Update flash image option to weights

README.md

@@ -81,10 +81,10 @@ xf.convert("example_int8_model.tflite", "xcore_optimised_int8_model.tflite", {
 })
 ```
 
-To create a parameters file and a tflite model suitable for loading to flash, use the "xcore-flash-image-file" option.
+To create a parameters file and a tflite model suitable for loading to flash, use the "xcore-weights-file" option.
 ```python
 xf.convert("example_int8_model.tflite", "xcore_optimised_int8_flash_model.tflite", {
-    "xcore-flash-image-file ": "./xcore_params.params",
+    "xcore-weights-file ": "./xcore_params.params",
 })
 ```
 

docs/rst/options.rst

@@ -70,7 +70,7 @@ Name of the file where to place the optimized TFLITE model
 Number of threads to translate for (max=5). Defaults to 1.
 
 
-``xcore-flash-image-file filename``
+``xcore-weights-file filename``
 +++++++++++++++++++++++++++++++++++
 
 File to place the learned parameters in. If this option is not specified,
@@ -86,7 +86,7 @@ be slower but allows large numbers of learned parameters to be used.
 Sets a threshold under which to not place learned parameters in flash. The
 default is set to 96 bytes. If less than 96 bytes, the overhead of lowering to flash is 
 more than the benefit gained. This option is only meaningful if
-``xcore-flash-image-file`` has been used. You can experiment with this
+``xcore-weights-file`` has been used. You can experiment with this
 parameter to get a different trade-off between speed and memory requirements.
 
 ``xcore-reduce-memory true|false``

examples/app_flash_single_model/README.rst

@@ -5,7 +5,7 @@ Please consult `here <../../docs/rst/flow.rst>`_ on how to install the tools.
 
 In order to compile and run this example follow these steps::
 
-  xcore-opt --xcore-flash-image-file=model.params vww_quant.tflite -o model.tflite
+  xcore-opt --xcore-weights-file=model.params vww_quant.tflite -o model.tflite
   mv model.tflite.cpp model.tflite.h src
   xmake
   python -c 'from xmos_ai_tools import xformer as xf; xf.generate_flash(

examples/app_flash_two_models/README.rst

@@ -9,10 +9,10 @@ sets of learned parameters into a single flash image.
 
 In order to compile and run this example follow these steps::
 
-  xcore-opt --xcore-flash-image-file=model1.params \
+  xcore-opt --xcore-weights-file=model1.params \
             --xcore-naming-prefix=model1_ \
             vww_quant1.tflite -o model1.tflite
-  xcore-opt --xcore-flash-image-file=model2.params \
+  xcore-opt --xcore-weights-file=model2.params \
             --xcore-naming-prefix=model2_ \
             vww_quant2.tflite -o model2.tflite
   mv model1.tflite.cpp model1.tflite.h src

examples/app_flash_two_models_one_arena/README.rst

@@ -20,10 +20,10 @@ The differences with ``app_flash_two_models`` example are minimal:
 
 In order to compile and run this example follow these steps::
 
-  xcore-opt --xcore-flash-image-file=model1.params \
+  xcore-opt --xcore-weights-file=model1.params \
             --xcore-naming-prefix=model1_ \
             vww_quant1.tflite -o model1.tflite
-  xcore-opt --xcore-flash-image-file=model2.params \
+  xcore-opt --xcore-weights-file=model2.params \
             --xcore-naming-prefix=model2_ \
             vww_quant2.tflite -o model2.tflite
   mv model1.tflite.cpp model1.tflite.h src

examples/app_mobilenetv2/obtain_and_optimize_mobilenetv2.py

@@ -33,7 +33,7 @@
     TFLITE_MODEL_PATH,
     OPT_MODEL_PATH,
     {
-        "xcore-flash-image-file": OPT_PARAMS_PATH,
+        "xcore-weights-file": OPT_PARAMS_PATH,
         "xcore-thread-count": "5",
         "xcore-naming-prefix": NAMING_PREFIX,
         "xcore-op-split-tensor-arena": "True",

examples/app_yolov8_classification/obtain_and_optimize_yolov8_cls.py

@@ -29,7 +29,7 @@
     TFLITE_MODEL_PATH,
     OPT_MODEL_PATH,
     {
-        "xcore-flash-image-file": OPT_PARAMS_PATH,
+        "xcore-weights-file": OPT_PARAMS_PATH,
         "xcore-thread-count": "5",
         "xcore-naming-prefix": NAMING_PREFIX,
     },

python/setup.py

@@ -107,17 +107,17 @@ def finalize_options(self):
         "Intended Audience :: Science/Research",
         "Programming Language :: Python :: 3",
         "Programming Language :: Python :: 3 :: Only",
-        "Programming Language :: Python :: 3.7",
         "Programming Language :: Python :: 3.8",
         "Programming Language :: Python :: 3.9",
+        "Programming Language :: Python :: 3.10",
         "Topic :: Scientific/Engineering",
         "Topic :: Scientific/Engineering :: Mathematics",
         "Topic :: Scientific/Engineering :: Artificial Intelligence",
         "Topic :: Software Development",
         "Topic :: Software Development :: Libraries",
         "Topic :: Software Development :: Libraries :: Python Modules",
     ],
-    python_requires=">=3.7",
+    python_requires=">=3.8",
     packages=find_namespace_packages(),
     install_requires=REQUIRED_PACKAGES,
     package_data=package_files,

python/xmos_ai_tools/runtime/buildfiles/aitoolslib.cmake

@@ -5,5 +5,9 @@ set(XMOS_AITOOLSLIB_DEFINITIONS
 		"NO_INTERPRETER"
     )
 
-set(XMOS_AITOOLSLIB_LIBRARIES "${CMAKE_CURRENT_LIST_DIR}/../lib/libxtflitemicro.a")
+if(NOT ${CMAKE_SYSTEM_PROCESSOR} STREQUAL XCORE_XS3A)
+	set(XMOS_AITOOLSLIB_LIBRARIES "${CMAKE_CURRENT_LIST_DIR}/../lib/libhost_xtflitemicro.a")
+else()
+	set(XMOS_AITOOLSLIB_LIBRARIES "${CMAKE_CURRENT_LIST_DIR}/../lib/libxtflitemicro.a")
+endif()
 set(XMOS_AITOOLSLIB_INCLUDES "${CMAKE_CURRENT_LIST_DIR}/../include")

xformer/Test/loadconstantop.mlir

@@ -1,6 +1,6 @@
-// RUN: xcore-opt --mlir-io %s --xcore-apply-loadconstantop-patterns --xcore-flash-image-file=/dev/null --xcore-load-externally-if-larger=0 | FileCheck %s
+// RUN: xcore-opt --mlir-io %s --xcore-apply-loadconstantop-patterns --xcore-weights-file=/dev/null --xcore-load-externally-if-larger=0 | FileCheck %s
 
-// RUN: xcore-opt --mlir-io %s --xcore-apply-loadconstantop-patterns --xcore-flash-image-file=/dev/null --xcore-load-externally-if-larger=16 | FileCheck %s -check-prefix=LARGER-CHECK
+// RUN: xcore-opt --mlir-io %s --xcore-apply-loadconstantop-patterns --xcore-weights-file=/dev/null --xcore-load-externally-if-larger=16 | FileCheck %s -check-prefix=LARGER-CHECK
 
 // CHECK-LABEL: valid
 func.func @valid(%arg0: tensor<?x4x8x1x!quant.uniform<i8:f32, 0.0078160231932997704>>) -> tensor<?x32x!quant.uniform<i8:f32, 0.037329975515604019:-13>> attributes {tf.entry_function = {inputs = "flatten_input", outputs = "Identity"}} {

xformer/Test/loadflashop.mlir

@@ -1,4 +1,4 @@
-// RUN: xcore-opt --mlir-io %s --xcore-write-flash-image --xcore-flash-image-file=/dev/null | FileCheck %s
+// RUN: xcore-opt --mlir-io %s --xcore-write-flash-image --xcore-weights-file=/dev/null | FileCheck %s
 
 // CHECK-LABEL: valid
 func.func @valid(%arg0: tensor<?x4x8x1x!quant.uniform<i8:f32, 0.0078160231932997704>>) -> tensor<?x32x!quant.uniform<i8:f32, 0.037329975515604019:-13>> attributes {tf.entry_function = {inputs = "flatten_input", outputs = "Identity"}} {

xformer/Transforms/ApplyLoadConstantOpPatterns.cpp

@@ -62,7 +62,7 @@ bool isNotUsedByLoadConstantOp(Value result) {
 
 void ApplyLoadConstantOpPatterns::runOnOperation() {
   func::FuncOp f = getOperation();
-  if (flashImageFilenameOption.empty()) {
+  if (weightsFilenameOption.empty()) {
     f.emitError("Flash image file option should be provided to run this pass!");
     signalPassFailure();
     return;

xformer/Transforms/Options.h

@@ -11,7 +11,7 @@ namespace xcore {
 
 extern llvm::cl::opt<bool> enableBetaFloatOption;
 extern llvm::cl::opt<unsigned> threadCountOption;
-extern llvm::cl::opt<std::string> flashImageFilenameOption;
+extern llvm::cl::opt<std::string> weightsFilenameOption;
 extern llvm::cl::opt<unsigned> loadExternallyIfLargerOption;
 extern llvm::cl::opt<bool> tileLoadOption;
 extern llvm::cl::opt<unsigned> maxLoadExternalSizeOption;

xformer/Transforms/Passes.cpp

@@ -35,9 +35,9 @@ void buildXCorePassPipeline(OpPassManager &pm) {
   pm.addPass(createReplaceConv2DPass());
   pm.addPass(createApplyXCPatternsPass());
   // Add to pipeline only if flash image file option is provided
-  if (!flashImageFilenameOption.empty()) {
+  if (!weightsFilenameOption.empty()) {
     pm.addPass(createApplyLoadConstantOpPatternsPass());
-    pm.addPass(createWriteFlashImagePass());
+    pm.addPass(createWriteWeightsPass());
   }
   // Run canonicalization, which includes combining Reshapes
   pm.addPass(mlir::createCanonicalizerPass());

xformer/Transforms/Passes.h

@@ -36,7 +36,7 @@ std::unique_ptr<OperationPass<func::FuncOp>> createReplaceConv2DPass();
 std::unique_ptr<OperationPass<func::FuncOp>> createApplyXCPatternsPass();
 std::unique_ptr<OperationPass<func::FuncOp>>
 createApplyLoadConstantOpPatternsPass();
-std::unique_ptr<OperationPass<func::FuncOp>> createWriteFlashImagePass();
+std::unique_ptr<OperationPass<func::FuncOp>> createWriteWeightsPass();
 std::unique_ptr<OperationPass<func::FuncOp>> createPlanMemoryPass();
 std::unique_ptr<OperationPass<func::FuncOp>> createTranslateToCustomOpPass();
 

xformer/Transforms/WriteFlashImage.cpp

@@ -17,9 +17,9 @@ namespace xcore {
 
 namespace {
 // Write flash image
-struct WriteFlashImage
-    : public PassWrapper<WriteFlashImage, OperationPass<func::FuncOp>> {
-  MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(WriteFlashImage)
+struct WriteWeights
+    : public PassWrapper<WriteWeights, OperationPass<func::FuncOp>> {
+  MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(WriteWeights)
 
   void getDependentDialects(DialectRegistry &registry) const final {
     registry.insert<XCoreDialect>();
@@ -29,9 +29,9 @@ struct WriteFlashImage
   void runOnOperation() override;
 };
 
-struct WriteFlashImagePattern : public OpRewritePattern<LoadConstantOp> {
-  WriteFlashImagePattern(std::vector<std::vector<char>> *tensorsVec,
-                         MLIRContext *context)
+struct WriteWeightsPattern : public OpRewritePattern<LoadConstantOp> {
+  WriteWeightsPattern(std::vector<std::vector<char>> *tensorsVec,
+                      MLIRContext *context)
       : OpRewritePattern<LoadConstantOp>(context), tensorsVec_(tensorsVec) {}
 
   std::vector<char> getTensorData(LoadConstantOp loadOp) const {
@@ -129,9 +129,9 @@ struct WriteFlashImagePattern : public OpRewritePattern<LoadConstantOp> {
   std::vector<std::vector<char>> *tensorsVec_;
 };
 
-void WriteFlashImage::runOnOperation() {
+void WriteWeights::runOnOperation() {
   func::FuncOp f = getOperation();
-  if (flashImageFilenameOption.empty()) {
+  if (weightsFilenameOption.empty()) {
     f.emitError("Flash image file option should be provided to run this pass!");
     signalPassFailure();
     return;
@@ -143,33 +143,33 @@ void WriteFlashImage::runOnOperation() {
   // with a LoadFlashOp
   std::vector<std::vector<char>> tensorsVec;
   RewritePatternSet patterns(ctx);
-  patterns.insert<WriteFlashImagePattern>(&tensorsVec, ctx);
+  patterns.insert<WriteWeightsPattern>(&tensorsVec, ctx);
   (void)applyPatternsAndFoldGreedily(func, std::move(patterns));
 
   if (tileLoadOption) {
-    if (failed(utils::writeTileServerDataToFile(flashImageFilenameOption,
+    if (failed(utils::writeTileServerDataToFile(weightsFilenameOption,
                                                 tensorsVec))) {
       f.emitError("Failed to write tile data!");
       signalPassFailure();
       return;
     }
   }
   // Write tensor data to flash image file
-  else if (failed(utils::writeFlashImageToFile(flashImageFilenameOption,
-                                               tensorsVec))) {
+  else if (failed(
+               utils::writeWeightsToFile(weightsFilenameOption, tensorsVec))) {
     f.emitError("Failed to write flash image!");
     signalPassFailure();
     return;
   }
 }
 } // namespace
 
-// Creates an instance of the WriteFlashImage pass.
-std::unique_ptr<OperationPass<func::FuncOp>> createWriteFlashImagePass() {
-  return std::make_unique<WriteFlashImage>();
+// Creates an instance of the WriteWeights pass.
+std::unique_ptr<OperationPass<func::FuncOp>> createWriteWeightsPass() {
+  return std::make_unique<WriteWeights>();
 }
 
-static PassRegistration<WriteFlashImage> pass;
+static PassRegistration<WriteWeights> pass;
 
 } // namespace xcore
 } // namespace mlir