| Message ID | 20220903190147.196927-3-remi@remlab.net |
|---|---|
| State | New |
| Headers | show |
| Series | Float DSP for RISC-V Vector extension - part I | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | fail | Make fate failed |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | fail | Make fate failed |
Sep 3, 2022, 21:01 by remi@remlab.net: > From: Rémi Denis-Courmont <remi@remlab.net> > > This is based on existing code from the VLC git tree, though the size > and scalar arguments are swapped. > --- > libavutil/float_dsp.c | 2 ++ > libavutil/float_dsp.h | 1 + > libavutil/riscv/Makefile | 4 ++- > libavutil/riscv/float_dsp_init.c | 42 ++++++++++++++++++++++ > libavutil/riscv/float_dsp_rvv.S | 60 ++++++++++++++++++++++++++++++++ > 5 files changed, 108 insertions(+), 1 deletion(-) > create mode 100644 libavutil/riscv/float_dsp_init.c > create mode 100644 libavutil/riscv/float_dsp_rvv.S > > diff --git a/libavutil/float_dsp.c b/libavutil/float_dsp.c > index 8676c8b0f8..742dd679d2 100644 > --- a/libavutil/float_dsp.c > +++ b/libavutil/float_dsp.c > @@ -156,6 +156,8 @@ av_cold AVFloatDSPContext *avpriv_float_dsp_alloc(int bit_exact) > ff_float_dsp_init_arm(fdsp); > #elif ARCH_PPC > ff_float_dsp_init_ppc(fdsp, bit_exact); > +#elif ARCH_RISCV > + ff_float_dsp_init_riscv(fdsp); > #elif ARCH_X86 > ff_float_dsp_init_x86(fdsp); > #elif ARCH_MIPS > diff --git a/libavutil/float_dsp.h b/libavutil/float_dsp.h > index 9c664592bd..7cad9fc622 100644 > --- a/libavutil/float_dsp.h > +++ b/libavutil/float_dsp.h > @@ -205,6 +205,7 @@ float avpriv_scalarproduct_float_c(const float *v1, const float *v2, int len); > void ff_float_dsp_init_aarch64(AVFloatDSPContext *fdsp); > void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp); > void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int strict); > +void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp); > void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp); > void ff_float_dsp_init_mips(AVFloatDSPContext *fdsp); > > diff --git a/libavutil/riscv/Makefile b/libavutil/riscv/Makefile > index 1f818043dc..6bf8243e8d 100644 > --- a/libavutil/riscv/Makefile > +++ b/libavutil/riscv/Makefile > @@ -1 +1,3 @@ > -OBJS += riscv/cpu.o > +OBJS += riscv/cpu.o \ > + riscv/float_dsp_init.o \ > + riscv/float_dsp_rvv.o > diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c > new file mode 100644 > index 0000000000..9a5b981917 > --- /dev/null > +++ b/libavutil/riscv/float_dsp_init.c > @@ -0,0 +1,42 @@ > +/* > + * 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/float_dsp.h" > + > +void ff_vector_fmul_scalar_rvv(float *dst, const float *src, float mul, > + int len); > + > +void ff_vector_dmul_scalar_rvv(double *dst, const double *src, double mul, > + int len); > + > +av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp) > +{ > + int flags = av_get_cpu_flags(); > + > + if (flags & AV_CPU_FLAG_ZVE32F) { > + fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_rvv; > + > + if (flags & AV_CPU_FLAG_ZVE64D) { > + fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_rvv;H > + } > Style. > + } > +} > diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S > new file mode 100644 > index 0000000000..54ea1d9d6d > --- /dev/null > +++ b/libavutil/riscv/float_dsp_rvv.S > @@ -0,0 +1,60 @@ > +/* > + * 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 "config.h" > +#include "asm.h" > + > + .option arch, +v > + > +func ff_vector_fmul_scalar_rvv > +#if !defined (__riscv_float_abi_soft) > + srli a2, a2, 2 > +#else > + fmv.w.x fa0, a2 > + srli a2, a3, 2 > +#endif > Can't this be handled by a macro, like it's done by arm64 and x86?
Le lauantaina 3. syyskuuta 2022, 22.11.26 EEST Lynne a écrit : > > diff --git a/libavutil/riscv/float_dsp_rvv.S > > b/libavutil/riscv/float_dsp_rvv.S new file mode 100644 > > index 0000000000..54ea1d9d6d > > --- /dev/null > > +++ b/libavutil/riscv/float_dsp_rvv.S > > @@ -0,0 +1,60 @@ > > +/* > > + * 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 "config.h" > > +#include "asm.h" > > + > > + .option arch, +v > > + > > +func ff_vector_fmul_scalar_rvv > > +#if !defined (__riscv_float_abi_soft) > > + srli a2, a2, 2 > > +#else > > + fmv.w.x fa0, a2 > > + srli a2, a3, 2 > > +#endif > > Can't this be handled by a macro, like it's done by arm64 and x86? Err, from a quick glance, the float DSP code for AArch64 just assumes a hardware floating ABI, and has no conditionals, so I'm not sure what you mean by that. Do you mean something like VFP/NOVFP on AArch32?
Sep 3, 2022, 21:34 by remi@remlab.net: > Le lauantaina 3. syyskuuta 2022, 22.11.26 EEST Lynne a écrit : > >> > diff --git a/libavutil/riscv/float_dsp_rvv.S >> > b/libavutil/riscv/float_dsp_rvv.S new file mode 100644 >> > index 0000000000..54ea1d9d6d >> > --- /dev/null >> > +++ b/libavutil/riscv/float_dsp_rvv.S >> > @@ -0,0 +1,60 @@ >> > +/* >> > + * 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 "config.h" >> > +#include "asm.h" >> > + >> > + .option arch, +v >> > + >> > +func ff_vector_fmul_scalar_rvv >> > +#if !defined (__riscv_float_abi_soft) >> > + srli a2, a2, 2 >> > +#else >> > + fmv.w.x fa0, a2 >> > + srli a2, a3, 2 >> > +#endif >> >> Can't this be handled by a macro, like it's done by arm64 and x86? >> > > Err, from a quick glance, the float DSP code for AArch64 just assumes a > hardware floating ABI, and has no conditionals, so I'm not sure what you mean > by that. Do you mean something like VFP/NOVFP on AArch32? > I meant all ABI stuff to be handled by macros, either `func` or the instructions themselves.
Le lauantaina 3. syyskuuta 2022, 22.48.45 EEST Lynne a écrit : > Sep 3, 2022, 21:34 by remi@remlab.net: > > Le lauantaina 3. syyskuuta 2022, 22.11.26 EEST Lynne a écrit : > >> > diff --git a/libavutil/riscv/float_dsp_rvv.S > >> > b/libavutil/riscv/float_dsp_rvv.S new file mode 100644 > >> > index 0000000000..54ea1d9d6d > >> > --- /dev/null > >> > +++ b/libavutil/riscv/float_dsp_rvv.S > >> > @@ -0,0 +1,60 @@ > >> > +/* > >> > + * 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 "config.h" > >> > +#include "asm.h" > >> > + > >> > + .option arch, +v > >> > + > >> > +func ff_vector_fmul_scalar_rvv > >> > +#if !defined (__riscv_float_abi_soft) > >> > + srli a2, a2, 2 > >> > +#else > >> > + fmv.w.x fa0, a2 > >> > + srli a2, a3, 2 > >> > +#endif > >> > >> Can't this be handled by a macro, like it's done by arm64 and x86? > > > > Err, from a quick glance, the float DSP code for AArch64 just assumes a > > hardware floating ABI, and has no conditionals, so I'm not sure what you > > mean by that. Do you mean something like VFP/NOVFP on AArch32? > > I meant all ABI stuff to be handled by macros, either `func` or the > instructions themselves. I don't think that's possible here, at least not with reasonably low complexity. AArch32 has the same challenge and the func/endfunc macros don't handle it.
diff --git a/libavutil/float_dsp.c b/libavutil/float_dsp.c index 8676c8b0f8..742dd679d2 100644 --- a/libavutil/float_dsp.c +++ b/libavutil/float_dsp.c @@ -156,6 +156,8 @@ av_cold AVFloatDSPContext *avpriv_float_dsp_alloc(int bit_exact) ff_float_dsp_init_arm(fdsp); #elif ARCH_PPC ff_float_dsp_init_ppc(fdsp, bit_exact); +#elif ARCH_RISCV + ff_float_dsp_init_riscv(fdsp); #elif ARCH_X86 ff_float_dsp_init_x86(fdsp); #elif ARCH_MIPS diff --git a/libavutil/float_dsp.h b/libavutil/float_dsp.h index 9c664592bd..7cad9fc622 100644 --- a/libavutil/float_dsp.h +++ b/libavutil/float_dsp.h @@ -205,6 +205,7 @@ float avpriv_scalarproduct_float_c(const float *v1, const float *v2, int len); void ff_float_dsp_init_aarch64(AVFloatDSPContext *fdsp); void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp); void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int strict); +void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp); void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp); void ff_float_dsp_init_mips(AVFloatDSPContext *fdsp); diff --git a/libavutil/riscv/Makefile b/libavutil/riscv/Makefile index 1f818043dc..6bf8243e8d 100644 --- a/libavutil/riscv/Makefile +++ b/libavutil/riscv/Makefile @@ -1 +1,3 @@ -OBJS += riscv/cpu.o +OBJS += riscv/cpu.o \ + riscv/float_dsp_init.o \ + riscv/float_dsp_rvv.o diff --git a/libavutil/riscv/float_dsp_init.c b/libavutil/riscv/float_dsp_init.c new file mode 100644 index 0000000000..9a5b981917 --- /dev/null +++ b/libavutil/riscv/float_dsp_init.c @@ -0,0 +1,42 @@ +/* + * 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/float_dsp.h" + +void ff_vector_fmul_scalar_rvv(float *dst, const float *src, float mul, + int len); + +void ff_vector_dmul_scalar_rvv(double *dst, const double *src, double mul, + int len); + +av_cold void ff_float_dsp_init_riscv(AVFloatDSPContext *fdsp) +{ + int flags = av_get_cpu_flags(); + + if (flags & AV_CPU_FLAG_ZVE32F) { + fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_rvv; + + if (flags & AV_CPU_FLAG_ZVE64D) { + fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_rvv; + } + } +} diff --git a/libavutil/riscv/float_dsp_rvv.S b/libavutil/riscv/float_dsp_rvv.S new file mode 100644 index 0000000000..54ea1d9d6d --- /dev/null +++ b/libavutil/riscv/float_dsp_rvv.S @@ -0,0 +1,60 @@ +/* + * 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 "config.h" +#include "asm.h" + + .option arch, +v + +func ff_vector_fmul_scalar_rvv +#if !defined (__riscv_float_abi_soft) + srli a2, a2, 2 +#else + fmv.w.x fa0, a2 + srli a2, a3, 2 +#endif +1: vsetvli t0, a2, e32, m8, ta, ma + slli t1, t0, 2 + vle32.v v16, (a1) + add a1, a1, t1 + vfmul.vf v16, v16, fa0 + sub a2, a2, t0 + vse32.v v16, (a0) + add a0, a0, t1 + bnez a2, 1b + ret +endfunc + +func ff_vector_dmul_scalar_rvv +#if !(defined (__riscv_float_abi_soft) || defined (__riscv_float_abi_single)) + srli a2, a2, 3 +#else + fmv.d.x fa0, a2 + srli a2, a3, 3 +#endif +1: vsetvli t0, a2, e64, m8, ta, ma + slli t1, t0, 3 + vle64.v v16, (a1) + add a1, a1, t1 + vfmul.vf v16, v16, fa0 + sub a2, a2, t0 + vse64.v v16, (a0) + add a0, a0, t1 + bnez a2, 1b + ret +endfunc
From: Rémi Denis-Courmont <remi@remlab.net> This is based on existing code from the VLC git tree, though the size and scalar arguments are swapped. --- libavutil/float_dsp.c | 2 ++ libavutil/float_dsp.h | 1 + libavutil/riscv/Makefile | 4 ++- libavutil/riscv/float_dsp_init.c | 42 ++++++++++++++++++++++ libavutil/riscv/float_dsp_rvv.S | 60 ++++++++++++++++++++++++++++++++ 5 files changed, 108 insertions(+), 1 deletion(-) create mode 100644 libavutil/riscv/float_dsp_init.c create mode 100644 libavutil/riscv/float_dsp_rvv.S