[FFmpeg-devel,2/2] avutil/swscale: add RGBF16 pix_fmt and input support

Message ID	20241013131607.860688-4-ms+git@mur.at
State	New
Headers	show Delivered-To: ffmpegpatchwork2@gmail.com Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; From: Martin Schitter <ms+git@mur.at> To: ffmpeg-devel@ffmpeg.org Date: Sun, 13 Oct 2024 15:16:08 +0200 Message-ID: <20241013131607.860688-4-ms+git@mur.at> In-Reply-To: <20241013131607.860688-2-ms+git@mur.at> References: <20241013131607.860688-2-ms+git@mur.at> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH 2/2] avutil/swscale: add RGBF16 pix_fmt and input support Precedence: list Reply-To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org> Cc: Martin Schitter <ms+git@mur.at> Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" <ffmpeg-devel-bounces@ffmpeg.org>
Series	RGB Float16/32 improvements \| expand [FFmpeg-devel,0/2] RGB Float16/32 improvements [FFmpeg-devel,1/2] swscale/input: add input support for RGBF32 [FFmpeg-devel,2/2] avutil/swscale: add RGBF16 pix_fmt and input support

Context	Check	Description
yinshiyou/make_loongarch64	success	Make finished
yinshiyou/make_fate_loongarch64	fail	Make fate failed

diff --git a/libavutil/pixdesc.c b/libavutil/pixdesc.c index 43b9c08..e73600f 100644 --- a/libavutil/pixdesc.c +++ b/libavutil/pixdesc.c @@ -2630,6 +2630,31 @@ static const AVPixFmtDescriptor av_pix_fmt_descriptors[AV_PIX_FMT_NB] = { { 0, 4, 0, 0, 8 }, /* V */ }, }, + [AV_PIX_FMT_RGBF16BE] = { + .name = "rgbf16be", + .nb_components = 3, + .log2_chroma_w = 0, + .log2_chroma_h = 0, + .comp = { + { 0, 6, 0, 0, 16 }, /* R */ + { 0, 6, 2, 0, 16 }, /* G */ + { 0, 6, 4, 0, 16 }, /* B */ + }, + .flags = AV_PIX_FMT_FLAG_BE | AV_PIX_FMT_FLAG_RGB | + AV_PIX_FMT_FLAG_FLOAT, + }, + [AV_PIX_FMT_RGBF16LE] = { + .name = "rgbf16le", + .nb_components = 3, + .log2_chroma_w = 0, + .log2_chroma_h = 0, + .comp = { + { 0, 6, 0, 0, 16 }, /* R */ + { 0, 6, 2, 0, 16 }, /* G */ + { 0, 6, 4, 0, 16 }, /* B */ + }, + .flags = AV_PIX_FMT_FLAG_RGB | AV_PIX_FMT_FLAG_FLOAT + }, [AV_PIX_FMT_RGBAF16BE] = { .name = "rgbaf16be", .nb_components = 4, diff --git a/libavutil/pixfmt.h b/libavutil/pixfmt.h index 3caa183..4efc4cb 100644 --- a/libavutil/pixfmt.h +++ b/libavutil/pixfmt.h @@ -400,6 +400,9 @@ enum AVPixelFormat { AV_PIX_FMT_VUYA, ///< packed VUYA 4:4:4:4, 32bpp (1 Cr & Cb sample per 1x1 Y & A samples), VUYAVUYA... + AV_PIX_FMT_RGBF16BE, ///< IEEE-754 half precision packed RGB 16:16:16, 48bpp, RGBRGB..., big-endian + AV_PIX_FMT_RGBF16LE, ///< IEEE-754 half precision packed RGB 16:16:16, 48bpp, RGBRGB..., little-endian + AV_PIX_FMT_RGBAF16BE, ///< IEEE-754 half precision packed RGBA 16:16:16:16, 64bpp, RGBARGBA..., big-endian AV_PIX_FMT_RGBAF16LE, ///< IEEE-754 half precision packed RGBA 16:16:16:16, 64bpp, RGBARGBA..., little-endian @@ -553,6 +556,7 @@ enum AVPixelFormat { #define AV_PIX_FMT_P216 AV_PIX_FMT_NE(P216BE, P216LE) #define AV_PIX_FMT_P416 AV_PIX_FMT_NE(P416BE, P416LE) +#define AV_PIX_FMT_RGBF16 AV_PIX_FMT_NE(RGBF16BE, RGBF16LE) #define AV_PIX_FMT_RGBAF16 AV_PIX_FMT_NE(RGBAF16BE, RGBAF16LE) #define AV_PIX_FMT_RGBF32 AV_PIX_FMT_NE(RGBF32BE, RGBF32LE) diff --git a/libavutil/version.h b/libavutil/version.h index 7eca46c..772addf 100644 --- a/libavutil/version.h +++ b/libavutil/version.h @@ -79,7 +79,7 @@ */ #define LIBAVUTIL_VERSION_MAJOR 59 -#define LIBAVUTIL_VERSION_MINOR 42 +#define LIBAVUTIL_VERSION_MINOR 43 #define LIBAVUTIL_VERSION_MICRO 100 #define LIBAVUTIL_VERSION_INT AV_VERSION_INT(LIBAVUTIL_VERSION_MAJOR, \ diff --git a/libswscale/input.c b/libswscale/input.c index e2af1d5..9beb72b 100644 --- a/libswscale/input.c +++ b/libswscale/input.c @@ -1369,6 +1369,57 @@ static av_always_inline void rgbaf16ToA_endian(uint16_t *dst, const uint16_t *sr } } +static av_always_inline void rgbf16ToUV_half_endian(uint16_t *dstU, uint16_t *dstV, int is_be, + const uint16_t *src, int width, + int32_t *rgb2yuv, Half2FloatTables *h2f_tbl) +{ + int32_t ru = rgb2yuv[RU_IDX], gu = rgb2yuv[GU_IDX], bu = rgb2yuv[BU_IDX]; + int32_t rv = rgb2yuv[RV_IDX], gv = rgb2yuv[GV_IDX], bv = rgb2yuv[BV_IDX]; + int i; + for (i = 0; i < width; i++) { + int r = (lrintf(av_clipf(65535.0f * rdpx(src[i*6+0]), 0.0f, 65535.0f)) + + lrintf(av_clipf(65535.0f * rdpx(src[i*6+3]), 0.0f, 65535.0f))) >> 1; + int g = (lrintf(av_clipf(65535.0f * rdpx(src[i*6+1]), 0.0f, 65535.0f)) + + lrintf(av_clipf(65535.0f * rdpx(src[i*6+4]), 0.0f, 65535.0f))) >> 1; + int b = (lrintf(av_clipf(65535.0f * rdpx(src[i*6+2]), 0.0f, 65535.0f)) + + lrintf(av_clipf(65535.0f * rdpx(src[i*6+5]), 0.0f, 65535.0f))) >> 1; + + dstU[i] = (ru*r + gu*g + bu*b + (0x10001<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; + dstV[i] = (rv*r + gv*g + bv*b + (0x10001<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; + } +} + +static av_always_inline void rgbf16ToUV_endian(uint16_t *dstU, uint16_t *dstV, int is_be, + const uint16_t *src, int width, + int32_t *rgb2yuv, Half2FloatTables *h2f_tbl) +{ + int32_t ru = rgb2yuv[RU_IDX], gu = rgb2yuv[GU_IDX], bu = rgb2yuv[BU_IDX]; + int32_t rv = rgb2yuv[RV_IDX], gv = rgb2yuv[GV_IDX], bv = rgb2yuv[BV_IDX]; + int i; + for (i = 0; i < width; i++) { + int r = lrintf(av_clipf(65535.0f * rdpx(src[i*3+0]), 0.0f, 65535.0f)); + int g = lrintf(av_clipf(65535.0f * rdpx(src[i*3+1]), 0.0f, 65535.0f)); + int b = lrintf(av_clipf(65535.0f * rdpx(src[i*3+2]), 0.0f, 65535.0f)); + + dstU[i] = (ru*r + gu*g + bu*b + (0x10001<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; + dstV[i] = (rv*r + gv*g + bv*b + (0x10001<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; + } +} + +static av_always_inline void rgbf16ToY_endian(uint16_t *dst, const uint16_t *src, int is_be, + int width, int32_t *rgb2yuv, Half2FloatTables *h2f_tbl) +{ + int32_t ry = rgb2yuv[RY_IDX], gy = rgb2yuv[GY_IDX], by = rgb2yuv[BY_IDX]; + int i; + for (i = 0; i < width; i++) { + int r = lrintf(av_clipf(65535.0f * rdpx(src[i*3+0]), 0.0f, 65535.0f)); + int g = lrintf(av_clipf(65535.0f * rdpx(src[i*3+1]), 0.0f, 65535.0f)); + int b = lrintf(av_clipf(65535.0f * rdpx(src[i*3+2]), 0.0f, 65535.0f)); + + dst[i] = (ry*r + gy*g + by*b + (0x2001<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT; + } +} + #undef rdpx #define rgbaf16_funcs_endian(endian_name, endian) \ @@ -1408,7 +1459,37 @@ static void rgbaf16##endian_name##ToA_c(uint8_t *_dst, const uint8_t *_src, cons const uint16_t *src = (const uint16_t*)_src; \ uint16_t *dst = (uint16_t*)_dst; \ rgbaf16ToA_endian(dst, src, endian, width, opq); \ -} +} \ +static void rgbf16##endian_name##ToUV_half_c(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *unused, \ + const uint8_t *src1, const uint8_t *src2, \ + int width, uint32_t *_rgb2yuv, void *opq) \ +{ \ + const uint16_t *src = (const uint16_t*)src1; \ + uint16_t *dstU = (uint16_t*)_dstU; \ + uint16_t *dstV = (uint16_t*)_dstV; \ + int32_t *rgb2yuv = (int32_t*)_rgb2yuv; \ + av_assert1(src1==src2); \ + rgbf16ToUV_half_endian(dstU, dstV, endian, src, width, rgb2yuv, opq); \ +} \ +static void rgbf16##endian_name##ToUV_c(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *unused, \ + const uint8_t *src1, const uint8_t *src2, \ + int width, uint32_t *_rgb2yuv, void *opq) \ +{ \ + const uint16_t *src = (const uint16_t*)src1; \ + uint16_t *dstU = (uint16_t*)_dstU; \ + uint16_t *dstV = (uint16_t*)_dstV; \ + int32_t *rgb2yuv = (int32_t*)_rgb2yuv; \ + av_assert1(src1==src2); \ + rgbf16ToUV_endian(dstU, dstV, endian, src, width, rgb2yuv, opq); \ +} \ +static void rgbf16##endian_name##ToY_c(uint8_t *_dst, const uint8_t *_src, const uint8_t *unused0, \ + const uint8_t *unused1, int width, uint32_t *_rgb2yuv, void *opq) \ +{ \ + const uint16_t *src = (const uint16_t*)_src; \ + uint16_t *dst = (uint16_t*)_dst; \ + int32_t *rgb2yuv = (int32_t*)_rgb2yuv; \ + rgbf16ToY_endian(dst, src, endian, width, rgb2yuv, opq); \ +} \ rgbaf16_funcs_endian(le, 0) rgbaf16_funcs_endian(be, 1) @@ -1729,6 +1810,12 @@ av_cold void ff_sws_init_input_funcs(SwsContext *c, case AV_PIX_FMT_RGBAF16LE: *chrToYV12 = rgbaf16leToUV_half_c; break; + case AV_PIX_FMT_RGBF16BE: + *chrToYV12 = rgbf16beToUV_half_c; + break; + case AV_PIX_FMT_RGBF16LE: + *chrToYV12 = rgbf16leToUV_half_c; + break; } } else { switch (srcFormat) { @@ -1822,6 +1909,12 @@ av_cold void ff_sws_init_input_funcs(SwsContext *c, case AV_PIX_FMT_RGBAF16LE: *chrToYV12 = rgbaf16leToUV_c; break; + case AV_PIX_FMT_RGBF16BE: + *chrToYV12 = rgbf16beToUV_c; + break; + case AV_PIX_FMT_RGBF16LE: + *chrToYV12 = rgbf16leToUV_c; + break; } } @@ -2150,6 +2243,12 @@ av_cold void ff_sws_init_input_funcs(SwsContext *c, case AV_PIX_FMT_RGBAF16LE: *lumToYV12 = rgbaf16leToY_c; break; + case AV_PIX_FMT_RGBF16BE: + *lumToYV12 = rgbf16beToY_c; + break; + case AV_PIX_FMT_RGBF16LE: + *lumToYV12 = rgbf16leToY_c; + break; case AV_PIX_FMT_RGBF32LE: *lumToYV12 = rgbf32le_to_y_c; break; diff --git a/libswscale/utils.c b/libswscale/utils.c index 3ed65a1..992d7f0 100644 --- a/libswscale/utils.c +++ b/libswscale/utils.c @@ -266,6 +266,8 @@ static const FormatEntry format_entries[] = { [AV_PIX_FMT_VUYX] = { 1, 1 }, [AV_PIX_FMT_RGBAF16BE] = { 1, 0 }, [AV_PIX_FMT_RGBAF16LE] = { 1, 0 }, + [AV_PIX_FMT_RGBF16BE] = { 1, 0 }, + [AV_PIX_FMT_RGBF16LE] = { 1, 0 }, [AV_PIX_FMT_RGBF32BE] = { 1, 0 }, [AV_PIX_FMT_RGBF32LE] = { 1, 0 }, [AV_PIX_FMT_XV30LE] = { 1, 1 }, diff --git a/libswscale/version.h b/libswscale/version.h index c13db31..f573bef 100644 --- a/libswscale/version.h +++ b/libswscale/version.h @@ -28,7 +28,7 @@ #include "version_major.h" -#define LIBSWSCALE_VERSION_MINOR 5 +#define LIBSWSCALE_VERSION_MINOR 6 #define LIBSWSCALE_VERSION_MICRO 100 #define LIBSWSCALE_VERSION_INT AV_VERSION_INT(LIBSWSCALE_VERSION_MAJOR, \

[FFmpeg-devel,2/2] avutil/swscale: add RGBF16 pix_fmt and input support

Checks

Commit Message

Patch