Hello, I have received the K230, and then installed Debian following your method. Therefore, I have updated the benchmark of K230 in the patch of this reply. k230 vc1dsp.vc1_inv_trans_4x4_dc_c: 125.7 vc1dsp.vc1_inv_trans_4x4_dc_rvv_i32: 53.5 vc1dsp.vc1_inv_trans_4x8_dc_c: 230.7 vc1dsp.vc1_inv_trans_4x8_dc_rvv_i32: 65.5 vc1dsp.vc1_inv_trans_8x4_dc_c: 228.7 vc1dsp.vc1_inv_trans_8x4_dc_rvv_i64: 64.5 vc1dsp.vc1_inv_trans_8x8_dc_c: 476.5 vc1dsp.vc1_inv_trans_8x8_dc_rvv_i64: 80.2 Rémi Denis-Courmont 于2023年12月4日周一 01:17写道： > Le sunnuntaina 3. joulukuuta 2023, 16.40.08 EET flow gg a écrit : > > c910 > > vc1dsp.vc1_inv_trans_4x4_dc_c: 84.0 > > vc1dsp.vc1_inv_trans_4x4_dc_rvv_i32: 74.0 > > vc1dsp.vc1_inv_trans_4x8_dc_c: 150.2 > > vc1dsp.vc1_inv_trans_4x8_dc_rvv_i32: 83.5 > > vc1dsp.vc1_inv_trans_8x4_dc_c: 129.0 > > vc1dsp.vc1_inv_trans_8x4_dc_rvv_i64: 75.7 > > vc1dsp.vc1_inv_trans_8x8_dc_c: 254.7 > > vc1dsp.vc1_inv_trans_8x8_dc_rvv_i64: 90.5 > > The code below uses fractional multipliers, so I infer that the > benchmarked > code was significantly different, and the measurements are not really > worth the > bother. > > I know that supply is a problem at the moment, but I if you are going to > keep > this up, I would hope that ISCAS can get you access to an RVV 1.0 board. > > In-line... > > > diff --git a/libavcodec/riscv/vc1dsp_init.c > b/libavcodec/riscv/vc1dsp_init.c > > new file mode 100644 > > index 0000000000..88e0434f0e > > --- /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 > > + > > +#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) { > > + 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; > > + } > > + if (flags & AV_CPU_FLAG_RVV_I32) { > > + 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; > > + } > > Probably missing VLENB checks. > > > +#endif > > +} > > diff --git a/libavcodec/riscv/vc1dsp_rvv.S > b/libavcodec/riscv/vc1dsp_rvv.S > > new file mode 100644 > > index 0000000000..8a6b27192a > > --- /dev/null > > +++ b/libavcodec/riscv/vc1dsp_rvv.S > > @@ -0,0 +1,123 @@ > > +/* > > + * 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) > > + li t1, 3 > > + mul t2, t2, t1 > > You can multiply by 3, 5 or 9 with shift-and-add. By 12 with shift-and-add > then shift, and by 17 with shift then add. You don't need multiplications. > > > + addi t2, t2, 1 > > + srai t2, t2, 1 > > + mul t2, t2, t1 > > + 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 > > + vmv.v.x v8, t2 > > Do you really need to splat? Can't .vx or .wx be used instead? > > > + vsetvli zero, t0, e8, m4, ta, ma > > + vwaddu.wv v8, v8, v0 > > + vsetvli zero, t0, e16, m8, ta, ma > > + vmax.vx v8, v8, zero > > + vsetvli zero, t0, e8, m4, ta, ma > > + 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) > > + li t1, 17 > > + mul t2, t2, t1 > > + addi t2, t2, 4 > > + srai t2, t2, 3 > > + li t1, 12 > > + mul t2, t2, t1 > > + 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 > > + vmv.v.x v4, t2 > > + vsetvli zero, t0, e8, m2, ta, ma > > + vwaddu.wv v4, v4, v0 > > + vsetvli zero, t0, e16, m4, ta, ma > > + 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) > > + li t1, 3 > > + mul t2, t2, t1 > > + addi t2, t2, 1 > > + srai t2, t2, 1 > > + li t1, 17 > > + mul 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 > > + vmv.v.x v4, t2 > > + vsetvli zero, t0, e8, m2, ta, ma > > + vwaddu.wv v4, v4, v0 > > + vsetvli zero, t0, e16, m4, ta, ma > > + 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) > > + li t1, 17 > > + mul t2, t2, t1 > > + addi t2, t2, 4 > > + srai t2, t2, 3 > > + mul 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 > > + vmv.v.x v2, t2 > > + vsetvli zero, t0, e8, m1, ta, ma > > + vwaddu.wv v2, v2, v0 > > + vsetvli zero, t0, e16, m2, ta, ma > > + 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); > > -- > レミ・デニ-クールモン > http://www.remlab.net/ > > > > _______________________________________________ > 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". >