Readme updated

.github/workflows/ci.yml

@@ -180,7 +180,8 @@ jobs:
     - name: Use ARMv6 container
       run: |
         docker run --rm -v $PWD:/app -w /app --platform linux/arm/v6 ghcr.io/nonannet/armv6_test:1 \
-        bash -lc "pip install .[mindev] && \
+        bash -lc "set -x && \
+                  pip install .[mindev] && \
                   mkdir -p build/runner && \
                   gcc -O3 -static -DENABLE_LOGGING -o build/runner/coparun src/coparun/runmem.c \
                   src/coparun/coparun.c src/coparun/mem_man.c && \
@@ -209,7 +210,8 @@ jobs:
     - name: Use ARMv7 container
       run: |
         docker run --rm -v $PWD:/app -w /app --platform linux/arm/v7 ghcr.io/nonannet/armv7_test:1 \
-        bash -lc "pip install .[mindev] && \
+        bash -lc "set -x && \
+                  pip install .[mindev] && \
                   mkdir -p build/runner && \
                   gcc -march=armv7-a -mfpu=neon-vfpv3 -mfloat-abi=hard -marm -static \
                   -Wall -Wextra -Wconversion -Wsign-conversion \
@@ -242,7 +244,8 @@ jobs:
     - name: Use ARMv7 container
       run: |
         docker run --rm -v $PWD:/app -w /app --platform linux/arm/v7 ghcr.io/nonannet/armv7_test:1 \
-        bash -lc "pip install .[mindev] && \
+        bash -lc "set -x && \
+                  pip install .[mindev] && \
                   mkdir -p build/runner && \
                   gcc -march=armv7-a -mfpu=neon-vfpv3 -mfloat-abi=hard -marm -static \
                   -Wall -Wextra -Wconversion -Wsign-conversion \
@@ -276,7 +279,8 @@ jobs:
     - name: Use ARMv7 container
       run: |
         docker run --rm -v $PWD:/app -w /app --platform linux/arm/v7 ghcr.io/nonannet/armv7_test:1 \
-        bash -lc "pip install .[mindev] && \
+        bash -lc "set -x && \
+                  pip install .[mindev] && \
                   mkdir -p build/runner && \
                   gcc -march=armv7-a -mfpu=neon-vfpv3 -mfloat-abi=hard -marm -static \
                   -Wall -Wextra -Wconversion -Wsign-conversion \
@@ -342,7 +346,7 @@ jobs:
         path: build/runner/*
 
   release-stencils:
-    needs: [build_stencils, build-ubuntu, build-windows, build-arm64, build-armv6, build-armv7]
+    needs: [build_stencils, build-ubuntu, build-windows, build-arm64, build-armv6, build-armv7, build-armv7thumb, build-armv7mthumb]
     runs-on: ubuntu-latest
     if: github.ref == 'refs/heads/main' && github.event_name == 'push'
     permissions:
@@ -375,8 +379,6 @@ jobs:
           cp tmp/runner-linux-arm64/coparun release/coparun-aarch64
           cp tmp/runner-linux-armv6/coparun release/coparun-armv6
           cp tmp/runner-linux-armv7/coparun release/coparun-armv7
-          cp tmp/runner-linux-armv7thumb/coparun release/coparun-armv7thumb
-          cp tmp/runner-linux-armv7mthumb/coparun release/coparun-armv7mthumb
           cp tmp/runner-win/coparun*.exe release/
 
           TAG="${{ steps.version.outputs.version }}"

README.md

@@ -13,7 +13,7 @@ The main features can be summarized as:
 - Memory and type safety with a minimal set of runtime errors
 - Deterministic execution
 - Automatic differentiation for efficient realtime optimization (reverse-mode)
-- Optimized machine code for x86_64, ARMv6, ARMv7 and AArch64
+- Optimized machine code for x86_64, 32 Bit ARM (Cortex-A and Cortex-M) and AArch64
 - Highly portable to new architectures
 - Small Python package with minimal dependencies and no cross-compile toolchain required
 
@@ -31,7 +31,6 @@ While hardware I/O is obviously a core aspect of the project, it is not yet avai
 
 Currently in development:
 - Array stencils for handling very large arrays and generating SIMD-optimized code - e.g., for machine vision and neural network applications
-- Support for Thumb instructions required by ARM*-M targets (for MCUs)
 - Constant regrouping for further symbolic optimization of the computation graph
 
 Despite missing SIMD-optimization, benchmark performance shows promising numbers. The following chart plots the results in comparison to NumPy 2.3.5:
@@ -253,4 +252,4 @@ This project is licensed under the MIT license - see the [LICENSE](LICENSE) file
 
 [^2]: The compiler must support tail-call optimization (TCO). Currently, GCC is supported. Porting to a new architecture requires implementing a subset of relocation types used by that architecture.
 
-[^3]: Supported architectures: x86_64, AArch64, ARMv6 and 7 (non-Thumb). ARMv6/7-M (Thumb) support is in development. Code for x86 32-bit exists but has unresolved issues and a low priority.
+[^3]: Supported architectures: x86_64, AArch64, ARMv6/7 (non-Thumb) and ARMv7 Thumb for Cortex-A and Cortex-M. Code for x86 32-bit exists but has unresolved issues and a low priority.

src/copapy/_stencils.py

@@ -343,6 +343,9 @@ class stencil_database():
         elif pr.type.endswith('_ABS32'):
             # R_ARM_ABS32
             # S + A (replaces full 32 bit)
+            assert not patch_offset % 4, 'R_ARM_ABS32 patched data like literals needs to be 4 Byte aligned'
+            # This might be caused by the call in entry_function_shell if not aligned
+
             patch_value = symbol_address + pr.fields['r_addend']
             symbol_type = symbol_type + 0x03  # Relative to data section
 

stencils/generate_stencils.py

@@ -57,8 +57,8 @@ def norm_indent(f: Callable[..., str]) -> Callable[..., str]:
 def get_entry_function_shell() -> str:
     return f"""
     {entry_func_prefix}int entry_function_shell(){{
-        volatile char stack_place_holder[{stack_size}];
+        //volatile char stack_place_holder[{stack_size}];
-        stack_place_holder[0] = 0;
+        //stack_place_holder[0] = 0;
         result_int(0);
         return 1;
     }}

tools/create_asm.sh

@@ -10,7 +10,6 @@ cparch=$(python3 -c "import copapy; print(copapy._stencils.detect_process_arch()
 # Disassemble stencil object file
 objdump -d -x src/copapy/obj/stencils_${cparch}_O3.o > build/runner/stencils.asm
 
-# Create example code disassembly
 python3 tools/make_example.py
 build/runner/coparun build/runner/test.copapy build/runner/test.copapy.bin
 
@@ -28,6 +27,10 @@ fi
 
 echo "Archtitecture: '$cparch'"
 
-objdump -D -b binary -m $cparch --adjust-vma=0x10000 build/runner/test.copapy.bin > build/runner/example.asm
+if [[ "$cparch" == *"thumb"* ]]; then
+	objdump -D -b binary -marm -M force-thumb --adjust-vma=0x10000 build/runner/test.copapy.bin > build/runner/example.asm
+else
+	objdump -D -b binary -m $cparch --adjust-vma=0x10000 build/runner/test.copapy.bin > build/runner/example.asm
+fi
 
 rm build/runner/test.copapy.bin

tools/test_example_code.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+# Build arm-v7 runner and stencils
+bash tools/build.sh arm-v7
+
+# Build arm-v7-thumb stencils
+bash tools/build.sh arm-v7-thumb
+
+# Build arm-v7-thumb example code
+export CP_TARGET_ARCH=armv7thumb
+python3 tools/make_example.py
+build/runner/coparun-armv7 build/runner/test.copapy build/runner/test.copapy.bin
+
+arm-none-eabi-objdump -D -b binary -marm -M force-thumb --adjust-vma=0x1000000 build/runner/test.copapy.bin > build/runner/test.copapy-example-armv7thumb.asm
+
+# Build arm-v7-thumb example code
+export CP_TARGET_ARCH=armv7
+python3 tools/make_example.py
+build/runner/coparun-armv7 build/runner/test.copapy build/runner/test.copapy.bin
+
+arm-none-eabi-objdump -D -b binary -marm --adjust-vma=0x1000000 build/runner/test.copapy.bin > build/runner/test.copapy-example-armv7.asm