[FFmpeg-devel] [RFC] [PATCH 0/3] RISC-V CPU detection

[FFmpeg-devel] [RFC] [PATCH 0/3] RISC-V CPU detection

[Rémi Denis-Courmont]

Hi,

This small series introduces the same CPU detection and assembler macros
as the earlier V extension stuff but sticking to a scalar use case.

Benchmark results are included in the last patch.

----------------------------------------------------------------
Rémi Denis-Courmont (3):
      lavu: detect RISC-V F extension (i.e. float)
      lavu/riscv: initial common header for assembler macros
      lavc/audiodsp: add RISC-V F float vector clip

 libavcodec/audiodsp.c            |  2 ++
 libavcodec/audiodsp.h            |  1 +
 libavcodec/riscv/Makefile        |  2 ++
 libavcodec/riscv/audiodsp_init.c | 31 +++++++++++++++++
 libavcodec/riscv/audiodsp_rvf.S  | 46 +++++++++++++++++++++++++
 libavutil/cpu.c                  |  4 +++
 libavutil/cpu.h                  |  3 ++
 libavutil/cpu_internal.h         |  1 +
 libavutil/riscv/Makefile         |  1 +
 libavutil/riscv/asm.S            | 74 ++++++++++++++++++++++++++++++++++++++++
 libavutil/riscv/cpu.c            | 44 ++++++++++++++++++++++++
 tests/checkasm/checkasm.c        |  2 ++
 12 files changed, 211 insertions(+)
 create mode 100644 libavcodec/riscv/Makefile
 create mode 100644 libavcodec/riscv/audiodsp_init.c
 create mode 100644 libavcodec/riscv/audiodsp_rvf.S
 create mode 100644 libavutil/riscv/Makefile
 create mode 100644 libavutil/riscv/asm.S
 create mode 100644 libavutil/riscv/cpu.c

-- 
Реми Дёни-Курмон
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".

[FFmpeg-devel] [PATCH 1/3] lavu: detect RISC-V F extension (i.e. float)

[remi]

From: Rémi Denis-Courmont <remi@remlab.net>

This introduces compile-tim and run-time CPU detection on RISC-V. In
practice, I doubt that FFmpeg will ever see a RISC-V CPU without the F
extension, and if it does, it probably won't have run-time detection.
So the flag is essentially always set.

But as things stand, checkasm wants it that way, and we are nowhere
near running short on CPU flag bits on that platform.
---
 libavutil/cpu.c           |  4 ++++
 libavutil/cpu.h           |  3 +++
 libavutil/cpu_internal.h  |  1 +
 libavutil/riscv/Makefile  |  1 +
 libavutil/riscv/cpu.c     | 44 +++++++++++++++++++++++++++++++++++++++
 tests/checkasm/checkasm.c |  2 ++
 6 files changed, 55 insertions(+)
 create mode 100644 libavutil/riscv/Makefile
 create mode 100644 libavutil/riscv/cpu.c

diff --git a/libavutil/cpu.c b/libavutil/cpu.c
index 0035e927a5..6e9b8c5f58 100644
--- a/libavutil/cpu.c
+++ b/libavutil/cpu.c
@@ -62,6 +62,8 @@ static int get_cpu_flags(void)
     return ff_get_cpu_flags_arm();
 #elif ARCH_PPC
     return ff_get_cpu_flags_ppc();
+#elif ARCH_RISCV
+    return ff_get_cpu_flags_riscv();
 #elif ARCH_X86
     return ff_get_cpu_flags_x86();
 #elif ARCH_LOONGARCH
@@ -178,6 +180,8 @@ int av_parse_cpu_caps(unsigned *flags, const char *s)
 #elif ARCH_LOONGARCH
         { "lsx",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_LSX      },    .unit = "flags" },
         { "lasx",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_LASX     },    .unit = "flags" },
+#elif ARCH_RISCV
+        { "float",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_F        },    .unit = "flags" },
 #endif
         { NULL },
     };
diff --git a/libavutil/cpu.h b/libavutil/cpu.h
index 9711e574c5..71ae70bcbd 100644
--- a/libavutil/cpu.h
+++ b/libavutil/cpu.h
@@ -78,6 +78,9 @@
 #define AV_CPU_FLAG_LSX          (1 << 0)
 #define AV_CPU_FLAG_LASX         (1 << 1)
 
+// RISC-V Vector extension
+#define AV_CPU_FLAG_F            (1 << 0)
+
 /**
  * Return the flags which specify extensions supported by the CPU.
  * The returned value is affected by av_force_cpu_flags() if that was used
diff --git a/libavutil/cpu_internal.h b/libavutil/cpu_internal.h
index 650d47fc96..634f28bac4 100644
--- a/libavutil/cpu_internal.h
+++ b/libavutil/cpu_internal.h
@@ -48,6 +48,7 @@ int ff_get_cpu_flags_mips(void);
 int ff_get_cpu_flags_aarch64(void);
 int ff_get_cpu_flags_arm(void);
 int ff_get_cpu_flags_ppc(void);
+int ff_get_cpu_flags_riscv(void);
 int ff_get_cpu_flags_x86(void);
 int ff_get_cpu_flags_loongarch(void);
 
diff --git a/libavutil/riscv/Makefile b/libavutil/riscv/Makefile
new file mode 100644
index 0000000000..1f818043dc
--- /dev/null
+++ b/libavutil/riscv/Makefile
@@ -0,0 +1 @@
+OBJS += riscv/cpu.o
diff --git a/libavutil/riscv/cpu.c b/libavutil/riscv/cpu.c
new file mode 100644
index 0000000000..6fc30f73c6
--- /dev/null
+++ b/libavutil/riscv/cpu.c
@@ -0,0 +1,44 @@
+/*
+ * 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/cpu.h"
+#include "libavutil/cpu_internal.h"
+#include "config.h"
+
+#if HAVE_GETAUXVAL
+#include <sys/auxv.h>
+#endif
+
+#define HWCAP_RV(letter) (1ul << ((letter) - 'A'))
+
+int ff_get_cpu_flags_riscv(void)
+{
+    int ret = 0;
+#if HAVE_GETAUXVAL
+    const unsigned long hwcap = getauxval(AT_HWCAP);
+
+    if (hwcap & HWCAP_RV('F'))
+        ret |= AV_CPU_FLAG_F;
+#endif
+
+#ifdef __riscv_flen
+    ret |= AV_CPU_FLAG_F;
+#endif
+
+    return ret;
+}
diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index e56fd3850e..4f6edfe6a3 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -226,6 +226,8 @@ static const struct {
     { "ALTIVEC",  "altivec",  AV_CPU_FLAG_ALTIVEC },
     { "VSX",      "vsx",      AV_CPU_FLAG_VSX },
     { "POWER8",   "power8",   AV_CPU_FLAG_POWER8 },
+#elif ARCH_RISCV
+    { "F",        "f",        AV_CPU_FLAG_F },
 #elif ARCH_MIPS
     { "MMI",      "mmi",      AV_CPU_FLAG_MMI },
     { "MSA",      "msa",      AV_CPU_FLAG_MSA },
-- 
2.37.2

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 1/3] lavu: detect RISC-V F extension (i.e. float)

[Lynne]

Sep 14, 2022, 19:50 by remi@remlab.net:

> From: Rémi Denis-Courmont <remi@remlab.net>
>
> This introduces compile-tim and run-time CPU detection on RISC-V. In
> practice, I doubt that FFmpeg will ever see a RISC-V CPU without the F
> extension, and if it does, it probably won't have run-time detection.
> So the flag is essentially always set.
>
> But as things stand, checkasm wants it that way, and we are nowhere
> near running short on CPU flag bits on that platform.
> ---
>  libavutil/cpu.c           |  4 ++++
>  libavutil/cpu.h           |  3 +++
>  libavutil/cpu_internal.h  |  1 +
>  libavutil/riscv/Makefile  |  1 +
>  libavutil/riscv/cpu.c     | 44 +++++++++++++++++++++++++++++++++++++++
>  tests/checkasm/checkasm.c |  2 ++
>  6 files changed, 55 insertions(+)
>  create mode 100644 libavutil/riscv/Makefile
>  create mode 100644 libavutil/riscv/cpu.c
>
> diff --git a/libavutil/cpu.c b/libavutil/cpu.c
> index 0035e927a5..6e9b8c5f58 100644
> --- a/libavutil/cpu.c
> +++ b/libavutil/cpu.c
> @@ -62,6 +62,8 @@ static int get_cpu_flags(void)
>  return ff_get_cpu_flags_arm();
>  #elif ARCH_PPC
>  return ff_get_cpu_flags_ppc();
> +#elif ARCH_RISCV
> +    return ff_get_cpu_flags_riscv();
>  #elif ARCH_X86
>  return ff_get_cpu_flags_x86();
>  #elif ARCH_LOONGARCH
> @@ -178,6 +180,8 @@ int av_parse_cpu_caps(unsigned *flags, const char *s)
>  #elif ARCH_LOONGARCH
>  { "lsx",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_LSX      },    .unit = "flags" },
>  { "lasx",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_LASX     },    .unit = "flags" },
> +#elif ARCH_RISCV
> +        { "float",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_F        },    .unit = "flags" },
>  #endif
>  { NULL },
>  };
> diff --git a/libavutil/cpu.h b/libavutil/cpu.h
> index 9711e574c5..71ae70bcbd 100644
> --- a/libavutil/cpu.h
> +++ b/libavutil/cpu.h
> @@ -78,6 +78,9 @@
>  #define AV_CPU_FLAG_LSX          (1 << 0)
>  #define AV_CPU_FLAG_LASX         (1 << 1)
>  
> +// RISC-V Vector extension
> +#define AV_CPU_FLAG_F            (1 << 0)
>

Can you prefix that with RV (FLAG_RVF) like the function?

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 1/3] lavu: detect RISC-V F extension (i.e. float)

[Rémi Denis-Courmont]

Le keskiviikkona 14. syyskuuta 2022, 22.28.01 EEST Lynne a écrit :
> Sep 14, 2022, 19:50 by remi@remlab.net:
> > From: Rémi Denis-Courmont <remi@remlab.net>
> > 
> > This introduces compile-tim and run-time CPU detection on RISC-V. In
> > practice, I doubt that FFmpeg will ever see a RISC-V CPU without the F
> > extension, and if it does, it probably won't have run-time detection.
> > So the flag is essentially always set.
> > 
> > But as things stand, checkasm wants it that way, and we are nowhere
> > near running short on CPU flag bits on that platform.
> > ---
> > 
> >  libavutil/cpu.c           |  4 ++++
> >  libavutil/cpu.h           |  3 +++
> >  libavutil/cpu_internal.h  |  1 +
> >  libavutil/riscv/Makefile  |  1 +
> >  libavutil/riscv/cpu.c     | 44 +++++++++++++++++++++++++++++++++++++++
> >  tests/checkasm/checkasm.c |  2 ++
> >  6 files changed, 55 insertions(+)
> >  create mode 100644 libavutil/riscv/Makefile
> >  create mode 100644 libavutil/riscv/cpu.c
> > 
> > diff --git a/libavutil/cpu.c b/libavutil/cpu.c
> > index 0035e927a5..6e9b8c5f58 100644
> > --- a/libavutil/cpu.c
> > +++ b/libavutil/cpu.c
> > @@ -62,6 +62,8 @@ static int get_cpu_flags(void)
> > 
> >  return ff_get_cpu_flags_arm();
> >  #elif ARCH_PPC
> >  return ff_get_cpu_flags_ppc();
> > 
> > +#elif ARCH_RISCV
> > +    return ff_get_cpu_flags_riscv();
> > 
> >  #elif ARCH_X86
> >  return ff_get_cpu_flags_x86();
> >  #elif ARCH_LOONGARCH
> > 
> > @@ -178,6 +180,8 @@ int av_parse_cpu_caps(unsigned *flags, const char *s)
> > 
> >  #elif ARCH_LOONGARCH
> >  { "lsx",      NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_LSX     
> >  },    .unit = "flags" }, { "lasx",     NULL, 0, AV_OPT_TYPE_CONST, {
> >  .i64 = AV_CPU_FLAG_LASX     },    .unit = "flags" },> 
> > +#elif ARCH_RISCV
> > +        { "float",    NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_F 
> >       },    .unit = "flags" },> 
> >  #endif
> >  { NULL },
> >  };
> > 
> > diff --git a/libavutil/cpu.h b/libavutil/cpu.h
> > index 9711e574c5..71ae70bcbd 100644
> > --- a/libavutil/cpu.h
> > +++ b/libavutil/cpu.h
> > @@ -78,6 +78,9 @@
> > 
> >  #define AV_CPU_FLAG_LSX          (1 << 0)
> >  #define AV_CPU_FLAG_LASX         (1 << 1)
> > 
> > +// RISC-V Vector extension
> > +#define AV_CPU_FLAG_F            (1 << 0)
> 
> Can you prefix that with RV (FLAG_RVF) like the function?

I agree that the name is a bit unfortunate, and that's essentially because 
RISC-V is not making up cool marketing names for their ISA extensions.

But it is consistent with what other architectures do, and RVF wouldn't be. So 
I have mixed feelings here. Arguably, the correct approach would be to only 
define the flags on their respective target architectures (and leave it as just 
"F").

I guess I will change it to RVF if nobody else expresses a contradictory 
opinion.

-- 
Реми Дёни-Курмон
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".

[FFmpeg-devel] [PATCH 2/3] lavu/riscv: initial common header for assembler macros

[remi]

From: Rémi Denis-Courmont <remi@remlab.net>

---
 libavutil/riscv/asm.S | 74 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 libavutil/riscv/asm.S

diff --git a/libavutil/riscv/asm.S b/libavutil/riscv/asm.S
new file mode 100644
index 0000000000..7623c161cf
--- /dev/null
+++ b/libavutil/riscv/asm.S
@@ -0,0 +1,74 @@
+/*
+ * 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"
+
+#if defined (__riscv_float_abi_soft)
+#define NOHWF
+#define NOHWD
+#define HWF   #
+#define HWD   #
+#elif defined (__riscv_float_abi_single)
+#define NOHWF #
+#define NOHWD
+#define HWF
+#define HWD   #
+#else
+#define NOHWF #
+#define NOHWD #
+#define HWF
+#define HWD
+#endif
+
+        .macro func sym, ext=
+            .text
+            .align 2
+
+            .option push
+            .ifnb \ext
+            .option arch, +\ext
+            .endif
+
+            .global \sym
+            .hidden \sym
+            .type   \sym, %function
+            \sym:
+
+            .macro endfunc
+                .size   \sym, . - \sym
+                .option pop
+                .previous
+                .purgem endfunc
+            .endm
+        .endm
+
+        .macro const sym, align=3, relocate=0
+            .if \relocate
+                .pushsection .data.rel.ro
+            .else
+                .pushsection .rodata
+            .endif
+            .align \align
+            \sym:
+
+            .macro endconst
+                .size  \sym, . - \sym
+                .popsection
+                .purgem endconst
+            .endm
+        .endm
-- 
2.37.2

_______________________________________________
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".

[FFmpeg-devel] [PATCH 3/3] lavc/audiodsp: add RISC-V F float vector clip

[remi]

From: Rémi Denis-Courmont <remi@remlab.net>

RV64G supports MIN & MAX instructions natively only on floating point
registers, not general purpose ones. The later would require the Zbb
extension. Due to that, it is actually faster to perform the clipping
"properly" in FPU.

Benchmarked on SiFive U74-MC:

audiodsp.vector_clipf_c: 29551.5
audiodsp.vector_clipf_f: 17871.0

Also tried unrolling with 2 or 8 elements but it gets worse either way.
---
 libavcodec/audiodsp.c            |  2 ++
 libavcodec/audiodsp.h            |  1 +
 libavcodec/riscv/Makefile        |  2 ++
 libavcodec/riscv/audiodsp_init.c | 31 +++++++++++++++++++++
 libavcodec/riscv/audiodsp_rvf.S  | 46 ++++++++++++++++++++++++++++++++
 5 files changed, 82 insertions(+)
 create mode 100644 libavcodec/riscv/Makefile
 create mode 100644 libavcodec/riscv/audiodsp_init.c
 create mode 100644 libavcodec/riscv/audiodsp_rvf.S

diff --git a/libavcodec/audiodsp.c b/libavcodec/audiodsp.c
index ff43e87dce..eba6e809fd 100644
--- a/libavcodec/audiodsp.c
+++ b/libavcodec/audiodsp.c
@@ -113,6 +113,8 @@ av_cold void ff_audiodsp_init(AudioDSPContext *c)
     ff_audiodsp_init_arm(c);
 #elif ARCH_PPC
     ff_audiodsp_init_ppc(c);
+#elif ARCH_RISCV
+    ff_audiodsp_init_riscv(c);
 #elif ARCH_X86
     ff_audiodsp_init_x86(c);
 #endif
diff --git a/libavcodec/audiodsp.h b/libavcodec/audiodsp.h
index aa6fa7898b..485b512839 100644
--- a/libavcodec/audiodsp.h
+++ b/libavcodec/audiodsp.h
@@ -55,6 +55,7 @@ typedef struct AudioDSPContext {
 void ff_audiodsp_init(AudioDSPContext *c);
 void ff_audiodsp_init_arm(AudioDSPContext *c);
 void ff_audiodsp_init_ppc(AudioDSPContext *c);
+void ff_audiodsp_init_riscv(AudioDSPContext *c);
 void ff_audiodsp_init_x86(AudioDSPContext *c);
 
 #endif /* AVCODEC_AUDIODSP_H */
diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
new file mode 100644
index 0000000000..a1f67ed55b
--- /dev/null
+++ b/libavcodec/riscv/Makefile
@@ -0,0 +1,2 @@
+OBJS += riscv/audiodsp_init.o \
+        riscv/audiodsp_rvf.o
diff --git a/libavcodec/riscv/audiodsp_init.c b/libavcodec/riscv/audiodsp_init.c
new file mode 100644
index 0000000000..7ffd7e8162
--- /dev/null
+++ b/libavcodec/riscv/audiodsp_init.c
@@ -0,0 +1,31 @@
+/*
+ * 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/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavcodec/audiodsp.h"
+
+void ff_vector_clipf_rvf(float *dst, const float *src, int len, float min, float max);
+
+av_cold void ff_audiodsp_init_riscv(AudioDSPContext *c)
+{
+    int flags = av_get_cpu_flags();
+
+    if (flags & AV_CPU_FLAG_F)
+        c->vector_clipf = ff_vector_clipf_rvf;
+}
diff --git a/libavcodec/riscv/audiodsp_rvf.S b/libavcodec/riscv/audiodsp_rvf.S
new file mode 100644
index 0000000000..148af96ea2
--- /dev/null
+++ b/libavcodec/riscv/audiodsp_rvf.S
@@ -0,0 +1,46 @@
+/*
+ * 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_vector_clipf_rvf, f
+NOHWF   fmv.w.x fa0, a3
+NOHWF   fmv.w.v fa1, a4
+1:
+        flw     ft0,   (a1)
+        flw     ft1,  4(a1)
+        fmax.s  ft0, ft0, fa0
+        flw     ft2,  8(a1)
+        fmax.s  ft1, ft1, fa0
+        flw     ft3, 12(a1)
+        fmax.s  ft2, ft2, fa0
+        addi    a2, a2, -4
+        fmax.s  ft3, ft3, fa0
+        addi    a1, a1, 16
+        fmin.s  ft0, ft0, fa1
+        fmin.s  ft1, ft1, fa1
+        fsw     ft0,   (a0)
+        fmin.s  ft2, ft2, fa1
+        fsw     ft1,  4(a0)
+        fmin.s  ft3, ft3, fa1
+        fsw     ft2,  8(a0)
+        fsw     ft3, 12(a0)
+        addi    a0, a0, 16
+        bnez    a2, 1b
+        ret
+endfunc
-- 
2.37.2

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 3/3] lavc/audiodsp: add RISC-V F float vector clip

[Rémi Denis-Courmont]

Le keskiviikkona 14. syyskuuta 2022, 20.50.31 EEST remi@remlab.net a écrit :
> From: Rémi Denis-Courmont <remi@remlab.net>
> 
> RV64G supports MIN & MAX instructions natively only on floating point
> registers, not general purpose ones. The later would require the Zbb
> extension. Due to that, it is actually faster to perform the clipping
> "properly" in FPU.
> 
> Benchmarked on SiFive U74-MC:
> 
> audiodsp.vector_clipf_c: 29551.5
> audiodsp.vector_clipf_f: 17871.0
> 
> Also tried unrolling with 2 or 8 elements but it gets worse either way.
> ---
>  libavcodec/audiodsp.c            |  2 ++
>  libavcodec/audiodsp.h            |  1 +
>  libavcodec/riscv/Makefile        |  2 ++
>  libavcodec/riscv/audiodsp_init.c | 31 +++++++++++++++++++++
>  libavcodec/riscv/audiodsp_rvf.S  | 46 ++++++++++++++++++++++++++++++++
>  5 files changed, 82 insertions(+)
>  create mode 100644 libavcodec/riscv/Makefile
>  create mode 100644 libavcodec/riscv/audiodsp_init.c
>  create mode 100644 libavcodec/riscv/audiodsp_rvf.S
> 
> diff --git a/libavcodec/audiodsp.c b/libavcodec/audiodsp.c
> index ff43e87dce..eba6e809fd 100644
> --- a/libavcodec/audiodsp.c
> +++ b/libavcodec/audiodsp.c
> @@ -113,6 +113,8 @@ av_cold void ff_audiodsp_init(AudioDSPContext *c)
>      ff_audiodsp_init_arm(c);
>  #elif ARCH_PPC
>      ff_audiodsp_init_ppc(c);
> +#elif ARCH_RISCV
> +    ff_audiodsp_init_riscv(c);
>  #elif ARCH_X86
>      ff_audiodsp_init_x86(c);
>  #endif
> diff --git a/libavcodec/audiodsp.h b/libavcodec/audiodsp.h
> index aa6fa7898b..485b512839 100644
> --- a/libavcodec/audiodsp.h
> +++ b/libavcodec/audiodsp.h
> @@ -55,6 +55,7 @@ typedef struct AudioDSPContext {
>  void ff_audiodsp_init(AudioDSPContext *c);
>  void ff_audiodsp_init_arm(AudioDSPContext *c);
>  void ff_audiodsp_init_ppc(AudioDSPContext *c);
> +void ff_audiodsp_init_riscv(AudioDSPContext *c);
>  void ff_audiodsp_init_x86(AudioDSPContext *c);
> 
>  #endif /* AVCODEC_AUDIODSP_H */
> diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
> new file mode 100644
> index 0000000000..a1f67ed55b
> --- /dev/null
> +++ b/libavcodec/riscv/Makefile
> @@ -0,0 +1,2 @@
> +OBJS += riscv/audiodsp_init.o \
> +        riscv/audiodsp_rvf.o
> diff --git a/libavcodec/riscv/audiodsp_init.c
> b/libavcodec/riscv/audiodsp_init.c new file mode 100644
> index 0000000000..7ffd7e8162
> --- /dev/null
> +++ b/libavcodec/riscv/audiodsp_init.c
> @@ -0,0 +1,31 @@
> +/*
> + * 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/attributes.h"
> +#include "libavutil/cpu.h"
> +#include "libavcodec/audiodsp.h"
> +
> +void ff_vector_clipf_rvf(float *dst, const float *src, int len, float min,
> float max); +
> +av_cold void ff_audiodsp_init_riscv(AudioDSPContext *c)
> +{
> +    int flags = av_get_cpu_flags();
> +
> +    if (flags & AV_CPU_FLAG_F)
> +        c->vector_clipf = ff_vector_clipf_rvf;
> +}
> diff --git a/libavcodec/riscv/audiodsp_rvf.S
> b/libavcodec/riscv/audiodsp_rvf.S new file mode 100644
> index 0000000000..148af96ea2
> --- /dev/null
> +++ b/libavcodec/riscv/audiodsp_rvf.S
> @@ -0,0 +1,46 @@
> +/*
> + * 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_vector_clipf_rvf, f
> +NOHWF   fmv.w.x fa0, a3
> +NOHWF   fmv.w.v fa1, a4

This should be fmv.w.x, though I don't know a distro that would use the soft 
float ABI. Need more sleep.

> +1:
> +        flw     ft0,   (a1)
> +        flw     ft1,  4(a1)
> +        fmax.s  ft0, ft0, fa0
> +        flw     ft2,  8(a1)
> +        fmax.s  ft1, ft1, fa0
> +        flw     ft3, 12(a1)
> +        fmax.s  ft2, ft2, fa0
> +        addi    a2, a2, -4
> +        fmax.s  ft3, ft3, fa0
> +        addi    a1, a1, 16
> +        fmin.s  ft0, ft0, fa1
> +        fmin.s  ft1, ft1, fa1
> +        fsw     ft0,   (a0)
> +        fmin.s  ft2, ft2, fa1
> +        fsw     ft1,  4(a0)
> +        fmin.s  ft3, ft3, fa1
> +        fsw     ft2,  8(a0)
> +        fsw     ft3, 12(a0)
> +        addi    a0, a0, 16
> +        bnez    a2, 1b
> +        ret
> +endfunc


-- 
Реми Дёни-Курмон
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".