- Circumvent Aarch64 miscompilations of atomics.

- Fixes to ByteBuffer allocation/free.
- Fix issue where compiling both for asm and object file would corrupt the obj file output.

.github/workflows/main.yml

@@ -87,15 +87,16 @@ jobs:
           ..\build\${{ matrix.build_type }}\c3c.exe compile --lib raylib5 --print-linking examples\raylib\raylib_snake.c3
           ..\build\${{ matrix.build_type }}\c3c.exe compile --lib raylib5 --print-linking examples\raylib\raylib_tetris.c3
 
-      - name: run compiler tests
+      - name: Compile run unit tests
         run: |
           cd test
-          python3.exe src/tester.py ..\build\${{ matrix.build_type }}\c3c.exe test_suite/
+          ..\build\${{ matrix.build_type }}\c3c.exe compile-test unit -O1
 
-      - name: Compile run unit tests
+      - name: run compiler tests
         run: |
           cd test
-          ..\build\${{ matrix.build_type }}\c3c.exe compile-test unit -O1
+          python3.exe src/tester.py ..\build\${{ matrix.build_type }}\c3c.exe test_suite/
+
 
       - name: Test python script
         run: |
@@ -486,7 +487,7 @@ jobs:
       - name: Compile run unit tests
         run: |
           cd test
-          ../build/c3c compile-test unit
+          ../build/c3c compile-test unit --sanitize=address
 
       - name: Build testproject
         run: |
@@ -660,7 +661,7 @@ jobs:
       - name: Compile run unit tests
         run: |
           cd test
-          ../build/c3c compile-test unit
+          ../build/c3c compile-test unit -O1
 
       - name: Test WASM
         run: |

lib/std/io/stream/bytebuffer.c3

@@ -16,15 +16,15 @@ struct ByteBuffer (InStream, OutStream)
  max_read defines how many bytes might be kept before its internal buffer is shrinked.
  @require self.bytes.len == 0 "Buffer already initialized."
 *>
-fn ByteBuffer*! ByteBuffer.new_init(&self, usz max_read, usz initial_capacity = 16, Allocator allocator = allocator::heap())
+fn ByteBuffer* ByteBuffer.new_init(&self, usz max_read, usz initial_capacity = 16, Allocator allocator = allocator::heap())
 {
 	*self = { .allocator = allocator, .max_read = max_read };
 	initial_capacity = max(initial_capacity, 16);
-	self.grow(initial_capacity)!;
+	self.grow(initial_capacity);
 	return self;
 }
 
-fn ByteBuffer*! ByteBuffer.temp_init(&self, usz max_read, usz initial_capacity = 16)
+fn ByteBuffer* ByteBuffer.temp_init(&self, usz max_read, usz initial_capacity = 16)
 {
 	return self.new_init(max_read, initial_capacity, allocator::temp());
 }
@@ -33,7 +33,7 @@ fn ByteBuffer*! ByteBuffer.temp_init(&self, usz max_read, usz initial_capacity =
  @require buf.len > 0
  @require self.bytes.len == 0 "Buffer already initialized."
 *>
-fn ByteBuffer*! ByteBuffer.init_with_buffer(&self, char[] buf)
+fn ByteBuffer* ByteBuffer.init_with_buffer(&self, char[] buf)
 {
 	*self = { .max_read = buf.len, .bytes = buf };
 	return self;
@@ -48,7 +48,7 @@ fn void ByteBuffer.free(&self)
 fn usz! ByteBuffer.write(&self, char[] bytes) @dynamic
 {
 	usz cap = self.bytes.len - self.write_idx;
-	if (cap < bytes.len) self.grow(bytes.len)!;
+	if (cap < bytes.len) self.grow(bytes.len);
 	self.bytes[self.write_idx:bytes.len] = bytes[..];
 	self.write_idx += bytes.len;
 	return bytes.len;
@@ -57,7 +57,7 @@ fn usz! ByteBuffer.write(&self, char[] bytes) @dynamic
 fn void! ByteBuffer.write_byte(&self, char c) @dynamic
 {
 	usz cap = self.bytes.len - self.write_idx;
-	if (cap == 0) self.grow(1)!;
+	if (cap == 0) self.grow(1);
 	self.bytes[self.write_idx] = c;
 	self.write_idx++;
 }
@@ -128,10 +128,10 @@ fn usz! ByteBuffer.available(&self) @inline @dynamic
 	return self.write_idx - self.read_idx;
 }
 
-fn void! ByteBuffer.grow(&self, usz n)
+fn void ByteBuffer.grow(&self, usz n)
 {
 	n = math::next_power_of_2(n);
-	char* p = allocator::realloc_aligned(self.allocator, self.bytes, n, alignment: char.alignof)!;
+	char* p = allocator::realloc(self.allocator, self.bytes, n);
 	self.bytes = p[:n];
 }
 

releasenotes.md

@@ -63,6 +63,9 @@
 - Fix issue where aligned bitstructs did not store/load with the given alignment.
 - Fix issue in GrowableBitSet with sanitizers.
 - Fix issue in List with sanitizers.
+- Circumvent Aarch64 miscompilations of atomics.
+- Fixes to ByteBuffer allocation/free.
+- Fix issue where compiling both for asm and object file would corrupt the obj file output.
 
 ### Stdlib changes
 - Added '%h' and '%H' for printing out binary data in hexadecimal using the formatter.

src/compiler/llvm_codegen.c

@@ -664,47 +664,48 @@ static void gencontext_verify_ir(GenContext *context)
 	}
 }
 
-
-static void gencontext_emit_object_file(GenContext *context)
+static void llvm_emit_file(GenContext *c, const char *filename, LLVMCodeGenFileType llvm_codegen_type, bool clone_module)
 {
-	char *err = "";
 	DEBUG_LOG("Target: %s", compiler.platform.target_triple);
-	LLVMSetTarget(context->module, compiler.platform.target_triple);
-	char *layout = LLVMCopyStringRepOfTargetData(context->target_data);
-	LLVMSetDataLayout(context->module, layout);
+	LLVMModuleRef module = clone_module ? LLVMCloneModule(c->module) : c->module;
+	LLVMSetTarget(module, compiler.platform.target_triple);
+	char *layout = LLVMCopyStringRepOfTargetData(c->target_data);
+	LLVMSetDataLayout(module, layout);
 	LLVMDisposeMessage(layout);
 
-	if (context->asm_filename)
+	char *err = "";
+	FILE *file = NULL;
+	LLVMMemoryBufferRef buffer = NULL;
+	if (LLVMTargetMachineEmitToMemoryBuffer(c->machine, module, llvm_codegen_type, &err, &buffer))
 	{
-		// Generate .s file
-		if (LLVMTargetMachineEmitToFile(context->machine, context->module, (char *)context->asm_filename, LLVMAssemblyFile, &err))
-		{
-			error_exit("Could not emit asm file: %s", err);
-		}
+		goto ERR;
 	}
-
-	// Generate .o or .obj file
-	if (LLVMTargetMachineEmitToFile(context->machine, context->module, (char *)context->object_filename, LLVMObjectFile, &err))
+	file = fopen(filename, "wb");
+	if (!file)
 	{
-		error_exit("Could not emit object file: %s", err);
+		err = "File could not be opened";
+		goto ERR;
 	}
-
-}
-
-static void llvm_emit_asm_file(GenContext *context)
-{
-	char *err = "";
-	DEBUG_LOG("Target: %s", compiler.platform.target_triple);
-	LLVMSetTarget(context->module, compiler.platform.target_triple);
-	char *layout = LLVMCopyStringRepOfTargetData(context->target_data);
-	LLVMSetDataLayout(context->module, layout);
-	LLVMDisposeMessage(layout);
-
-	// Generate .s file
-	if (LLVMTargetMachineEmitToFile(context->machine, context->module, (char *)context->asm_filename, LLVMAssemblyFile, &err))
+	size_t len = LLVMGetBufferSize(buffer);
+	const char *ptr = LLVMGetBufferStart(buffer);
+	while (len > 0)
 	{
-		error_exit("Could not emit asm file: %s", err);
+		size_t written = fwrite(ptr, 1, len, file);
+		if (written == 0)
+		{
+			err = "Failed to write to file";
+			goto ERR;
+		}
+		ptr += written;
+		len -= written;
 	}
+	fclose(file);
+	LLVMDisposeMemoryBuffer(buffer);
+	return;
+ERR:
+	if (file) fclose(file);
+	if (buffer) LLVMDisposeMemoryBuffer(buffer);
+	error_exit("Could not emit '%s': %s", filename, err);
 }
 
 void gencontext_print_llvm_ir(GenContext *context)
@@ -1067,15 +1068,16 @@ const char *llvm_codegen(void *context)
 	}
 
 	const char *object_name = NULL;
-	if (compiler.build.emit_object_files)
+	if (compiler.build.emit_asm)
 	{
-		gencontext_emit_object_file(c);
-		object_name = c->object_filename;
+		// Clone if there will be object file output.
+		llvm_emit_file(c, c->asm_filename, LLVMAssemblyFile, compiler.build.emit_object_files);
 	}
 
-	if (compiler.build.emit_asm)
+	if (compiler.build.emit_object_files)
 	{
-		llvm_emit_asm_file(context);
+		llvm_emit_file(c, c->object_filename, LLVMObjectFile, false);
+		object_name = c->object_filename;
 	}
 
 	gencontext_end_module(c);

src/compiler/target.c

@@ -159,6 +159,16 @@ static bool os_target_signed_c_char_type(OsType os, ArchType arch)
 	}
 }
 
+static inline void target_setup_aarch64_abi(void)
+{
+	compiler.platform.aarch.is_darwin = os_is_apple(compiler.platform.os);
+	compiler.platform.aarch.is_win32 = compiler.platform.os == OS_TYPE_WIN32;
+	compiler.platform.abi = ABI_AARCH64;
+	// TODO improve Aarch64 functionality support.
+	scratch_buffer_clear();
+	scratch_buffer_append("+crc,+lse,+rdm,+fp-armv8,+neon");
+	compiler.platform.features = scratch_buffer_copy();
+}
 static inline void target_setup_arm_abi(void)
 {
 	compiler.platform.abi = ABI_ARM;
@@ -836,8 +846,7 @@ static void x86_features_from_host(X86Features *cpu_features)
 #if LLVM_AVAILABLE
 	char *features = LLVMGetHostCPUFeatures();
 	INFO_LOG("Detected the following host features: %s", features);
-	INFO_LOG("For %s",
-			 LLVMGetHostCPUName());
+	INFO_LOG("For %s", LLVMGetHostCPUName());
 
 	char *tok = strtok(features, ",");
 	*cpu_features = x86_feature_zero;
@@ -1994,9 +2003,7 @@ void target_setup(BuildTarget *target)
 			break;
 		case ARCH_TYPE_AARCH64:
 		case ARCH_TYPE_AARCH64_BE:
-			compiler.platform.aarch.is_darwin = os_is_apple(compiler.platform.os);
-			compiler.platform.aarch.is_win32 = compiler.platform.os == OS_TYPE_WIN32;
-			compiler.platform.abi = ABI_AARCH64;
+			target_setup_aarch64_abi();
 			break;
 		case ARCH_TYPE_XTENSA:
 			compiler.platform.abi = ABI_XTENSA;

test/unit/stdlib/io/bytebuffer.c3

@@ -4,7 +4,7 @@ import std::io;
 fn void write_read()
 {
 	ByteBuffer buffer;
-	buffer.new_init(0)!!;
+	buffer.new_init(0);
 	defer buffer.free();
 	buffer.write("hello")!!;
 

test/unit/stdlib/io/varint.c3

@@ -4,7 +4,7 @@ import std::io;
 fn void write_read()
 {
 	ByteBuffer buf;
-	buf.temp_init(16)!!;
+	buf.temp_init(16);
 	usz n;
 	uint x;
 	uint y;