Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 version
> -Original Message- > From: ffmpeg-devel On Behalf Of Fu, > Ting > Sent: Monday, February 10, 2020 05:54 PM > To: FFmpeg development discussions and patches de...@ffmpeg.org> > Subject: Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add > ssse3 version > > > > > -Original Message- > > From: ffmpeg-devel On Behalf Of > Paul > > B Mahol > > Sent: Monday, February 10, 2020 04:56 PM > > To: FFmpeg development discussions and patches > de...@ffmpeg.org> > > Subject: Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add > > ssse3 version > > > > Does this pass fate? > > If yes i will apply. > > Hi Paul, > > It has passed the fate. > > Thank you, > Ting Fu Hi Paul, After a further check, I found there may be no fate for yuv2rgb. I just run 'make fate' with no error yesterday, sorry for that. And I have checked the md5 of the output of ssse3 and mmx already (100 frames with and without apply my patch). They are the same. Here are examples with 5 frames below. I think this can also prove that the ssse3 outputs the same with mmx. Command: ./ffmpeg -pix_fmt yuv420p -s 1920*1080 -i ArashRawYuv420.yuv -vcodec rawvideo -s 1920*1080 -pix_fmt FMT -vframes 5 -f framemd5 FMT.md5 (FMT = {RGB24, BGR24, RGB32, BGR32, RGB555, RGB565}) YUV420P to RGB24 mmx: 2c891b51088497776cd85ab8bdf7d9c0 fa8df95a630096a96ee06048b446f7cd 21048f651cfa5e5309b7838d119070bd 3149361521068d00862ec9f88224c92d 7790cd3d7a966546a6c245738dc7430b ssse3: 2c891b51088497776cd85ab8bdf7d9c0 fa8df95a630096a96ee06048b446f7cd 21048f651cfa5e5309b7838d119070bd 3149361521068d00862ec9f88224c92d 7790cd3d7a966546a6c245738dc7430b YUV420P to BGR24 mmx: 46f19629d8f34fe4dba3df99c49668fa 000e251995a0269dd0dce77c31711e98 edccc926fbee775b00e2a5da4a7365e1 d46e286699c0361fa21832c33c662120 a79321d438111dae575158aefca4c87f ssse3: 46f19629d8f34fe4dba3df99c49668fa 000e251995a0269dd0dce77c31711e98 edccc926fbee775b00e2a5da4a7365e1 d46e286699c0361fa21832c33c662120 a79321d438111dae575158aefca4c87f YUV420P to RGB32 mmx: 51e99fab05a84fb08c62ddde3898648e 4c078b41498d1fde73a8b636ba3c c60448161ed3662c79445e24d5e6b834 4a4ae471a2e5448833fbbf41a7de5c42 0200b2a18f1212c5adb5d6f59b513503 ssse3: 51e99fab05a84fb08c62ddde3898648e 4c078b41498d1fde73a8b636ba3c c60448161ed3662c79445e24d5e6b834 4a4ae471a2e5448833fbbf41a7de5c42 0200b2a18f1212c5adb5d6f59b513503 YUV420P to BGR32 mmx: a494abb5a4fa9f51a228d13d2452cc87 0fb30a71f70eb977580968a96cefca43 77feb78a9355601d2d36dfe15a6a3993 b64b7717e3baaa012d98a019f64b55d8 1abca5233e66e0dca63c8476c08e830f ssse3: a494abb5a4fa9f51a228d13d2452cc87 0fb30a71f70eb977580968a96cefca43 77feb78a9355601d2d36dfe15a6a3993 b64b7717e3baaa012d98a019f64b55d8 1abca5233e66e0dca63c8476c08e830f YUV420P to BGR555 mmx: 1874dea6f6f7afb872d748530c3e7116 af1f98567c1d7c3fa9705d08b7873231 2f4e21ab5dd6499c57db0932bbdc2498 4eccd6dcf69dba8e1e5365cad2cdcf84 0c20aaf1602522a012b1543f0cef7169 ssse3: 1874dea6f6f7afb872d748530c3e7116 af1f98567c1d7c3fa9705d08b7873231 2f4e21ab5dd6499c57db0932bbdc2498 4eccd6dcf69dba8e1e5365cad2cdcf84 0c20aaf1602522a012b1543f0cef7169 YUV420P to BGR565 mmx: 6cac48711756152b3ca815b6985b2882 93264a7683ff0761a8ef300590674e40 fc0242e2f7dacbcd9a5f9ac7229a8d9f dbc5acb000fa5a9d468ab8db03431c39 af7a512c00c1b177d4f65e6d056ea97c ssse3: 6cac48711756152b3ca815b6985b2882 93264a7683ff0761a8ef300590674e40 fc0242e2f7dacbcd9a5f9ac7229a8d9f dbc5acb000fa5a9d468ab8db03431c39 af7a512c00c1b177d4f65e6d056ea97c Command: ./ffmpeg -pix_fmt yuva420p -s 1920*1080 -i Allegro_HEVC_Main444_MT41_Funny_00_1920x1080_r2260.yuva420.yuv -vcodec rawvideo -s 1920*1080 -pix_fmt FMT -vframes 5 -f framemd5 FMT.md5 (FMT = {RGB32, BGR32}) YUVA420P to RGB32 mmx: 602cfb10a7eadebd7f85c8937ab43509 210ed4c10d021c754b2ab9a434b0f0c2 88eef228f50264c6340ae5490412a4aa 8fcda91158c4c26309da6f76cf79c4a7 6cd6d747ce91726b33c649df8b04f15c ssse3: 602cfb10a7eadebd7f85c8937ab43509
Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 version
> -Original Message- > From: ffmpeg-devel On Behalf Of Paul > B Mahol > Sent: Monday, February 10, 2020 04:56 PM > To: FFmpeg development discussions and patches de...@ffmpeg.org> > Subject: Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add > ssse3 version > > Does this pass fate? > If yes i will apply. Hi Paul, It has passed the fate. Thank you, Ting Fu > > On 2/10/20, Fu, Ting wrote: > > > > > >> -Original Message- > >> From: ffmpeg-devel On Behalf Of > >> Ting Fu > >> Sent: Sunday, January 19, 2020 11:51 AM > >> To: ffmpeg-devel@ffmpeg.org > >> Subject: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add > >> ssse3 version > >> > >> Tested using this command: > >> /ffmpeg -pix_fmt yuv420p -s 1920*1080 -i ArashRawYuv420.yuv \ -vcodec > >> rawvideo -s 1920*1080 -pix_fmt rgb24 -f null /dev/null > >> > >> The fps increase from 389 to 640 on Intel(R) Core(TM) i7-8700K CPU @ > >> 3.70GHz > >> > >> Signed-off-by: Ting Fu > >> --- > >> libswscale/x86/yuv2rgb.c | 38 + > >> libswscale/x86/yuv_2_rgb.asm | 145 > +++-- > >> -- > >> > > [...] > >> + > >> +INIT_XMM ssse3 > >> +yuv2rgb_fn yuv, rgb, 24 > >> +yuv2rgb_fn yuv, bgr, 24 > >> +yuv2rgb_fn yuv, rgb, 32 > >> +yuv2rgb_fn yuv, bgr, 32 > >> +yuv2rgb_fn yuva, rgb, 32 > >> +yuv2rgb_fn yuva, bgr, 32 > >> +yuv2rgb_fn yuv, rgb, 15 > >> +yuv2rgb_fn yuv, rgb, 16 > >> -- > >> 2.17.1 > > > > Hi all, > > > > Any further comment for this single patch? > > This version has only changed a little compared with V4 (reviewed by > > Paul), which is that the check of SIMD has been moved from the wrapper > > function (in > > libswscale/x86/yuv2rgb_template.c) to format check (in > > libswscale/x86/yuv2rgb.c). > > While assembly code is exactly same with the V4. > > > > Thank you, > > Ting Fu. > > > >> > >> ___ > >> ffmpeg-devel mailing list > >> ffmpeg-devel@ffmpeg.org > >> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > >> > >> To unsubscribe, visit link above, or email > >> ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". > > ___ > > ffmpeg-devel mailing list > > ffmpeg-devel@ffmpeg.org > > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > > > To unsubscribe, visit link above, or email > > ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". > ___ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org > with subject "unsubscribe". ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 version
Does this pass fate? If yes i will apply. On 2/10/20, Fu, Ting wrote: > > >> -Original Message- >> From: ffmpeg-devel On Behalf Of Ting >> Fu >> Sent: Sunday, January 19, 2020 11:51 AM >> To: ffmpeg-devel@ffmpeg.org >> Subject: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 >> version >> >> Tested using this command: >> /ffmpeg -pix_fmt yuv420p -s 1920*1080 -i ArashRawYuv420.yuv \ -vcodec >> rawvideo -s 1920*1080 -pix_fmt rgb24 -f null /dev/null >> >> The fps increase from 389 to 640 on Intel(R) Core(TM) i7-8700K CPU @ >> 3.70GHz >> >> Signed-off-by: Ting Fu >> --- >> libswscale/x86/yuv2rgb.c | 38 + >> libswscale/x86/yuv_2_rgb.asm | 145 +++-- >> -- >> > [...] >> + >> +INIT_XMM ssse3 >> +yuv2rgb_fn yuv, rgb, 24 >> +yuv2rgb_fn yuv, bgr, 24 >> +yuv2rgb_fn yuv, rgb, 32 >> +yuv2rgb_fn yuv, bgr, 32 >> +yuv2rgb_fn yuva, rgb, 32 >> +yuv2rgb_fn yuva, bgr, 32 >> +yuv2rgb_fn yuv, rgb, 15 >> +yuv2rgb_fn yuv, rgb, 16 >> -- >> 2.17.1 > > Hi all, > > Any further comment for this single patch? > This version has only changed a little compared with V4 (reviewed by Paul), > which is that the check of SIMD has been moved from the wrapper function (in > libswscale/x86/yuv2rgb_template.c) to format check (in > libswscale/x86/yuv2rgb.c). > While assembly code is exactly same with the V4. > > Thank you, > Ting Fu. > >> >> ___ >> ffmpeg-devel mailing list >> ffmpeg-devel@ffmpeg.org >> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel >> >> To unsubscribe, visit link above, or email >> ffmpeg-devel-requ...@ffmpeg.org >> with subject "unsubscribe". > ___ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe". ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
Re: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 version
> -Original Message- > From: ffmpeg-devel On Behalf Of Ting > Fu > Sent: Sunday, January 19, 2020 11:51 AM > To: ffmpeg-devel@ffmpeg.org > Subject: [FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 > version > > Tested using this command: > /ffmpeg -pix_fmt yuv420p -s 1920*1080 -i ArashRawYuv420.yuv \ -vcodec > rawvideo -s 1920*1080 -pix_fmt rgb24 -f null /dev/null > > The fps increase from 389 to 640 on Intel(R) Core(TM) i7-8700K CPU @ > 3.70GHz > > Signed-off-by: Ting Fu > --- > libswscale/x86/yuv2rgb.c | 38 + > libswscale/x86/yuv_2_rgb.asm | 145 +++-- > -- > [...] > + > +INIT_XMM ssse3 > +yuv2rgb_fn yuv, rgb, 24 > +yuv2rgb_fn yuv, bgr, 24 > +yuv2rgb_fn yuv, rgb, 32 > +yuv2rgb_fn yuv, bgr, 32 > +yuv2rgb_fn yuva, rgb, 32 > +yuv2rgb_fn yuva, bgr, 32 > +yuv2rgb_fn yuv, rgb, 15 > +yuv2rgb_fn yuv, rgb, 16 > -- > 2.17.1 Hi all, Any further comment for this single patch? This version has only changed a little compared with V4 (reviewed by Paul), which is that the check of SIMD has been moved from the wrapper function (in libswscale/x86/yuv2rgb_template.c) to format check (in libswscale/x86/yuv2rgb.c). While assembly code is exactly same with the V4. Thank you, Ting Fu. > > ___ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org > with subject "unsubscribe". ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
[FFmpeg-devel] [PATCH V8 2/2] libswscale/x86/yuv2rgb: add ssse3 version
Tested using this command: /ffmpeg -pix_fmt yuv420p -s 1920*1080 -i ArashRawYuv420.yuv \ -vcodec rawvideo -s 1920*1080 -pix_fmt rgb24 -f null /dev/null The fps increase from 389 to 640 on Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz Signed-off-by: Ting Fu --- libswscale/x86/yuv2rgb.c | 38 + libswscale/x86/yuv_2_rgb.asm | 145 +++ 2 files changed, 167 insertions(+), 16 deletions(-) diff --git a/libswscale/x86/yuv2rgb.c b/libswscale/x86/yuv2rgb.c index c7668f487c..c12e88cbb5 100644 --- a/libswscale/x86/yuv2rgb.c +++ b/libswscale/x86/yuv2rgb.c @@ -67,6 +67,15 @@ DECLARE_ASM_CONST(8, uint64_t, pb_07) = 0x0707070707070707ULL; #include "yuv2rgb_template.c" #endif /* HAVE_MMXEXT */ +//SSSE3 versions +#if HAVE_SSSE3 +#undef RENAME +#undef COMPILE_TEMPLATE_MMXEXT +#define COMPILE_TEMPLATE_MMXEXT 0 +#define RENAME(a) a ## _ssse3 +#include "yuv2rgb_template.c" +#endif + #endif /* HAVE_X86ASM */ av_cold SwsFunc ff_yuv2rgb_init_x86(SwsContext *c) @@ -74,6 +83,35 @@ av_cold SwsFunc ff_yuv2rgb_init_x86(SwsContext *c) #if HAVE_X86ASM int cpu_flags = av_get_cpu_flags(); +if (EXTERNAL_SSSE3(cpu_flags)) { +switch (c->dstFormat) { +case AV_PIX_FMT_RGB32: +if (c->srcFormat == AV_PIX_FMT_YUVA420P) { +#if CONFIG_SWSCALE_ALPHA +return yuva420_rgb32_ssse3; +#endif +break; +} else +return yuv420_rgb32_ssse3; +case AV_PIX_FMT_BGR32: +if (c->srcFormat == AV_PIX_FMT_YUVA420P) { +#if CONFIG_SWSCALE_ALPHA +return yuva420_bgr32_ssse3; +#endif +break; +} else +return yuv420_bgr32_ssse3; +case AV_PIX_FMT_RGB24: +return yuv420_rgb24_ssse3; +case AV_PIX_FMT_BGR24: +return yuv420_bgr24_ssse3; +case AV_PIX_FMT_RGB565: +return yuv420_rgb16_ssse3; +case AV_PIX_FMT_RGB555: +return yuv420_rgb15_ssse3; +} +} + if (EXTERNAL_MMXEXT(cpu_flags)) { switch (c->dstFormat) { case AV_PIX_FMT_RGB24: diff --git a/libswscale/x86/yuv_2_rgb.asm b/libswscale/x86/yuv_2_rgb.asm index a44ab1607b..e05bbb89f5 100644 --- a/libswscale/x86/yuv_2_rgb.asm +++ b/libswscale/x86/yuv_2_rgb.asm @@ -25,11 +25,18 @@ SECTION_RODATA -pw_00ff: times 4 dw 255 -pb_f8: times 8 db 248 -pb_e0: times 8 db 224 -pb_03: times 8 db 3 -pb_07: times 8 db 7 +; below variables are named like mask_dwXY, which means to preserve dword No.X & No.Y +mask_dw036 : db -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0 +mask_dw147 : db 0, 0, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, -1, -1 +mask_dw25 : db 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0 +rgb24_shuf1: db 0, 1, 6, 7, 12, 13, 2, 3, 8, 9, 14, 15, 4, 5, 10, 11 +rgb24_shuf2: db 10, 11, 0, 1, 6, 7, 12, 13, 2, 3, 8, 9, 14, 15, 4, 5 +rgb24_shuf3: db 4, 5, 10, 11, 0, 1, 6, 7, 12, 13, 2, 3, 8, 9, 14, 15 +pw_00ff: times 8 dw 255 +pb_f8: times 16 db 248 +pb_e0: times 16 db 224 +pb_03: times 16 db 3 +pb_07: times 16 db 7 mask_1101: dw -1, -1, 0, -1 mask_0010: dw 0, 0, -1, 0 @@ -49,7 +56,11 @@ SECTION .text ;- %macro MOV_H2L 1 -psrlq %1, 32 +%if mmsize == 8 +psrlq %1, 32 +%else ; mmsize == 16 +psrldq %1, 8 +%endif %endmacro %macro yuv2rgb_fn 3 @@ -77,6 +88,7 @@ psrlq %1, 32 %define m_blue m1 %endif +%if mmsize == 8 %define time_num 1 %define reg_num 8 %define y_offset [pointer_c_ditherq + 8 * 8] @@ -87,11 +99,45 @@ psrlq %1, 32 %define y_coff [pointer_c_ditherq + 3 * 8] %define ub_coff [pointer_c_ditherq + 5 * 8] %define vr_coff [pointer_c_ditherq + 4 * 8] +%elif mmsize == 16 +%define time_num 2 +%if ARCH_X86_32 +%define reg_num 8 +%define my_offset [pointer_c_ditherq + 8 * 8] +%define mu_offset [pointer_c_ditherq + 9 * 8] +%define mv_offset [pointer_c_ditherq + 10 * 8] +%define mug_coff [pointer_c_ditherq + 7 * 8] +%define mvg_coff [pointer_c_ditherq + 6 * 8] +%define my_coff [pointer_c_ditherq + 3 * 8] +%define mub_coff [pointer_c_ditherq + 5 * 8] +%define mvr_coff [pointer_c_ditherq + 4 * 8] +%else ; ARCH_X86_64 +%define reg_num 16 +%define y_offset m8 +%define u_offset m9 +%define v_offset m10 +%define ug_coff m11 +%define vg_coff m12 +%define y_coff m13 +%define ub_coff m14 +%define vr_coff m15 +%endif ; ARCH_X86_32/64 +%endif ; coeff define mmsize == 8/16 cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters %if ARCH_X86_64 movsxd indexq, indexd +%if mmsize == 16 +VBROADCASTSD y_offset, [pointer_c_ditherq + 8 * 8] +VBROADCASTSD u_offset, [pointer_c_ditherq + 9 * 8] +VBROADCASTSD v_offset, [pointer_c_ditherq + 10 * 8] +VBROADCASTSD ug_coff, [pointer_c_ditherq + 7 * 8] +VBROADCASTSD vg_coff, [pointer_c_ditherq + 6 * 8] +VBROADCASTSD y_coff,