Merge branch 'master' into macharena

1 files changed, 28 insertions, 8 deletions

diff --git a/src/types.cpp b/src/types.cpp
index 21dd6ad39..a9a7d6dda 100644
--- a/src/types.cpp
+++ b/src/types.cpp
@@ -1474,6 +1474,7 @@ gb_internal i64 matrix_align_of(Type *t, struct TypePath *tp) {
 	
 	Type *elem = t->Matrix.elem;
 	i64 row_count = gb_max(t->Matrix.row_count, 1);
+	i64 column_count = gb_max(t->Matrix.column_count, 1);
 
 	bool pop = type_path_push(tp, elem);
 	if (tp->failure) {
@@ -1491,7 +1492,7 @@ gb_internal i64 matrix_align_of(Type *t, struct TypePath *tp) {
 	// could be maximally aligned but as a compromise, having no padding will be
 	// beneficial to third libraries that assume no padding
 	
-	i64 total_expected_size = row_count*t->Matrix.column_count*elem_size;
+	i64 total_expected_size = row_count*column_count*elem_size;
 	// i64 min_alignment = prev_pow2(elem_align * row_count);
 	i64 min_alignment = prev_pow2(total_expected_size);
 	while (total_expected_size != 0 && (total_expected_size % min_alignment) != 0) {
@@ -1523,12 +1524,15 @@ gb_internal i64 matrix_type_stride_in_bytes(Type *t, struct TypePath *tp) {
 	i64 stride_in_bytes = 0;
 	
 	// NOTE(bill, 2021-10-25): The alignment strategy here is to have zero padding
-	// It would be better for performance to pad each column so that each column
+	// It would be better for performance to pad each column/row so that each column/row
 	// could be maximally aligned but as a compromise, having no padding will be
 	// beneficial to third libraries that assume no padding
-	i64 row_count = t->Matrix.row_count;
-	stride_in_bytes = elem_size*row_count;
-	
+
+	if (t->Matrix.is_row_major) {
+		stride_in_bytes = elem_size*t->Matrix.column_count;
+	} else {
+		stride_in_bytes = elem_size*t->Matrix.row_count;
+	}
 	t->Matrix.stride_in_bytes = stride_in_bytes;
 	return stride_in_bytes;
 }
@@ -2810,8 +2814,14 @@ gb_internal bool are_types_identical_internal(Type *x, Type *y, bool check_tuple
 
 	case Type_Union:
 		if (x->Union.variants.count == y->Union.variants.count &&
-		    x->Union.custom_align == y->Union.custom_align &&
 		    x->Union.kind == y->Union.kind) {
+
+			if (x->Union.custom_align != y->Union.custom_align) {
+				if (type_align_of(x) != type_align_of(y)) {
+					return false;
+				}
+			}
+
 			// NOTE(bill): zeroth variant is nullptr
 			for_array(i, x->Union.variants) {
 				if (!are_types_identical(x->Union.variants[i], y->Union.variants[i])) {
@@ -2827,10 +2837,16 @@ gb_internal bool are_types_identical_internal(Type *x, Type *y, bool check_tuple
 		    x->Struct.is_no_copy   == y->Struct.is_no_copy &&
 		    x->Struct.fields.count == y->Struct.fields.count &&
 		    x->Struct.is_packed    == y->Struct.is_packed &&
-		    x->Struct.custom_align == y->Struct.custom_align &&
 		    x->Struct.soa_kind == y->Struct.soa_kind &&
 		    x->Struct.soa_count == y->Struct.soa_count &&
 		    are_types_identical(x->Struct.soa_elem, y->Struct.soa_elem)) {
+
+			if (x->Struct.custom_align != y->Struct.custom_align) {
+				if (type_align_of(x) != type_align_of(y)) {
+					return false;
+				}
+			}
+
 			for_array(i, x->Struct.fields) {
 				Entity *xf = x->Struct.fields[i];
 				Entity *yf = y->Struct.fields[i];
@@ -4205,7 +4221,11 @@ gb_internal i64 type_size_of_internal(Type *t, TypePath *path) {
 	
 	case Type_Matrix: {
 		i64 stride_in_bytes = matrix_type_stride_in_bytes(t, path);
-		return stride_in_bytes * t->Matrix.column_count;
+		if (t->Matrix.is_row_major) {
+			return stride_in_bytes * t->Matrix.row_count;
+		} else {
+			return stride_in_bytes * t->Matrix.column_count;
+		}
 	}
 
 	case Type_BitField: