Hello February 16, 2022 2:14 PM, "Giovanni Mascellani" <gmascellani(a)codeweavers.com> wrote:
Hi,
the patch looks mostly ok, with a few comments.
Il 15/02/22 21:17, Francisco Casas ha scritto:
+static void initialize_numeric_var(struct hlsl_ctx *ctx, struct hlsl_ir_var *var, + struct parse_initializer *initializer, unsigned int reg_offset, struct hlsl_type *type, + unsigned int *initializer_offset) +{
Isn't "type" here the same thing as "var->data_type" in the only call to this function? If so, you could avoid passing "type" explicitly.
For now it is the same thing but 4 patches ahead (in the continuation of this batch) that will no longer be true. The idea is that var may also be a struct or an array, and type would be the type of the numeric field inside that we want to initialize.
+ unsigned int i; + + if (type->type == HLSL_CLASS_MATRIX) + hlsl_fixme(ctx, &var->loc, "Matrix initializer."); + + for (i = 0; i < type->dimx; i++) + { + struct hlsl_ir_store *store; + struct hlsl_ir_constant *c; + struct hlsl_ir_node *node; + + node = initializer->args[*initializer_offset]; + *initializer_offset += 1; + + if (!(node = add_implicit_conversion(ctx, initializer->instrs, node, + hlsl_get_scalar_type(ctx, type->base_type), &node->loc))) + return; + + if (hlsl_type_is_single_reg(var->data_type)) + { + if (!(c = hlsl_new_uint_constant(ctx, reg_offset, node->loc))) + return; + list_add_tail(initializer->instrs, &c->node.entry); + + if (!(store = hlsl_new_store(ctx, var, &c->node, node, (1u << i), node->loc))) + return; + } + else + { + if (!(c = hlsl_new_uint_constant(ctx, reg_offset + i, node->loc))) + return; + list_add_tail(initializer->instrs, &c->node.entry); + + if (!(store = hlsl_new_store(ctx, var, &c->node, node, 0, node->loc))) + return; + }
Again under the assumption that "type == var->data_type", here it seems that you're handling the matrix case, while it is marked as fixme above. I guess the reason for the fixme is that you're only iterating on dimx? Still, I find it a bit confusing. If you consider the matrix case not (yet) supported, I'd just avoid to add pieces of it until you're really supporting it. Though maybe I missing something.
As before, "type == var->data_type" won't hold after 33e27ff7 vkd3d-shader/hlsl: Use generic variable initializer for struct fields. which would have been the 7th patch of this batch if I hadn't split it. So it is not just for the matrix case. (I implemented matrix initializers too but they are 10 initializer patches ahead.) Best regards, Francisco