2025 +0200 avformat/mccenc: Hardcode codec names Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index 1fbeebfe0f..c64d53805d 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -437,9 +437,7 @@ static int mcc_init(AVFormatContext *avf) } else if (st->codecpar->codec_id != AV_CODEC_ID_SMPTE_436M_ANC) { av_log(avf, AV_LOG_ERROR, - "mcc muxer supports only codec %s or codec %s\n", - avcodec_get_name(AV_CODEC_ID_SMPTE_436M_ANC), - avcodec_get_name(AV_CODEC_ID_EIA_608)); + "mcc muxer supports only codec smpte_436m_anc or codec eia_608\n"); return AVERROR(EINVAL); } commit f65875c0825ec9900836fec37c75e6f106c43f7f Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 20:40:50 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:31:02 2025 +0200 avformat/mccdec: Avoid relocations Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccdec.c b/libavformat/mccdec.c index 8a9eff4c2a..71882d13a9 100644 --- a/libavformat/mccdec.c +++ b/libavformat/mccdec.c @@ -155,7 +155,7 @@ static int time_tracker_set_time(TimeTracker *tt, const MCCTimecode *tc, void *l struct ValidTimeCodeRate { AVRational rate; - const char *str; + char str[5]; }; static struct ValidTimeCodeRate valid_time_code_rates[] = { commit e965fe9057ff4436fe0778c8c5c396793f3f331c Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 20:36:18 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:59 2025 +0200 avformat/mccenc: Avoid relocations Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index d34b8625c4..1fbeebfe0f 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -144,7 +144,7 @@ static int mcc_write_header(AVFormatContext *avf) if (!localtime_r((time_t[1]){ timeval / 1000000 }, &tm)) return AVERROR(EINVAL); // we can't rely on having the C locale, so convert the date/time to a string ourselves: - static const char *const months[12] = { + static const char months[12][10] = { "January", "February", "March", @@ -162,7 +162,7 @@ static int mcc_write_header(AVFormatContext *avf) av_assert0(tm.tm_mon >= 0 && tm.tm_mon < FF_ARRAY_ELEMS(months)); const char *month = months[tm.tm_mon]; - static const char *const weekdays[7] = { + static const char weekdays[7][10] = { "Sunday", "Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday" }; // assert that values are sane so we don't index out of bounds commit 28f0b7a237f24b16d09d292d39aa190fed39077f Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 20:00:00 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:53 2025 +0200 avformat/mccenc: Add newlines to logmessages Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index 14f67e420a..d34b8625c4 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -129,15 +129,15 @@ static int mcc_write_header(AVFormatContext *avf) else creation_program = "FFmpeg version " FFMPEG_VERSION; } else if (strchr(creation_program, '\n')) { - av_log(avf, AV_LOG_FATAL, "creation_program must not contain multiple lines of text"); + av_log(avf, AV_LOG_FATAL, "creation_program must not contain multiple lines of text\n"); return AVERROR(EINVAL); } if (avf->flags & AVFMT_FLAG_BITEXACT && !av_strcasecmp(mcc->creation_time, "now")) - av_log(avf, AV_LOG_ERROR, "creation_time must be overridden for bit-exact output"); + av_log(avf, AV_LOG_ERROR, "creation_time must be overridden for bit-exact output\n"); int64_t timeval = 0; int ret = av_parse_time(&timeval, mcc->creation_time, 0); if (ret < 0) { - av_log(avf, AV_LOG_FATAL, "can't parse creation_time"); + av_log(avf, AV_LOG_FATAL, "can't parse creation_time\n"); return ret; } struct tm tm; @@ -326,7 +326,7 @@ static int mcc_write_packet(AVFormatContext *avf, AVPacket *pkt) default: av_log(avf, AV_LOG_WARNING, - "Unsupported SMPTE 436M ANC Wrapping Type %#x -- discarding ANC packet", + "Unsupported SMPTE 436M ANC Wrapping Type %#x -- discarding ANC packet\n", (unsigned)coded_anc.wrapping_type); continue; } @@ -344,7 +344,7 @@ static int mcc_write_packet(AVFormatContext *avf, AVPacket *pkt) av_log(avf, AV_LOG_WARNING, "MCC Version 1.0 doesn't support ANC packets where the field number (got %u) isn't 0 and " - "line number (got %u) isn't 9: discarding ANC packet", + "line number (got %u) isn't 9: discarding ANC packet\n", field_number, (unsigned)coded_anc.line_number); continue; @@ -416,7 +416,7 @@ static int mcc_init(AVFormatContext *avf) if (mcc->mcc_version == MCC_VERSION_1) { if (mcc->timecode.fps == 60 && mcc->timecode.flags & AV_TIMECODE_FLAG_DROPFRAME) { - av_log(avf, AV_LOG_FATAL, "MCC Version 1.0 doesn't support 60DF (59.94 fps drop-frame)"); + av_log(avf, AV_LOG_FATAL, "MCC Version 1.0 doesn't support 60DF (59.94 fps drop-frame)\n"); return AVERROR(EINVAL); } } commit 3c79219e93a37d671271d934e206ad5e8f63405b Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 19:55:51 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:33 2025 +0200 avformat/mccenc: Check version-timecode compatibility earlier Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index 1baa007d66..14f67e420a 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -122,14 +122,6 @@ static const AVRational valid_time_code_rates[] = { static int mcc_write_header(AVFormatContext *avf) { MCCContext *mcc = avf->priv_data; - switch ((MCCVersion)mcc->mcc_version) { - case MCC_VERSION_1: - if (mcc->timecode.fps == 60 && mcc->timecode.flags & AV_TIMECODE_FLAG_DROPFRAME) { - av_log(avf, AV_LOG_FATAL, "MCC Version 1.0 doesn't support 60DF (59.94 fps drop-frame)"); - return AVERROR(EINVAL); - } - break; - } const char *creation_program = mcc->creation_program; if (!creation_program) { if (avf->flags & AVFMT_FLAG_BITEXACT) @@ -422,6 +414,13 @@ static int mcc_init(AVFormatContext *avf) if (ret < 0) return ret; + if (mcc->mcc_version == MCC_VERSION_1) { + if (mcc->timecode.fps == 60 && mcc->timecode.flags & AV_TIMECODE_FLAG_DROPFRAME) { + av_log(avf, AV_LOG_FATAL, "MCC Version 1.0 doesn't support 60DF (59.94 fps drop-frame)"); + return AVERROR(EINVAL); + } + } + // get av_timecode to calculate how many frames are in 24hr ret = av_timecode_init_from_components(&twenty_four_hr, time_code_rate, timecode_flags, 24, 0, 0, 0, avf); if (ret < 0) commit 159330016c26d0a56ac3bf7b1a6b3aca3287511f Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 19:53:33 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:28 2025 +0200 avformat/mccenc: Deduplicate strings Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index c46fa033cc..1baa007d66 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -58,83 +58,47 @@ typedef enum MCCVersion MCC_VERSION_MAX = MCC_VERSION_2, } MCCVersion; -static const char mcc_header_v1[] = // - "File Format=MacCaption_MCC V1.0\n" - "\n" +#define MCC_HEADER \ + "File Format=MacCaption_MCC V%c.0\n" \ + "\n" \ + "///////////////////////////////////////////////////////////////////////////////////\n" \ + "// Computer Prompting and Captioning Company\n" \ + "// Ancillary Data Packet Transfer File\n" \ + "//\n" \ + "// Permission to generate this format is granted provided that\n" \ + "// 1. This ANC Transfer file format is used on an as-is basis and no warranty is given, and\n" \ + "// 2. This entire descriptive information text is included in a generated .mcc file.\n" \ + "//\n" \ + "// General file format:\n" \ + "// HH:MM:SS:FF(tab)[Hexadecimal ANC data in groups of 2 characters]\n" \ + "// Hexadecimal data starts with the Ancillary Data Packet DID (Data ID defined in S291M)\n" \ + "// and concludes with the Check Sum following the User Data Words.\n" \ + "// Each time code line must contain at most one complete ancillary data packet.\n" \ + "// To transfer additional ANC Data successive lines may contain identical time code.\n" \ + "// Time Code Rate=[24, 25, 30, 30DF, 50, 60%s]\n" \ + "//\n" \ + "// ANC data bytes may be represented by one ASCII character according to the following schema:\n" \ + "// G FAh 00h 00h\n" \ + "// H 2 x (FAh 00h 00h)\n" \ + "// I 3 x (FAh 00h 00h)\n" \ + "// J 4 x (FAh 00h 00h)\n" \ + "// K 5 x (FAh 00h 00h)\n" \ + "// L 6 x (FAh 00h 00h)\n" \ + "// M 7 x (FAh 00h 00h)\n" \ + "// N 8 x (FAh 00h 00h)\n" \ + "// O 9 x (FAh 00h 00h)\n" \ + "// P FBh 80h 80h\n" \ + "// Q FCh 80h 80h\n" \ + "// R FDh 80h 80h\n" \ + "// S 96h 69h\n" \ + "// T 61h 01h\n" \ + "// U E1h 00h 00h 00h\n" \ + "// Z 00h\n" \ + "//\n" \ "///////////////////////////////////////////////////////////////////////////////////\n" - "// Computer Prompting and Captioning Company\n" - "// Ancillary Data Packet Transfer File\n" - "//\n" - "// Permission to generate this format is granted provided that\n" - "// 1. This ANC Transfer file format is used on an as-is basis and no warranty is given, and\n" - "// 2. This entire descriptive information text is included in a generated .mcc file.\n" - "//\n" - "// General file format:\n" - "// HH:MM:SS:FF(tab)[Hexadecimal ANC data in groups of 2 characters]\n" - "// Hexadecimal data starts with the Ancillary Data Packet DID (Data ID defined in S291M)\n" - "// and concludes with the Check Sum following the User Data Words.\n" - "// Each time code line must contain at most one complete ancillary data packet.\n" - "// To transfer additional ANC Data successive lines may contain identical time code.\n" - "// Time Code Rate=[24, 25, 30, 30DF, 50, 60]\n" - "//\n" - "// ANC data bytes may be represented by one ASCII character according to the following schema:\n" - "// G FAh 00h 00h\n" - "// H 2 x (FAh 00h 00h)\n" - "// I 3 x (FAh 00h 00h)\n" - "// J 4 x (FAh 00h 00h)\n" - "// K 5 x (FAh 00h 00h)\n" - "// L 6 x (FAh 00h 00h)\n" - "// M 7 x (FAh 00h 00h)\n" - "// N 8 x (FAh 00h 00h)\n" - "// O 9 x (FAh 00h 00h)\n" - "// P FBh 80h 80h\n" - "// Q FCh 80h 80h\n" - "// R FDh 80h 80h\n" - "// S 96h 69h\n" - "// T 61h 01h\n" - "// U E1h 00h 00h 00h\n" - "// Z 00h\n" - "//\n" - "///////////////////////////////////////////////////////////////////////////////////\n"; - -static const char mcc_header_v2[] = // - "File Format=MacCaption_MCC V2.0\n" - "\n" - "///////////////////////////////////////////////////////////////////////////////////\n" - "// Computer Prompting and Captioning Company\n" - "// Ancillary Data Packet Transfer File\n" - "//\n" - "// Permission to generate this format is granted provided that\n" - "// 1. This ANC Transfer file format is used on an as-is basis and no warranty is given, and\n" - "// 2. This entire descriptive information text is included in a generated .mcc file.\n" - "//\n" - "// General file format:\n" - "// HH:MM:SS:FF(tab)[Hexadecimal ANC data in groups of 2 characters]\n" - "// Hexadecimal data starts with the Ancillary Data Packet DID (Data ID defined in S291M)\n" - "// and concludes with the Check Sum following the User Data Words.\n" - "// Each time code line must contain at most one complete ancillary data packet.\n" - "// To transfer additional ANC Data successive lines may contain identical time code.\n" - "// Time Code Rate=[24, 25, 30, 30DF, 50, 60, 60DF]\n" - "//\n" - "// ANC data bytes may be represented by one ASCII character according to the following schema:\n" - "// G FAh 00h 00h\n" - "// H 2 x (FAh 00h 00h)\n" - "// I 3 x (FAh 00h 00h)\n" - "// J 4 x (FAh 00h 00h)\n" - "// K 5 x (FAh 00h 00h)\n" - "// L 6 x (FAh 00h 00h)\n" - "// M 7 x (FAh 00h 00h)\n" - "// N 8 x (FAh 00h 00h)\n" - "// O 9 x (FAh 00h 00h)\n" - "// P FBh 80h 80h\n" - "// Q FCh 80h 80h\n" - "// R FDh 80h 80h\n" - "// S 96h 69h\n" - "// T 61h 01h\n" - "// U E1h 00h 00h 00h\n" - "// Z 00h\n" - "//\n" - "///////////////////////////////////////////////////////////////////////////////////\n"; + +#define MCC_HEADER_PRINTF_ARGS(mcc_version) (mcc_version) + '0', \ + (mcc_version) == MCC_VERSION_1 ? "" : ", 60DF" /** * generated with the bash command: @@ -158,7 +122,6 @@ static const AVRational valid_time_code_rates[] = { static int mcc_write_header(AVFormatContext *avf) { MCCContext *mcc = avf->priv_data; - const char *mcc_header = mcc_header_v1; switch ((MCCVersion)mcc->mcc_version) { case MCC_VERSION_1: if (mcc->timecode.fps == 60 && mcc->timecode.flags & AV_TIMECODE_FLAG_DROPFRAME) { @@ -166,9 +129,6 @@ static int mcc_write_header(AVFormatContext *avf) return AVERROR(EINVAL); } break; - case MCC_VERSION_2: - mcc_header = mcc_header_v2; - break; } const char *creation_program = mcc->creation_program; if (!creation_program) { @@ -218,13 +178,13 @@ static int mcc_write_header(AVFormatContext *avf) const char *weekday = weekdays[tm.tm_wday]; avio_printf(avf->pb, - "%s\n" + MCC_HEADER "\n" "UUID=%s\n" "Creation Program=%s\n" "Creation Date=%s, %s %d, %d\n" "Creation Time=%02d:%02d:%02d\n" "Time Code Rate=%u%s\n\n", - mcc_header, + MCC_HEADER_PRINTF_ARGS(mcc->mcc_version), mcc_ffmpeg_uuid, creation_program, weekday, commit d81a971b3503ec636fcb15655975e2ceb0e9a1ef Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 19:16:44 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:24 2025 +0200 avformat/mccenc: Constify read-only data Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index bdc8425408..c46fa033cc 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -145,7 +145,7 @@ static const char mcc_header_v2[] = // */ static const char mcc_ffmpeg_uuid[] = "0087C4F6-A6B4-5469-8C8E-BBF44950401D"; -static AVRational valid_time_code_rates[] = { +static const AVRational valid_time_code_rates[] = { { .num = 24, .den = 1 }, { .num = 25, .den = 1 }, { .num = 30000, .den = 1001 }, commit 62dd3e1e7db374400cf7d9f8be1cac11d95916c5 Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 19:16:06 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:20 2025 +0200 avformat/mccenc: Remove redundant setting of time base It has already been done in init. Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index 02541caeb3..bdc8425408 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -158,7 +158,6 @@ static AVRational valid_time_code_rates[] = { static int mcc_write_header(AVFormatContext *avf) { MCCContext *mcc = avf->priv_data; - avpriv_set_pts_info(avf->streams[0], 64, mcc->timecode.rate.den, mcc->timecode.rate.num); const char *mcc_header = mcc_header_v1; switch ((MCCVersion)mcc->mcc_version) { case MCC_VERSION_1: commit 0cd78bddc40f95a770700d2bec098cabdcbd51f1 Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 18:49:42 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:30:17 2025 +0200 avformat/mccenc: Fix assert check Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index f978c420a3..02541caeb3 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -208,7 +208,7 @@ static int mcc_write_header(AVFormatContext *avf) "December", }; // assert that values are sane so we don't index out of bounds - av_assert0(tm.tm_mon >= 0 && tm.tm_mon <= FF_ARRAY_ELEMS(months)); + av_assert0(tm.tm_mon >= 0 && tm.tm_mon < FF_ARRAY_ELEMS(months)); const char *month = months[tm.tm_mon]; static const char *const weekdays[7] = { commit a34a85ed54089682fdab4ee4ff93999d4a7b3776 Author: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> AuthorDate: Fri Sep 19 19:42:24 2025 +0200 Commit: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> CommitDate: Sat Sep 20 00:29:41 2025 +0200 avformat/mccenc: Remove redundant check This has already been checked in init. Reviewed-by: Jacob Lifshay <programmerjake(a)gmail.com> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt(a)outlook.com> diff --git a/libavformat/mccenc.c b/libavformat/mccenc.c index 298bc6dd1a..f978c420a3 100644 --- a/libavformat/mccenc.c +++ b/libavformat/mccenc.c @@ -158,10 +158,6 @@ static AVRational valid_time_code_rates[] = { static int mcc_write_header(AVFormatContext *avf) { MCCContext *mcc = avf->priv_data; - if (avf->nb_streams != 1) { - av_log(avf, AV_LOG_ERROR, "mcc muxer supports at most one stream\n"); - return AVERROR(EINVAL); - } avpriv_set_pts_info(avf->streams[0], 64, mcc->timecode.rate.den, mcc->timecode.rate.num); const char *mcc_header = mcc_header_v1; switch ((MCCVersion)mcc->mcc_version) { ----------------------------------------------------------------------- Summary of changes: libavformat/mccdec.c | 4 +- libavformat/mccenc.c | 166 ++++++++++++++++++--------------------------------- 2 files changed, 61 insertions(+), 109 deletions(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. 82a09aef05 avformat/urldecode: factorize core url decoding from ff_urldecode
by ffmpeg-git＠ffmpeg.org 19 Sep '25

19 Sep '25

The branch, master has been updated via 82a09aef05e5b738d6dec16a0e0080543ff30c87 (commit) via 23cf2c5f01d3465a139c7b6ea7b3abb26e3ae080 (commit) via b0314dc76e8ebebc627606313c3a6b5a4a9bd4af (commit) via 3a29702cb6b8f9dcb1aa66bb65d04b392f9de98d (commit) via f2be0e68eff7fc748a65f91ed4f28c5bffedd4e5 (commit) via b2184a97413c8b1c1794528f7112d31ec95e1db4 (commit) via 7d1e22b7eacd2142b23c19782d032d1f56de9f4b (commit) via ca8cba4eac89577e281515ee51d583650d15cfa3 (commit) via 3e7314fe30ecf356b71bf6e40223f82e4e8c138e (commit) from 1ce3f9fdabf0c769e86b585d9da197e869e1a44e (commit) - Log ----------------------------------------------------------------- commit 82a09aef05e5b738d6dec16a0e0080543ff30c87 Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Tue Aug 26 09:22:57 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/urldecode: factorize core url decoding from ff_urldecode Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/libavformat/urldecode.c b/libavformat/urldecode.c index 150c64c15c..e7fa27b3fa 100644 --- a/libavformat/urldecode.c +++ b/libavformat/urldecode.c @@ -32,20 +32,10 @@ #include "libavutil/avstring.h" #include "urldecode.h" -char *ff_urldecode(const char *url, int decode_plus_sign) +static size_t urldecode(char *dest, const char *url, size_t url_len, int decode_plus_sign) { - int s = 0, d = 0, url_len = 0; + size_t s = 0, d = 0; char c; - char *dest = NULL; - - if (!url) - return NULL; - - url_len = strlen(url) + 1; - dest = av_malloc(url_len); - - if (!dest) - return NULL; while (s < url_len) { c = url[s++]; @@ -82,5 +72,24 @@ char *ff_urldecode(const char *url, int decode_plus_sign) } + return d; +} + +char *ff_urldecode(const char *url, int decode_plus_sign) +{ + char *dest = NULL; + size_t url_len; + + if (!url) + return NULL; + + url_len = strlen(url) + 1; + dest = av_malloc(url_len); + + if (!dest) + return NULL; + + urldecode(dest, url, url_len, decode_plus_sign); + return dest; } commit 23cf2c5f01d3465a139c7b6ea7b3abb26e3ae080 Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Mon Sep 1 00:21:57 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/urldecode: fix decoding last char if it was percent encoded The length check was too strict and if the end of the string was a percent encoded sequence it did not decode correctly. Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/libavformat/urldecode.c b/libavformat/urldecode.c index 5261bcd0cd..150c64c15c 100644 --- a/libavformat/urldecode.c +++ b/libavformat/urldecode.c @@ -50,7 +50,7 @@ char *ff_urldecode(const char *url, int decode_plus_sign) while (s < url_len) { c = url[s++]; - if (c == '%' && s + 2 < url_len) { + if (c == '%' && s + 1 < url_len) { char c2 = url[s++]; char c3 = url[s++]; if (av_isxdigit(c2) && av_isxdigit(c3)) { commit b0314dc76e8ebebc627606313c3a6b5a4a9bd4af Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Mon Aug 25 21:40:40 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/rtpproto: use proper return error codes in rtp_open Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/libavformat/rtpproto.c b/libavformat/rtpproto.c index a8f718114d..69879b7fa8 100644 --- a/libavformat/rtpproto.c +++ b/libavformat/rtpproto.c @@ -240,6 +240,7 @@ static int rtp_open(URLContext *h, const char *uri, int flags) const char *p; int i, max_retry_count = 3; int rtcpflags; + int ret; av_url_split(NULL, 0, NULL, 0, hostname, sizeof(hostname), &rtp_port, path, sizeof(path), uri); @@ -296,8 +297,10 @@ static int rtp_open(URLContext *h, const char *uri, int flags) if (av_find_info_tag(buf, sizeof(buf), "localaddr", p)) { av_freep(&s->localaddr); s->localaddr = av_strdup(buf); - if (!s->localaddr) + if (!s->localaddr) { + ret = AVERROR(ENOMEM); goto fail; + } } } if (s->rw_timeout >= 0) @@ -308,10 +311,12 @@ static int rtp_open(URLContext *h, const char *uri, int flags) if (!(fec_protocol = av_get_token(&p, "="))) { av_log(h, AV_LOG_ERROR, "Failed to parse the FEC protocol value\n"); + ret = AVERROR(EINVAL); goto fail; } if (strcmp(fec_protocol, "prompeg")) { av_log(h, AV_LOG_ERROR, "Unsupported FEC protocol %s\n", fec_protocol); + ret = AVERROR(EINVAL); goto fail; } @@ -320,6 +325,7 @@ static int rtp_open(URLContext *h, const char *uri, int flags) if (av_dict_parse_string(&fec_opts, p, "=", ":", 0) < 0) { av_log(h, AV_LOG_ERROR, "Failed to parse the FEC options\n"); + ret = AVERROR(EINVAL); goto fail; } if (s->ttl > 0) { @@ -331,8 +337,9 @@ static int rtp_open(URLContext *h, const char *uri, int flags) build_udp_url(s, buf, sizeof(buf), hostname, s->localaddr, rtp_port, s->local_rtpport, sources, block); - if (ffurl_open_whitelist(&s->rtp_hd, buf, flags, &h->interrupt_callback, - NULL, h->protocol_whitelist, h->protocol_blacklist, h) < 0) + ret = ffurl_open_whitelist(&s->rtp_hd, buf, flags, &h->interrupt_callback, + NULL, h->protocol_whitelist, h->protocol_blacklist, h); + if (ret < 0) goto fail; s->local_rtpport = ff_udp_get_local_port(s->rtp_hd); if(s->local_rtpport == 65535) { @@ -356,8 +363,9 @@ static int rtp_open(URLContext *h, const char *uri, int flags) build_udp_url(s, buf, sizeof(buf), hostname, s->localaddr, s->rtcp_port, s->local_rtcpport, sources, block); - if (ffurl_open_whitelist(&s->rtcp_hd, buf, rtcpflags, &h->interrupt_callback, - NULL, h->protocol_whitelist, h->protocol_blacklist, h) < 0) + ret = ffurl_open_whitelist(&s->rtcp_hd, buf, rtcpflags, &h->interrupt_callback, + NULL, h->protocol_whitelist, h->protocol_blacklist, h); + if (ret < 0) goto fail; break; } @@ -365,8 +373,9 @@ static int rtp_open(URLContext *h, const char *uri, int flags) s->fec_hd = NULL; if (fec_protocol) { ff_url_join(buf, sizeof(buf), fec_protocol, NULL, hostname, rtp_port, NULL); - if (ffurl_open_whitelist(&s->fec_hd, buf, flags, &h->interrupt_callback, - &fec_opts, h->protocol_whitelist, h->protocol_blacklist, h) < 0) + ret = ffurl_open_whitelist(&s->fec_hd, buf, flags, &h->interrupt_callback, + &fec_opts, h->protocol_whitelist, h->protocol_blacklist, h); + if (ret < 0) goto fail; } @@ -390,7 +399,7 @@ static int rtp_open(URLContext *h, const char *uri, int flags) ffurl_closep(&s->fec_hd); av_free(fec_protocol); av_dict_free(&fec_opts); - return AVERROR(EIO); + return ret; } static int rtp_read(URLContext *h, uint8_t *buf, int size) commit 3a29702cb6b8f9dcb1aa66bb65d04b392f9de98d Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Mon Aug 25 23:00:33 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/rtpproto: add some URL options as AVOption aliases Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/doc/protocols.texi b/doc/protocols.texi index 130ce35f36..cd0726cc7a 100644 --- a/doc/protocols.texi +++ b/doc/protocols.texi @@ -1164,7 +1164,7 @@ rtp://@var{hostname}[:@var{port}][?@var{options}] @var{options} contains a list of &-separated options of the form @var{key}=@var{val}. -The following URL options are supported: +The following options are supported: @table @option @@ -1174,10 +1174,12 @@ Set the TTL (Time-To-Live) value (for multicast only). @item rtcpport=@var{n} Set the remote RTCP port to @var{n}. -@item localrtpport=@var{n} +@item localport, local_rtpport, localrtpport=@var{n} Set the local RTP port to @var{n}. -@item localrtcpport=@var{n}' +Using the localport option name is deprecated and should not be used. + +@item local_rtcpport, localrtcpport=@var{n}' Set the local RTCP port to @var{n}. @item pkt_size=@var{n} @@ -1200,12 +1202,6 @@ List disallowed (blocked) source IP addresses. Send packets to the source address of the latest received packet (if set to 1) or to a default remote address (if set to 0). -@item localport=@var{n} -Set the local RTP port to @var{n}. - -This is a deprecated option. Instead, @option{localrtpport} should be -used. - @item localaddr=@var{addr} Local IP address of a network interface used for sending packets or joining multicast groups. diff --git a/libavformat/rtpproto.c b/libavformat/rtpproto.c index 15d0050936..a8f718114d 100644 --- a/libavformat/rtpproto.c +++ b/libavformat/rtpproto.c @@ -66,12 +66,17 @@ typedef struct RTPContext { #define OFFSET(x) offsetof(RTPContext, x) #define D AV_OPT_FLAG_DECODING_PARAM #define E AV_OPT_FLAG_ENCODING_PARAM +#define DEPR AV_OPT_FLAG_DEPRECATED static const AVOption options[] = { { "ttl", "Time to live (multicast only)", OFFSET(ttl), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 255, .flags = D|E }, { "buffer_size", "Send/Receive buffer size (in bytes)", OFFSET(buffer_size), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, { "rtcp_port", "Custom rtcp port", OFFSET(rtcp_port), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, + { "rtcpport", "Custom rtcp port", OFFSET(rtcp_port), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, { "local_rtpport", "Local rtp port", OFFSET(local_rtpport), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, + { "localrtpport", "Local rtp port", OFFSET(local_rtpport), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, + { "localport", "Local rtp port", OFFSET(local_rtpport), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E|DEPR }, { "local_rtcpport", "Local rtcp port", OFFSET(local_rtcpport), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, + { "localrtcpport", "Local rtcp port", OFFSET(local_rtcpport), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, { "connect", "Connect socket", OFFSET(connect), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = D|E }, { "write_to_source", "Send packets to the source address of the latest received packet", OFFSET(write_to_source), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = D|E }, { "pkt_size", "Maximum packet size", OFFSET(pkt_size), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, .flags = D|E }, commit f2be0e68eff7fc748a65f91ed4f28c5bffedd4e5 Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Tue Aug 26 00:07:32 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/tls: move AVClass to TLSShared Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/libavformat/tls.h b/libavformat/tls.h index b72537c063..6c5f7bc15f 100644 --- a/libavformat/tls.h +++ b/libavformat/tls.h @@ -35,6 +35,7 @@ #define MAX_CERTIFICATE_SIZE 8192 typedef struct TLSShared { + const AVClass *class; char *ca_file; int verify; char *cert_file; diff --git a/libavformat/tls_gnutls.c b/libavformat/tls_gnutls.c index 53306872a0..fe2fd53291 100644 --- a/libavformat/tls_gnutls.c +++ b/libavformat/tls_gnutls.c @@ -44,7 +44,6 @@ GCRY_THREAD_OPTION_PTHREAD_IMPL; #endif typedef struct TLSContext { - const AVClass *class; TLSShared tls_shared; gnutls_session_t session; gnutls_certificate_credentials_t cred; diff --git a/libavformat/tls_libtls.c b/libavformat/tls_libtls.c index 22858d4867..4f011ad67a 100644 --- a/libavformat/tls_libtls.c +++ b/libavformat/tls_libtls.c @@ -32,7 +32,6 @@ #include <tls.h> typedef struct TLSContext { - const AVClass *class; TLSShared tls_shared; struct tls *ctx; } TLSContext; diff --git a/libavformat/tls_mbedtls.c b/libavformat/tls_mbedtls.c index e802c6b872..2bcd3cca63 100644 --- a/libavformat/tls_mbedtls.c +++ b/libavformat/tls_mbedtls.c @@ -40,7 +40,6 @@ #include "libavutil/avstring.h" typedef struct TLSContext { - const AVClass *class; TLSShared tls_shared; mbedtls_ssl_context ssl_context; mbedtls_ssl_config ssl_config; diff --git a/libavformat/tls_openssl.c b/libavformat/tls_openssl.c index db7147e491..edfd657a3f 100644 --- a/libavformat/tls_openssl.c +++ b/libavformat/tls_openssl.c @@ -418,7 +418,6 @@ static X509 *cert_from_pem_string(const char *pem_str) typedef struct TLSContext { - const AVClass *class; TLSShared tls_shared; SSL_CTX *ctx; SSL *ssl; diff --git a/libavformat/tls_schannel.c b/libavformat/tls_schannel.c index b854f484fa..6708302b65 100644 --- a/libavformat/tls_schannel.c +++ b/libavformat/tls_schannel.c @@ -587,7 +587,6 @@ end: } typedef struct TLSContext { - const AVClass *class; TLSShared tls_shared; char *cert_store_subject; diff --git a/libavformat/tls_securetransport.c b/libavformat/tls_securetransport.c index a5a3bd87a1..f52fc97db2 100644 --- a/libavformat/tls_securetransport.c +++ b/libavformat/tls_securetransport.c @@ -43,7 +43,6 @@ SecIdentityRef SecIdentityCreate(CFAllocatorRef allocator, SecCertificateRef cer #define ioErr -36 typedef struct TLSContext { - const AVClass *class; TLSShared tls_shared; SSLContextRef ssl_context; CFArrayRef ca_array; commit b2184a97413c8b1c1794528f7112d31ec95e1db4 Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Sun Aug 31 22:03:05 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/tls: use AV_OPT_TYPE_BOOL for some AVOptions Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/libavformat/tls.h b/libavformat/tls.h index 55361aab30..b72537c063 100644 --- a/libavformat/tls.h +++ b/libavformat/tls.h @@ -76,8 +76,8 @@ typedef struct TLSShared { #define FF_TLS_CLIENT_OPTIONS(pstruct, options_field) \ {"ca_file", "Certificate Authority database file", offsetof(pstruct, options_field . ca_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"cafile", "Certificate Authority database file", offsetof(pstruct, options_field . ca_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ - {"tls_verify", "Verify the peer certificate", offsetof(pstruct, options_field . verify), AV_OPT_TYPE_INT, { .i64 = TLS_VERIFY_DEFAULT }, 0, 1, .flags = TLS_OPTFL }, \ - {"verify", "Verify the peer certificate", offsetof(pstruct, options_field . verify), AV_OPT_TYPE_INT, { .i64 = TLS_VERIFY_DEFAULT }, 0, 1, .flags = TLS_OPTFL }, \ + {"tls_verify", "Verify the peer certificate", offsetof(pstruct, options_field . verify), AV_OPT_TYPE_BOOL, { .i64 = TLS_VERIFY_DEFAULT }, 0, 1, .flags = TLS_OPTFL }, \ + {"verify", "Verify the peer certificate", offsetof(pstruct, options_field . verify), AV_OPT_TYPE_BOOL, { .i64 = TLS_VERIFY_DEFAULT }, 0, 1, .flags = TLS_OPTFL }, \ {"cert_file", "Certificate file", offsetof(pstruct, options_field . cert_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"cert", "Certificate file", offsetof(pstruct, options_field . cert_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"key_file", "Private key file", offsetof(pstruct, options_field . key_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ @@ -87,8 +87,8 @@ typedef struct TLSShared { #define TLS_COMMON_OPTIONS(pstruct, options_field) \ {"listen", "Listen for incoming connections", offsetof(pstruct, options_field . listen), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, .flags = TLS_OPTFL }, \ {"http_proxy", "Set proxy to tunnel through", offsetof(pstruct, options_field . http_proxy), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ - {"external_sock", "Use external socket", offsetof(pstruct, options_field . external_sock), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, .flags = TLS_OPTFL }, \ - {"use_srtp", "Enable use_srtp DTLS extension", offsetof(pstruct, options_field . use_srtp), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, .flags = TLS_OPTFL }, \ + {"external_sock", "Use external socket", offsetof(pstruct, options_field . external_sock), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = TLS_OPTFL }, \ + {"use_srtp", "Enable use_srtp DTLS extension", offsetof(pstruct, options_field . use_srtp), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = TLS_OPTFL }, \ {"mtu", "Maximum Transmission Unit", offsetof(pstruct, options_field . mtu), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, .flags = TLS_OPTFL}, \ {"cert_pem", "Certificate PEM string", offsetof(pstruct, options_field . cert_buf), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"key_pem", "Private key PEM string", offsetof(pstruct, options_field . key_buf), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ commit 7d1e22b7eacd2142b23c19782d032d1f56de9f4b Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Mon Aug 25 23:13:36 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/tls: add some URL options as AVOption aliases Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/doc/protocols.texi b/doc/protocols.texi index 4b871dddb7..130ce35f36 100644 --- a/doc/protocols.texi +++ b/doc/protocols.texi @@ -1989,7 +1989,7 @@ need to be specified for verification to work, but not all libraries and setups have defaults built in. The file must be in OpenSSL PEM format. -@item tls_verify=@var{1|0} +@item tls_verify, verify=@var{1|0} If enabled, try to verify the peer that we are communicating with. Note, if using OpenSSL, this currently only makes sure that the peer certificate is signed by one of the root certificates in the CA @@ -2055,7 +2055,7 @@ need to be specified for verification to work, but not all libraries and setups have defaults built in. The file must be in OpenSSL PEM format. -@item tls_verify=@var{1|0} +@item tls_verify, verify=@var{1|0} If enabled, try to verify the peer that we are communicating with. Note, if using OpenSSL, this currently only makes sure that the peer certificate is signed by one of the root certificates in the CA diff --git a/libavformat/tls.h b/libavformat/tls.h index 2afa248f4c..55361aab30 100644 --- a/libavformat/tls.h +++ b/libavformat/tls.h @@ -77,8 +77,11 @@ typedef struct TLSShared { {"ca_file", "Certificate Authority database file", offsetof(pstruct, options_field . ca_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"cafile", "Certificate Authority database file", offsetof(pstruct, options_field . ca_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"tls_verify", "Verify the peer certificate", offsetof(pstruct, options_field . verify), AV_OPT_TYPE_INT, { .i64 = TLS_VERIFY_DEFAULT }, 0, 1, .flags = TLS_OPTFL }, \ + {"verify", "Verify the peer certificate", offsetof(pstruct, options_field . verify), AV_OPT_TYPE_INT, { .i64 = TLS_VERIFY_DEFAULT }, 0, 1, .flags = TLS_OPTFL }, \ {"cert_file", "Certificate file", offsetof(pstruct, options_field . cert_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ + {"cert", "Certificate file", offsetof(pstruct, options_field . cert_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"key_file", "Private key file", offsetof(pstruct, options_field . key_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ + {"key", "Private key file", offsetof(pstruct, options_field . key_file), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL }, \ {"verifyhost", "Verify against a specific hostname", offsetof(pstruct, options_field . host), AV_OPT_TYPE_STRING, .flags = TLS_OPTFL } #define TLS_COMMON_OPTIONS(pstruct, options_field) \ commit ca8cba4eac89577e281515ee51d583650d15cfa3 Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Mon Aug 25 21:02:54 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/udp: factorize warning unsupported options for builds without PTHREAD_CANCEL Also fix 'circular_buffer_size' parameter name in the message and the 'fifo_size' option description. Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/libavformat/udp.c b/libavformat/udp.c index ded7a1a85e..e1dec49010 100644 --- a/libavformat/udp.c +++ b/libavformat/udp.c @@ -145,7 +145,7 @@ static const AVOption options[] = { { "ttl", "Time to live (multicast only)", OFFSET(ttl), AV_OPT_TYPE_INT, { .i64 = 16 }, 0, 255, E }, { "dscp", "DSCP class for outgoing packets", OFFSET(dscp), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 63, E }, { "connect", "set if connect() should be called on socket", OFFSET(is_connected), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = D|E }, - { "fifo_size", "set the UDP receiving circular buffer size, expressed as a number of packets with size of 188 bytes", OFFSET(circular_buffer_size), AV_OPT_TYPE_INT, {.i64 = 7*4096}, 0, INT_MAX, D }, + { "fifo_size", "set the UDP circular buffer size (in 188-byte packets)", OFFSET(circular_buffer_size), AV_OPT_TYPE_INT, {.i64 = HAVE_PTHREAD_CANCEL ? 7*4096 : 0}, 0, INT_MAX, D }, { "overrun_nonfatal", "survive in case of UDP receiving circular buffer overrun", OFFSET(overrun_nonfatal), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, D }, { "timeout", "set raise error timeout, in microseconds (only in read mode)",OFFSET(timeout), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX, D }, { "sources", "Source list", OFFSET(sources), AV_OPT_TYPE_STRING, { .str = NULL }, .flags = D|E }, @@ -733,10 +733,6 @@ static int udp_open(URLContext *h, const char *uri, int flags) /* assume if no digits were found it is a request to enable it */ if (buf == endptr) s->overrun_nonfatal = 1; - if (!HAVE_PTHREAD_CANCEL) - av_log(h, AV_LOG_WARNING, - "'overrun_nonfatal' option was set but it is not supported " - "on this build (pthread support is required)\n"); } if (av_find_info_tag(buf, sizeof(buf), "ttl", p)) { s->ttl = strtol(buf, NULL, 10); @@ -766,17 +762,9 @@ static int udp_open(URLContext *h, const char *uri, int flags) } if (av_find_info_tag(buf, sizeof(buf), "fifo_size", p)) { s->circular_buffer_size = strtol(buf, NULL, 10); - if (!HAVE_PTHREAD_CANCEL) - av_log(h, AV_LOG_WARNING, - "'circular_buffer_size' option was set but it is not supported " - "on this build (pthread support is required)\n"); } if (av_find_info_tag(buf, sizeof(buf), "bitrate", p)) { s->bitrate = strtoll(buf, NULL, 10); - if (!HAVE_PTHREAD_CANCEL) - av_log(h, AV_LOG_WARNING, - "'bitrate' option was set but it is not supported " - "on this build (pthread support is required)\n"); } if (av_find_info_tag(buf, sizeof(buf), "burst_bits", p)) { s->burst_bits = strtoll(buf, NULL, 10); @@ -802,6 +790,16 @@ static int udp_open(URLContext *h, const char *uri, int flags) if (is_output && av_find_info_tag(buf, sizeof(buf), "broadcast", p)) s->is_broadcast = strtol(buf, NULL, 10); } + if (!HAVE_PTHREAD_CANCEL) { + int64_t optvals[] = {s->overrun_nonfatal, s->bitrate, s->circular_buffer_size}; + const char* optnames[] = { "overrun_nonfatal", "bitrate", "fifo_size"}; + for (unsigned i = 0; i < FF_ARRAY_ELEMS(optvals); i++) { + if (optvals[i]) + av_log(h, AV_LOG_WARNING, + "'%s' option was set but it is not supported " + "on this build (pthread support is required)\n", optnames[i]); + } + } /* handling needed to support options picking from both AVOption and URL */ s->circular_buffer_size *= 188; if (flags & AVIO_FLAG_WRITE) { commit 3e7314fe30ecf356b71bf6e40223f82e4e8c138e Author: Marton Balint <cus(a)passwd.hu> AuthorDate: Sun Aug 24 23:28:36 2025 +0200 Commit: Marton Balint <cus(a)passwd.hu> CommitDate: Fri Sep 19 09:59:26 2025 +0200 avformat/udp: add DSCP as a normal AVOption Previously this was an URL-only option. Signed-off-by: Marton Balint <cus(a)passwd.hu> diff --git a/doc/protocols.texi b/doc/protocols.texi index 6b582fde30..4b871dddb7 100644 --- a/doc/protocols.texi +++ b/doc/protocols.texi @@ -2162,6 +2162,9 @@ Explicitly allow or disallow reusing UDP sockets. @item ttl=@var{ttl} Set the time to live value (for multicast only). +@item dscp=@var{dscp} +Set the 6-bit DSCP field for outgoing packets. + @item connect=@var{1|0} Initialize the UDP socket with @code{connect()}. In this case, the destination address can't be changed with ff_udp_set_remote_url later. diff --git a/libavformat/udp.c b/libavformat/udp.c index 84f9d3e62e..ded7a1a85e 100644 --- a/libavformat/udp.c +++ b/libavformat/udp.c @@ -118,6 +118,7 @@ typedef struct UDPContext { int remaining_in_dg; char *localaddr; int timeout; + int dscp; struct sockaddr_storage local_addr_storage; char *sources; char *block; @@ -142,6 +143,7 @@ static const AVOption options[] = { { "reuse_socket", "explicitly allow reusing UDP sockets", OFFSET(reuse_socket), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, .flags = D|E }, { "broadcast", "explicitly allow or disallow broadcast destination", OFFSET(is_broadcast), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, E }, { "ttl", "Time to live (multicast only)", OFFSET(ttl), AV_OPT_TYPE_INT, { .i64 = 16 }, 0, 255, E }, + { "dscp", "DSCP class for outgoing packets", OFFSET(dscp), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 63, E }, { "connect", "set if connect() should be called on socket", OFFSET(is_connected), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, .flags = D|E }, { "fifo_size", "set the UDP receiving circular buffer size, expressed as a number of packets with size of 188 bytes", OFFSET(circular_buffer_size), AV_OPT_TYPE_INT, {.i64 = 7*4096}, 0, INT_MAX, D }, { "overrun_nonfatal", "survive in case of UDP receiving circular buffer overrun", OFFSET(overrun_nonfatal), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, D }, @@ -691,7 +693,7 @@ end: static int udp_open(URLContext *h, const char *uri, int flags) { char hostname[1024]; - int port, udp_fd = -1, tmp, bind_ret = -1, dscp = -1; + int port, udp_fd = -1, tmp, bind_ret = -1; UDPContext *s = h->priv_data; int is_output; const char *p; @@ -760,7 +762,7 @@ static int udp_open(URLContext *h, const char *uri, int flags) s->is_connected = strtol(buf, NULL, 10); } if (av_find_info_tag(buf, sizeof(buf), "dscp", p)) { - dscp = strtol(buf, NULL, 10); + s->dscp = strtol(buf, NULL, 10); } if (av_find_info_tag(buf, sizeof(buf), "fifo_size", p)) { s->circular_buffer_size = strtol(buf, NULL, 10); @@ -870,8 +872,8 @@ static int udp_open(URLContext *h, const char *uri, int flags) av_log(h, AV_LOG_WARNING, "socket option UDPLITE_RECV_CSCOV not available"); } - if (dscp >= 0) { - dscp <<= 2; + if (s->dscp >= 0) { + int dscp = s->dscp << 2; if (setsockopt (udp_fd, IPPROTO_IP, IP_TOS, &dscp, sizeof(dscp)) != 0) { ret = ff_neterrno(); goto fail; ----------------------------------------------------------------------- Summary of changes: doc/protocols.texi | 21 ++++++++++----------- libavformat/rtpproto.c | 30 ++++++++++++++++++++++-------- libavformat/tls.h | 10 +++++++--- libavformat/tls_gnutls.c | 1 - libavformat/tls_libtls.c | 1 - libavformat/tls_mbedtls.c | 1 - libavformat/tls_openssl.c | 1 - libavformat/tls_schannel.c | 1 - libavformat/tls_securetransport.c | 1 - libavformat/udp.c | 34 +++++++++++++++++----------------- libavformat/urldecode.c | 35 ++++++++++++++++++++++------------- 11 files changed, 78 insertions(+), 58 deletions(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. 1ce3f9fdab forgejo/workflows: run fate-build
by ffmpeg-git＠ffmpeg.org 18 Sep '25

18 Sep '25

The branch, master has been updated via 1ce3f9fdabf0c769e86b585d9da197e869e1a44e (commit) from c74181a04b5f4e650eae662231e56518daef64d4 (commit) - Log ----------------------------------------------------------------- commit 1ce3f9fdabf0c769e86b585d9da197e869e1a44e Author: Kacper Michajłow <kasper93(a)gmail.com> AuthorDate: Sat Aug 9 03:48:00 2025 +0200 Commit: Kacper Michajłow <kasper93(a)gmail.com> CommitDate: Thu Sep 18 16:53:05 2025 +0000 forgejo/workflows: run fate-build It's not included in make fate target. Signed-off-by: Kacper Michajłow <kasper93(a)gmail.com> diff --git a/.forgejo/workflows/test.yml b/.forgejo/workflows/test.yml index 6b5e483b39..79c5238a8b 100644 --- a/.forgejo/workflows/test.yml +++ b/.forgejo/workflows/test.yml @@ -47,7 +47,7 @@ jobs: path: fate-suite key: fate-suite-${{ steps.fate.outputs.hash }} - name: Run Fate - run: make fate SAMPLES=$PWD/fate-suite -j$(nproc) + run: make fate fate-build SAMPLES=$PWD/fate-suite -j$(nproc) compile_only: strategy: fail-fast: false @@ -66,3 +66,5 @@ jobs: --extra-libs="$FF_LIBS" --extra-ldflags="$FF_LDFLAGS" --extra-ldexeflags="$FF_LDEXEFLAGS" - name: Build run: make -j$(nproc) + - name: Run Fate + run: make -j$(nproc) fate-build ----------------------------------------------------------------------- Summary of changes: .forgejo/workflows/test.yml | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. c74181a04b avformat/tls_gnutls: add av_assert0() for tls_shared
by ffmpeg-git＠ffmpeg.org 18 Sep '25

18 Sep '25

The branch, master has been updated via c74181a04b5f4e650eae662231e56518daef64d4 (commit) via df7532b21d9cafd2b3527128eab252e81a81f21a (commit) from 647138334abd6ea001a16a768eb12cc4156db5f9 (commit) - Log ----------------------------------------------------------------- commit c74181a04b5f4e650eae662231e56518daef64d4 Author: Jack Lau <jacklau1222(a)qq.com> AuthorDate: Mon Sep 8 16:31:10 2025 +0800 Commit: stevenliu <lingjiujianke(a)gmail.com> CommitDate: Thu Sep 18 12:34:15 2025 +0000 avformat/tls_gnutls: add av_assert0() for tls_shared Signed-off-by: Jack Lau <jacklau1222(a)qq.com> diff --git a/libavformat/tls_gnutls.c b/libavformat/tls_gnutls.c index 272603f1dd..53306872a0 100644 --- a/libavformat/tls_gnutls.c +++ b/libavformat/tls_gnutls.c @@ -30,6 +30,7 @@ #include "os_support.h" #include "url.h" #include "tls.h" +#include "libavutil/avassert.h" #include "libavutil/opt.h" #include "libavutil/thread.h" @@ -156,6 +157,7 @@ static int tls_open(URLContext *h, const char *uri, int flags, AVDictionary **op TLSShared *s = &c->tls_shared; uint16_t gnutls_flags = 0; int ret; + av_assert0(s); ff_gnutls_init(); @@ -261,6 +263,7 @@ static int dtls_open(URLContext *h, const char *uri, int flags, AVDictionary **o { TLSContext *c = h->priv_data; TLSShared *s = &c->tls_shared; + av_assert0(s); s->is_dtls = 1; return tls_open(h, uri, flags, options); } commit df7532b21d9cafd2b3527128eab252e81a81f21a Author: Jack Lau <jacklau1222(a)qq.com> AuthorDate: Mon Sep 8 16:08:22 2025 +0800 Commit: stevenliu <lingjiujianke(a)gmail.com> CommitDate: Thu Sep 18 12:34:15 2025 +0000 avformat/tls_gnutls: add initial dtls support Set GNUTLS_DATAGRAM flag when is_dtls is true. Set mtu when it's specified. Modify the read/write function could use udp socket. There are more patches to make dtls really work. Signed-off-by: Jack Lau <jacklau1222(a)qq.com> diff --git a/libavformat/tls_gnutls.c b/libavformat/tls_gnutls.c index e894765d7c..272603f1dd 100644 --- a/libavformat/tls_gnutls.c +++ b/libavformat/tls_gnutls.c @@ -22,6 +22,7 @@ #include <errno.h> #include <gnutls/gnutls.h> +#include <gnutls/dtls.h> #include <gnutls/x509.h> #include "avformat.h" @@ -100,13 +101,15 @@ static int print_tls_error(URLContext *h, int ret) static int tls_close(URLContext *h) { TLSContext *c = h->priv_data; + TLSShared *s = &c->tls_shared; if (c->need_shutdown) gnutls_bye(c->session, GNUTLS_SHUT_WR); if (c->session) gnutls_deinit(c->session); if (c->cred) gnutls_certificate_free_credentials(c->cred); - ffurl_closep(&c->tls_shared.tcp); + if (!s->external_sock) + ffurl_closep(s->is_dtls ? &s->udp : &s->tcp); ff_gnutls_deinit(); return 0; } @@ -151,6 +154,7 @@ static int tls_open(URLContext *h, const char *uri, int flags, AVDictionary **op { TLSContext *c = h->priv_data; TLSShared *s = &c->tls_shared; + uint16_t gnutls_flags = 0; int ret; ff_gnutls_init(); @@ -158,7 +162,14 @@ static int tls_open(URLContext *h, const char *uri, int flags, AVDictionary **op if ((ret = ff_tls_open_underlying(s, h, uri, options)) < 0) goto fail; - gnutls_init(&c->session, s->listen ? GNUTLS_SERVER : GNUTLS_CLIENT); + if (s->is_dtls) + gnutls_flags |= GNUTLS_DATAGRAM; + + if (s->listen) + gnutls_flags |= GNUTLS_SERVER; + else + gnutls_flags |= GNUTLS_CLIENT; + gnutls_init(&c->session, gnutls_flags); if (!s->listen && !s->numerichost) gnutls_server_name_set(c->session, GNUTLS_NAME_DNS, s->host, strlen(s->host)); gnutls_certificate_allocate_credentials(&c->cred); @@ -190,6 +201,9 @@ static int tls_open(URLContext *h, const char *uri, int flags, AVDictionary **op gnutls_transport_set_pull_function(c->session, gnutls_url_pull); gnutls_transport_set_push_function(c->session, gnutls_url_push); gnutls_transport_set_ptr(c->session, c); + if (s->is_dtls) + if (s->mtu) + gnutls_dtls_set_mtu(c->session, s->mtu); gnutls_set_default_priority(c->session); do { if (ff_check_interrupt(&h->interrupt_callback)) { @@ -243,13 +257,23 @@ fail: return ret; } +static int dtls_open(URLContext *h, const char *uri, int flags, AVDictionary **options) +{ + TLSContext *c = h->priv_data; + TLSShared *s = &c->tls_shared; + s->is_dtls = 1; + return tls_open(h, uri, flags, options); +} + static int tls_read(URLContext *h, uint8_t *buf, int size) { TLSContext *c = h->priv_data; + TLSShared *s = &c->tls_shared; + URLContext *uc = s->is_dtls ? s->udp : s->tcp; int ret; // Set or clear the AVIO_FLAG_NONBLOCK on c->tls_shared.tcp - c->tls_shared.tcp->flags &= ~AVIO_FLAG_NONBLOCK; - c->tls_shared.tcp->flags |= h->flags & AVIO_FLAG_NONBLOCK; + uc->flags &= ~AVIO_FLAG_NONBLOCK; + uc->flags |= h->flags & AVIO_FLAG_NONBLOCK; ret = gnutls_record_recv(c->session, buf, size); if (ret > 0) return ret; @@ -261,10 +285,12 @@ static int tls_read(URLContext *h, uint8_t *buf, int size) static int tls_write(URLContext *h, const uint8_t *buf, int size) { TLSContext *c = h->priv_data; + TLSShared *s = &c->tls_shared; + URLContext *uc = s->is_dtls ? s->udp : s->tcp; int ret; // Set or clear the AVIO_FLAG_NONBLOCK on c->tls_shared.tcp - c->tls_shared.tcp->flags &= ~AVIO_FLAG_NONBLOCK; - c->tls_shared.tcp->flags |= h->flags & AVIO_FLAG_NONBLOCK; + uc->flags &= ~AVIO_FLAG_NONBLOCK; + uc->flags |= h->flags & AVIO_FLAG_NONBLOCK; ret = gnutls_record_send(c->session, buf, size); if (ret > 0) return ret; @@ -309,3 +335,23 @@ const URLProtocol ff_tls_protocol = { .flags = URL_PROTOCOL_FLAG_NETWORK, .priv_data_class = &tls_class, }; + +static const AVClass dtls_class = { + .class_name = "dtls", + .item_name = av_default_item_name, + .option = options, + .version = LIBAVUTIL_VERSION_INT, +}; + +const URLProtocol ff_dtls_protocol = { + .name = "dtls", + .url_open2 = dtls_open, + .url_read = tls_read, + .url_write = tls_write, + .url_close = tls_close, + .url_get_file_handle = tls_get_file_handle, + .url_get_short_seek = tls_get_short_seek, + .priv_data_size = sizeof(TLSContext), + .flags = URL_PROTOCOL_FLAG_NETWORK, + .priv_data_class = &dtls_class, +}; ----------------------------------------------------------------------- Summary of changes: libavformat/tls_gnutls.c | 61 +++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 55 insertions(+), 6 deletions(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. 647138334a avcodec/exif: check count in exif_decode_tag()
by ffmpeg-git＠ffmpeg.org 18 Sep '25

18 Sep '25

The branch, master has been updated via 647138334abd6ea001a16a768eb12cc4156db5f9 (commit) via 88e04205b39ebc2fb6d5c78733e490aa5f5e69bb (commit) via 742b0d4675a977c0cf67c306df95b4ef9aff7e36 (commit) from c373636f554579340c7b1f46c42cc3a6989a7da6 (commit) - Log ----------------------------------------------------------------- commit 647138334abd6ea001a16a768eb12cc4156db5f9 Author: Michael Niedermayer <michael(a)niedermayer.cc> AuthorDate: Thu Sep 18 02:25:32 2025 +0200 Commit: Leo Izen <leo.izen(a)gmail.com> CommitDate: Thu Sep 18 11:06:16 2025 +0000 avcodec/exif: check count in exif_decode_tag() Fixes: out of array access Fixes: integer overflow Fixes: poc_heap_bof Found-by: *2ourc3 (Salim LARGO) Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> diff --git a/libavcodec/exif.c b/libavcodec/exif.c index a953252da0..e7aa9b8d8f 100644 --- a/libavcodec/exif.c +++ b/libavcodec/exif.c @@ -477,7 +477,7 @@ static int exif_decode_tag(void *logctx, GetByteContext *gb, int le, "payload: %" PRIu32 "\n", entry->id, type, count, tell, payload); /* AV_TIFF_IFD is the largest, numerically */ - if (type > AV_TIFF_IFD) + if (type > AV_TIFF_IFD || count >= INT_MAX/8U) return AVERROR_INVALIDDATA; is_ifd = type == AV_TIFF_IFD || ff_tis_ifd(entry->id) || entry->id == MAKERNOTE_TAG; commit 88e04205b39ebc2fb6d5c78733e490aa5f5e69bb Author: Michael Niedermayer <michael(a)niedermayer.cc> AuthorDate: Sat Sep 13 13:53:53 2025 +0200 Commit: Leo Izen <leo.izen(a)gmail.com> CommitDate: Thu Sep 18 11:06:16 2025 +0000 avcodec/exif: Do not leave uninitialized pointers on errors in exif_clone_entry() No testcase, but this looks like it could free garbage pointers Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> diff --git a/libavcodec/exif.c b/libavcodec/exif.c index dbbd5acd1c..a953252da0 100644 --- a/libavcodec/exif.c +++ b/libavcodec/exif.c @@ -950,6 +950,8 @@ static int exif_clone_entry(AVExifEntry *dst, const AVExifEntry *src) { int ret = 0; + memset(dst, 0, sizeof(*dst)); + dst->count = src->count; dst->id = src->id; dst->type = src->type; commit 742b0d4675a977c0cf67c306df95b4ef9aff7e36 Author: Michael Niedermayer <michael(a)niedermayer.cc> AuthorDate: Sat Sep 13 14:00:55 2025 +0200 Commit: Leo Izen <leo.izen(a)gmail.com> CommitDate: Thu Sep 18 11:06:16 2025 +0000 avcodec/exif: Use av_fast_mallocz() in av_exif_clone_ifd() using fast realloc leaves the entries uninitialized and frees garbage pointers on errors Fixes: bug_triggering_file Found-by: *2ourc3, 5pider Signed-off-by: Michael Niedermayer <michael(a)niedermayer.cc> diff --git a/libavcodec/exif.c b/libavcodec/exif.c index f9ad3e1bdb..dbbd5acd1c 100644 --- a/libavcodec/exif.c +++ b/libavcodec/exif.c @@ -1151,7 +1151,7 @@ AVExifMetadata *av_exif_clone_ifd(const AVExifMetadata *ifd) size_t required_size; if (av_size_mult(ret->count, sizeof(*ret->entries), &required_size) < 0) goto fail; - ret->entries = av_fast_realloc(NULL, &ret->size, required_size); + av_fast_mallocz(&ret->entries, &ret->size, required_size); if (!ret->entries) goto fail; } ----------------------------------------------------------------------- Summary of changes: libavcodec/exif.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. c373636f55 avfilter/vsrc_gfxcapture: fix string format specifier
by ffmpeg-git＠ffmpeg.org 18 Sep '25

18 Sep '25

The branch, master has been updated via c373636f554579340c7b1f46c42cc3a6989a7da6 (commit) from 61835e1d8be9ace611e82b64f136a50f191a6e25 (commit) - Log ----------------------------------------------------------------- commit c373636f554579340c7b1f46c42cc3a6989a7da6 Author: Kacper Michajłow <kasper93(a)gmail.com> AuthorDate: Thu Sep 18 03:54:51 2025 +0200 Commit: Kacper Michajłow <kasper93(a)gmail.com> CommitDate: Thu Sep 18 03:55:54 2025 +0200 avfilter/vsrc_gfxcapture: fix string format specifier Signed-off-by: Kacper Michajłow <kasper93(a)gmail.com> diff --git a/libavfilter/vsrc_gfxcapture_winrt.cpp b/libavfilter/vsrc_gfxcapture_winrt.cpp index 1738fa23af..9cc38342b7 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.cpp +++ b/libavfilter/vsrc_gfxcapture_winrt.cpp @@ -603,7 +603,7 @@ static void wgc_thread_entry(AVFilterContext *avctx) noexcept { static const wchar_t name_prefix[] = L"wgc_winrt@0x"; wchar_t thread_name[FF_ARRAY_ELEMS(name_prefix) + sizeof(void*) * 2] = { 0 }; - swprintf(thread_name, FF_ARRAY_ELEMS(thread_name), L"%s%" PRIxPTR, name_prefix, (uintptr_t)avctx); + swprintf(thread_name, FF_ARRAY_ELEMS(thread_name), L"%ls%" PRIxPTR, name_prefix, (uintptr_t)avctx); ctx->fn.SetThreadDescription(GetCurrentThread(), thread_name); std::lock_guard init_lock(ctx->wgc_thread_init_mutex); ----------------------------------------------------------------------- Summary of changes: libavfilter/vsrc_gfxcapture_winrt.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) hooks/post-receive --

1 0

[ffmpeg] branch master updated. 61835e1d8b avfilter/vsrc_gfxcapture: keep cbdata object alive
by ffmpeg-git＠ffmpeg.org 18 Sep '25

18 Sep '25

The branch, master has been updated via 61835e1d8be9ace611e82b64f136a50f191a6e25 (commit) via c4e537793b5939262cabb708ff798a732b8aba33 (commit) via 1849068c657c23e979e804cf90e63ae2f7381ec5 (commit) via 4879c9f4e2f2afc269d305d941cf70d18f3883a8 (commit) via 23d9412ff9598b8989072e53d5474a41d2f14042 (commit) via 983fed34f8172228c2e5490882531668d9d35105 (commit) from 8accbbdf9a99a1999e7e0c19f6cc91085b3b1905 (commit) - Log ----------------------------------------------------------------- commit 61835e1d8be9ace611e82b64f136a50f191a6e25 Author: Timo Rothenpieler <timo(a)rothenpieler.org> AuthorDate: Wed Sep 17 22:03:28 2025 +0200 Commit: Timo Rothenpieler <timo(a)rothenpieler.org> CommitDate: Thu Sep 18 01:04:59 2025 +0000 avfilter/vsrc_gfxcapture: keep cbdata object alive Depending on the threading backend the stdlib uses, creating a mutex/condvar can be quite expensive. So keep this object alive in the ctx, on which we synchronize via the uninit mutex anyway. diff --git a/libavfilter/vsrc_gfxcapture_winrt.cpp b/libavfilter/vsrc_gfxcapture_winrt.cpp index 62b1a94802..1738fa23af 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.cpp +++ b/libavfilter/vsrc_gfxcapture_winrt.cpp @@ -149,6 +149,7 @@ struct GfxCaptureContextCpp { volatile int wgc_thread_init_res { INT_MAX }; std::recursive_mutex wgc_thread_uninit_mutex; volatile int wgc_thread_res { 0 }; + std::shared_ptr<void> wgc_thread_cb_data; HWND capture_hwnd { nullptr }; HMONITOR capture_hmonitor { nullptr }; @@ -716,11 +717,17 @@ static int run_on_wgc_thread(AVFilterContext *avctx, F &&cb) struct CBData { std::mutex mutex; std::condition_variable cond; - bool done { false }; - bool cancel { false }; - int ret { AVERROR_BUG }; + bool done; + bool cancel; + int ret; }; - auto cbdata = std::make_shared<CBData>(); + auto cbdata = ctx->wgc_thread_cb_data ? + std::static_pointer_cast<CBData>(ctx->wgc_thread_cb_data) : + std::make_shared<CBData>(); + ctx->wgc_thread_cb_data = cbdata; + + cbdata->done = cbdata->cancel = false; + cbdata->ret = AVERROR_BUG; boolean res = 0; CHECK_HR_RET(wgctx->dispatcher_queue->TryEnqueue( commit c4e537793b5939262cabb708ff798a732b8aba33 Author: Timo Rothenpieler <timo(a)rothenpieler.org> AuthorDate: Wed Sep 17 21:31:56 2025 +0200 Commit: Timo Rothenpieler <timo(a)rothenpieler.org> CommitDate: Thu Sep 18 01:04:59 2025 +0000 avfilter/vsrc_gfxcapture: fix leaking all callback handlers Fixes #20537 diff --git a/libavfilter/vsrc_gfxcapture_winrt.hpp b/libavfilter/vsrc_gfxcapture_winrt.hpp index a02e768c8f..ba1979b1c1 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.hpp +++ b/libavfilter/vsrc_gfxcapture_winrt.hpp @@ -159,9 +159,9 @@ private: template<class Iface, typename... Args, typename F> static Microsoft::WRL::ComPtr<Iface> create_cb_handler(F&& cb_func) { - return Microsoft::WRL::ComPtr<Iface>( - new FFTypedCBHandler<Iface, F, Args...>(std::forward<F>(cb_func)) - ); + Microsoft::WRL::ComPtr<Iface> res; + res.Attach(new FFTypedCBHandler<Iface, F, Args...>(std::forward<F>(cb_func))); + return res; } /****************************************** commit 1849068c657c23e979e804cf90e63ae2f7381ec5 Author: Timo Rothenpieler <timo(a)rothenpieler.org> AuthorDate: Wed Sep 17 20:23:37 2025 +0200 Commit: Timo Rothenpieler <timo(a)rothenpieler.org> CommitDate: Thu Sep 18 01:04:59 2025 +0000 avfilter/vsrc_gfxcapture: use free threaded capture frame pool Apparently, using a normal frame pool in a multithreaded environment leads to strange resource leaks on shutdown, which vanish when using a free threaded pool. diff --git a/libavfilter/vsrc_gfxcapture_winrt.cpp b/libavfilter/vsrc_gfxcapture_winrt.cpp index 47c3b2db88..62b1a94802 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.cpp +++ b/libavfilter/vsrc_gfxcapture_winrt.cpp @@ -327,7 +327,7 @@ static int wgc_setup_gfxcapture_session(AVFilterContext *avctx) std::unique_ptr<GfxCaptureContextWgc> &wgctx = ctx->wgc; int ret; - ComPtr<IDirect3D11CaptureFramePoolStatics> frame_pool_statics; + ComPtr<IDirect3D11CaptureFramePoolStatics2> frame_pool_statics; ComPtr<ID3D11Device> d3d11_device = ctx->device_hwctx->device; ComPtr<ID3D10Multithread> d3d10_multithread; ComPtr<IDXGIDevice> dxgi_device; @@ -350,8 +350,8 @@ static int wgc_setup_gfxcapture_session(AVFilterContext *avctx) CHECK_HR_RET(d3d11_device.As(&dxgi_device)); CHECK_HR_RET(ctx->fn.CreateDirect3D11DeviceFromDXGIDevice(dxgi_device.Get(), &wgctx->d3d_device)); - CHECK_HR_RET(get_activation_factory<IDirect3D11CaptureFramePoolStatics>(ctx, RuntimeClass_Windows_Graphics_Capture_Direct3D11CaptureFramePool, &frame_pool_statics)); - CHECK_HR_RET(frame_pool_statics->Create(wgctx->d3d_device.Get(), fmt, CAPTURE_POOL_SIZE, wgctx->cap_size, &wgctx->frame_pool)); + CHECK_HR_RET(get_activation_factory<IDirect3D11CaptureFramePoolStatics2>(ctx, RuntimeClass_Windows_Graphics_Capture_Direct3D11CaptureFramePool, &frame_pool_statics)); + CHECK_HR_RET(frame_pool_statics->CreateFreeThreaded(wgctx->d3d_device.Get(), fmt, CAPTURE_POOL_SIZE, wgctx->cap_size, &wgctx->frame_pool)); CHECK_HR_RET(wgctx->frame_pool->CreateCaptureSession(wgctx->capture_item.Get(), &wgctx->capture_session)); if (SUCCEEDED(wgctx->capture_session.As(&session2))) { commit 4879c9f4e2f2afc269d305d941cf70d18f3883a8 Author: Timo Rothenpieler <timo(a)rothenpieler.org> AuthorDate: Wed Sep 17 19:13:29 2025 +0200 Commit: Timo Rothenpieler <timo(a)rothenpieler.org> CommitDate: Thu Sep 18 01:04:59 2025 +0000 avfilter/vsrc_gfxcapture: stop capture session before initializing capture thread shutdown It might have things going on in the background that still need cleaned up, and this gives it a chance to enqueue them on the dispatch queue. diff --git a/libavfilter/vsrc_gfxcapture_winrt.cpp b/libavfilter/vsrc_gfxcapture_winrt.cpp index ade14d297f..47c3b2db88 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.cpp +++ b/libavfilter/vsrc_gfxcapture_winrt.cpp @@ -229,7 +229,7 @@ static void wgc_stop_capture_session(AVFilterContext *avctx) noexcept if (wgctx->capture_session) { ComPtr<IClosable> closable; if (SUCCEEDED(wgctx->capture_session.As(&closable))) { - closable->Close(); + CHECK_HR_LOG(closable->Close()); } else { av_log(avctx, AV_LOG_ERROR, "Failed to get capture session IClosable interface\n"); } @@ -243,6 +243,11 @@ static void wgc_stop_capture_session(AVFilterContext *avctx) noexcept av_log(avctx, AV_LOG_ERROR, "Failed to get frame pool IClosable interface\n"); } } + + wgctx->capture_session.Reset(); + wgctx->frame_pool.Reset(); + wgctx->capture_item.Reset(); + wgctx->d3d_device.Reset(); } static int wgc_calculate_client_area(AVFilterContext *avctx) @@ -557,6 +562,9 @@ static int wgc_thread_worker(AVFilterContext *avctx) if (!msg.hwnd && msg.message == WM_WGC_THREAD_SHUTDOWN) { av_log(avctx, AV_LOG_DEBUG, "Initializing WGC thread shutdown\n"); + + wgc_stop_capture_session(avctx); + if (FAILED(wgctx->dispatcher_queue_controller->ShutdownQueueAsync(&async))) { av_log(avctx, AV_LOG_ERROR, "Failed to shutdown dispatcher queue\n"); return AVERROR_EXTERNAL; commit 23d9412ff9598b8989072e53d5474a41d2f14042 Author: Timo Rothenpieler <timo(a)rothenpieler.org> AuthorDate: Wed Sep 17 19:12:53 2025 +0200 Commit: Timo Rothenpieler <timo(a)rothenpieler.org> CommitDate: Thu Sep 18 01:04:59 2025 +0000 avfilter/vsrc_gfxcapture: fix re-using ret variable from outside of lambda scope diff --git a/libavfilter/vsrc_gfxcapture_winrt.cpp b/libavfilter/vsrc_gfxcapture_winrt.cpp index b037dc89aa..ade14d297f 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.cpp +++ b/libavfilter/vsrc_gfxcapture_winrt.cpp @@ -1394,9 +1394,9 @@ static int process_frame_if_exists(AVFilterLink *outlink) ComPtr<ID3D11Texture2D> frame_texture; TimeSpan frame_time = { 0 }; - ret = wgc_try_get_next_frame(avctx, capture_frame); - if (ret < 0) - return ret; + int res = wgc_try_get_next_frame(avctx, capture_frame); + if (res < 0) + return res; CHECK_HR_RET(capture_frame->get_SystemRelativeTime(&frame_time)); commit 983fed34f8172228c2e5490882531668d9d35105 Author: Timo Rothenpieler <timo(a)rothenpieler.org> AuthorDate: Wed Sep 17 18:31:11 2025 +0200 Commit: Timo Rothenpieler <timo(a)rothenpieler.org> CommitDate: Thu Sep 18 01:04:59 2025 +0000 avfilter/vsrc_gfxcapture: don't pass pointer to ComPtr While it does appear to work fine, with all the operator overloads, it at least has potential for surprises, so pass it by reference instead. diff --git a/libavfilter/vsrc_gfxcapture_winrt.cpp b/libavfilter/vsrc_gfxcapture_winrt.cpp index 9562863d7f..b037dc89aa 100644 --- a/libavfilter/vsrc_gfxcapture_winrt.cpp +++ b/libavfilter/vsrc_gfxcapture_winrt.cpp @@ -436,7 +436,7 @@ static int wgc_setup_gfxcapture_capture(AVFilterContext *avctx) return 0; } -static int wgc_try_get_next_frame(AVFilterContext *avctx, ComPtr<IDirect3D11CaptureFrame> *capture_frame) +static int wgc_try_get_next_frame(AVFilterContext *avctx, ComPtr<IDirect3D11CaptureFrame> &capture_frame) { GfxCaptureContext *cctx = CCTX(avctx->priv); GfxCaptureContextCpp *ctx = cctx->ctx; @@ -447,11 +447,11 @@ static int wgc_try_get_next_frame(AVFilterContext *avctx, ComPtr<IDirect3D11Capt ComPtr<ID3D11Texture2D> frame_texture; SizeInt32 frame_size = { 0, 0 }; - CHECK_HR_RET(wgctx->frame_pool->TryGetNextFrame(capture_frame->ReleaseAndGetAddressOf())); - if (!capture_frame->Get()) + CHECK_HR_RET(wgctx->frame_pool->TryGetNextFrame(&capture_frame)); + if (!capture_frame) return AVERROR(EAGAIN); - CHECK_HR_RET(capture_frame->Get()->get_ContentSize(&frame_size)); + CHECK_HR_RET(capture_frame->get_ContentSize(&frame_size)); if (frame_size.Width != wgctx->cap_size.Width || frame_size.Height != wgctx->cap_size.Height) { av_log(avctx, AV_LOG_VERBOSE, "Capture size changed to %dx%d\n", frame_size.Width, frame_size.Height); @@ -1394,7 +1394,7 @@ static int process_frame_if_exists(AVFilterLink *outlink) ComPtr<ID3D11Texture2D> frame_texture; TimeSpan frame_time = { 0 }; - ret = wgc_try_get_next_frame(avctx, &capture_frame); + ret = wgc_try_get_next_frame(avctx, capture_frame); if (ret < 0) return ret; ----------------------------------------------------------------------- Summary of changes: libavfilter/vsrc_gfxcapture_winrt.cpp | 45 +++++++++++++++++++++++------------ libavfilter/vsrc_gfxcapture_winrt.hpp | 6 ++--- 2 files changed, 33 insertions(+), 18 deletions(-) hooks/post-receive --

1 0