Re: [FFmpeg-devel] [PATCH 2/2] lavc/flacdsp: implement wasted32 DSP function for VSX on POWER

[Sean McGovern <gseanmcg@gmail.com>]

Hi Rémi,

First of all, thanks for the review.

On Thu, Jul 4, 2024, 07:15 Rémi Denis-Courmont <remi@remlab.net> wrote:

>
>
> Le 4 juillet 2024 04:23:30 GMT+03:00, Sean McGovern <gseanmcg@gmail.com>
> a écrit :
> >RaptorCS POWER9 (8c4t) @ 2.2GHz:
> >flac_wasted_32_c: 50.1
> >flac_wasted_32_vsx: 17.3
> >---
> > libavcodec/flacdsp.c          |  2 ++
> > libavcodec/flacdsp.h          |  1 +
> > libavcodec/ppc/Makefile       |  2 ++
> > libavcodec/ppc/flacdsp_init.c | 38 ++++++++++++++++++++++++++++
> > libavcodec/ppc/flacdsp_vsx.c  | 47 +++++++++++++++++++++++++++++++++++
> > 5 files changed, 90 insertions(+)
> > create mode 100644 libavcodec/ppc/flacdsp_init.c
> > create mode 100644 libavcodec/ppc/flacdsp_vsx.c
> >
> >diff --git a/libavcodec/flacdsp.c b/libavcodec/flacdsp.c
> >index f5362bf66f..b63d55ddcd 100644
> >--- a/libavcodec/flacdsp.c
> >+++ b/libavcodec/flacdsp.c
> >@@ -156,5 +156,7 @@ av_cold void ff_flacdsp_init(FLACDSPContext *c, enum
> AVSampleFormat fmt, int cha
> >     ff_flacdsp_init_riscv(c, fmt, channels);
> > #elif ARCH_X86
> >     ff_flacdsp_init_x86(c, fmt, channels);
> >+#elif ARCH_PPC
> >+    ff_flacdsp_init_ppc(c, fmt, channels);
> > #endif
> > }
> >diff --git a/libavcodec/flacdsp.h b/libavcodec/flacdsp.h
> >index 3b7b35a112..941536ef16 100644
> >--- a/libavcodec/flacdsp.h
> >+++ b/libavcodec/flacdsp.h
> >@@ -45,5 +45,6 @@ void ff_flacdsp_init(FLACDSPContext *c, enum
> AVSampleFormat fmt, int channels);
> > void ff_flacdsp_init_arm(FLACDSPContext *c, enum AVSampleFormat fmt, int
> channels);
> > void ff_flacdsp_init_riscv(FLACDSPContext *c, enum AVSampleFormat fmt,
> int channels);
> > void ff_flacdsp_init_x86(FLACDSPContext *c, enum AVSampleFormat fmt, int
> channels);
> >+void ff_flacdsp_init_ppc(FLACDSPContext *c, enum AVSampleFormat fmt, int
> channels);
> >
> > #endif /* AVCODEC_FLACDSP_H */
> >diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile
> >index 10b9ca60da..7f81a8aa34 100644
> >--- a/libavcodec/ppc/Makefile
> >+++ b/libavcodec/ppc/Makefile
> >@@ -2,6 +2,8 @@
> > OBJS-$(CONFIG_AUDIODSP)                += ppc/audiodsp.o
> > OBJS-$(CONFIG_BLOCKDSP)                += ppc/blockdsp.o
> > OBJS-$(CONFIG_FDCTDSP)                 += ppc/fdctdsp.o
> >+OBJS-$(CONFIG_FLAC_DECODER)            += ppc/flacdsp_init.o
> >+VSX-OBJS-$(CONFIG_FLAC_DECODER)        += ppc/flacdsp_vsx.o
> > OBJS-$(CONFIG_FMTCONVERT)              += ppc/fmtconvert_altivec.o
> > OBJS-$(CONFIG_H264CHROMA)              += ppc/h264chroma_init.o
> > OBJS-$(CONFIG_H264DSP)                 += ppc/h264dsp.o
> ppc/hpeldsp_altivec.o
> >diff --git a/libavcodec/ppc/flacdsp_init.c b/libavcodec/ppc/flacdsp_init.c
> >new file mode 100644
> >index 0000000000..526bddddbf
> >--- /dev/null
> >+++ b/libavcodec/ppc/flacdsp_init.c
> >@@ -0,0 +1,38 @@
> >+/*
> >+ * Copyright (c) 2024 Sean McGovern <gseanmcg@gmail.com>
> >+ *
> >+ * 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 "libavutil/ppc/cpu.h"
> >+#include "libavcodec/flacdsp.h"
> >+
> >+void ff_flac_wasted32_vsx(int32_t *, int wasted, int len);
> >+
> >+av_cold void ff_flacdsp_init_ppc(FLACDSPContext *c, enum AVSampleFormat
> fmt, int channels)
> >+{
> >+#if HAVE_VSX
> >+    if (!PPC_VSX(av_get_cpu_flags()))
> >+        return;
> >+
> >+    c->wasted32 = ff_flac_wasted32_vsx;
> >+#endif
> >+}
> >diff --git a/libavcodec/ppc/flacdsp_vsx.c b/libavcodec/ppc/flacdsp_vsx.c
> >new file mode 100644
> >index 0000000000..c31566458c
> >--- /dev/null
> >+++ b/libavcodec/ppc/flacdsp_vsx.c
> >@@ -0,0 +1,47 @@
> >+/*
> >+ * Copyright (c) 2024 Sean McGovern <gseanmcg@gmail.com>
> >+ *
> >+ * 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 "libavutil/ppc/cpu.h"
> >+#include "libavutil/ppc/util_altivec.h"
> >+#include "libavcodec/flacdsp.h"
> >+
> >+void ff_flac_wasted32_vsx(int32_t *, int wasted, int len);
> >+
> >+#if HAVE_VSX
> >+
> >+void ff_flac_wasted32_vsx(int32_t *decoded, int wasted, int len)
> >+{
> >+    register vec_s32 vec1;
> >+    const register vec_u32 vec2 = vec_splats((unsigned)wasted);
> >+    register vec_s32 shifted;
> >+
> >+    for (int i = 0; i < len; i+=sizeof(int32_t)) {
>
> len in elements, so why is the increment in bytes? Looks like
> heterogeneous measurement units.
>

I wondered about that, this is my first attempt at vectorization -- it
seems I have much to learn.


> >+        vec1 = vec_xl(0, decoded);
> >+        shifted = vec_sl(vec1, vec2);
> >+        vec_xst(shifted, 0, decoded);
>
> Judging by other AltiVec code, this lacks unrolling.
>

Is that correlated with the comment above re: len? Or is it more general
that I should unroll until I've exhausted the available vector registers?


> >+        decoded += sizeof(int32_t);
> >+    }
> >+}
> >+
> >+#endif /* HAVE_VSX */
> _______________________________________________
> 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".
>

-- Sean McGovern

>
_______________________________________________
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".