Skip to content

[General]: Some helpers assert their parameter domains instead of declaring set types #110

Open

Description

@shuklaayush

Problem

The same kind of declaration is written two ways. Outside the Std:: library, 123
function signatures give a parameter a set type, e.g. Calculate_Bit_Offset
(implicit operand_size : {16, 32, 64}), and most of their bodies do not repeat it. Other helpers declare the parameter Integer and restrict
it with assert statements at the start of the body, e.g. Mem_Fetch:

function Mem_Fetch?(rip : Bits(64), size : Integer) -> Bits(size)
begin
    assert size in {8, 16, 32, 64};

In these 15 helpers, the leading asserts are the first statements, are pure, and check the
parameters in parameter order:

helper parameter: set
Calculate_Offset address_size : {16, 32, 64}
Compress_Mask size : {128, 256, 512}, element_size : {8, 16, 32, 64}
Create_Physical_Address size : {8, 16, 32, 64, 128, 256, 384, 512, 1024}
Is_Linear_Aligned size : {8, 16, 32, 64, 80, 128, 256, 384, 512, 1024}
Linear_Mem_Read? size : {8, 16, 32, 64, 128, 256, 384, 512, 1024}
Linear_Mem_Write? size : {8, 16, 32, 64, 128, 256, 384, 512, 1024}
Logical_Mem_Read? size : {8, 16, 32, 64, 80, 112, 128, 224, 256, 384, 512, 1024}
Masked_Vector_Mem_Read? size : {16, 32, 64, 128, 256, 512}, element_size : {8, 16, 32, 64}
Masked_Vector_Mem_Write? size : {16, 32, 64, 128, 256, 512}, element_size : {8, 16, 32, 64}
Mem_Fetch? size : {8, 16, 32, 64}
Packed_Vector_Mem_Read? size : {128, 256, 512}, element_size : {8, 16, 32, 64}
Packed_Vector_Mem_Write? size : {16, 32, 64, 128, 256, 512}, element_size : {8, 16, 32, 64}
Pop_Stack? size : {16, 32, 64}, pop_size : {16, 32, 64}
Push_Stack? size : {16, 32, 64}, push_size : {16, 32, 64}
Vector_Mem_Read? size : {128, 256, 512}, element_size : {8, 16, 32, 64}

Seventeen helpers do both: they declare the set type and then repeat it as the first
statement, e.g. Dot_Product(implicit N : {16, 32, 64}, …) followed by
assert N in {16, 32, 64};. They are Dot_Product, FP::Arithmetic, FP::Class,
FP::Compare, FP::Compare5, FP::FMA, FP::Get_Exponent, FP::Get_Mantissa,
FP::Min_Max, FP::Reciprocal, FP::Reciprocal_14, FP::Reciprocal_Sqrt,
FP::Reciprocal_Sqrt_14, FP::Reduce, FP::Round_To_Fraction, FP::Scale and FP::Sqrt.

The manual says "checking of set types is performed dynamically", but describes the
check only for assignment (ISA_LRM, "Types"): "… a runtime check that the value
being assigned is a member of the set". It does not say whether binding an argument to a
set-typed parameter is checked, when that check happens, or what a failed check does.

Proposed fix

  1. In the manual, state that binding an argument to a parameter of a set type checks
    membership after all arguments have been evaluated, and that a failed check fails as a
    failed assert does.
  2. With that rule, move each leading assert in the table into the parameter's type, and
    drop the 17 repeated asserts. Each change keeps the same checks in the same order, so
    no result changes; the parameter domains become part of the signatures.

AI disclosure

Assisted-by: Claude

Claude assisted with source analysis, the proposed correction and drafting this report.
Reviewed with assistance from Codex.

Pinned source: definitions.xml.gz at revision 35eface1b7ee003a1cda95005ec5fd83de658a64.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions