Re: [PATCH olang 5/5] codegen: perform mov instructions based on variable type

[Johnny Richard <johnny@johnnyrichard.com>]

On Sat, Sep 21, 2024 at 08:25:03AM GMT, Carlos Maniero wrote:
> There are two function that was introduced to performe the translations
> bellow:
> 
> type | type_to_bytes | bytes_to_rax
> -----------------------------------
>   u8 | 1             | ah
>  u16 | 2             | ax
>  u32 | 4             | eax
>  u64 | 8             | rax
> 
> I opted to create *bytes_to_rax* instead of *type_to_rax* mainly because
> we may use the same function to mov literals once we extend the literals
> as well (We say that all literals are u32 but we actually handle them as
> u64 on codegen).

Yeah, you are very right here.  I always mess with bytes sizes LOL.

> diff --git a/src/checker.c b/src/checker.c
> index f5068e0..3a78a59 100644
> --- a/src/checker.c
> +++ b/src/checker.c
> @@ -133,11 +133,26 @@ populate_scope(checker_t *checker, scope_t *scope, ast_node_t *ast)
>  static void
>  evaluate_type(type_t *type)
>  {
> +    if (string_view_eq_to_cstr(type->id, "u8")) {
> +        type->kind = TYPE_PRIMITIVE;
> +        type->as_primitive = TYPE_U8;
> +        return;
> +    }
> +    if (string_view_eq_to_cstr(type->id, "u16")) {
> +        type->kind = TYPE_PRIMITIVE;
> +        type->as_primitive = TYPE_U16;
> +        return;
> +    }
>      if (string_view_eq_to_cstr(type->id, "u32")) {
>          type->kind = TYPE_PRIMITIVE;
>          type->as_primitive = TYPE_U32;
>          return;
>      }
> +    if (string_view_eq_to_cstr(type->id, "u64")) {
> +        type->kind = TYPE_PRIMITIVE;
> +        type->as_primitive = TYPE_U64;
> +        return;
> +    }
>  
>      type->kind = TYPE_USER_DEFINED;
>  }

Is "as_primitive" property used only for calculate the size in bytes or
you see any other need in the future?

> diff --git a/src/codegen_linux_x86_64.c b/src/codegen_linux_x86_64.c
> index 415c81b..fa2a082 100644
> --- a/src/codegen_linux_x86_64.c
> +++ b/src/codegen_linux_x86_64.c
> @@ -38,6 +38,12 @@ codegen_linux_x86_64_emit_function(codegen_x86_64_t *codegen, ast_fn_definition_
>  static size_t
>  type_to_bytes(type_t *type);
>  
> +static char *
> +bytes_to_mov(size_t bytes);

Is this necessary?  I mean, isn't GAS smart enough to pick the right mov
instruction based on register size?

If we have to pick the right size mov explicitly, what do you think
about renaming it to **get_mov_inst_for(size_t)** or something alike to
express better the intention?

> +static char *
> +bytes_to_rax(size_t bytes);

This is similar to MOV instruction above, I was thinking about naming it
to **get_accumulator_reg_for(size_t)** or something else like it.  What
do you think?  The current name uses the 64bit version of the
accumulator register.

> +
>  void
>  codegen_linux_x86_64_init(codegen_x86_64_t *codegen, arena_t *arena, FILE *out)
>  {
> @@ -474,7 +502,7 @@ calculate_fn_local_size(scope_t *scope)
>  static void
>  codegen_linux_x86_64_emit_function(codegen_x86_64_t *codegen, ast_fn_definition_t *fn)
>  {
> -    codegen->base_offset = 0;
> +    codegen->base_offset = 8;

Nicely spotted!

>      ast_node_t *block_node = fn->block;
>      fprintf(codegen->out, "" SV_FMT ":\n", SV_ARG(fn->identifier));
>