[FFmpeg-devel] [PATCH] lavc/vc1dsp: R-V V inv_trans

flow gg hlefthleft at gmail.com
Tue Dec 5 21:25:12 EET 2023


> This block can be folded into the next. You don't need to check VLENB
twice.

Changed.

> Instruction scheduling could be better, especially on in-order CPUs.

I put the vload at the front, and then proceeded with the t2 operation, but
I'm not sure...

> You don't need to reset the AVL here, just pass zero.

Changed.

> vsetivli

Changed.

Rémi Denis-Courmont <remi at remlab.net> 于2023年12月6日周三 00:41写道:

>         Hi,
>
> > diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
> > index 2d0e6c19c8..442c5961ea 100644
> > --- a/libavcodec/riscv/Makefile
> > +++ b/libavcodec/riscv/Makefile
> > @@ -39,5 +39,7 @@ OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_init.o
> \
> >  RVV-OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_rvv.o
> >  OBJS-$(CONFIG_UTVIDEO_DECODER) += riscv/utvideodsp_init.o
> >  RVV-OBJS-$(CONFIG_UTVIDEO_DECODER) += riscv/utvideodsp_rvv.o
> > +OBJS-$(CONFIG_VC1DSP) += riscv/vc1dsp_init.o
> > +RVV-OBJS-$(CONFIG_VC1DSP) += riscv/vc1dsp_rvv.o
> >  OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_init.o
> >  RVV-OBJS-$(CONFIG_VORBIS_DECODER) += riscv/vorbisdsp_rvv.o
> > diff --git a/libavcodec/riscv/vc1dsp_init.c
> b/libavcodec/riscv/vc1dsp_init.c
> > new file mode 100644
> > index 0000000000..6a04bea6d6
> > --- /dev/null
> > +++ b/libavcodec/riscv/vc1dsp_init.c
> > @@ -0,0 +1,47 @@
> > +/*
> > + * Copyright (c) 2023 Institue of Software Chinese Academy of Sciences
> > (ISCAS).
> > + *
> > + * 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 <stdint.h>
> > +
> > +#include "libavutil/attributes.h"
> > +#include "libavutil/cpu.h"
> > +#include "libavutil/riscv/cpu.h"
> > +#include "libavcodec/vc1.h"
> > +
> > +void ff_vc1_inv_trans_8x8_dc_rvv(uint8_t *dest, ptrdiff_t stride,
> int16_t
> > *block);
> > +void ff_vc1_inv_trans_4x8_dc_rvv(uint8_t *dest, ptrdiff_t stride,
> int16_t
> > *block);
> > +void ff_vc1_inv_trans_8x4_dc_rvv(uint8_t *dest, ptrdiff_t stride,
> int16_t
> > *block);
> > +void ff_vc1_inv_trans_4x4_dc_rvv(uint8_t *dest, ptrdiff_t stride,
> int16_t
> > *block);
> > +
> > +av_cold void ff_vc1dsp_init_riscv(VC1DSPContext *dsp)
> > +{
> > +#if HAVE_RVV
> > +    int flags = av_get_cpu_flags();
> > +
> > +    if ((flags & AV_CPU_FLAG_RVV_I64) && ff_get_rv_vlenb() >= 16) {
> > +        dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_rvv;
> > +        dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_rvv;
> > +    }
>
> This block can be folded into the next. You don't need to check VLENB
> twice.
>
> > +    if ((flags & AV_CPU_FLAG_RVV_I32) && ff_get_rv_vlenb() >= 16) {
> > +        dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_rvv;
> > +        dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_rvv;
> > +    }
> > +#endif
> > +}
> > diff --git a/libavcodec/riscv/vc1dsp_rvv.S
> b/libavcodec/riscv/vc1dsp_rvv.S
> > new file mode 100644
> > index 0000000000..b865bd2cbc
> > --- /dev/null
> > +++ b/libavcodec/riscv/vc1dsp_rvv.S
> > @@ -0,0 +1,114 @@
> > +/*
> > + * Copyright (c) 2023 Institue of Software Chinese Academy of Sciences
> > (ISCAS).
> > + *
> > + * 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/riscv/asm.S"
> > +
> > +func ff_vc1_inv_trans_8x8_dc_rvv, zve64x
> > +        lh            t2, (a2)
> > +        sh1add        t2, t2, t2
> > +        addi          t2, t2, 1
> > +        srai          t2, t2, 1
> > +        sh1add        t2, t2, t2
> > +        addi          t2, t2, 16
> > +        srai          t2, t2, 5
> > +        vsetivli      zero, 8, e8, mf2, ta, ma
> > +        vlse64.v      v0, (a0), a1
> > +        li            t0, 8*8
> > +        vsetvli       zero, t0, e16, m8, ta, ma
>
> Instruction scheduling could be better, especially on in-order CPUs.
>
> > +        vzext.vf2     v8, v0
> > +        vadd.vx       v8, v8, t2
> > +        vmax.vx       v8, v8, zero
> > +        vsetvli       zero, t0, e8, m4, ta, ma
>
> You don't need to reset the AVL here, just pass zero.
>
> > +        vnclipu.wi    v0, v8, 0
> > +        vsetivli      zero, 8, e8, mf2, ta, ma
> > +        vsse64.v      v0, (a0), a1
> > +        ret
> > +endfunc
> > +
> > +func ff_vc1_inv_trans_4x8_dc_rvv, zve32x
> > +        lh            t2, (a2)
> > +        slli          t1, t2, 4
> > +        add           t2, t2, t1
> > +        addi          t2, t2, 4
> > +        srai          t2, t2, 3
> > +        sh1add        t2, t2, t2
> > +        slli          t2, t2, 2
> > +        addi          t2, t2, 64
> > +        srai          t2, t2, 7
> > +        vsetivli      zero, 8, e8, mf2, ta, ma
> > +        vlse32.v      v0, (a0), a1
> > +        li            t0, 4*8
> > +        vsetvli       zero, t0, e16, m4, ta, ma
> > +        vzext.vf2     v4, v0
> > +        vadd.vx       v4, v4, t2
> > +        vmax.vx       v4, v4, zero
> > +        vsetvli       zero, t0, e8, m2, ta, ma
> > +        vnclipu.wi    v0, v4, 0
> > +        vsetivli      zero, 8, e8, mf2, ta, ma
> > +        vsse32.v      v0, (a0), a1
> > +        ret
> > +endfunc
> > +
> > +func ff_vc1_inv_trans_8x4_dc_rvv, zve64x
> > +        lh            t2, (a2)
> > +        sh1add        t2, t2, t2
> > +        addi          t2, t2, 1
> > +        srai          t2, t2, 1
> > +        slli          t1, t2, 4
> > +        add           t2, t2, t1
> > +        addi          t2, t2, 64
> > +        srai          t2, t2, 7
> > +        vsetivli      zero, 8, e8, mf2, ta, ma
> > +        vlse64.v      v0, (a0), a1
> > +        li            t0, 8*4
> > +        vsetvli       zero, t0, e16, m4, ta, ma
> > +        vzext.vf2     v4, v0
> > +        vadd.vx       v4, v4, t2
> > +        vmax.vx       v4, v4, zero
> > +        vsetvli       zero, t0, e8, m2, ta, ma
> > +        vnclipu.wi    v0, v4, 0
> > +        vsetivli      zero, 8, e8, mf2, ta, ma
> > +        vsse64.v      v0, (a0), a1
> > +        ret
> > +endfunc
> > +
> > +func ff_vc1_inv_trans_4x4_dc_rvv, zve32x
> > +        lh            t2, (a2)
> > +        slli          t1, t2, 4
> > +        add           t2, t2, t1
> > +        addi          t2, t2, 4
> > +        srai          t2, t2, 3
> > +        slli          t1, t2, 4
> > +        add           t2, t2, t1
> > +        addi          t2, t2, 64
> > +        srai          t2, t2, 7
> > +        vsetivli      zero, 4, e8, mf2, ta, ma
> > +        vlse32.v      v0, (a0), a1
> > +        li            t0, 4*4
> > +        vsetvli       zero, t0, e16, m2, ta, ma
>
> vsetivli
>
> > +        vzext.vf2     v2, v0
> > +        vadd.vx       v2, v2, t2
> > +        vmax.vx       v2, v2, zero
> > +        vsetvli       zero, t0, e8, m1, ta, ma
> > +        vnclipu.wi    v0, v2, 0
> > +        vsetivli      zero, 4, e8, mf2, ta, ma
> > +        vsse32.v      v0, (a0), a1
> > +        ret
> > +endfunc
> > diff --git a/libavcodec/vc1dsp.c b/libavcodec/vc1dsp.c
> > index 62c8eb21fa..2caa3c6863 100644
> > --- a/libavcodec/vc1dsp.c
> > +++ b/libavcodec/vc1dsp.c
> > @@ -1039,6 +1039,8 @@ av_cold void ff_vc1dsp_init(VC1DSPContext *dsp)
> >      ff_vc1dsp_init_arm(dsp);
> >  #elif ARCH_PPC
> >      ff_vc1dsp_init_ppc(dsp);
> > +#elif ARCH_RISCV
> > +    ff_vc1dsp_init_riscv(dsp);
> >  #elif ARCH_X86
> >      ff_vc1dsp_init_x86(dsp);
> >  #elif ARCH_MIPS
> > diff --git a/libavcodec/vc1dsp.h b/libavcodec/vc1dsp.h
> > index 7ed1776ca7..e3b90d2b62 100644
> > --- a/libavcodec/vc1dsp.h
> > +++ b/libavcodec/vc1dsp.h
> > @@ -89,6 +89,7 @@ void ff_vc1dsp_init(VC1DSPContext* c);
> >  void ff_vc1dsp_init_aarch64(VC1DSPContext* dsp);
> >  void ff_vc1dsp_init_arm(VC1DSPContext* dsp);
> >  void ff_vc1dsp_init_ppc(VC1DSPContext *c);
> > +void ff_vc1dsp_init_riscv(VC1DSPContext *c);
> >  void ff_vc1dsp_init_x86(VC1DSPContext* dsp);
> >  void ff_vc1dsp_init_mips(VC1DSPContext* dsp);
> >  void ff_vc1dsp_init_loongarch(VC1DSPContext* dsp);
>
>
> --
> Rémi Denis-Courmont
> http://www.remlab.net/
>
>
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lavc-vc1dsp-R-V-V-inv_trans_nosplat.patch
Type: text/x-patch
Size: 8679 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20231206/8ad5e23a/attachment.bin>


More information about the ffmpeg-devel mailing list