Added new_aligned and alloc_aligned functions to prevent accident…

…al under-alignment when allocating simd.

lib/std/core/mem.c3

@@ -536,6 +536,7 @@ fn void* tmalloc(usz size, usz alignment = 0, usz offset = 0) @builtin @inline @
 /**
  * @require $vacount < 2 : "Too many arguments."
  * @require $or($vacount == 0, $assignable($vaexpr(0), $Type)) : "The second argument must be an initializer for the type"
+ * @require $Type.alignof <= DEFAULT_MEM_ALIGNMENT : "Types with alignment exceeding the default must use 'alloc_aligned' instead"
  **/
 macro new($Type, ...) @nodiscard
 {
@@ -548,11 +549,40 @@ macro new($Type, ...) @nodiscard
 	$endif
 }
 
+/**
+ * Allocate using an aligned allocation. This is necessary for types with a default memory alignment
+ * exceeding DEFAULT_MEM_ALIGNMENT. IMPORTANT! It must be freed using free_aligned.
+ * @require $vacount < 2 : "Too many arguments."
+ * @require $or($vacount == 0, $assignable($vaexpr(0), $Type)) : "The second argument must be an initializer for the type"
+ **/
+macro new_aligned($Type, ...) @nodiscard
+{
+	$if $vacount == 0:
+		return ($Type*)calloc_aligned($Type.sizeof, $Type.alignof);
+	$else
+		$Type* val = malloc_aligned($Type.sizeof, $Type.alignof);
+		*val = $vaexpr(0);
+		return val;
+	$endif
+}
+
+/**
+ * @require $Type.alignof <= DEFAULT_MEM_ALIGNMENT : "Types with alignment exceeding the default must use 'alloc_aligned' instead"
+ **/
 macro alloc($Type) @nodiscard
 {
 	return ($Type*)malloc($Type.sizeof);
 }
 
+/**
+ * Allocate using an aligned allocation. This is necessary for types with a default memory alignment
+ * exceeding DEFAULT_MEM_ALIGNMENT. IMPORTANT! It must be freed using free_aligned.
+ **/
+macro alloc_aligned($Type) @nodiscard
+{
+	return ($Type*)malloc_aligned($Type.sizeof, $Type.alignof);
+}
+
 macro new_clear($Type) @deprecated("Use mem::new")
 {
 	return new($Type);
@@ -588,16 +618,42 @@ macro new_temp_clear($Type) @deprecated("use mem::temp_new")
 	return tcalloc($Type.sizeof);
 }
 
+
+
+/**
+ * @require $Type.alignof <= DEFAULT_MEM_ALIGNMENT : "Types with alignment exceeding the default must use 'new_array_aligned' instead"
+ **/
 macro new_array($Type, usz elements) @nodiscard
 {
 	return allocator::new_array(allocator::heap(), $Type, elements);
 }
 
+/**
+ * Allocate using an aligned allocation. This is necessary for types with a default memory alignment
+ * exceeding DEFAULT_MEM_ALIGNMENT. IMPORTANT! It must be freed using free_aligned.
+ **/
+macro new_array_aligned($Type, usz elements) @nodiscard
+{
+	return allocator::new_array_aligned(allocator::heap(), $Type, elements);
+}
+
+/**
+ * @require $Type.alignof <= DEFAULT_MEM_ALIGNMENT : "Types with alignment exceeding the default must use 'alloc_array_aligned' instead"
+ **/
 macro alloc_array($Type, usz elements) @nodiscard
 {
 	return allocator::alloc_array(allocator::heap(), $Type, elements);
 }
 
+/**
+ * Allocate using an aligned allocation. This is necessary for types with a default memory alignment
+ * exceeding DEFAULT_MEM_ALIGNMENT. IMPORTANT! It must be freed using free_aligned.
+ **/
+macro alloc_array_aligned($Type, usz elements) @nodiscard
+{
+	return allocator::alloc_array(allocator::heap(), $Type, elements);
+}
+
 macro talloc_array($Type, usz elements) @nodiscard @deprecated("use mem::temp_alloc_array")
 {
 	return temp_alloc_array($Type, elements);
@@ -633,6 +689,11 @@ fn void* calloc(usz size) @builtin @inline @nodiscard
 	return allocator::calloc(allocator::heap(), size);
 }
 
+fn void* calloc_aligned(usz size, usz alignment) @builtin @inline @nodiscard
+{
+	return allocator::calloc_aligned(allocator::heap(), size, alignment)!!;
+}
+
 fn void* tcalloc(usz size, usz alignment = 0, usz offset = 0) @builtin @inline @nodiscard
 {
 	return allocator::temp().acquire(size, false, alignment, offset)!!;
@@ -643,11 +704,21 @@ fn void* realloc(void *ptr, usz new_size) @builtin @inline @nodiscard
 	return allocator::realloc(allocator::heap(), ptr, new_size);
 }
 
+fn void* realloc_aligned(void *ptr, usz new_size, usz alignment) @builtin @inline @nodiscard
+{
+	return allocator::realloc_aligned(allocator::heap(), ptr, new_size, alignment)!!;
+}
+
 fn void free(void* ptr) @builtin @inline
 {
 	return allocator::free(allocator::heap(), ptr);
 }
 
+fn void free_aligned(void* ptr) @builtin @inline
+{
+	return allocator::free_aligned(allocator::heap(), ptr);
+}
+
 fn void* trealloc(void* ptr, usz size, usz alignment = mem::DEFAULT_MEM_ALIGNMENT) @builtin @inline @nodiscard
 {
 	return allocator::temp().resize(ptr, size, alignment, 0)!!;

lib/std/core/mem_allocator.c3

@@ -166,11 +166,21 @@ macro new_array_try(Allocator* allocator, $Type, usz elements) @nodiscard
 	return (($Type*)calloc_try(allocator, $Type.sizeof * elements))[:elements];
 }
 
+macro new_array_aligned(Allocator* allocator, $Type, usz elements) @nodiscard
+{
+	return ((Type*)calloc_aligned(allocator, $Type.sizeof * elements, $Type.alignof))[:elements]!!;
+}
+
 macro alloc_array(Allocator* allocator, $Type, usz elements) @nodiscard
 {
 	return alloc_array_try(allocator, $Type, elements)!!;
 }
 
+macro alloc_array_aligned(Allocator* allocator, $Type, usz elements) @nodiscard
+{
+	return ((Type*)malloc_aligned(allocator, $Type.sizeof * elements, $Type.alignof))[:elements]!!;
+}
+
 macro alloc_array_try(Allocator* allocator, $Type, usz elements) @nodiscard
 {
 	return (($Type*)malloc_try(allocator, $Type.sizeof * elements))[:elements];

releasenotes.md

@@ -10,7 +10,7 @@
 - `math::pow` will now correctly promote integer arguments.
 
 ### Stdlib changes
-None
+- Added `new_aligned` and `alloc_aligned` functions to prevent accidental under-alignment when allocating simd.
 
 ## 0.5.4 Change list
 

src/compiler/target.c

@@ -4,7 +4,7 @@
 #include "compiler_internal.h"
 
 extern void LLVMSetTargetMachineUseInitArray(LLVMTargetMachineRef ref, bool use_init_array);
-
+static bool x64features_contains(X86Features *cpu_features, X86Feature feature);
 static ObjectFormatType object_format_from_os(OsType os, ArchType arch_type);
 static unsigned arch_pointer_bit_width(OsType os, ArchType arch);
 static ArchType arch_from_llvm_string(StringSlice string);
@@ -47,7 +47,7 @@ bool arch_is_wasm(ArchType type)
 	return type == ARCH_TYPE_WASM32 || type == ARCH_TYPE_WASM64;
 }
 
-static AlignSize os_arch_max_alignment_of_vector(OsType os, ArchType arch, EnvironmentType type, ARMVariant variant)
+static AlignSize os_arch_max_alignment_of_vector(OsType os, ArchType arch, EnvironmentType type, ARMVariant variant, X86Features* features)
 {
 	switch (arch)
 	{
@@ -71,12 +71,14 @@ static AlignSize os_arch_max_alignment_of_vector(OsType os, ArchType arch, Envir
 		case ARCH_TYPE_X86:
 			if (os == OS_TYPE_WIN32) /* COFF */
 			{
-				return 8192;
+				return 8192 / 8;
 			}
 			if (os_is_apple(os))
 			{
 				// With AVX512 - 512, AVX - 256 otherwise AVX - 128
-				return 256;
+				if (x64features_contains(features, X86_FEAT_AVX512F)) return 512 / 8;
+				if (x64features_contains(features, X86_FEAT_AVX)) return 256 / 8;
+				return 128 / 8;
 			}
 			break;
 		default:
@@ -393,6 +395,7 @@ static char *x86_feature_name[] = {
 		[X86_FEAT_AVXVNNIINT8] = "avxvnniint8",
 		[X86_FEAT_AVXVNNIINT16] = "avxvnniint16",
 };
+
 static X86Feature x86feature_from_string(const char *str)
 {
 	for (int i = 0; i <= X86_FEATURE_LAST; i++)
@@ -1928,7 +1931,11 @@ void target_setup(BuildTarget *target)
 			platform_target.abi = ABI_UNKNOWN;
 			break;
 	}
-	platform_target.align_max_vector = os_arch_max_alignment_of_vector(platform_target.os, platform_target.arch, platform_target.environment_type, platform_target.arm.variant);
+	platform_target.align_max_vector = os_arch_max_alignment_of_vector(platform_target.os,
+																	   platform_target.arch,
+																	   platform_target.environment_type,
+																	   platform_target.arm.variant,
+																	   &platform_target.x64.features);
 	platform_target.align_max_tls = os_arch_max_alignment_of_tls(platform_target.os,
 																 platform_target.arch,
 																 platform_target.environment_type);