On Fri, Dec 1, 2017 at 3:49 PM, Khem Raj <raj.k...@gmail.com> wrote: > On Fri, Dec 1, 2017 at 3:08 AM, Andreas Müller <schnitzelt...@gmail.com> > wrote: > > * Use floats instead of double for sound calculations. This improves > > performance notable and was default for version 1.1.6 using autotools. > > * Fix buffer overrun when using floats > > * Make use of ARM NEON for multithreading enabled > > > > Performance and sound correctnes was tested with qtractor and a reworked > > version of fluidsynth-dssi [1-2]. Tests were performed for single- and > > multithreading enabled. > > > > [1] https://github.com/schnitzeltony/fluidsynth-dssi/commit/ > bad09c6f5c5508c5f5330aa5188510f975e50c50 > > [2] https://github.com/schnitzeltony/meta-qt5-extra/ > blob/master/recipes-misc/recipes-multimedia/fluidsynth/ > fluidsynth-dssi_1.0.0.bb > > > > Signed-off-by: Andreas Müller <schnitzelt...@gmail.com> > > --- > > ...uffer-overrun-in-fluid_synth_nwrite_float.patch | 32 +++++++++ > > ...N-accelaration-for-float-multithreaded-se.patch | 76 > ++++++++++++++++++++++ > > .../fluidsynth/fluidsynth_1.1.8.bb | 8 ++- > > 3 files changed, 114 insertions(+), 2 deletions(-) > > create mode 100644 meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0001-avoid-buffer-overrun-in-fluid_synth_nwrite_float.patch > > create mode 100644 meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0002-Use-ARM-NEON-accelaration-for-float-multithreaded-se.patch > > > > diff --git a/meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0001-avoid-buffer-overrun-in-fluid_synth_nwrite_float.patch > b/meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0001-avoid-buffer-overrun-in-fluid_synth_nwrite_float.patch > > new file mode 100644 > > index 0000000..dda76cf > > --- /dev/null > > +++ b/meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0001-avoid-buffer-overrun-in-fluid_synth_nwrite_float.patch > > @@ -0,0 +1,32 @@ > > +From a13cb63103aa56b5e8bad816c7d13d6e01c0cd9f Mon Sep 17 00:00:00 2001 > > +From: derselbst <tom.m...@googlemail.com> > > +Date: Sun, 26 Nov 2017 22:12:12 +0100 > > +Subject: [PATCH 1/2] avoid buffer overrun in fluid_synth_nwrite_float() > > + > > +Upstream-Status: Backport [1] > > + > > +[1] https://github.com/FluidSynth/fluidsynth/commit/ > a13cb63103aa56b5e8bad816c7d13d6e01c0cd9f > > +--- > > + src/synth/fluid_synth.c | 4 ++-- > > + 1 file changed, 2 insertions(+), 2 deletions(-) > > + > > +diff --git a/src/synth/fluid_synth.c b/src/synth/fluid_synth.c > > +index 266d759..14f6b21 100644 > > +--- a/src/synth/fluid_synth.c > > ++++ b/src/synth/fluid_synth.c > > +@@ -2752,10 +2752,10 @@ fluid_synth_nwrite_float(fluid_synth_t* synth, > int len, > > + { > > + #ifdef WITH_FLOAT > > + if(fx_left != NULL) > > +- FLUID_MEMCPY(fx_left[i + count], fx_left_in[i], bytes); > > ++ FLUID_MEMCPY(fx_left[i] + count, fx_left_in[i], bytes); > > + > > + if(fx_right != NULL) > > +- FLUID_MEMCPY(fx_right[i + count], fx_right_in[i], bytes); > > ++ FLUID_MEMCPY(fx_right[i] + count, fx_right_in[i], bytes); > > + #else //WITH_FLOAT > > + int j; > > + if(fx_left != NULL) { > > +-- > > +2.9.5 > > + > > diff --git a/meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0002-Use-ARM-NEON-accelaration-for-float-multithreaded-se.patch > b/meta-multimedia/recipes-multimedia/fluidsynth/files/0002-Use-ARM-NEON- > accelaration-for-float-multithreaded-se.patch > > new file mode 100644 > > index 0000000..0e1846e > > --- /dev/null > > +++ b/meta-multimedia/recipes-multimedia/fluidsynth/files/ > 0002-Use-ARM-NEON-accelaration-for-float-multithreaded-se.patch > > @@ -0,0 +1,76 @@ > > +From 2de7e128fbdf528716b500cf27ed9a4358c931c9 Mon Sep 17 00:00:00 2001 > > +From: =?UTF-8?q?Andreas=20M=C3=BCller?= <schnitzelt...@gmail.com> > > +Date: Fri, 24 Nov 2017 00:05:35 +0100 > > +Subject: [PATCH 2/2] Use ARM-NEON accelaration for float-multithreaded > setups > > +MIME-Version: 1.0 > > +Content-Type: text/plain; charset=UTF-8 > > +Content-Transfer-Encoding: 8bit > > + > > +Upstream-Status: Pending > > + > > +Signed-off-by: Andreas Müller <schnitzelt...@gmail.com> > > +--- > > + src/rvoice/fluid_rvoice_mixer.c | 26 ++++++++++++++++++++++++++ > > + 1 file changed, 26 insertions(+) > > + > > +diff --git a/src/rvoice/fluid_rvoice_mixer.c b/src/rvoice/fluid_rvoice_ > mixer.c > > +index 9616518..dbf8057 100644 > > +--- a/src/rvoice/fluid_rvoice_mixer.c > > ++++ b/src/rvoice/fluid_rvoice_mixer.c > > +@@ -27,6 +27,10 @@ > > + #include "fluid_ladspa.h" > > + #include "fluid_synth.h" > > + > > ++#if defined(__ARM_NEON__) > > ++#include "arm_neon.h" > > ++#endif > > ++ > > + > > + #define ENABLE_MIXER_THREADS 1 > > + > > +@@ -794,20 +798,42 @@ fluid_mixer_buffers_mix(fluid_mixer_buffers_t* > dest, fluid_mixer_buffers_t* src) > > + if (minbuf > src->buf_count) > > + minbuf = src->buf_count; > > + for (i=0; i < minbuf; i++) { > > ++#if defined(__ARM_NEON__) && defined(WITH_FLOAT) > > ++ for (j=0; j < scount; j+=4) { > > ++ float32x4_t vleft = vld1q_f32(&dest->left_buf[i][j]); > > ++ float32x4_t vright = vld1q_f32(&dest->right_buf[i][j]); > > ++ vleft = vaddq_f32(vleft, vld1q_f32(&src->left_buf[i][j])); > > ++ vright = vaddq_f32(vright, vld1q_f32(&src->right_buf[i][j])); > > ++ vst1q_f32(&dest->left_buf[i][j], vleft); > > ++ vst1q_f32(&dest->right_buf[i][j], vright); > > ++ } > > > I wonder if gcc could be initiated to see if it can vectorize the code, > here > > have you tried setting cmdline options or may be pragma hints > > I've sent patches like these to some projects. If I remember correctly first one was jack2 [1]. For that I wrote some performance tests [2] and tested vectorizing either but with disappointing results. This is ~ a year ago and I think intrinsics are still the option performing better.
[1] https://github.com/jackaudio/jack2/commit/77bb8be12e0d856fbc004cf57185ab36a2df04c2 [2] https://github.com/jackaudio/jack2/commit/c32b823860bc8e887774ab10bc3e9bd76e85e3f3 Andreas -- _______________________________________________ Openembedded-devel mailing list Openembedded-devel@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-devel