1
1
mirror of https://github.com/OpenVoiceOS/OpenVoiceOS synced 2024-12-12 08:56:25 +01:00

Extend the whole TensorFlow-Lite runtime infrastructure

- TFlite-runtime GPU delegate support
- TFlite-runtime external delegate support
- ARM Compute Library and ARMnn Delegate
- Whisper-TFlite updates.
This commit is contained in:
j1nx 2023-01-25 14:22:22 +01:00
parent 5e277d5e24
commit 876ee82daa
16 changed files with 226 additions and 53 deletions

View File

@ -1,7 +1,9 @@
menu "Additional drivers, libraries and/or applications"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/armnn/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/baresip/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/btspeaker/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/clog/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/compute-library/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/cpuinfo/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/deepfilternet-ladspa/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/docbook-xml/Config.in"
@ -98,6 +100,7 @@ endmenu
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/lottie-qml/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/ncpamixer/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/neon2sse/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/opencl_headers/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/openfec/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/openfst/Config.in"
source "$BR2_EXTERNAL_OPENVOICEOS_PATH/package/ovos-bus-server/Config.in"

View File

@ -543,8 +543,10 @@ BR2_PACKAGE_HOST_PYTHON_SIX=y
BR2_PACKAGE_HOST_PYTHON_XLRD=y
BR2_PACKAGE_HOST_RUSTC=y
BR2_PACKAGE_HOST_RUST=y
BR2_PACKAGE_ARMNN=y
BR2_PACKAGE_BARESIP=y
BR2_PACKAGE_BTSPEAKER=y
BR2_PACKAGE_COMPUTE_LIBRARY=y
BR2_PACKAGE_DEEPFILTERNET_LADSPA=y
BR2_PACKAGE_FANN=y
BR2_PACKAGE_GROWDISK_SERVICE=y

View File

@ -543,8 +543,10 @@ BR2_PACKAGE_HOST_PYTHON_SIX=y
BR2_PACKAGE_HOST_PYTHON_XLRD=y
BR2_PACKAGE_HOST_RUSTC=y
BR2_PACKAGE_HOST_RUST=y
BR2_PACKAGE_ARMNN=y
BR2_PACKAGE_BARESIP=y
BR2_PACKAGE_BTSPEAKER=y
BR2_PACKAGE_COMPUTE_LIBRARY=y
BR2_PACKAGE_DEEPFILTERNET_LADSPA=y
BR2_PACKAGE_FANN=y
BR2_PACKAGE_GROWDISK_SERVICE=y

View File

@ -0,0 +1,11 @@
config BR2_PACKAGE_ARMNN
select BR2_PACKAGE_COMPUTE_LIBRARY
select BR2_PACKAGE_TENSORFLOW_LITE
bool "armnn"
help
Arm NN is the most performant machine learning (ML)
inference engine for Android and Linux, accelerating
ML on Arm Cortex-A CPUs and Arm Mali GPUs.
https://github.com/ARM-software/armnn

View File

@ -0,0 +1,34 @@
################################################################################
#
# armnn
#
################################################################################
ARMNN_VERSION = v22.11.01
ARMNN_SITE = $(call github,ARM-software,armnn,$(ARMNN_VERSION))
ARMNN_LICENSE = MIT
ARMNN_INSTALL_STAGING = YES
ARMNN_DEPENDENCIES = host-pkgconf compute-library tensorflow-lite
ARMNN_SUPPORTS_IN_SOURCE_BUILD = NO
ARMNN_CONF_OPTS = \
-DCMAKE_C_FLAGS="$(TARGET_CFLAGS) -fPIC -Wno-error=missing-field-initializers" \
-DCMAKE_CXX_FLAGS="$(TARGET_CXXFLAGS) -fPIC -Wno-error=missing-field-initializers" \
-DCMAKE_SHARED_LINKER_FLAGS="$(TARGET_LDFLAGS) -fPIC -Wno-error=missing-field-initializers" \
-DCMAKE_POSITION_INDEPENDENT_CODE=ON \
-DBUILD_SHARED_LIBS=ON \
-DBUILD_TF_LITE_PARSER=1 \
-DBUILD_ARMNN_TFLITE_DELEGATE=ON \
-DARMCOMPUTENEON=ON \
-DARMCOMPUTECL=1 \
-DARMNNREF=1 \
-DARMCOMPUTE_ROOT="$(BUILD_DIR)/compute-library-$(COMPUTE_LIBRARY_VERSION)" \
-DARMCOMPUTE_BUILD_DIR="$(BUILD_DIR)/compute-library-$(COMPUTE_LIBRARY_VERSION)/buildroot-build" \
-DTENSORFLOW_ROOT="$(BUILD_DIR)/tensorflow-lite-$(TENSORFLOW_LITE_VERSION)" \
-DTF_LITE_SCHEMA_INCLUDE_PATH="$(BUILD_DIR)/tensorflow-lite-$(TENSORFLOW_LITE_VERSION)/tensorflow/lite/schema" \
-DTFLITE_LIB_ROOT="$(STAGING_DIR)/usr/lib" \
-DTfLite_Schema_INCLUDE_PATH="$(BUILD_DIR)/tensorflow-lite-$(TENSORFLOW_LITE_VERSION)/tensorflow/lite/schema" \
-DTfLite_LIB="$(STAGING_DIR)/usr/lib/libtensorflow-lite.so"
$(eval $(cmake-package))

View File

@ -0,0 +1,9 @@
config BR2_PACKAGE_COMPUTE_LIBRARY
bool "compute-library"
help
The Compute Library is a set of computer vision and
machine learning functions optimised for both Arm CPUs
and GPUs using SIMD technologies.
https://github.com/ARM-software/ComputeLibrary

View File

@ -0,0 +1,65 @@
################################################################################
#
# compute-library
#
################################################################################
COMPUTE_LIBRARY_VERSION = v22.11
COMPUTE_LIBRARY_SITE = $(call github,ARM-software,ComputeLibrary,$(COMPUTE_LIBRARY_VERSION))
COMPUTE_LIBRARY_LICENSE = MIT
COMPUTE_LIBRARY_LICENSE_FILES = LICENSE
COMPUTE_LIBRARY_INSTALL_STAGING = YES
COMPUTE_LIBRARY_DEPENDENCIES = \
host-scons \
host-pkgconf
COMPUTE_LIBRARY_LDFLAGS = "$(TARGET_LDFLAGS)"
COMPUTE_LIBRARY_CFLAGS = " $(TARGET_CFLAGS)"
COMPUTE_LIBRARY_CXXFLAGS = "$(TARGET_CXXFLAGS)"
COMPUTE_LIBRARY_SCONS_ENV = $(TARGET_CONFIGURE_OPTS)
COMPUTE_LIBRARY_SCONS_OPTS = \
arch=arm64-v8a \
Werror=0 \
debug=0 \
asserts=0 \
neon=1 \
os=linux \
examples=0 \
build=cross_compile \
toolchain_prefix="" \
embed_kernels=1 \
extra_cxx_flags="-fPIC"
COMPUTE_LIBRARY_SCONS_ENV += \
LDFLAGS=$(COMPUTE_LIBRARY_LDFLAGS) \
CFLAGS=$(COMPUTE_LIBRARY_CFLAGS) \
CCFLAGS=$(COMPUTE_LIBRARY_CFLAGS) \
CXXFLAGS=$(COMPUTE_LIBRARY_CXXFLAGS)
define COMPUTE_LIBRARY_BUILD_CMDS
(cd $(@D); \
$(COMPUTE_LIBRARY_SCONS_ENV) \
$(SCONS) \
$(COMPUTE_LIBRARY_SCONS_OPTS))
endef
define COMPUTE_LIBRARY_INSTALL_STAGING_CMDS
(cd $(@D); \
$(COMPUTE_LIBRARY_SCONS_ENV) \
$(SCONS) \
$(COMPUTE_LIBRARY_SCONS_OPTS) \
install_dir="$(STAGING_DIR)/usr")
endef
define COMPUTE_LIBRARY_INSTALL_TARGET_CMDS
(cd $(@D); \
$(COMPUTE_LIBRARY_SCONS_ENV) \
$(SCONS) \
$(COMPUTE_LIBRARY_SCONS_OPTS) \
install_dir="$(TARGET_DIR)/usr")
endef
$(eval $(generic-package))

View File

@ -0,0 +1,6 @@
config BR2_PACKAGE_OPENCL_HEADERS
bool "opencl_headers"
help
Header-only library for Khronos OpenCL-Headers.
https://github.com/KhronosGroup/OpenCL-Headers

View File

@ -0,0 +1 @@
sha256 ca8090359654e94f2c41e946b7e9d826253d795ae809ce7c83a7d3c859624693 opencl_headers-dcd5bede6859d26833cd85f0d6bbcee7382dc9b3.tar.gz

View File

@ -0,0 +1,23 @@
################################################################################
#
# opencl_headers
#
################################################################################
OPENCL_HEADERS_VERSION = dcd5bede6859d26833cd85f0d6bbcee7382dc9b3
OPENCL_HEADERS_SITE = $(call github,KhronosGroup,OpenCL-Headers,$(OPENCL_HEADERS_VERSION))
OPENCL_HEADERS_LICENSE = BSD-2-Clause
OPENCL_HEADERS_LICENSE_FILES = LICENSE
OPENCL_HEADERS_SUPPORTS_IN_SOURCE_BUILD = NO
OPENCL_HEADERS_DEPENDENCIES = psimd
OPENCL_HEADERS_CONF_OPTS = -DCMAKE_SYSTEM_NAME=Linux \
-DCMAKE_SYSTEM_PROCESSOR=aarch64 \
-DOPENCL_HEADERS_BUILD_TESTING=OFF \
-DOPENCL_HEADERS_BUILD_CXX_TESTS=OFF
OPENCL_HEADERS_INSTALL_STAGING = YES
OPENCL_HEADERS_INSTALL_TARGET = NO
$(eval $(cmake-package))

View File

@ -16,6 +16,8 @@ config BR2_PACKAGE_TENSORFLOW_LITE
select BR2_PACKAGE_NEON2SSE
select BR2_PACKAGE_RUY
select BR2_PACKAGE_XNNPACK
select BR2_PACKAGE_OPENCL_HEADERS
select BR2_PACKAGE_VULKAN_HEADERS
help
Tensorflow Lite dynamic library and headers: Inference engine to run previously trained machine learning models.

View File

@ -24,7 +24,9 @@ TENSORFLOW_LITE_DEPENDENCIES += \
libabseil-cpp \
neon2sse \
ruy \
xnnpack
xnnpack \
opencl_headers \
vulkan-headers
TENSORFLOW_LITE_CONF_OPTS = \
-DCMAKE_C_FLAGS="$(TARGET_CFLAGS) -funsafe-math-optimizations \
@ -50,17 +52,25 @@ TENSORFLOW_LITE_CONF_OPTS = \
-DFlatBuffers_DIR=$(STAGING_DIR)/usr/lib/cmake/flatbuffers \
-DNEON_2_SSE_DIR=$(STAGING_DIR)/usr/lib/cmake/NEON_2_SSE \
-DTFLITE_ENABLE_EXTERNAL_DELEGATE=ON \
-DTFLITE_ENABLE_GPU=OFF \
-DTFLITE_ENABLE_GPU=ON \
-DTFLITE_ENABLE_INSTALL=ON \
-DTFLITE_ENABLE_MMAP=ON \
-DTFLITE_ENABLE_NNAPI=ON \
-DTFLITE_ENABLE_RUY=ON \
-DTFLITE_ENABLE_XNNPACK=ON
TENSORFLOW_LITE_MAKE_OPTS += _pywrap_tensorflow_interpreter_wrapper
TENSORFLOW_LITE_MAKE_OPTS += _pywrap_tensorflow_interpreter_wrapper benchmark_model
TENSORFLOW_LITE_POST_INSTALL_STAGING_HOOKS = TENSORFLOW_LITE_INSTALL_VERSION_HEADER
TENSORFLOW_LITE_POST_INSTALL_TARGET_HOOKS = TENSORFLOW_LITE_INSTALL_TFLITE_RUNTIME
define TENSORFLOW_LITE_INSTALL_VERSION_HEADER
mkdir -p $(STAGING_DIR)/usr/include/tensorflow/core/public
$(INSTALL) -D -m 644 $(@D)/tensorflow/core/public/version.h \
$(STAGING_DIR)/usr/include/tensorflow/core/public/
endef
define TENSORFLOW_LITE_INSTALL_TFLITE_RUNTIME
mkdir -p $(TARGET_DIR)/usr/lib/python$(PYTHON3_VERSION_MAJOR)/site-packages/tflite_runtime

View File

@ -1,25 +0,0 @@
From 4aa7797f8f043df84104b76ceca423832824ea13 Mon Sep 17 00:00:00 2001
From: j1nx <p.steenbergen@j1nx.nl>
Date: Fri, 9 Dec 2022 20:25:46 +0100
Subject: [PATCH 1/1] Add multithread support
---
tflite_minimal/minimal.cc | 2 ++
1 file changed, 2 insertions(+)
diff --git a/tflite_minimal/minimal.cc b/tflite_minimal/minimal.cc
index cd045e0..82e9b11 100644
--- a/tflite_minimal/minimal.cc
+++ b/tflite_minimal/minimal.cc
@@ -200,6 +200,8 @@ int main(int argc, char* argv[]) {
else if (argc == 3) {
memcpy(input, mel.data.data(), mel.n_mel*mel.n_len*sizeof(float));
}
+ interpreter->SetNumThreads(2);
+
// Fill input buffers
// TODO(user): Insert code to fill input tensors.
// Note: The buffer of the input tensor with index `i` of type T can
--
2.34.1

View File

@ -0,0 +1,24 @@
From e2ebeaca6a16d981b0d130c06cfe38592623fbe3 Mon Sep 17 00:00:00 2001
From: j1nx <p.steenbergen@j1nx.nl>
Date: Fri, 20 Jan 2023 15:44:40 +0100
Subject: [PATCH 1/1] Use 4 threads
---
tflite_minimal/minimal.cc | 1 +
1 file changed, 1 insertion(+)
diff --git a/tflite_minimal/minimal.cc b/tflite_minimal/minimal.cc
index cd045e0..88e21fe 100644
--- a/tflite_minimal/minimal.cc
+++ b/tflite_minimal/minimal.cc
@@ -186,6 +186,7 @@ int main(int argc, char* argv[]) {
tflite::InterpreterBuilder builder(*model, resolver);
std::unique_ptr<tflite::Interpreter> interpreter;
builder(&interpreter);
+ interpreter->SetNumThreads(4);
TFLITE_MINIMAL_CHECK(interpreter != nullptr);
// Allocate tensor buffers.
--
2.34.1

View File

@ -1,25 +0,0 @@
From 624f7abc65cf1dee85d803c398b36a9b35106720 Mon Sep 17 00:00:00 2001
From: j1nx <p.steenbergen@j1nx.nl>
Date: Fri, 20 Jan 2023 14:22:23 +0100
Subject: [PATCH 1/1] Multithread support
---
tflite_minimal/minimal.cc | 2 ++
1 file changed, 2 insertions(+)
diff --git a/tflite_minimal/minimal.cc b/tflite_minimal/minimal.cc
index 82e9b11..e7d0187 100644
--- a/tflite_minimal/minimal.cc
+++ b/tflite_minimal/minimal.cc
@@ -186,6 +186,8 @@ int main(int argc, char* argv[]) {
tflite::InterpreterBuilder builder(*model, resolver);
std::unique_ptr<tflite::Interpreter> interpreter;
builder(&interpreter);
+ const auto processor_count = std:: thread ::hardware_concurrency();
+ interpreter->SetNumThreads(processor_count);
TFLITE_MINIMAL_CHECK(interpreter != nullptr);
// Allocate tensor buffers.
--
2.34.1

View File

@ -0,0 +1,31 @@
From 6c4e6023d5b65447e23b737ebddff359668e64c4 Mon Sep 17 00:00:00 2001
From: j1nx <p.steenbergen@j1nx.nl>
Date: Wed, 25 Jan 2023 14:13:37 +0100
Subject: [PATCH 1/1] Bump libabseil CXX standard
---
package/libabseil-cpp/libabseil-cpp.mk | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/package/libabseil-cpp/libabseil-cpp.mk b/package/libabseil-cpp/libabseil-cpp.mk
index 7c6bda7659..61ef26c11a 100644
--- a/package/libabseil-cpp/libabseil-cpp.mk
+++ b/package/libabseil-cpp/libabseil-cpp.mk
@@ -11,12 +11,12 @@ LIBABSEIL_CPP_LICENSE_FILES = LICENSE
LIBABSEIL_CPP_INSTALL_STAGING = YES
LIBABSEIL_CPP_CONF_OPTS = \
- -DCMAKE_CXX_STANDARD=11 \
+ -DCMAKE_CXX_STANDARD=17 \
-DABSL_ENABLE_INSTALL=ON \
-DABSL_USE_GOOGLETEST_HEAD=OFF
HOST_LIBABSEIL_CPP_CONF_OPTS = \
- -DCMAKE_CXX_STANDARD=11 \
+ -DCMAKE_CXX_STANDARD=17 \
-DABSL_ENABLE_INSTALL=ON \
-DABSL_USE_GOOGLETEST_HEAD=OFF
--
2.34.1