CI: Testing for ARMv7 extended to armv7thumb and armv7mthumb

.github/workflows/ci.yml

@@ -214,6 +214,10 @@ jobs:
                   gcc -O3 -static -DENABLE_LOGGING -o build/runner/coparun src/coparun/runmem.c \
                   src/coparun/coparun.c src/coparun/mem_man.c && \
                   pytest && \
+                  export CP_TARGET_ARCH=armv7thumb && \
+                  pytest && \
+                  export CP_TARGET_ARCH=armv7mthumb && \
+                  pytest && \
                   bash tools/create_asm.sh"
 
     - uses: actions/upload-artifact@v4

src/copapy/_binwrite.py

@@ -13,6 +13,7 @@ Command = Enum('Command', [('ALLOCATE_DATA', 1), ('COPY_DATA', 2),
                            ('PATCH_OBJECT_ABS', 0x2002),
                            ('PATCH_OBJECT_REL', 0x2003),
                            ('PATCH_OBJECT_ARM32_ABS', 0x2004),
+                           ('PATCH_OBJECT_ARM32_ABS_THM', 0x2006),
                            ('ENTRY_POINT', 7),
                            ('RUN_PROG', 64), ('READ_DATA', 65),
                            ('END_COM', 256), ('FREE_MEMORY', 257), ('DUMP_CODE', 258)])

src/copapy/_compiler.py

@@ -393,14 +393,14 @@ def compile_to_dag(node_list: Iterable[Node], sdb: stencil_database) -> tuple[bi
     # assemble stencils to main program and patch stencils
     data = sdb.get_function_code('entry_function_shell', 'start')
     data_list.append(data)
-    print(f"* entry_function_shell (0) " + ' '.join(f'{d:02X}' for d in data))
+    #print(f"* entry_function_shell (0) " + ' '.join(f'{d:02X}' for d in data))
     offset = aux_func_len + len(data)
 
     for associated_net, node in extended_output_ops:
         assert node.name in sdb.stencil_definitions, f"- Warning: {node.name} stencil not found"
         data = sdb.get_stencil_code(node.name)
         data_list.append(data)
-        print(f"* {node.name} ({offset}) " + ' '.join(f'{d:02X}' for d in data))
+        #print(f"* {node.name} ({offset}) " + ' '.join(f'{d:02X}' for d in data))
 
         for reloc in sdb.get_relocations(node.name, stencil=True):
             if reloc.target_symbol_info in ('STT_OBJECT', 'STT_NOTYPE', 'STT_SECTION'):
@@ -453,7 +453,7 @@ def compile_to_dag(node_list: Iterable[Node], sdb: stencil_database) -> tuple[bi
         for reloc in sdb.get_relocations(name):
 
             if not reloc.target_section_index:
-                assert reloc.pelfy_reloc.type == 'R_ARM_V4BX', (reloc.pelfy_reloc.type, name)
+                assert reloc.pelfy_reloc.type == 'R_ARM_V4BX', (reloc.pelfy_reloc.type, name, reloc.pelfy_reloc.symbol.name)
 
             elif reloc.target_symbol_info in {'STT_OBJECT', 'STT_NOTYPE', 'STT_SECTION'}:
                 # Patch constants/variable addresses on heap

src/copapy/_math.py

@@ -310,7 +310,7 @@ def get_42(x: value[Any]) -> value[float]: ...
 def get_42(x: NumLike) -> value[float] | float:
     """Returns the value representing the constant 42"""
     if isinstance(x, value):
-        return add_op('get_42', [x, x])
+        return add_op('get_42', [x])
     return float((int(x) * 3.0 + 42.0) * 5.0 + 21.0)
 
 

src/copapy/_stencils.py

@@ -2,6 +2,7 @@ from dataclasses import dataclass
 from typing import Generator, Literal, Iterable, TYPE_CHECKING
 import struct
 import platform
+import os
 
 if TYPE_CHECKING:
     import pelfy
@@ -46,6 +47,10 @@ class patch_entry:
 
 
 def detect_process_arch() -> str:
+    cp_target_arch = os.environ.get("CP_TARGET_ARCH")
+    if cp_target_arch:
+        return cp_target_arch
+
     bits = struct.calcsize("P") * 8
     arch = platform.machine().lower()
 
@@ -305,21 +310,20 @@ class stencil_database():
             scale = 8
             #print(f" *> {patch_value=} {symbol_address=} {pr.fields['r_addend']=}, {function_offset=}")
 
-        elif pr.type.endswith('_MOVW_ABS_NC'):
+        elif pr.type == 'R_ARM_MOVW_ABS_NC':
-            # R_ARM_MOVW_ABS_NC
             # (S + A) & 0xFFFF
             mask = 0xFFFF
             patch_value = symbol_address + pr.fields['r_addend']
             symbol_type = symbol_type + 0x04  # Absolut value
             #print(f" *> {pr.type} {patch_value=} {symbol_address=}, {function_offset=}")
 
-        elif pr.type.endswith('_MOVT_ABS'):
+        elif pr.type =='R_ARM_MOVT_ABS':
-            # R_ARM_MOVT_ABS
             # (S + A) & 0xFFFF0000
             mask = 0xFFFF0000
             patch_value = symbol_address + pr.fields['r_addend']
             symbol_type = symbol_type + 0x04  # Absolut value
             scale = 0x10000
+            #print(f" *> {pr.type} {patch_value=} {symbol_address=}, {function_offset=}, {pr.fields['r_addend']=}")
 
         elif pr.type.endswith('_ABS32'):
             # R_ARM_ABS32
@@ -330,9 +334,24 @@ class stencil_database():
         elif pr.type.endswith('_THM_JUMP24') or pr.type.endswith('_THM_CALL'):
             # R_ARM_THM_JUMP24
             # S + A - P
-            #assert pr.fields['r_addend'] == 0, pr.fields['r_addend']
+            patch_value = symbol_address - patch_offset  + pr.fields['r_addend']
-            patch_value = symbol_address - (patch_offset + 4)  #+ pr.fields['r_addend']
             symbol_type = symbol_type + 0x05  # PATCH_FUNC_ARM32_THM
+            #print(f" *> {pr.type} {patch_value=} {symbol_address=} {pr.fields['r_addend']=} {pr.bits=}, {function_offset=} {patch_offset=}")
+
+        elif pr.type == 'R_ARM_THM_MOVW_ABS_NC':
+            # (S + A) & 0xFFFF
+            mask = 0xFFFF
+            patch_value = symbol_address + pr.fields['r_addend']
+            symbol_type = symbol_type + 0x06  # PATCH_OBJECT_ARM32_ABS_THM
+            #print(f" *> {pr.type} {patch_value=} {symbol_address=}, {function_offset=}, {pr.fields['r_addend']=}")
+        
+        elif pr.type == 'R_ARM_THM_MOVT_ABS':
+            # (S + A) & 0xFFFF0000
+            mask = 0xFFFF0000
+            patch_value = symbol_address + pr.fields['r_addend']
+            symbol_type = symbol_type + 0x06  # PATCH_OBJECT_ARM32_ABS_THM
+            scale = 0x10000
+            #print(f" *> {pr.type} {patch_value=} {symbol_address=}, {function_offset=}, {pr.fields['r_addend']=}")
 
         else:
             raise NotImplementedError(f"Relocation type {pr.type} in {relocation.pelfy_reloc.target_section.name} pointing to {relocation.pelfy_reloc.symbol.name} not implemented")

src/coparun/runmem.c

@@ -50,6 +50,32 @@ void patch_arm32_abs(uint8_t *patch_addr, uint32_t imm16)
     *((uint32_t *)patch_addr) = instr;
 }
 
+void patch_arm_thm_abs(uint8_t *patch_addr, uint32_t imm16)
+{
+    // Thumb MOVW (T3) / MOVT (T1) encoding
+    
+    uint16_t *instr16 = (uint16_t *)patch_addr;
+    uint16_t first_half = instr16[0];
+    uint16_t second_half = instr16[1];
+
+    // Extract fields from imm16
+    uint32_t imm4 = (imm16 >> 12) & 0xF;
+    uint32_t i    = (imm16 >> 11) & 0x1;
+    uint32_t imm3 = (imm16 >> 8) & 0x7;
+    uint32_t imm8 = imm16 & 0xFF;
+
+    // Clear bits
+    first_half &= (uint16_t)(~(0x000F | (1 << 10)));
+    second_half &= (uint16_t)(~(0x00FF | (0x7 << 12)));
+
+    // Set new fields
+    first_half |= (uint16_t)((imm4 << 0) | (i << 10));
+    second_half |= (uint16_t)(imm8 | (imm3 << 12));
+
+    instr16[0] = first_half;
+    instr16[1] = second_half;
+}
+
 void patch_arm_thm_jump24(uint8_t *patch_addr, int32_t imm24)
 {
     // Read the 32-bit instruction (two halfwords)
@@ -77,8 +103,8 @@ void patch_arm_thm_jump24(uint8_t *patch_addr, int32_t imm24)
     second_half &= 0xD000;           // Keep upper 5 bits
 
     // Set new imm fields
-    first_half  |= (S << 10) | imm10;
+    first_half  |= (uint16_t)((S << 10) | imm10);
-    second_half |= (J1 << 13) | (J2 << 11) | imm11;
+    second_half |= (uint16_t)((J1 << 13) | (J2 << 11) | imm11);
 
     // Write back
     instr16[0] = first_half;
@@ -225,6 +251,16 @@ int parse_commands(runmem_t *context, uint8_t *bytes) {
                 patch_arm_thm_jump24(context->executable_memory + offs, value);
                 break;
 
+            case PATCH_OBJECT_ARM32_ABS_THM:
+                offs = *(uint32_t*)bytes; bytes += 4;
+                patch_mask = *(uint32_t*)bytes; bytes += 4;
+                patch_scale = *(int32_t*)bytes; bytes += 4;
+                value = *(int32_t*)bytes; bytes += 4;
+                LOG("PATCH_OBJECT_ARM32_ABS_THM patch_offs=%i patch_mask=%#08x scale=%i value=%i imm16=%#04x\n",
+                    offs, patch_mask, patch_scale, value, (uint32_t)((uintptr_t)(context->data_memory + value) & patch_mask) / (uint32_t)patch_scale);
+                patch_arm_thm_abs(context->executable_memory + offs, (uint32_t)((uintptr_t)(context->data_memory + value) & patch_mask) / (uint32_t)patch_scale);
+                break;
+
             case ENTRY_POINT:
                 rel_entr_point = *(uint32_t*)bytes; bytes += 4;
                 context->entr_point = (entry_point_t)(context->executable_memory + rel_entr_point);

src/coparun/runmem.h

@@ -26,6 +26,7 @@
 #define PATCH_OBJECT_ABS  0x2002
 #define PATCH_OBJECT_REL  0x2003
 #define PATCH_OBJECT_ARM32_ABS 0x2004
+#define PATCH_OBJECT_ARM32_ABS_THM 0x2006
 #define ENTRY_POINT       7
 #define RUN_PROG         64
 #define READ_DATA        65

tests/test_ops_armv7thumb.py

@@ -55,6 +55,9 @@ def function1(c1: NumLike) -> list[NumLike]:
             c1 + 4, c1 - 4,
             c1 > 2, c1 > 100, c1 < 4, c1 < 100]
 
+def function1ex(c1: NumLike) -> list[NumLike]:
+    return [c1 // 4]
+
 
 def function2(c1: NumLike) -> list[NumLike]:
     return [c1 * 4.44, c1 * -4.44]
@@ -90,24 +93,21 @@ def test_compile():
     c_f = value(1.111)
     c_b = value(True)
 
-    #ret_test = function1(c_i) + function1(c_f) + function2(c_i) + function2(c_f) + function3(c_i) + function4(c_i) + function5(c_b) + [value(9) % 2] + iiftests(c_i) + iiftests(c_f) + [cp.asin(c_i/10)]
+    ret_test = function1(c_i) + function1(c_f) + function2(c_i) + function2(c_f) + function3(c_i) + function4(c_i) + function5(c_b) + [value(9) % 2] + iiftests(c_i) + iiftests(c_f) + [cp.asin(c_i/10)]
-    #ret_ref = function1(9) + function1(1.111) + function2(9) + function2(1.111) + function3(9) + function4(9) + function5(True) + [9 % 2] + iiftests(9) + iiftests(1.111) + [cp.asin(9/10)]
+    ret_ref = function1(9) + function1(1.111) + function2(9) + function2(1.111) + function3(9) + function4(9) + function5(True) + [9 % 2] + iiftests(9) + iiftests(1.111) + [cp.asin(9/10)]
-
-    ret_test = (c_i * 100 + 5,)
-    ret_ref = (9 * 100 + 5,)
 
     out = [Store(r) for r in ret_test]
 
-    sdb = backend.stencil_db_from_package('armv7thumb')
+    sdb = backend.stencil_db_from_package('armv7mthumb')
     dw, variables = compile_to_dag(out, sdb)
 
     #dw.write_com(_binwrite.Command.READ_DATA)
     #dw.write_int(0)
     #dw.write_int(28)
 
-    # run program command
+    du = dw.copy()
-    #dw.write_com(_binwrite.Command.RUN_PROG)
+    dw.write_com(_binwrite.Command.RUN_PROG)
-    dw.write_com(_binwrite.Command.DUMP_CODE)
+    du.write_com(_binwrite.Command.DUMP_CODE)
 
     for v in ret_test:
         assert isinstance(v, value)
@@ -118,11 +118,13 @@ def test_compile():
     #dw.write_int(28)
 
     dw.write_com(_binwrite.Command.END_COM)
+    du.write_com(_binwrite.Command.END_COM)
 
     #print('* Data to runner:')
     #dw.print()
 
     dw.to_file('build/runner/test-armv7thumb.copapy')
+    du.to_file('build/runner/test-armv7thumb-dump.copapy')
 
     if not check_for_qemu():
         warnings.warn("qemu-armv7 not found, armv7 test skipped!", UserWarning)
@@ -131,12 +133,14 @@ def test_compile():
         warnings.warn("armv7thumb runner not found, armv7thumb test skipped!", UserWarning)
         return
 
-    command = qemu_command + ['build/runner/coparun-armv7thumb', 'build/runner/test-armv7thumb.copapy'] + ['build/runner/test.copapy-armv7thumb.bin']
+    print('----- Dump code...')
-    #try:
+    command = qemu_command + ['build/runner/coparun-armv7thumb', 'build/runner/test-armv7thumb-dump.copapy', 'build/runner/test.copapy-armv7thumb.bin']
     result = run_command(command)
-    #except FileNotFoundError:
+
-    #    warnings.warn(f"Test skipped, executable not found.", UserWarning)
+    print('----- Run code...')
-    #    return
+    command = qemu_command + ['build/runner/coparun-armv7thumb', 'build/runner/test-armv7thumb.copapy']
+    result = run_command(command)
+
 
     print('* Output from runner:\n--')
     print(result)
@@ -167,5 +171,14 @@ def test_compile():
 
 
 if __name__ == "__main__":
-    #test_example()
+    test_compile()
-    test_slow_31bit_int_list_hash()
+
+
+"""
+qemu-arm -d in_asm,exec,cpu_reset -D qemu.log build/runner/coparun-armv7thumb build/runner/test-armv7thumb.copapy build/runner/test.copapy-armv7thumb.bin
+
+qemu-arm -d in_asm,exec -D qemu_trace.log \
+  -global driver=pl011.audiomaddr,property=addr,value=0xff7ec000 \
+  -global driver=pl011.audiomaddr,property=size,value=0x100000 \
+  your_binary
+"""

tools/build.sh

@@ -4,10 +4,10 @@ set -eu
 ARCH=${1:-x86_64}
 
 case "$ARCH" in
-    (x86_64|arm-v6|arm-v7|arm-v7-thumb|arm-v7m-thumb|all)
+    (x86_64|arm64|arm-v6|arm-v7|arm-v7-thumb|arm-v7m-thumb|all)
         ;;
     (*)
-        echo "Usage: $0 [x86_64|arm-v6|arm-v7|arm-v6-thumb|arm-v7m-thumb|all]"
+        echo "Usage: $0 [x86_64|arm64|arm-v6|arm-v7|arm-v6-thumb|arm-v7m-thumb|all]"
         exit 1
         ;;
 esac
@@ -42,13 +42,44 @@ if [[ "$ARCH" == "x86_64" || "$ARCH" == "all" ]]; then
         -o build/runner/coparun
 fi
 
+#######################################
+# ARM 64
+#######################################
+if [[ "$ARCH" == "arm64" || "$ARCH" == "all" ]]; then
+    echo "--------------arm64----------------"
+
+    LIBGCC=$(aarch64-linux-gnu-gcc -print-libgcc-file-name)
+
+    aarch64-linux-gnu-gcc -fno-pic -ffunction-sections \
+        -c $SRC -O3 -o build/stencils/stencils.o
+
+    aarch64-linux-gnu-ld -r \
+        build/stencils/stencils.o \
+        build/musl/musl_objects_arm64.o \
+        $LIBGCC \
+        -o $DEST/stencils_arm64_O3.o
+
+    aarch64-linux-gnu-objdump -d -x \
+        $DEST/stencils_arm64_O3.o \
+        > build/stencils/stencils_arm64_O3.asm
+
+    aarch64-linux-gnu-gcc \
+        -Wall -Wextra -Wconversion -Wsign-conversion -static \
+        -Wshadow -Wstrict-overflow -O3 \
+        -DENABLE_LOGGING \
+        src/coparun/runmem.c \
+        src/coparun/coparun.c \
+        src/coparun/mem_man.c \
+        -o build/runner/coparun-arm64
+fi
+
 #######################################
 # ARM v6
 #######################################
 if [[ "$ARCH" == "arm-v6" || "$ARCH" == "all" ]]; then
     echo "--------------arm-v6 32 bit----------------"
 
-    LIBGCC=$(arm-none-eabi-gcc -print-libgcc-file-name)
+    LIBGCC=$(arm-none-eabi-gcc -march=armv6 -mfpu=vfp -mfloat-abi=hard -marm -print-libgcc-file-name)
 
     arm-none-eabi-gcc -fno-pic -ffunction-sections \
         -march=armv6 -mfpu=vfp -mfloat-abi=hard -marm \
@@ -81,7 +112,7 @@ fi
 if [[ "$ARCH" == "arm-v7" || "$ARCH" == "all" ]]; then
     echo "--------------arm-v7 32 bit----------------"
 
-    LIBGCC=$(arm-none-eabi-gcc -print-libgcc-file-name)
+    LIBGCC=$(arm-none-eabi-gcc -march=armv7-a -mfpu=neon-vfpv3 -mfloat-abi=hard -marm -print-libgcc-file-name)
 
     arm-none-eabi-gcc -fno-pic -ffunction-sections \
         -march=armv7-a -mfpu=neon-vfpv3 -mfloat-abi=hard -marm \
@@ -114,7 +145,7 @@ fi
 if [[ "$ARCH" == "arm-v7-thumb" || "$ARCH" == "all" ]]; then
     echo "--------------arm-v7a-thumb 32 bit----------------"
 
-    LIBGCC=$(arm-none-eabi-gcc -print-libgcc-file-name)
+    LIBGCC=$(arm-none-eabi-gcc -march=armv7 -mfpu=vfp3 -mthumb -print-libgcc-file-name)
 
     arm-none-eabi-gcc -fno-pic -ffunction-sections \
         -march=armv7-a -mfpu=neon-vfpv3 -mfloat-abi=hard -mthumb \
@@ -122,7 +153,7 @@ if [[ "$ARCH" == "arm-v7-thumb" || "$ARCH" == "all" ]]; then
 
     arm-none-eabi-ld -r \
         build/stencils/stencils.o \
-        build/musl/musl_objects_armv7.o \
+        build/musl/musl_objects_armv7thumb.o \
         $LIBGCC \
         -o $DEST/stencils_armv7thumb_O3.o
 
@@ -147,7 +178,7 @@ fi
 if [[ "$ARCH" == "arm-v7m-thumb" || "$ARCH" == "all" ]]; then
     echo "--------------arm-v7m-thumb 32 bit----------------"
 
-    LIBGCC=$(arm-none-eabi-gcc -print-libgcc-file-name)
+    LIBGCC=$(arm-none-eabi-gcc -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard -mthumb -print-libgcc-file-name)
 
     arm-none-eabi-gcc -fno-pic -ffunction-sections \
         -march=armv7e-m -mfpu=fpv4-sp-d16 -mfloat-abi=hard -mthumb \

tools/test_thumb_stancils.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+
+bash tools/build.sh arm-v7-thumb
+python tests/test_ops_armv7thumb.py
+qemu-arm -d in_asm -D qemu.log build/runner/coparun-armv7thumb build/runner/test-armv7thumb.copapy build/runner/test.copapy-armv7thumb.bin
+arm-none-eabi-objdump -D -b binary -marm -M force-thumb --adjust-vma=0xff7ed000 build/runner/test.copapy-armv7thumb.bin > build/runner/test.copapy-armv7thumb.asm