Setup automated tests with QEMU

.github/workflows/build.yaml

@@ -47,7 +47,9 @@ jobs:
       with:
         submodules: true
     - name: Install OS dependencies
-      run: sudo add-apt-repository universe
+      run: |
+        sudo add-apt-repository universe
+        sudo apt-get install qemu-system-arm
     - uses: actions/setup-python@v5
       with:
         python-version: '3.10'
@@ -71,6 +73,12 @@ jobs:
       run: make dist ARCH=armv7m V=1
     - name: Build module armv7emsp
       run: make dist ARCH=armv7emsp V=1
+    - name: Build module armv7emdp
+      run: make dist ARCH=armv7emdp V=1
+    - name: Build QEMU MicroPython image
+      run: make qemu_build V=1
+    - name: Run tests under QEMU
+      run: make check_qemu V=1
     - name: Archive dist artifacts
       uses: actions/upload-artifact@v4
       with:

Makefile

@@ -80,23 +80,35 @@ $(MODULES_PATH)/emlearn_cnn_int8.mpy: $(MODULES_PATH)/emlearn_cnn_fp32.mpy
 # Generate list of .mpy files
 MODULE_MPYS = $(addprefix $(MODULES_PATH)/,$(addsuffix .mpy,$(MODULES)))
 
-# Build dynamic native module (without forced clean)
-$(MODULES_PATH)/%.mpy:
+# Build dynamic native module
+$(MODULES_PATH)/%.mpy: $(ARCH_SENTINEL)
 	$(MAKE) -C $(or $($(*)_SRC),src/$*) \
 		ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) CFLAGS_EXTRA=$(CFLAGS_EXTRA) \
 		V=1 $($(*)_CONFIG) dist
 
+# A file to track ARCH changes etc, to nuke old artifact
+ARCH_SENTINEL := .sentinel_arch_$(ARCH)
+$(ARCH_SENTINEL):
+	@echo "ARCH changed or not set — cleaning old build artifacts"
+	@rm -f .sentinel_arch_*
+	find src -name ".mpy_ld_cache" -type d -exec rm -rf {} + ; \
+	find src -name "build" -type d -exec rm -rf {} + ; \
+	find src -name "*.mpy" -delete ; \
+	find src -name "*.o" -delete
+	@mkdir -p $(MODULES_PATH)
+	@touch $@
+
 # CNN modules need clean build due to shared build directory
 # They must also build sequentially (fp32 first, then int8)
-$(MODULES_PATH)/emlearn_cnn_fp32.mpy:
+$(MODULES_PATH)/emlearn_cnn_fp32.mpy: $(ARCH_SENTINEL)
 	$(MAKE) -C src/tinymaix_cnn \
 		ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) CFLAGS_EXTRA=$(CFLAGS_EXTRA) \
 		V=1 CONFIG=fp32 clean
 	$(MAKE) -C src/tinymaix_cnn \
 		ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) CFLAGS_EXTRA=$(CFLAGS_EXTRA) \
 		V=1 CONFIG=fp32 dist
 
-$(MODULES_PATH)/emlearn_cnn_int8.mpy: $(MODULES_PATH)/emlearn_cnn_fp32.mpy
+$(MODULES_PATH)/emlearn_cnn_int8.mpy: $(MODULES_PATH)/emlearn_cnn_fp32.mpy $(ARCH_SENTINEL)
 	$(MAKE) -C src/tinymaix_cnn \
 		ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) CFLAGS_EXTRA=$(CFLAGS_EXTRA) \
 		V=1 CONFIG=int8 clean
@@ -159,10 +171,7 @@ codesize:
 	python3 tools/code_size.py $(MPY_DIR_ABS)/ports/unix/build-standard
 
 clean:
-	make -C src/emlearn_trees/ ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) V=1 clean
-	make -C src/emlearn_neighbors/ ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) V=1 clean
-	make -C src/emlearn_iir/ ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) V=1 clean
-	make -C src/emlearn_logreg/ ARCH=$(ARCH) MPY_DIR=$(MPY_DIR_ABS) V=1 clean
+	git clean -dfx src/
 	rm -rf ./dist
 
 RELEASE_NAME = emlearn-micropython-$(VERSION)
@@ -177,4 +186,24 @@ check: check_unix_natmod
 
 dist: $(MODULE_MPYS)
 
+# =============================================================================
+# QEMU MicroPython targets
+# =============================================================================
+
+QEMU_PORT_DIR = $(MPY_DIR)/ports/qemu
+QEMU_BOARD ?= MPS2_AN500
+QEMU_ARCH ?= armv7emdp
+QEMU_FIRMWARE = $(QEMU_PORT_DIR)/build-$(QEMU_BOARD)/firmware.elf
+
+# Build firmware for QEMU
+.PHONY: qemu_build
+qemu_build:
+	$(MAKE) -C $(QEMU_PORT_DIR) BOARD=$(QEMU_BOARD) MICROPY_HEAP_SIZE=1024000
+
+# Run tests/test_all.py on QEMU using mpremote mount
+# Usage: make check_qemu QEMU_BOARD=MPS2_AN500 QEMU_ARCH=armv7emdp
+.PHONY: check_qemu
+check_qemu: $(QEMU_FIRMWARE)
+	python3 $(abspath tools/run_qemu_tests.py) --board $(QEMU_BOARD) --arch $(QEMU_ARCH) --abi-version $(MPY_ABI_VERSION) --mount $(abspath .)
+
 

tests/npyfile.py

@@ -271,23 +271,21 @@ def write_values(self, arr, typecode=None):
         #print('write', self.written_bytes)
 
 
-def load(filelike) -> tuple[tuple, array.array]:
+def load(filelike, chunk_size=64) -> tuple[tuple, array.array]:
     """
     Load array from .npy file
-
     Convenience function for doing it in one shot.
     For streaming, use npyfile.Reader instead
     """    
-
-    chunks = []
     with Reader(filelike) as reader:
-        # Just read everything in one chunk
         total_items = compute_items(reader.shape)
-        for c in reader.read_data_chunks(total_items):
-            chunks.append(c)
+        data = array.array(reader.typecode, (0 for _ in range(total_items)))
+        idx = 0
+        for chunk in reader.read_data_chunks(chunk_size):
+            data[idx:idx + len(chunk)] = chunk
+            idx += len(chunk)
+    return reader.shape, data
 
-    assert len(chunks) == 1
-    return reader.shape, chunks[0]
 
 def save(filelike, arr : array.array, shape=None, typecode=None):
     """

tests/test_all.py

@@ -1,5 +1,6 @@
 
 import sys
+import gc
 
 # Find the module path (architecture+version specific)
 sys_mpy = sys.implementation._mpy
@@ -63,6 +64,16 @@ def main():
     passed = 0
     failed = 0
 
+    # Test markers for external test runners (mpremote, etc.)
+    # These allow the runner to know when tests are complete without relying on timeouts
+    print('\n=== TEST START ===')
+
+    free = gc.mem_free()
+    alloc = gc.mem_alloc()
+    print(f"RAM free={free} used={alloc} total={free+alloc}")
+
+    print('sys.path', sys.path)
+
     for module_name in modules:
         mod = None
         print(f'{module_name}:')
@@ -75,6 +86,9 @@ def main():
             failed += 1
             continue
 
+        # Try to free space
+        gc.collect()
+
         module_attributes = dir(mod)
         tests = [ o for o in module_attributes if o.startswith('test_') ]
         for test_name in tests:
@@ -92,8 +106,12 @@ def main():
             print(f'\t PASS')
             passed += 1
 
+            # Try to free space
+            gc.collect()
+
     print(f'Passed: {passed}')
     print(f'Failed: {failed}')
+    print('\n=== TEST END ===')
 
     # Let status code reflect number of failures
     return failed

tests/test_extratrees_cancer.py

@@ -27,14 +27,11 @@ def load_npy_labels_int16(filename):
     return array.array('h', labels)
 
 def test_real_dataset():
-    print("=== REAL DATASET TEST ===")
-
     X_train_flat = load_npy_features_int16(DATA_FILES['X_train'])
     y_train = load_npy_labels_int16(DATA_FILES['y_train']) 
     X_test_flat = load_npy_features_int16(DATA_FILES['X_test'])
     y_test = load_npy_labels_int16(DATA_FILES['y_test'])
 
-
     n_features = 30
     n_train = len(y_train)
     n_test = len(y_test)

tests/test_extratrees_wine.py

@@ -19,28 +19,22 @@ def load_npy_features_int16(filename):
     """Load .npy file and convert to int16 array (scaled from float32)"""
     shape, data = npyfile.load(filename)
     # Scale float32 data to int16 range (multiply by 1000 and convert)
-    scaled = [int(v * 1000) for v in data]
-    return array.array('h', scaled)
+    return array.array('h', (int(v * 1000) for v in data))
 
 def load_npy_labels_int16(filename):
     """Load .npy file and convert to int16 array (labels, no scaling)"""
     shape, data = npyfile.load(filename)
     # Labels are already integers (0.0, 1.0, 2.0), just convert directly
-    labels = [int(v) for v in data]
-    return array.array('h', labels)
+    return array.array('h', (int(v) for v in data))
 
 def test_wine():
     print("=== WINE DATASET TEST (3-class) ===")
 
     # Load preprocessed data
-    try:
-        X_train_flat = load_npy_features_int16(DATA_FILES['X_train'])
-        y_train = load_npy_labels_int16(DATA_FILES['y_train'])
-        X_test_flat = load_npy_features_int16(DATA_FILES['X_test'])
-        y_test = load_npy_labels_int16(DATA_FILES['y_test'])
-    except:
-        print("Error: Run wine/prepare.py first")
-        return
+    X_train_flat = load_npy_features_int16(DATA_FILES['X_train'])
+    y_train = load_npy_labels_int16(DATA_FILES['y_train'])
+    X_test_flat = load_npy_features_int16(DATA_FILES['X_test'])
+    y_test = load_npy_labels_int16(DATA_FILES['y_test'])
 
     n_features = 13  # 13 wine features (alcohol, malic_acid, ash, etc.)
     n_train = len(y_train)
@@ -54,12 +48,13 @@ def test_wine():
     print(f"Classes: {n_classes} (wine cultivars 0, 1, 2)")
     print("Task: Classify wine cultivar")
 
+    max_samples = n_train
     # Create model
     model = emlearn_extratrees.new(
         n_features, n_classes,
-        n_trees=20, max_depth=12, min_samples_leaf=2,
-        n_thresholds=15, subsample_ratio=0.8, feature_subsample_ratio=1.0,
-        max_nodes=3000, max_samples=500, rng_seed=42
+        n_trees=5, max_depth=5, min_samples_leaf=2,
+        n_thresholds=8, subsample_ratio=0.8, feature_subsample_ratio=1.0,
+        max_nodes=1000, max_samples=max_samples, rng_seed=42
     )
 
     train_start = time.ticks_ms()