2013-05-24 20:24:16 +00:00
|
|
|
PROGNAME=dump1090
|
|
|
|
|
|
2020-08-07 06:28:00 +00:00
|
|
|
DUMP1090_VERSION ?= unknown
|
2020-08-03 06:57:31 +00:00
|
|
|
|
2021-12-07 07:12:49 +00:00
|
|
|
CFLAGS ?= -O3 -g
|
|
|
|
|
DUMP1090_CFLAGS := -std=c11 -fno-common -Wall -Wmissing-declarations -Werror -W
|
2022-01-12 17:46:41 +00:00
|
|
|
DUMP1090_CPPFLAGS := -I. -D_POSIX_C_SOURCE=200112L -DMODES_DUMP1090_VERSION=\"$(DUMP1090_VERSION)\" -DMODES_DUMP1090_VARIANT=\"dump1090-fa\"
|
2015-01-22 13:19:15 +00:00
|
|
|
|
2019-03-31 21:01:06 +00:00
|
|
|
LIBS = -lpthread -lm
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
SDR_OBJ = cpu.o sdr.o fifo.o sdr_ifile.o dsp/helpers/tables.o
|
2020-08-07 06:28:00 +00:00
|
|
|
|
|
|
|
|
# Try to autodetect available libraries via pkg-config if no explicit setting was used
|
|
|
|
|
PKGCONFIG=$(shell pkg-config --version >/dev/null 2>&1 && echo "yes" || echo "no")
|
|
|
|
|
ifeq ($(PKGCONFIG), yes)
|
|
|
|
|
ifndef RTLSDR
|
|
|
|
|
ifdef RTLSDR_PREFIX
|
|
|
|
|
RTLSDR := yes
|
|
|
|
|
else
|
|
|
|
|
RTLSDR := $(shell pkg-config --exists librtlsdr && echo "yes" || echo "no")
|
|
|
|
|
endif
|
|
|
|
|
endif
|
|
|
|
|
|
|
|
|
|
ifndef BLADERF
|
|
|
|
|
BLADERF := $(shell pkg-config --exists libbladeRF && echo "yes" || echo "no")
|
|
|
|
|
endif
|
|
|
|
|
|
|
|
|
|
ifndef HACKRF
|
|
|
|
|
HACKRF := $(shell pkg-config --exists libhackrf && echo "yes" || echo "no")
|
2020-08-03 06:57:31 +00:00
|
|
|
endif
|
|
|
|
|
|
2020-08-07 06:28:00 +00:00
|
|
|
ifndef LIMESDR
|
|
|
|
|
LIMESDR := $(shell pkg-config --exists LimeSuite && echo "yes" || echo "no")
|
|
|
|
|
endif
|
|
|
|
|
else
|
|
|
|
|
# pkg-config not available. Only use explicitly enabled libraries.
|
|
|
|
|
RTLSDR ?= no
|
|
|
|
|
BLADERF ?= no
|
|
|
|
|
HACKRF ?= no
|
|
|
|
|
LIMESDR ?= no
|
2020-08-03 06:57:31 +00:00
|
|
|
endif
|
|
|
|
|
|
2021-09-23 05:23:17 +00:00
|
|
|
HOST_UNAME := $(shell uname)
|
|
|
|
|
HOST_ARCH := $(shell uname -m)
|
|
|
|
|
|
|
|
|
|
UNAME ?= $(HOST_UNAME)
|
|
|
|
|
ARCH ?= $(HOST_ARCH)
|
2019-03-31 21:01:06 +00:00
|
|
|
|
|
|
|
|
ifeq ($(UNAME), Linux)
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -D_DEFAULT_SOURCE
|
2019-03-31 23:57:48 +00:00
|
|
|
LIBS += -lrt
|
|
|
|
|
LIBS_USB += -lusb-1.0
|
2021-02-10 04:44:30 +00:00
|
|
|
LIBS_CURSES := -lncurses
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
CPUFEATURES ?= yes
|
2020-08-03 06:57:31 +00:00
|
|
|
endif
|
2017-01-30 21:55:20 +00:00
|
|
|
|
2019-03-31 21:01:06 +00:00
|
|
|
ifeq ($(UNAME), Darwin)
|
|
|
|
|
ifneq ($(shell sw_vers -productVersion | egrep '^10\.([0-9]|1[01])\.'),) # Mac OS X ver <= 10.11
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DMISSING_GETTIME
|
2019-03-31 21:01:06 +00:00
|
|
|
COMPAT += compat/clock_gettime/clock_gettime.o
|
|
|
|
|
endif
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DMISSING_NANOSLEEP
|
2019-03-31 21:01:06 +00:00
|
|
|
COMPAT += compat/clock_nanosleep/clock_nanosleep.o
|
2019-03-31 23:57:48 +00:00
|
|
|
LIBS_USB += -lusb-1.0
|
2021-02-10 04:44:30 +00:00
|
|
|
LIBS_CURSES := -lncurses
|
2021-12-07 07:18:58 +00:00
|
|
|
# cpufeatures reportedly does not work (yet) on darwin arm64
|
2022-01-14 18:23:30 +00:00
|
|
|
ifneq ($(ARCH),arm64)
|
2021-12-07 07:18:58 +00:00
|
|
|
CPUFEATURES ?= yes
|
|
|
|
|
endif
|
2020-08-03 07:16:32 +00:00
|
|
|
endif
|
2017-01-30 21:55:20 +00:00
|
|
|
|
2019-03-31 21:01:06 +00:00
|
|
|
ifeq ($(UNAME), OpenBSD)
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DMISSING_NANOSLEEP
|
2019-03-31 21:01:06 +00:00
|
|
|
COMPAT += compat/clock_nanosleep/clock_nanosleep.o
|
2019-03-31 23:57:48 +00:00
|
|
|
LIBS_USB += -lusb-1.0
|
2021-02-10 04:44:30 +00:00
|
|
|
LIBS_CURSES := -lncurses
|
2019-03-31 23:57:48 +00:00
|
|
|
endif
|
2020-08-05 07:02:33 +00:00
|
|
|
|
2019-03-31 23:57:48 +00:00
|
|
|
ifeq ($(UNAME), FreeBSD)
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -D_DEFAULT_SOURCE
|
2019-03-31 23:57:48 +00:00
|
|
|
LIBS += -lrt
|
|
|
|
|
LIBS_USB += -lusb
|
2021-02-10 04:44:30 +00:00
|
|
|
LIBS_CURSES := -lncurses
|
|
|
|
|
endif
|
|
|
|
|
|
|
|
|
|
ifeq ($(UNAME), NetBSD)
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -D_DEFAULT_SOURCE
|
2021-02-10 04:44:30 +00:00
|
|
|
LIBS += -lrt
|
|
|
|
|
LIBS_USB += -lusb-1.0
|
|
|
|
|
LIBS_CURSES := -lcurses
|
2019-03-31 21:01:06 +00:00
|
|
|
endif
|
2015-01-22 13:19:15 +00:00
|
|
|
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
CPUFEATURES ?= no
|
|
|
|
|
|
|
|
|
|
ifeq ($(CPUFEATURES),yes)
|
|
|
|
|
include Makefile.cpufeatures
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DENABLE_CPUFEATURES -Icpu_features/include
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
endif
|
|
|
|
|
|
2019-03-31 23:57:48 +00:00
|
|
|
RTLSDR ?= yes
|
|
|
|
|
BLADERF ?= yes
|
2015-09-03 03:41:00 +00:00
|
|
|
|
2017-01-30 21:55:20 +00:00
|
|
|
ifeq ($(RTLSDR), yes)
|
|
|
|
|
SDR_OBJ += sdr_rtlsdr.o
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DENABLE_RTLSDR
|
2017-01-30 21:55:20 +00:00
|
|
|
|
|
|
|
|
ifdef RTLSDR_PREFIX
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -I$(RTLSDR_PREFIX)/include
|
2019-04-17 03:58:59 +00:00
|
|
|
ifeq ($(STATIC), yes)
|
2020-08-07 06:28:00 +00:00
|
|
|
LIBS_SDR += -L$(RTLSDR_PREFIX)/lib -Wl,-Bstatic -lrtlsdr -Wl,-Bdynamic $(LIBS_USB)
|
2019-04-17 03:58:59 +00:00
|
|
|
else
|
2020-08-07 06:28:00 +00:00
|
|
|
LIBS_SDR += -L$(RTLSDR_PREFIX)/lib -lrtlsdr $(LIBS_USB)
|
2019-04-17 03:58:59 +00:00
|
|
|
endif
|
2017-01-30 21:55:20 +00:00
|
|
|
else
|
2019-04-17 03:58:59 +00:00
|
|
|
# some packaged .pc files are massively broken, try to handle it
|
2020-08-07 06:43:44 +00:00
|
|
|
|
|
|
|
|
# FreeBSD's librtlsdr.pc includes -std=gnu89 in cflags
|
2021-07-26 12:25:35 +00:00
|
|
|
# some linux librtlsdr packages return a bare -I/ with no path in --cflags
|
2020-08-07 06:43:44 +00:00
|
|
|
RTLSDR_CFLAGS := $(shell pkg-config --cflags librtlsdr)
|
2021-07-26 12:25:35 +00:00
|
|
|
RTLSDR_CFLAGS := $(filter-out -std=%,$(RTLSDR_CFLAGS))
|
|
|
|
|
RTLSDR_CFLAGS := $(filter-out -I/,$(RTLSDR_CFLAGS))
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CFLAGS += $(RTLSDR_CFLAGS)
|
2020-08-07 06:43:44 +00:00
|
|
|
|
|
|
|
|
# some linux librtlsdr packages return a bare -L with no path in --libs
|
|
|
|
|
# which horribly confuses things because it eats the next option on the command line
|
2019-04-17 03:58:59 +00:00
|
|
|
RTLSDR_LFLAGS := $(shell pkg-config --libs-only-L librtlsdr)
|
|
|
|
|
ifeq ($(RTLSDR_LFLAGS),-L)
|
|
|
|
|
LIBS_SDR += $(shell pkg-config --libs-only-l --libs-only-other librtlsdr)
|
|
|
|
|
else
|
|
|
|
|
LIBS_SDR += $(shell pkg-config --libs librtlsdr)
|
|
|
|
|
endif
|
2017-01-30 21:55:20 +00:00
|
|
|
endif
|
2015-09-03 03:41:00 +00:00
|
|
|
endif
|
|
|
|
|
|
2017-01-30 21:55:20 +00:00
|
|
|
ifeq ($(BLADERF), yes)
|
|
|
|
|
SDR_OBJ += sdr_bladerf.o
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DENABLE_BLADERF
|
|
|
|
|
DUMP1090_CFLAGS += $(shell pkg-config --cflags libbladeRF)
|
2017-01-30 21:55:20 +00:00
|
|
|
LIBS_SDR += $(shell pkg-config --libs libbladeRF)
|
|
|
|
|
endif
|
2017-01-27 17:44:42 +00:00
|
|
|
|
2019-12-09 21:55:40 +00:00
|
|
|
ifeq ($(HACKRF), yes)
|
|
|
|
|
SDR_OBJ += sdr_hackrf.o
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DENABLE_HACKRF
|
|
|
|
|
DUMP1090_CFLAGS += $(shell pkg-config --cflags libhackrf)
|
2019-12-09 21:55:40 +00:00
|
|
|
LIBS_SDR += $(shell pkg-config --libs libhackrf)
|
|
|
|
|
endif
|
|
|
|
|
|
2020-07-26 13:52:48 +00:00
|
|
|
ifeq ($(LIMESDR), yes)
|
|
|
|
|
SDR_OBJ += sdr_limesdr.o
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DENABLE_LIMESDR
|
|
|
|
|
DUMP1090_CFLAGS += $(shell pkg-config --cflags LimeSuite)
|
2020-07-26 13:52:48 +00:00
|
|
|
LIBS_SDR += $(shell pkg-config --libs LimeSuite)
|
|
|
|
|
endif
|
|
|
|
|
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
|
|
|
|
|
##
|
|
|
|
|
## starch (runtime DSP code selection) mix, architecture-specific
|
|
|
|
|
##
|
|
|
|
|
|
|
|
|
|
ifneq ($(CPUFEATURES),yes)
|
|
|
|
|
# need to be able to detect CPU features at runtime to enable any non-standard compiler flags
|
|
|
|
|
STARCH_MIX := generic
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DSTARCH_MIX_GENERIC
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
else
|
2021-02-08 00:07:17 +00:00
|
|
|
ifeq ($(ARCH),x86_64)
|
|
|
|
|
# AVX, AVX2
|
|
|
|
|
STARCH_MIX := x86
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DSTARCH_MIX_X86
|
2021-12-07 05:16:20 +00:00
|
|
|
else ifeq ($(findstring aarch,$(ARCH)),aarch)
|
|
|
|
|
STARCH_MIX := aarch64
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DSTARCH_MIX_AARCH64
|
2021-12-07 05:16:20 +00:00
|
|
|
else ifeq ($(findstring arm64,$(ARCH)),arm64)
|
|
|
|
|
# Apple calls this arm64, not aarch64
|
|
|
|
|
STARCH_MIX := aarch64
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DSTARCH_MIX_AARCH64
|
2021-02-08 00:07:17 +00:00
|
|
|
else ifeq ($(findstring arm,$(ARCH)),arm)
|
|
|
|
|
# ARMv7 NEON
|
|
|
|
|
STARCH_MIX := arm
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DSTARCH_MIX_ARM
|
2021-02-08 00:07:17 +00:00
|
|
|
else
|
|
|
|
|
STARCH_MIX := generic
|
2021-12-07 07:12:49 +00:00
|
|
|
DUMP1090_CPPFLAGS += -DSTARCH_MIX_GENERIC
|
2021-02-08 00:07:17 +00:00
|
|
|
endif
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
endif
|
|
|
|
|
all: showconfig dump1090 view1090 starch-benchmark
|
|
|
|
|
|
2021-12-07 07:12:49 +00:00
|
|
|
ALL_CCFLAGS := $(CPPFLAGS) $(DUMP1090_CPPFLAGS) $(CFLAGS) $(DUMP1090_CFLAGS)
|
|
|
|
|
|
|
|
|
|
STARCH_COMPILE := $(CC) $(ALL_CCFLAGS) -c
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
include dsp/generated/makefile.$(STARCH_MIX)
|
2020-08-03 06:57:31 +00:00
|
|
|
|
|
|
|
|
showconfig:
|
|
|
|
|
@echo "Building with:" >&2
|
|
|
|
|
@echo " Version string: $(DUMP1090_VERSION)" >&2
|
2021-09-23 05:23:17 +00:00
|
|
|
@echo " Architecture: $(ARCH)" >&2
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
@echo " DSP mix: $(STARCH_MIX)" >&2
|
2020-08-03 06:57:31 +00:00
|
|
|
@echo " RTLSDR support: $(RTLSDR)" >&2
|
|
|
|
|
@echo " BladeRF support: $(BLADERF)" >&2
|
|
|
|
|
@echo " HackRF support: $(HACKRF)" >&2
|
2020-08-03 07:16:32 +00:00
|
|
|
@echo " LimeSDR support: $(LIMESDR)" >&2
|
|
|
|
|
|
2015-02-08 17:58:46 +00:00
|
|
|
%.o: %.c *.h
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -c $< -o $@
|
2015-06-26 16:49:21 +00:00
|
|
|
|
2021-06-29 12:11:13 +00:00
|
|
|
dump1090: dump1090.o anet.o interactive.o mode_ac.o mode_s.o comm_b.o net_io.o crc.o demod_2400.o stats.o cpr.o icao_filter.o track.o util.o convert.o ais_charset.o adaptive.o $(SDR_OBJ) $(COMPAT) $(CPUFEATURES_OBJS) $(STARCH_OBJS)
|
2021-02-10 04:44:30 +00:00
|
|
|
$(CC) -g -o $@ $^ $(LDFLAGS) $(LIBS) $(LIBS_SDR) $(LIBS_CURSES)
|
2013-01-05 12:52:25 +00:00
|
|
|
|
2021-06-29 12:11:13 +00:00
|
|
|
view1090: view1090.o anet.o interactive.o mode_ac.o mode_s.o comm_b.o net_io.o crc.o stats.o cpr.o icao_filter.o track.o util.o ais_charset.o sdr_stub.o $(COMPAT)
|
2021-02-10 04:44:30 +00:00
|
|
|
$(CC) -g -o $@ $^ $(LDFLAGS) $(LIBS) $(LIBS_CURSES)
|
Implement a remote interactive screen
No changes to dump1090, (except the version number)
Include a sample Linux batch start file called dump1090.sh for use when
running dump1090 headless. This file needs to be copied to the
/etc/init.d/ subdirectory on your raspberry pi, and marked as
executable. Then when you re-start your RPi, dump1090 will start-up
auto-magically and run as a sort of server to allow both local and
remote connection to it's various internet ports.
Modified the Makefile to build a new headless helper application called
view1090
Added view1090. This is an executable that allows you to connect to
dump1090 when it is running and 'see' the interactive screen display.
The default is to try and connect to dump1090 on IP address 127.0.0.1
port 30005. This should work if you are running on the same RPi as
dump1090 and using the default dump1090 port settings. However, if
you're running on a different machine you will have to specify the IP
address of the RPi running dump1090 using the --net-bo-ipaddr switch.
Something like "view1090 --net-bo-ipaddr 192.168.2.65" . You may also
have to sudo it, depending on your privilige settings.
I've also compiled view1090 as a Wiin32 exe, so you should be able to
run it under any 32 bit version of Microsoft Windows - i.e. Win95, Win
2K, Win XP, Win 7 etc. It may work on Win 8 and 64 bit Windows, but I
haven't tried it. The Win32 version is compiled from the same source, so
takes all the same command line switches.
2013-09-24 17:37:54 +00:00
|
|
|
|
2021-06-29 12:11:13 +00:00
|
|
|
faup1090: faup1090.o anet.o mode_ac.o mode_s.o comm_b.o net_io.o crc.o stats.o cpr.o icao_filter.o track.o util.o ais_charset.o sdr_stub.o $(COMPAT)
|
2016-09-16 16:16:28 +00:00
|
|
|
$(CC) -g -o $@ $^ $(LDFLAGS) $(LIBS)
|
2015-06-26 16:54:23 +00:00
|
|
|
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
starch-benchmark: cpu.o dsp/helpers/tables.o $(CPUFEATURES_OBJS) $(STARCH_OBJS) $(STARCH_BENCHMARK_OBJ)
|
|
|
|
|
$(CC) -g -o $@ $^ $(LDFLAGS) $(LIBS)
|
|
|
|
|
|
2013-01-05 12:52:25 +00:00
|
|
|
clean:
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
rm -f *.o oneoff/*.o compat/clock_gettime/*.o compat/clock_nanosleep/*.o cpu_features/src/*.o dsp/generated/*.o dsp/helpers/*.o $(CPUFEATURES_OBJS) dump1090 view1090 faup1090 cprtests crctests oneoff/convert_benchmark oneoff/decode_comm_b oneoff/dsp_error_measurement oneoff/uc8_capture_stats starch-benchmark
|
2015-01-20 18:41:44 +00:00
|
|
|
|
|
|
|
|
test: cprtests
|
2015-02-19 12:03:08 +00:00
|
|
|
./cprtests
|
2015-01-20 18:41:44 +00:00
|
|
|
|
|
|
|
|
cprtests: cpr.o cprtests.o
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -g -o $@ $^ -lm
|
2015-01-20 23:04:05 +00:00
|
|
|
|
|
|
|
|
crctests: crc.c crc.h
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -g -DCRCDEBUG -o $@ $<
|
2017-01-29 17:15:26 +00:00
|
|
|
|
2020-08-04 12:38:03 +00:00
|
|
|
benchmarks: oneoff/convert_benchmark
|
|
|
|
|
oneoff/convert_benchmark
|
2017-01-29 17:15:26 +00:00
|
|
|
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
oneoff/convert_benchmark: oneoff/convert_benchmark.o convert.o util.o dsp/helpers/tables.o cpu.o $(CPUFEATURES_OBJS) $(STARCH_OBJS)
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -g -o $@ $^ -lm -lpthread
|
2019-05-02 09:29:46 +00:00
|
|
|
|
|
|
|
|
oneoff/decode_comm_b: oneoff/decode_comm_b.o comm_b.o ais_charset.o
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -g -o $@ $^ -lm
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
|
|
|
|
|
oneoff/dsp_error_measurement: oneoff/dsp_error_measurement.o dsp/helpers/tables.o cpu.o $(CPUFEATURES_OBJS) $(STARCH_OBJS)
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -g -o $@ $^ -lm
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
|
|
|
|
|
oneoff/uc8_capture_stats: oneoff/uc8_capture_stats.o
|
2021-12-07 07:12:49 +00:00
|
|
|
$(CC) $(ALL_CCFLAGS) -g -o $@ $^ -lm
|
Move all converters to starch-based implementations (#97)
* Switch all conversion routines to use starch.
main user-visible changes:
* ensure you check out submodules ('git clone --recurse-submodules")
* --version shows the CPU features and DSP implementations in use
* --wisdom allows overriding of the built-in architecture wisdom
* --dcfilter no longer supported
* "starch-benchmark" binary will benchmark all options on the
current machine and can produce a wisdom file to feed to
the --wisdom option
If you have a usecase for --dcfilter, please get in touch and
let me know - it's an edge case and for now there's no starch/DSP
support for it, but support can be written if needed.
In almost all cases the new conversion routines are slightly or
substantially faster than the old conversion routines. The only case
that is slower is SC16/SC16Q11 on a Pi 0, which is around 10% slower
due to changing from heavily approximated lookup tables to higher
quality results (but SC16 is probably already out of reach of a Pi 0)
* No need to build with SC16Q11_TABLE_BITS any more
* Add oneoff/uc8_capture_stats
(reads a UC8 capture; measures min/max/mean I and Q)
* Switch UC8 conversion to 127.4 center, 128 range.
Looking at actual UC8 captures from a RTL2832, the mean I and Q
are actually at 127.4, so use that as the zero point.
This means that the resulting I/Q maximum values could be as large as
127.6. Switch to 128 for simplicity.
* Switch to the new UC8 zero offset in benchmarks, fix some bugs
* Fix some bugs in SC16/SC16Q11 validation, tighten the max error requirements
* Ditch UC8 approximation path, add a NEON VRQSQRTE path.
* Tweak the SC16 exact path, add a new impl that uses a mix of
u32 & floats.
* SC16Q11 impl tweaks:
* add a u32->float exact path
* ditch the approximation path
* add a NEON VRSQRTE path
* add a 12-bit table path (using the full signed I/Q value, not absolute value)
* Ditch SC16 approximation path, add NEON vrsqrte path
* Add oneoff/dsp_error_measurement
This runs sample input through the DSP functions that are
allowed to be inexact and dumps the results as a TSV suitable for
feeding to gnuplot to look at the actual errors.
* Update make clean, make wisdom targets
* Update wisdom based on benchmarking
* Preserve the raw wisdom benchmark data
* Update to latest starch
* Update .gitignore for new wisdom files
* Update starch generated code
* Build starch-benchmark as part of the 'all' target
* Use wisdom from /etc/dump1090-fa/wisdom.local if present
* Package starch-benchmark and a helper script to generate local wisdom data
* Remove submodules in preparation for importing them directly
* Import cpu_features v0.6.0 from https://github.com/google/cpu_features/releases/tag/v0.6.0
* Import starch at commit a725c8491dc33a321565d451b385131e589d8490
from https://github.com/flightaware/starch
2021-01-21 11:45:00 +00:00
|
|
|
|
|
|
|
|
starchgen:
|
|
|
|
|
dsp/starchgen.py .
|
|
|
|
|
|
|
|
|
|
.PHONY: wisdom.local
|
|
|
|
|
wisdom.local: starch-benchmark
|
2021-03-08 07:18:35 +00:00
|
|
|
./starch-benchmark -i 5 -o wisdom.local mean_power_u16 mean_power_u16_aligned magnitude_uc8 magnitude_uc8_aligned
|
|
|
|
|
./starch-benchmark -i 5 -r wisdom.local -o wisdom.local
|
