Allow swizzling assign, eg. abc.xz += { 5, 10 };

releasenotes.md

@@ -16,6 +16,7 @@
 - Remove `.allocator = allocator` syntax for functions.
 - Remove `@operator(construct)`.
 - Removal of "any-switch".
+- Allow swizzling assign, eg. `abc.xz += { 5, 10 };`
 
 ### Fixes
 - Fix address sanitizer to work on MachO targets (e.g. MacOS).

src/compiler/codegen_internal.h

@@ -113,10 +113,10 @@ UNUSED static inline bool abi_type_is_promotable_integer_or_bool(AbiType type)
 	return false;
 }
 
-static inline bool expr_is_vector_index(Expr *expr)
+static inline bool expr_is_vector_index_or_swizzle(Expr *expr)
 {
-	return expr->expr_kind == EXPR_SUBSCRIPT
-		   && type_lowering(exprtype(expr->subscript_expr.expr))->type_kind == TYPE_VECTOR;
+	return (expr->expr_kind == EXPR_SUBSCRIPT && type_lowering(exprtype(expr->subscript_expr.expr))->type_kind == TYPE_VECTOR)
+			|| (expr->expr_kind == EXPR_SWIZZLE && type_lowering(exprtype(expr->swizzle_expr.parent))->type_kind == TYPE_VECTOR);
 }
 
 const char *codegen_create_asm(Ast *ast);

src/compiler/compiler_internal.h

@@ -1066,6 +1066,7 @@ typedef struct
 typedef struct
 {
 	ExprId parent;
+	bool is_overlapping;
 	const char *swizzle;
 } ExprSwizzle;
 

src/compiler/llvm_codegen_expr.c

@@ -16,6 +16,7 @@ static LLVMValueRef llvm_emit_dynamic_search(GenContext *c, LLVMValueRef type_id
 static inline void llvm_emit_bitassign_array(GenContext *c, LLVMValueRef result, BEValue parent, Decl *parent_decl, Decl *member);
 static inline void llvm_emit_builtin_access(GenContext *c, BEValue *be_value, Expr *expr);
 static inline void llvm_emit_const_initialize_reference(GenContext *c, BEValue *ref, Expr *expr);
+static void llvm_emit_swizzle_from_value(GenContext *c, LLVMValueRef vector_value, BEValue *value, Expr *expr);
 
 static inline void llvm_emit_optional(GenContext *c, BEValue *be_value, Expr *expr);
 static inline void llvm_emit_inc_dec_change(GenContext *c, BEValue *addr, BEValue *after, BEValue *before, Expr *expr, int diff,
@@ -4343,17 +4344,58 @@ static void llvm_emit_vector_assign_expr(GenContext *c, BEValue *be_value, Expr
 	Expr *left = exprptr(expr->binary_expr.left);
 	BinaryOp binary_op = expr->binary_expr.operator;
 	BEValue addr;
-	BEValue index;
 
+	bool is_swizzle = left->expr_kind == EXPR_SWIZZLE;
+
+	if (left->expr_kind == EXPR_SWIZZLE)
+	{
+		// Emit the variable
+		llvm_emit_exprid(c, &addr, left->swizzle_expr.parent);
+	}
+	else
+	{
+		// Emit the variable
+		llvm_emit_exprid(c, &addr, left->subscript_expr.expr);
+	}
 	// Emit the variable
-	llvm_emit_exprid(c, &addr, left->subscript_expr.expr);
 	llvm_value_addr(c, &addr);
-	LLVMValueRef vector_value = llvm_load_value_store(c, &addr);
 
+	LLVMValueRef vector_value = llvm_load_value_store(c, &addr);
+	if (is_swizzle)
+	{
+		if (addr.type->array.base == type_bool) vector_value = llvm_emit_trunc_bool(c, vector_value);
+		if (binary_op > BINARYOP_ASSIGN)
+		{
+			BEValue lhs;
+			llvm_emit_swizzle_from_value(c, llvm_load_value_store(c, &addr), &lhs, left);
+			BinaryOp base_op = binaryop_assign_base_op(binary_op);
+			ASSERT(base_op != BINARYOP_ERROR);
+			llvm_value_rvalue(c, &lhs);
+			llvm_emit_binary(c, be_value, expr, &lhs, base_op);
+		}
+		else
+		{
+			llvm_emit_expr(c, be_value, exprptr(expr->binary_expr.right));
+		}
+		llvm_value_rvalue(c, be_value);
+		const char *sw_ptr = left->swizzle_expr.swizzle;
+		unsigned vec_len = be_value->type->array.len;
+		LLVMValueRef result = be_value->value;
+		for (unsigned i = 0; i < vec_len; i++)
+		{
+			int index = (swizzle[(int)sw_ptr[i]] - 1) & 0xF;
+			LLVMValueRef val = llvm_emit_extract_value(c, result, i);
+			vector_value = llvm_emit_insert_value(c, vector_value, val, index);
+		}
+		llvm_value_set(be_value, vector_value, addr.type);
+		llvm_store(c, &addr, be_value);
+		llvm_value_set(be_value, result, expr->type);
+		return;
+	}
 	// Emit the index
+	BEValue index;
 	llvm_emit_exprid(c, &index, left->subscript_expr.index.expr);
 	LLVMValueRef index_val = llvm_load_value_store(c, &index);
-
 	if (binary_op > BINARYOP_ASSIGN)
 	{
 		BinaryOp base_op = binaryop_assign_base_op(binary_op);
@@ -4367,15 +4409,17 @@ static void llvm_emit_vector_assign_expr(GenContext *c, BEValue *be_value, Expr
 		llvm_emit_expr(c, be_value, exprptr(expr->binary_expr.right));
 	}
 
-	LLVMValueRef new_value = LLVMBuildInsertElement(c->builder, vector_value, llvm_load_value_store(c, be_value), index_val, "elemset");
+	LLVMValueRef new_value = LLVMBuildInsertElement(c->builder, vector_value, llvm_load_value_store(c, be_value),
+	                                                index_val, "elemset");
 	llvm_store_raw(c, &addr, new_value);
+
 }
 
 static void llvm_emit_binary_expr(GenContext *c, BEValue *be_value, Expr *expr)
 {
 	BinaryOp binary_op = expr->binary_expr.operator;
 	// Vector assign is handled separately.
-	if (binary_op >= BINARYOP_ASSIGN && expr_is_vector_index(exprptr(expr->binary_expr.left)))
+	if (binary_op >= BINARYOP_ASSIGN && expr_is_vector_index_or_swizzle(exprptr(expr->binary_expr.left)))
 	{
 		llvm_emit_vector_assign_expr(c, be_value, expr);
 		return;
@@ -6675,12 +6719,8 @@ static void llmv_emit_test_hook(GenContext *c, BEValue *value, Expr *expr)
 	llvm_value_set_address_abi_aligned(value, get_global, expr->type);
 }
 
-
-static void llvm_emit_swizzle(GenContext *c, BEValue *value, Expr *expr)
+static void llvm_emit_swizzle_from_value(GenContext *c, LLVMValueRef vector_value, BEValue *value, Expr *expr)
 {
-	llvm_emit_exprid(c, value, expr->swizzle_expr.parent);
-	llvm_value_rvalue(c, value);
-	LLVMValueRef parent = value->value;
 	LLVMTypeRef result_type = llvm_get_type(c, expr->type);
 	unsigned vec_len = LLVMGetVectorSize(result_type);
 	LLVMValueRef mask_val[4];
@@ -6691,10 +6731,17 @@ static void llvm_emit_swizzle(GenContext *c, BEValue *value, Expr *expr)
 		int index = (swizzle[(int)sw_ptr[i]] - 1) & 0xF;
 		mask_val[i] = llvm_const_int(c, type_uint, index);
 	}
-	LLVMValueRef res = LLVMBuildShuffleVector(c->builder, parent, LLVMGetUndef(LLVMTypeOf(parent)), LLVMConstVector(mask_val, vec_len), sw_ptr);
+	LLVMValueRef res = LLVMBuildShuffleVector(c->builder, vector_value, LLVMGetUndef(LLVMTypeOf(vector_value)), LLVMConstVector(mask_val, vec_len), sw_ptr);
 	llvm_value_set(value, res, expr->type);
 }
 
+static void llvm_emit_swizzle(GenContext *c, BEValue *value, Expr *expr)
+{
+	llvm_emit_exprid(c, value, expr->swizzle_expr.parent);
+	llvm_value_rvalue(c, value);
+	llvm_emit_swizzle_from_value(c, value->value, value, expr);
+}
+
 static void llvm_emit_default_arg(GenContext *c, BEValue *value, Expr *expr)
 {
 	if (llvm_use_debug(c))

src/compiler/sema_expr.c

@@ -481,7 +481,11 @@ static bool sema_binary_is_expr_lvalue(SemaContext *context, Expr *top_expr, Exp
 			UNREACHABLE
 		case EXPR_SWIZZLE:
 			if (failed_ref) goto FAILED_REF;
-			RETURN_SEMA_ERROR(expr, "You cannot use swizzling to assign to multiple elements, use element-wise assign instead.");
+			if (expr->swizzle_expr.is_overlapping)
+			{
+				RETURN_SEMA_ERROR(expr, "You cannot use swizzling to assign to multiple elements if they are overlapping");
+			}
+			return sema_binary_is_expr_lvalue(context, top_expr, exprptr(expr->swizzle_expr.parent), failed_ref);
 		case EXPR_LAMBDA:
 			if (failed_ref) goto FAILED_REF;
 			RETURN_SEMA_ERROR(expr, "This expression is a value and cannot be assigned to.");
@@ -4892,6 +4896,7 @@ static inline bool sema_expr_analyse_swizzle(SemaContext *context, Expr *expr, E
 	if (is_lvalue) check = CHECK_VALUE;
 	ASSERT_SPAN(expr, len > 0);
 	int index;
+	bool is_overlapping = false;
 	for (unsigned i = 0; i < len; i++)
 	{
 		char val = swizzle[(int)kw[i]] - 1;
@@ -4907,6 +4912,18 @@ static inline bool sema_expr_analyse_swizzle(SemaContext *context, Expr *expr, E
 		{
 			RETURN_SEMA_ERROR(expr, "Mixing [xyzw] and [rgba] is not permitted, you will need to select one of them.");
 		}
+		if (!is_overlapping)
+		{
+			for (int j = 0; j < i; j++)
+			{
+				char prev = swizzle[(int)kw[j]] - 1;
+				if (val == prev)
+				{
+					is_overlapping = true;
+					break;
+				}
+			}
+		}
 	}
 	index &= 0xF;
 	if (len == 1)
@@ -4941,7 +4958,8 @@ static inline bool sema_expr_analyse_swizzle(SemaContext *context, Expr *expr, E
 	}
 	Type *result = type_get_vector(indexed_type, len);
 	expr->expr_kind = EXPR_SWIZZLE;
-	expr->swizzle_expr = (ExprSwizzle) { exprid(parent), kw };
+	expr->swizzle_expr = (ExprSwizzle) { .parent = exprid(parent), .swizzle = kw, .is_overlapping = is_overlapping };
+
 	expr->type = result;
 	return true;
 }
@@ -6045,10 +6063,11 @@ static bool sema_expr_analyse_op_assign(SemaContext *context, Expr *expr, Expr *
 	Type *no_fail = type_no_optional(left->type);
 	Type *flat = type_flatten(no_fail);
 
-	// 3. If this is only defined for ints (*%, ^= |= &= %=) verify that this is an int.
+	// 3. If this is only defined for ints (^= |= &= %=) verify that this is an int.
 	if (int_only && !type_flat_is_intlike(flat))
 	{
-		if (is_bit_op && flat->type_kind == TYPE_BITSTRUCT) goto BITSTRUCT_OK;
+		if (is_bit_op && (flat->type_kind == TYPE_BITSTRUCT || flat == type_bool || type_flat_is_bool_vector(flat))) goto BITSTRUCT_OK;
+
 		RETURN_SEMA_ERROR(left, "Expected an integer here, not a value of type %s.", type_quoted_error_string(left->type));
 	}
 

test/test_suite/vector/vector_conversion_scalar.c3

@@ -4,7 +4,7 @@ fn void main()
 {
   int[<2>] y = 1;
   y[..] = 3;
-  y.xy = 3;       // #error: cannot use swizzling
+  y.xxy = 3;       // #error: cannot use swizzling
   y *= 2;
   y = 3;          // #error: explicit cast
   test2(3);       // #error: explicit cast

test/unit/regression/vector_ops.c3

@@ -10,6 +10,34 @@ fn void test_int_mod() @test
 	assert((int[<2>]){ 10, 99 } / { 3, 5 } == { 3, 19 });
 }
 
+fn void test_swizzle_assign() @test
+{
+	int[<3>] abc;
+	abc.xy = { 3, 4 };
+	assert(abc == { 3, 4, 0 });
+	abc.xz += { 2, 7 };
+	assert(abc == { 5, 4, 7 });
+	abc.xy += { 4, -100 };
+	assert(abc == { 9, -96, 7 });
+	abc.xz = { 0, 0 };
+	assert(abc == { 0, -96, 0});
+}
+
+fn void test_swizzle_assign_bool() @test
+{
+	bool[<3>] abc;
+	abc.xy = { true, true };
+	assert(abc == { true, true, false });
+	abc.yz ^= { true, true };
+	assert(abc == { true, false, true });
+	abc ^= true;
+	assert(abc == { false, true, false });
+	abc.xy ^= true;
+	assert(abc == { true, false, false });
+	assert((abc.yx ^= true) == { true, false });
+	assert(abc == { false, true, false });
+}
+
 fn void test_conv() @test
 {
     float[<4>] y = { 1, 2, 3, 4 };