- Detect unaligned loads #1951.

- Fix issue where aligned bitstructs did not store/load with the given alignment.

lib/std/compression/qoi.c3

@@ -471,7 +471,7 @@ bitstruct OpDiff : char
 	char diff_green : 2..3;
 	char diff_blue : 0..1;
 }
-bitstruct OpLuma : ushort
+bitstruct OpLuma : ushort @align(1)
 {
 	char tag : 6..7;
 	char diff_green : 0..5;

lib/std/hash/md5.c3

@@ -108,7 +108,7 @@ macro @i(x, y, z) => y ^ (x | ~z);
 
 macro @step(#f, a, b, c, d, ptr, n, t, s)
 {
-	*a += #f(b, c, d) + *(uint *)&ptr[n * 4] + t;
+	*a += #f(b, c, d) + @unaligned_load(*(uint *)&ptr[n * 4], 2) + t;
 	*a = (*a << s) | ((*a & 0xffffffff) >> (32 - s));
 	*a += b;
 }

releasenotes.md

@@ -22,6 +22,7 @@
 - Allow `fn int test() => @pool() { return 1; }` short function syntax usage #1906.
 - Test runner will also check for leaks.
 - Improve inference on `??` #1943.
+- Detect unaligned loads #1951.
 
 ### Fixes
 - Fix issue requiring prefix on a generic interface declaration.
@@ -59,6 +60,7 @@
 - `write` of qoi would leak memory.
 - Issue when having an empty `Path` or just "."
 - `set_env` would leak memory.
+- Fix issue where aligned bitstructs did not store/load with the given alignment.
 
 ### Stdlib changes
 - Added '%h' and '%H' for printing out binary data in hexadecimal using the formatter.

src/compiler/codegen_internal.h

@@ -13,7 +13,6 @@ enum IntrospectIndex
 };
 
 bool type_is_homogenous_aggregate(Type *type, Type **base, unsigned *elements);
-static inline Type *type_reduced_from_expr(Expr *expr);
 static inline bool abi_type_is_type(AbiType type);
 
 static inline bool abi_type_is_valid(AbiType type);
@@ -90,11 +89,6 @@ static inline Type *type_lowering(Type *type)
 	}
 }
 
-static inline Type *type_reduced_from_expr(Expr *expr)
-{
-	return type_lowering(expr->type);
-}
-
 static inline bool abi_type_is_type(AbiType type)
 {
 	return !(type.int_bits_plus_1 & 0x01);

src/compiler/llvm_codegen_builtins.c

@@ -160,12 +160,15 @@ INLINE void llvm_emit_atomic_store(GenContext *c, BEValue *result_value, Expr *e
 
 INLINE void llvm_emit_unaligned_store(GenContext *c, BEValue *result_value, Expr *expr)
 {
+	bool emit_check = c->emitting_load_store_check;
+	c->emitting_load_store_check = true;
 	BEValue value;
 	llvm_emit_expr(c, &value, expr->call_expr.arguments[0]);
 	llvm_emit_expr(c, result_value, expr->call_expr.arguments[1]);
 	llvm_value_deref(c, &value);
 	value.alignment = expr->call_expr.arguments[2]->const_expr.ixx.i.low;
 	llvm_store(c, &value, result_value);
+	c->emitting_load_store_check = emit_check;
 }
 
 INLINE void llvm_emit_atomic_fetch(GenContext *c, BuiltinFunction func, BEValue *result_value, Expr *expr)
@@ -241,10 +244,13 @@ INLINE void llvm_emit_atomic_load(GenContext *c, BEValue *result_value, Expr *ex
 
 INLINE void llvm_emit_unaligned_load(GenContext *c, BEValue *result_value, Expr *expr)
 {
+	bool emit_check = c->emitting_load_store_check;
+	c->emitting_load_store_check = true;
 	llvm_emit_expr(c, result_value, expr->call_expr.arguments[0]);
 	llvm_value_deref(c, result_value);
 	result_value->alignment = expr->call_expr.arguments[1]->const_expr.ixx.i.low;
 	llvm_value_rvalue(c, result_value);
+	c->emitting_load_store_check = emit_check;
 }
 
 static inline LLVMValueRef llvm_syscall_asm(LLVMTypeRef func_type, char *call)

src/compiler/llvm_codegen_expr.c

@@ -2286,6 +2286,7 @@ static inline void llvm_emit_deref(GenContext *c, BEValue *value, Expr *inner, T
 	}
 	llvm_emit_expr(c, value, inner);
 	llvm_value_rvalue(c, value);
+	AlignSize alignment = type_abi_alignment(type);
 	if (safe_mode_enabled())
 	{
 		LLVMValueRef check = LLVMBuildICmp(c->builder, LLVMIntEQ, value->value, llvm_get_zero(c, inner->type), "checknull");
@@ -2294,11 +2295,28 @@ static inline void llvm_emit_deref(GenContext *c, BEValue *value, Expr *inner, T
 		span_to_scratch(inner->span);
 		scratch_buffer_append("' was null.");
 		llvm_emit_panic_on_true(c, check, scratch_buffer_to_string(), inner->span, NULL, NULL, NULL);
+		if (alignment > 1 && !c->emitting_load_store_check)
+		{
+			LLVMValueRef as_int = LLVMBuildPtrToInt(c->builder, value->value, llvm_get_type(c, type_usz), "");
+			LLVMValueRef align = llvm_const_int(c, type_usz, alignment);
+			LLVMValueRef rem = LLVMBuildURem(c->builder, as_int, align, "");
+			LLVMValueRef is_not_zero = LLVMBuildICmp(c->builder, LLVMIntNE, rem, llvm_get_zero(c, type_usz), "");
+			c->emitting_load_store_check = true;
+			BEValue value1;
+			BEValue value2;
+			if (inner->type->name )
+			llvm_value_set(&value1, align, type_usz);
+			llvm_value_set(&value2, rem, type_usz);
+			llvm_emit_panic_on_true(c, is_not_zero, "Unaligned pointer access detected", inner->span, "Unaligned access: ptr %% %s = %s, use @unaligned_load / @unaligned_store for unaligned access.",
+				&value1, &value2);
+			c->emitting_load_store_check = false;
+		}
+
 	}
 	// Convert pointer to address
 	value->kind = BE_ADDRESS;
-	value->type = type;
-	value->alignment = type_abi_alignment(type);
+	value->type = type_lowering(type);
+	value->alignment = alignment;
 }
 
 /**
@@ -2342,7 +2360,7 @@ static void llvm_emit_dynamic_method_addr(GenContext *c, BEValue *value, Expr *e
 
 static void llvm_emit_unary_expr(GenContext *c, BEValue *value, Expr *expr)
 {
-	Type *type = type_reduced_from_expr(expr->unary_expr.expr);
+	Type *type = type_lowering(expr->unary_expr.expr->type);
 	Expr *inner = expr->unary_expr.expr;
 
 	switch (expr->unary_expr.operator)
@@ -2430,7 +2448,7 @@ static void llvm_emit_unary_expr(GenContext *c, BEValue *value, Expr *expr)
 			value->type = type_lowering(expr->type);
 			return;
 		case UNARYOP_DEREF:
-			llvm_emit_deref(c, value, inner, type_lowering(expr->type));
+			llvm_emit_deref(c, value, inner, expr->type);
 			return;
 		case UNARYOP_INC:
 			llvm_emit_pre_inc_dec(c, value, inner, 1, !expr->unary_expr.no_wrap);
@@ -4506,7 +4524,7 @@ static inline void llvm_emit_const_initializer_list_expr(GenContext *c, BEValue
 
 static void llvm_emit_const_expr(GenContext *c, BEValue *be_value, Expr *expr)
 {
-	Type *type = type_reduced_from_expr(expr)->canonical;
+	Type *type = type_lowering(expr->type)->canonical;
 	bool is_bytes = false;
 	switch (expr->const_expr.const_kind)
 	{

src/compiler/llvm_codegen_internal.h

@@ -84,6 +84,7 @@ typedef struct GenContext_
 	bool shared_context;
 	bool in_init_ref;
 	bool weaken;
+	bool emitting_load_store_check;
 	LLVMModuleRef module;
 	LLVMBuilderRef global_builder;
 	LLVMTargetMachineRef machine;

src/compiler/types.c

@@ -683,6 +683,7 @@ AlignSize type_abi_alignment(Type *type)
 		case TYPE_WILDCARD:
 			UNREACHABLE;
 		case TYPE_BITSTRUCT:
+			if (type->decl->alignment) return type->decl->alignment;
 			type = type->decl->strukt.container_type->type;
 			goto RETRY;
 		case TYPE_INFERRED_VECTOR:

test/test_suite/bitstruct/bitstruct_align.c3t

@@ -0,0 +1,40 @@
+// #target: macos-x64
+module test;
+import std::io;
+
+Foo y;
+bitstruct Foo : ushort @align(1)
+{
+	int a : 0..10;
+}
+fn int main(String[] args)
+{
+	Foo z = y;
+	Foo* y = &z;
+	y.a = 123;
+	return 0;
+}
+
+/* #expect: test.ll
+
+@test.y = local_unnamed_addr global i16 0, align 1
+
+define i32 @test.main(ptr %0, i64 %1) #0 {
+entry:
+  %args = alloca %"char[][]", align 8
+  %z = alloca i16, align 1
+  %y = alloca ptr, align 8
+  store ptr %0, ptr %args, align 8
+  %ptradd = getelementptr inbounds i8, ptr %args, i64 8
+  store i64 %1, ptr %ptradd, align 8
+  %2 = load i16, ptr @test.y, align 1
+  store i16 %2, ptr %z, align 1
+  store ptr %z, ptr %y, align 8
+  %3 = load ptr, ptr %y, align 8
+  %4 = load i16, ptr %3, align 1
+  %5 = and i16 %4, -2048
+  %6 = or i16 %5, 123
+  store i16 %6, ptr %3, align 1
+  ret i32 0
+}
+

test/test_suite/builtins/unaligned_access.c3t

@@ -0,0 +1,82 @@
+// #target: macos-x64
+// #safe: yes
+module test;
+import std::io;
+
+fn int main()
+{
+    char[10] buf;
+
+    int* i = (int*)(&buf[1]);
+    int z = *i;
+    int z2 = @unaligned_load(*i, 1);
+    *i = 5;
+    @unaligned_store(*i, 6, 1);
+    int y = *i;
+
+    return 0;
+}
+
+/* #expect: test.ll
+
+  store ptr %ptradd9, ptr %i, align 8
+  %0 = load ptr, ptr %i, align 8
+  %checknull = icmp eq ptr %0, null
+  %1 = call i1 @llvm.expect.i1(i1 %checknull, i1 false)
+  br i1 %1, label %panic, label %checkok
+
+checkok:                                          ; preds = %entry
+  %2 = ptrtoint ptr %0 to i64
+  %3 = urem i64 %2, 4
+  %4 = icmp ne i64 %3, 0
+  %5 = call i1 @llvm.expect.i1(i1 %4, i1 false)
+  br i1 %5, label %panic10, label %checkok13
+
+checkok13:                                        ; preds = %checkok
+  %6 = load i32, ptr %0, align 4
+  store i32 %6, ptr %z, align 4
+  %7 = load ptr, ptr %i, align 8
+  %checknull14 = icmp eq ptr %7, null
+  %8 = call i1 @llvm.expect.i1(i1 %checknull14, i1 false)
+  br i1 %8, label %panic15, label %checkok16
+
+checkok16:                                        ; preds = %checkok13
+  %9 = load i32, ptr %7, align 1
+  store i32 %9, ptr %z2, align 4
+  %10 = load ptr, ptr %i, align 8
+  %checknull17 = icmp eq ptr %10, null
+  %11 = call i1 @llvm.expect.i1(i1 %checknull17, i1 false)
+  br i1 %11, label %panic18, label %checkok19
+
+checkok19:                                        ; preds = %checkok16
+  %12 = ptrtoint ptr %10 to i64
+  %13 = urem i64 %12, 4
+  %14 = icmp ne i64 %13, 0
+  %15 = call i1 @llvm.expect.i1(i1 %14, i1 false)
+  br i1 %15, label %panic20, label %checkok27
+
+checkok27:                                        ; preds = %checkok19
+  store i32 5, ptr %10, align 4
+  %16 = load ptr, ptr %i, align 8
+  %checknull28 = icmp eq ptr %16, null
+  %17 = call i1 @llvm.expect.i1(i1 %checknull28, i1 false)
+  br i1 %17, label %panic29, label %checkok30
+
+checkok30:                                        ; preds = %checkok27
+  store i32 6, ptr %16, align 1
+  %18 = load ptr, ptr %i, align 8
+  %checknull31 = icmp eq ptr %18, null
+  %19 = call i1 @llvm.expect.i1(i1 %checknull31, i1 false)
+  br i1 %19, label %panic32, label %checkok33
+
+checkok33:                                        ; preds = %checkok30
+  %20 = ptrtoint ptr %18 to i64
+  %21 = urem i64 %20, 4
+  %22 = icmp ne i64 %21, 0
+  %23 = call i1 @llvm.expect.i1(i1 %22, i1 false)
+  br i1 %23, label %panic34, label %checkok41
+
+checkok41:                                        ; preds = %checkok33
+  %24 = load i32, ptr %18, align 4
+  store i32 %24, ptr %y, align 4
+  ret i32 0

test/test_suite/safe/deref.c3t

@@ -14,18 +14,47 @@ define void @foo.main() #0 {
 entry:
   %x = alloca ptr, align 8
   %y = alloca i32, align 4
+  %taddr = alloca i64, align 8
+  %taddr2 = alloca i64, align 8
+  %varargslots = alloca [2 x %any], align 16
+  %indirectarg = alloca %"any[]", align 8
   store ptr null, ptr %x, align 8
   %0 = load ptr, ptr %x, align 8
   %checknull = icmp eq ptr %0, null
   %1 = call i1 @llvm.expect.i1(i1 %checknull, i1 false)
   br i1 %1, label %panic, label %checkok
+
 checkok:                                          ; preds = %entry
-  %2 = load i32, ptr %0, align 4
-  store i32 %2, ptr %y, align 4
+  %2 = ptrtoint ptr %0 to i64
+  %3 = urem i64 %2, 4
+  %4 = icmp ne i64 %3, 0
+  %5 = call i1 @llvm.expect.i1(i1 %4, i1 false)
+  br i1 %5, label %panic1, label %checkok3
+
+checkok3:                                         ; preds = %checkok
+  %6 = load i32, ptr %0, align 4
+  store i32 %6, ptr %y, align 4
   ret void
+
 panic:                                            ; preds = %entry
-  %3 = load ptr, ptr @std.core.builtin.panic, align 8
-  call void %3(ptr @.panic_msg, i64 42, ptr @.file, i64 8, ptr @.func, i64 4, i32 6)
+  %7 = load ptr, ptr @std.core.builtin.panic, align 8
+  call void %7(ptr @.panic_msg, i64 42, ptr @.file, i64 8, ptr @.func, i64 4, i32 6) #2
+  unreachable
+
+panic1:                                           ; preds = %checkok
+  store i64 4, ptr %taddr, align 8
+  %8 = insertvalue %any undef, ptr %taddr, 0
+  %9 = insertvalue %any %8, i64 ptrtoint (ptr @"$ct.ulong" to i64), 1
+  store i64 %3, ptr %taddr2, align 8
+  %10 = insertvalue %any undef, ptr %taddr2, 0
+  %11 = insertvalue %any %10, i64 ptrtoint (ptr @"$ct.ulong" to i64), 1
+  store %any %9, ptr %varargslots, align 16
+  %ptradd = getelementptr inbounds i8, ptr %varargslots, i64 16
+  store %any %11, ptr %ptradd, align 16
+  %12 = insertvalue %"any[]" undef, ptr %varargslots, 0
+  %"$$temp" = insertvalue %"any[]" %12, i64 2, 1
+  store %"any[]" %"$$temp", ptr %indirectarg, align 8
+  call void @std.core.builtin.panicf(ptr @.panic_msg.1, i64 94, ptr @.file, i64 8, ptr @.func, i64 4, i32 6, ptr byval(%"any[]") align 8 %indirectarg) #2
   unreachable
 }