From: Lynne <dev@lynne.ee> To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org> Subject: [FFmpeg-devel] [PATCH 2/2] lavu/tx: add a half-complex R2C transform Date: Tue, 25 Jul 2023 11:43:46 +0200 (CEST) Message-ID: <NaBQigo--N-9@lynne.ee> (raw) In-Reply-To: <NaBQCTh--3-9@lynne.ee-NaBQFrM----9> [-- Attachment #1: Type: text/plain, Size: 48 bytes --] Required for DCT-I and DST-I. Patch attached. [-- Attachment #2: 0002-lavu-tx-add-a-half-complex-R2C-transform.patch --] [-- Type: text/x-diff, Size: 7571 bytes --] From 2e95c3a7128b7f826db8394eb238c5405c47864b Mon Sep 17 00:00:00 2001 From: Lynne <dev@lynne.ee> Date: Tue, 25 Jul 2023 10:57:21 +0200 Subject: [PATCH 2/2] lavu/tx: add a half-complex R2C transform Required for DCT-I and DST-I. --- doc/APIchanges | 3 ++ libavutil/tx.c | 13 +++++++-- libavutil/tx.h | 9 ++++++ libavutil/tx_template.c | 65 +++++++++++++++++++++++++++++++++++++++-- libavutil/version.h | 2 +- 5 files changed, 86 insertions(+), 6 deletions(-) diff --git a/doc/APIchanges b/doc/APIchanges index 5afe8bcb75..0797dcc372 100644 --- a/doc/APIchanges +++ b/doc/APIchanges @@ -2,6 +2,9 @@ The last version increases of all libraries were on 2023-02-09 API changes, most recent first: +2023-07-xx - xxxxxxxxxx - lavu 58.15.100 - tx.h + Add AV_TX_HALF_COMPLEX + 2023-07-xx - xxxxxxxxxx - lavc 60 - avcodec.h Deprecate AV_CODEC_FLAG_DROPCHANGED without replacement. diff --git a/libavutil/tx.c b/libavutil/tx.c index e25abf998f..d6ab8ee485 100644 --- a/libavutil/tx.c +++ b/libavutil/tx.c @@ -437,7 +437,8 @@ int ff_tx_decompose_length(int dst[TX_MAX_DECOMPOSITIONS], enum AVTXType type, /* Check direction for non-orthogonal codelets */ if (((cd->flags & FF_TX_FORWARD_ONLY) && inv) || - ((cd->flags & (FF_TX_INVERSE_ONLY | AV_TX_FULL_IMDCT)) && !inv)) + ((cd->flags & (FF_TX_INVERSE_ONLY | AV_TX_FULL_IMDCT)) && !inv) || + ((cd->flags & (FF_TX_FORWARD_ONLY | AV_TX_HALF_COMPLEX)) && inv)) continue; /* Check if the CPU supports the required ISA */ @@ -560,6 +561,8 @@ static void print_flags(AVBPrint *bp, uint64_t f) av_bprintf(bp, "%spreshuf", prev > 1 ? sep : ""); if ((f & AV_TX_FULL_IMDCT) && ++prev) av_bprintf(bp, "%simdct_full", prev > 1 ? sep : ""); + if ((f & AV_TX_HALF_COMPLEX) && ++prev) + av_bprintf(bp, "%shalf_complex", prev > 1 ? sep : ""); if ((f & FF_TX_ASM_CALL) && ++prev) av_bprintf(bp, "%sasm_call", prev > 1 ? sep : ""); av_bprintf(bp, "]"); @@ -717,7 +720,10 @@ av_cold int ff_tx_init_subtx(AVTXContext *s, enum AVTXType type, uint64_t req_flags = flags; /* Flags the codelet may require to be present */ - uint64_t inv_req_mask = AV_TX_FULL_IMDCT | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL; + uint64_t inv_req_mask = AV_TX_FULL_IMDCT | + AV_TX_HALF_COMPLEX | + FF_TX_PRESHUFFLE | + FF_TX_ASM_CALL; /* Unaligned codelets are compatible with the aligned flag */ if (req_flags & FF_TX_ALIGNED) @@ -742,7 +748,8 @@ av_cold int ff_tx_init_subtx(AVTXContext *s, enum AVTXType type, /* Check direction for non-orthogonal codelets */ if (((cd->flags & FF_TX_FORWARD_ONLY) && inv) || - ((cd->flags & (FF_TX_INVERSE_ONLY | AV_TX_FULL_IMDCT)) && !inv)) + ((cd->flags & (FF_TX_INVERSE_ONLY | AV_TX_FULL_IMDCT)) && !inv) || + ((cd->flags & (FF_TX_FORWARD_ONLY | AV_TX_HALF_COMPLEX)) && inv)) continue; /* Check if the requested flags match from both sides */ diff --git a/libavutil/tx.h b/libavutil/tx.h index 064edbc097..63fba8e8b4 100644 --- a/libavutil/tx.h +++ b/libavutil/tx.h @@ -149,6 +149,15 @@ enum AVTXFlags { * Ignored for all transforms but inverse MDCTs. */ AV_TX_FULL_IMDCT = 1ULL << 2, + + /** + * Perform a real to half-complex RDFT or FFT. + * Only the real coefficients will + * be output. Only available for forward RDFTs. + * Output array must have enough space to hold N complex values + * (regular size for a real to complex transform). + */ + AV_TX_HALF_COMPLEX = 1ULL << 3, }; /** diff --git a/libavutil/tx_template.c b/libavutil/tx_template.c index a4040454db..bdec53f335 100644 --- a/libavutil/tx_template.c +++ b/libavutil/tx_template.c @@ -1617,6 +1617,8 @@ static av_cold int TX_NAME(ff_tx_rdft_init)(AVTXContext *s, s->scale_d = *((SCALE_TYPE *)scale); s->scale_f = s->scale_d; + flags &= ~AV_TX_HALF_COMPLEX; + if ((ret = ff_tx_init_subtx(s, TX_TYPE(FFT), flags, NULL, len >> 1, inv, scale))) return ret; @@ -1700,8 +1702,51 @@ static void TX_NAME(ff_tx_rdft_ ##name)(AVTXContext *s, void *_dst, \ } \ } -DECL_RDFT(r2c, 0) -DECL_RDFT(c2r, 1) +DECL_RDFT(r2c, 0) +DECL_RDFT(c2r, 1) + +static void TX_NAME(ff_tx_rdft_r2hc)(AVTXContext *s, void *_dst, + void *_src, ptrdiff_t stride) +{ + const int len = s->len; + const int len2 = len >> 1; + const int len4 = len >> 2; + const TXSample *fact = (void *)s->exp; + const TXSample *tcos = fact + 8; + const TXSample *tsin = tcos + len4; + TXComplex *data = _dst; + TXSample *out = _dst; /* Half-complex is forward-only */ + TXSample tmp_dc; + + s->fn[0](&s->sub[0], _dst, _src, sizeof(TXComplex)); + + tmp_dc = data[0].re; + + /* NOTE: we can avoid all multiplies here as they're all 1.0, except + * for one, -1.0, which is not needed */ + out[0] = tmp_dc + data[0].im; + tmp_dc -= data[0].im; + + /* NOTE: unrolling this breaks non-mod8 lengths */ + for (int i = 1; i <= len4; i++) { + TXSample tmp[4]; + TXComplex sf = data[i]; + TXComplex sl = data[len2 - i]; + + tmp[0] = MULT(fact[4], (sf.re + sl.re)); + tmp[1] = MULT(fact[6], (sf.im + sl.im)); + tmp[2] = MULT(fact[7], (sf.re - sl.re)); + tmp[3] = tmp[1]*tcos[i] - tmp[2]*tsin[i]; + + out[i] = tmp[0] + tmp[3]; + out[len - i] = tmp[0] - tmp[3]; + } + + for (int i = 1; i < len4; i++) + out[len2 - i - 0] = out[len - i]; + + out[len2] = tmp_dc; +} static const FFTXCodelet TX_NAME(ff_tx_rdft_r2c_def) = { .name = TX_NAME_STR("rdft_r2c"), @@ -1733,6 +1778,21 @@ static const FFTXCodelet TX_NAME(ff_tx_rdft_c2r_def) = { .prio = FF_TX_PRIO_BASE, }; +static const FFTXCodelet TX_NAME(ff_tx_rdft_r2hc_def) = { + .name = TX_NAME_STR("rdft_r2hc"), + .function = TX_NAME(ff_tx_rdft_r2hc), + .type = TX_TYPE(RDFT), + .flags = AV_TX_UNALIGNED | AV_TX_HALF_COMPLEX | AV_TX_INPLACE | + FF_TX_OUT_OF_PLACE | FF_TX_FORWARD_ONLY, + .factors[0] = 2, + .nb_factors = 1, + .min_len = 2, + .max_len = TX_LEN_UNLIMITED, + .init = TX_NAME(ff_tx_rdft_init), + .cpu_flags = FF_TX_CPU_FLAGS_ALL, + .prio = FF_TX_PRIO_BASE, +}; + static av_cold int TX_NAME(ff_tx_dct_init)(AVTXContext *s, const FFTXCodelet *cd, uint64_t flags, @@ -1997,6 +2057,7 @@ const FFTXCodelet * const TX_NAME(ff_tx_codelet_list)[] = { &TX_NAME(ff_tx_mdct_naive_inv_def), &TX_NAME(ff_tx_mdct_inv_full_def), &TX_NAME(ff_tx_rdft_r2c_def), + &TX_NAME(ff_tx_rdft_r2hc_def), &TX_NAME(ff_tx_rdft_c2r_def), &TX_NAME(ff_tx_dctII_def), &TX_NAME(ff_tx_dctIII_def), diff --git a/libavutil/version.h b/libavutil/version.h index 24af520e08..9e798b0e3f 100644 --- a/libavutil/version.h +++ b/libavutil/version.h @@ -79,7 +79,7 @@ */ #define LIBAVUTIL_VERSION_MAJOR 58 -#define LIBAVUTIL_VERSION_MINOR 14 +#define LIBAVUTIL_VERSION_MINOR 15 #define LIBAVUTIL_VERSION_MICRO 100 #define LIBAVUTIL_VERSION_INT AV_VERSION_INT(LIBAVUTIL_VERSION_MAJOR, \ -- 2.40.1 [-- Attachment #3: Type: text/plain, Size: 251 bytes --] _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
next prev parent reply other threads:[~2023-07-25 9:43 UTC|newest] Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top 2023-07-25 9:41 [FFmpeg-devel] [PATCH 1/2] lavu/tx: disable odd-length RDFTs Lynne [not found] ` <NaBQCTh--3-9@lynne.ee-NaBQFrM----9> 2023-07-25 9:43 ` Lynne [this message] 2023-07-27 9:59 ` Lynne
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=NaBQigo--N-9@lynne.ee \ --to=dev@lynne.ee \ --cc=ffmpeg-devel@ffmpeg.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel This inbox may be cloned and mirrored by anyone: git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git # If you have public-inbox 1.1+ installed, you may # initialize and index your mirror using the following commands: public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \ ffmpegdev@gitmailbox.com public-inbox-index ffmpegdev Example config snippet for mirrors. AGPL code for this site: git clone https://public-inbox.org/public-inbox.git