| Message ID | 20231208174652.21299-1-remi@remlab.net |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel] lavc/lpc: R-V V apply_welch_window | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
Quoting Rémi Denis-Courmont (2023-12-08 18:46:51) > +#if __riscv_xlen >= 64 > +func ff_lpc_apply_welch_window_rvv, zve64d > + vsetvli t0, zero, e64, m8, ta, ma > + vid.v v0 > + addi t2, a1, -1 > + vfcvt.f.xu.v v0, v0 > + li t3, 2 > + fcvt.d.l ft2, t2 > + srai t1, a1, 1 > + fcvt.d.l ft3, t3 > + li t4, 1 > + fdiv.d ft0, ft3, ft2 # ft0 = c = 2. / (len - 1) > + fcvt.d.l fa1, t4 # fa1 = 1. > + fsub.d ft1, ft0, fa1 > + vfrsub.vf v0, v0, ft1 # v0[i] = c - i - 1. > +1: > + vsetvli t0, t1, e64, m8, ta, ma > + vfmul.vv v16, v0, v0 # no fused multipy-add as v0 is reused > + sub t1, t1, t0 > + vle32.v v8, (a0) > + fcvt.d.l ft2, t0 > + vfrsub.vf v16, v16, fa1 # v16 = 1. - w * w > + sh2add a0, t0, a0 > + vsetvli zero, zero, e32, m4, ta, ma > + vfwcvt.f.x.v v24, v8 > + vsetvli zero, zero, e64, m8, ta, ma > + vfsub.vf v0, v0, ft2 # v0 -= vl > + vfmul.vv v8, v24, v16 > + vse64.v v8, (a2) > + sh3add a2, t0, a2 > + bnez t1, 1b > + > + andi t1, a1, 1 > + beqz t1, 2f > + > + sd zero, (a2) > + addi a0, a0, 4 > + addi a2, a2, 8 > +2: > + vsetvli t0, zero, e64, m8, ta, ma > + vid.v v0 > + srai t1, a1, 1 > + vfcvt.f.xu.v v0, v0 > + fcvt.d.l ft1, t1 > + fsub.d ft1, ft0, ft1 # ft1 = c - (len / 2) > + vfadd.vf v0, v0, ft1 # v0[i] = c - (len / 2) + i > +3: > + vsetvli t0, t1, e64, m8, ta, ma > + vfmul.vv v16, v0, v0 > + sub t1, t1, t0 > + vle32.v v8, (a0) > + fcvt.d.l ft2, t0 > + vfrsub.vf v16, v16, fa1 # v16 = 1. - w * w > + sh2add a0, t0, a0 > + vsetvli zero, zero, e32, m4, ta, ma > + vfwcvt.f.x.v v24, v8 > + vsetvli zero, zero, e64, m8, ta, ma > + vfadd.vf v0, v0, ft2 # v0 += vl > + vfmul.vv v8, v24, v16 > + vse64.v v8, (a2) > + sh3add a2, t0, a2 > + bnez t1, 3b I think it'd look a lot less like base64 < /dev/random if you vertically aligned the first operands.
Le 11 décembre 2023 11:11:28 GMT+02:00, Anton Khirnov <anton@khirnov.net> a écrit : >Quoting Rémi Denis-Courmont (2023-12-08 18:46:51) >> +#if __riscv_xlen >= 64 >> +func ff_lpc_apply_welch_window_rvv, zve64d >> + vsetvli t0, zero, e64, m8, ta, ma >> + vid.v v0 >> + addi t2, a1, -1 >> + vfcvt.f.xu.v v0, v0 >> + li t3, 2 >> + fcvt.d.l ft2, t2 >> + srai t1, a1, 1 >> + fcvt.d.l ft3, t3 >> + li t4, 1 >> + fdiv.d ft0, ft3, ft2 # ft0 = c = 2. / (len - 1) >> + fcvt.d.l fa1, t4 # fa1 = 1. >> + fsub.d ft1, ft0, fa1 >> + vfrsub.vf v0, v0, ft1 # v0[i] = c - i - 1. >> +1: >> + vsetvli t0, t1, e64, m8, ta, ma >> + vfmul.vv v16, v0, v0 # no fused multipy-add as v0 is reused >> + sub t1, t1, t0 >> + vle32.v v8, (a0) >> + fcvt.d.l ft2, t0 >> + vfrsub.vf v16, v16, fa1 # v16 = 1. - w * w >> + sh2add a0, t0, a0 >> + vsetvli zero, zero, e32, m4, ta, ma >> + vfwcvt.f.x.v v24, v8 >> + vsetvli zero, zero, e64, m8, ta, ma >> + vfsub.vf v0, v0, ft2 # v0 -= vl >> + vfmul.vv v8, v24, v16 >> + vse64.v v8, (a2) >> + sh3add a2, t0, a2 >> + bnez t1, 1b >> + >> + andi t1, a1, 1 >> + beqz t1, 2f >> + >> + sd zero, (a2) >> + addi a0, a0, 4 >> + addi a2, a2, 8 >> +2: >> + vsetvli t0, zero, e64, m8, ta, ma >> + vid.v v0 >> + srai t1, a1, 1 >> + vfcvt.f.xu.v v0, v0 >> + fcvt.d.l ft1, t1 >> + fsub.d ft1, ft0, ft1 # ft1 = c - (len / 2) >> + vfadd.vf v0, v0, ft1 # v0[i] = c - (len / 2) + i >> +3: >> + vsetvli t0, t1, e64, m8, ta, ma >> + vfmul.vv v16, v0, v0 >> + sub t1, t1, t0 >> + vle32.v v8, (a0) >> + fcvt.d.l ft2, t0 >> + vfrsub.vf v16, v16, fa1 # v16 = 1. - w * w >> + sh2add a0, t0, a0 >> + vsetvli zero, zero, e32, m4, ta, ma >> + vfwcvt.f.x.v v24, v8 >> + vsetvli zero, zero, e64, m8, ta, ma >> + vfadd.vf v0, v0, ft2 # v0 += vl >> + vfmul.vv v8, v24, v16 >> + vse64.v v8, (a2) >> + sh3add a2, t0, a2 >> + bnez t1, 3b > >I think it'd look a lot less like base64 < /dev/random if you vertically >aligned the first operands. They are aligned to the 17th column. Problem is that quite a few vector mnemonics are larger than 7 characters. > >-- >Anton Khirnov >_______________________________________________ >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".
Quoting Rémi Denis-Courmont (2023-12-11 10:50:53) > Le 11 décembre 2023 11:11:28 GMT+02:00, Anton Khirnov <anton@khirnov.net> a écrit : > >I think it'd look a lot less like base64 < /dev/random if you vertically > >aligned the first operands. > > They are aligned to the 17th column. Problem is that quite a few vector mnemonics are larger than 7 characters. Align to 25 or 33 then?
Le 11 décembre 2023 11:57:50 GMT+02:00, Anton Khirnov <anton@khirnov.net> a écrit : >Quoting Rémi Denis-Courmont (2023-12-11 10:50:53) >> Le 11 décembre 2023 11:11:28 GMT+02:00, Anton Khirnov <anton@khirnov.net> a écrit : >> >I think it'd look a lot less like base64 < /dev/random if you vertically >> >aligned the first operands. >> >> They are aligned to the 17th column. Problem is that quite a few vector mnemonics are larger than 7 characters. > >Align to 25 or 33 then? IMO that's even worse. The operands end up too far off from most mnemonics that it hurts legibility more than it improves. I initially aligned to the longest mnemonics but that turned out badly whenever revectoring (for obvious reasons). > >-- >Anton Khirnov >_______________________________________________ >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".
diff --git a/libavcodec/lpc.c b/libavcodec/lpc.c index dc6a3060ce..9e2fd0f128 100644 --- a/libavcodec/lpc.c +++ b/libavcodec/lpc.c @@ -320,7 +320,9 @@ av_cold int ff_lpc_init(LPCContext *s, int blocksize, int max_order, s->lpc_apply_welch_window = lpc_apply_welch_window_c; s->lpc_compute_autocorr = lpc_compute_autocorr_c; -#if ARCH_X86 +#if ARCH_RISCV + ff_lpc_init_riscv(s); +#elif ARCH_X86 ff_lpc_init_x86(s); #endif diff --git a/libavcodec/lpc.h b/libavcodec/lpc.h index 467d0b2830..0200baea5c 100644 --- a/libavcodec/lpc.h +++ b/libavcodec/lpc.h @@ -109,6 +109,7 @@ double ff_lpc_calc_ref_coefs_f(LPCContext *s, const float *samples, int len, */ int ff_lpc_init(LPCContext *s, int blocksize, int max_order, enum FFLPCType lpc_type); +void ff_lpc_init_riscv(LPCContext *s); void ff_lpc_init_x86(LPCContext *s); /** diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile index e9825c0856..1d4572fbc5 100644 --- a/libavcodec/riscv/Makefile +++ b/libavcodec/riscv/Makefile @@ -33,6 +33,8 @@ OBJS-$(CONFIG_LLVIDDSP) += riscv/llviddsp_init.o RVV-OBJS-$(CONFIG_LLVIDDSP) += riscv/llviddsp_rvv.o OBJS-$(CONFIG_LLVIDENCDSP) += riscv/llvidencdsp_init.o RVV-OBJS-$(CONFIG_LLVIDENCDSP) += riscv/llvidencdsp_rvv.o +OBJS-$(CONFIG_LPC) += riscv/lpc_init.o +RVV-OBJS-$(CONFIG_LPC) += riscv/lpc_rvv.o OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_init.o RVV-OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_rvv.o OBJS-$(CONFIG_PIXBLOCKDSP) += riscv/pixblockdsp_init.o diff --git a/libavcodec/riscv/lpc_init.c b/libavcodec/riscv/lpc_init.c new file mode 100644 index 0000000000..c16e5745f0 --- /dev/null +++ b/libavcodec/riscv/lpc_init.c @@ -0,0 +1,37 @@ +/* + * Copyright © 2022 Rémi Denis-Courmont. + * + * 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 "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavcodec/lpc.h" + +void ff_lpc_apply_welch_window_rvv(const int32_t *, ptrdiff_t, double *); + +av_cold void ff_lpc_init_riscv(LPCContext *c) +{ +#if HAVE_RVV && (__riscv_xlen >= 64) + int flags = av_get_cpu_flags(); + + if ((flags & AV_CPU_FLAG_RVV_F64) && (flags & AV_CPU_FLAG_RVB_ADDR)) + c->lpc_apply_welch_window = ff_lpc_apply_welch_window_rvv; +#endif +} diff --git a/libavcodec/riscv/lpc_rvv.S b/libavcodec/riscv/lpc_rvv.S new file mode 100644 index 0000000000..2bc729d400 --- /dev/null +++ b/libavcodec/riscv/lpc_rvv.S @@ -0,0 +1,88 @@ +/* + * Copyright © 2023 Rémi Denis-Courmont. + * + * 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" + +#if __riscv_xlen >= 64 +func ff_lpc_apply_welch_window_rvv, zve64d + vsetvli t0, zero, e64, m8, ta, ma + vid.v v0 + addi t2, a1, -1 + vfcvt.f.xu.v v0, v0 + li t3, 2 + fcvt.d.l ft2, t2 + srai t1, a1, 1 + fcvt.d.l ft3, t3 + li t4, 1 + fdiv.d ft0, ft3, ft2 # ft0 = c = 2. / (len - 1) + fcvt.d.l fa1, t4 # fa1 = 1. + fsub.d ft1, ft0, fa1 + vfrsub.vf v0, v0, ft1 # v0[i] = c - i - 1. +1: + vsetvli t0, t1, e64, m8, ta, ma + vfmul.vv v16, v0, v0 # no fused multipy-add as v0 is reused + sub t1, t1, t0 + vle32.v v8, (a0) + fcvt.d.l ft2, t0 + vfrsub.vf v16, v16, fa1 # v16 = 1. - w * w + sh2add a0, t0, a0 + vsetvli zero, zero, e32, m4, ta, ma + vfwcvt.f.x.v v24, v8 + vsetvli zero, zero, e64, m8, ta, ma + vfsub.vf v0, v0, ft2 # v0 -= vl + vfmul.vv v8, v24, v16 + vse64.v v8, (a2) + sh3add a2, t0, a2 + bnez t1, 1b + + andi t1, a1, 1 + beqz t1, 2f + + sd zero, (a2) + addi a0, a0, 4 + addi a2, a2, 8 +2: + vsetvli t0, zero, e64, m8, ta, ma + vid.v v0 + srai t1, a1, 1 + vfcvt.f.xu.v v0, v0 + fcvt.d.l ft1, t1 + fsub.d ft1, ft0, ft1 # ft1 = c - (len / 2) + vfadd.vf v0, v0, ft1 # v0[i] = c - (len / 2) + i +3: + vsetvli t0, t1, e64, m8, ta, ma + vfmul.vv v16, v0, v0 + sub t1, t1, t0 + vle32.v v8, (a0) + fcvt.d.l ft2, t0 + vfrsub.vf v16, v16, fa1 # v16 = 1. - w * w + sh2add a0, t0, a0 + vsetvli zero, zero, e32, m4, ta, ma + vfwcvt.f.x.v v24, v8 + vsetvli zero, zero, e64, m8, ta, ma + vfadd.vf v0, v0, ft2 # v0 += vl + vfmul.vv v8, v24, v16 + vse64.v v8, (a2) + sh3add a2, t0, a2 + bnez t1, 3b + + ret +endfunc +#endif