Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Specification language to describe flat conversions and aggregate splats. #358

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
+49 −0
Open

Specification language to describe flat conversions and aggregate splats. #358

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
49 changes: 49 additions & 0 deletions specs/language/conversions.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,19 @@
of boolean type. A zero value is converted to \texttt{false}; all other values
are converted to \texttt{true}.

\Sec{Flat conversion}{Conv.flat}

\p There is a flattened order of subobjects for an aggregate, \texttt{A},
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this language is fine for now, but we should probably have some common definitions and unify this with the language in Decl.Init.Agg.

I filed #372 to track that separately.

\texttt{A$^\prime$}, as described in section \ref{Decl.Init.Agg}. An aggregate,
\texttt{A}, can be flat converted to an aggregate, \texttt{B}, whose flattened
form is \texttt{B$^\prime$}, if the following conditions are true:
\begin{itemize}
\item The length of \texttt{A$^\prime$}, \texttt{N}, is greater than or equal to the
length of \texttt{B$^\prime$}, \texttt{M}.
\item For every index \texttt{I} from \texttt{0} to \texttt{M-1}, there exists an
implicit conversion from \texttt{A$^\prime$[I]} to \texttt{B$^\prime$[I]}.
\end{itemize}

\Sec{Vector splat conversion}{Conv.vsplat}

\p A glvalue of type \texttt{T} can be converted to a cxvalue of type
Expand All @@ -130,6 +143,42 @@
prvalue of type \texttt{matrix<T,x,y>}. The destination value is the source
value replicated into each element of the destination.

\p A glvalue of type \texttt{vector<T,1>} can be converted to a cxvalue of type
\texttt{vector<T,x>} or a prvalue of type \texttt{vector<T,1>} can be converted
to a prvalue of type \texttt{vector<T,x>}. The destination value is the value in
the source vector replicated into each element of the destination.

\Sec{Array and Class splat conversion}{Conv.asplat}

\p A glvalue of type \texttt{T} can be converted to a cxvalue of type
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe that this conversion can only produce a prvalue, not a cxvalue. The only place cxvalues are created is for inout parameters, and I believe this code is invalid:

struct S {
    float V[4];
};

void fn(inout float4 F) {}

export void call() {
    S V = {1.xxxx};
    fn((float4)V);
    fn(V);
}

see: https://godbolt.org/z/Ms4ocrGMf

\texttt{struct S} or a prvalue of type \texttt{T} can be converted to a
prvalue of type \texttt{struct S}, if there are valid conversions from
\texttt{T} to each \texttt{U} in flattened \texttt{S}, see section
\ref{Decl.init.Agg} for description of a flattened ordering. The destination
value is the source value \texttt{T} replicated into each element of the
flattened \texttt{S}.

\p A glvalue of type \texttt{vector<T,1>} can be converted to a cxvalue of type
\texttt{struct S} or a prvalue of type \texttt{vector<T,1>} can be converted to a
prvalue of type \texttt{struct S}, if there are valid conversions from
\texttt{T} to each \texttt{U} in flattened \texttt{S}. The destination value is
the value in the source vector replicated into each element of the flattened
\texttt{S}.

\p A glvalue of type \texttt{T} can be converted to a cxvalue of type
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There is an odd case for vector<->array conversions where we can create a cxvalue, but only from an explicit cast:

void fn(inout float4 F) {}

export void call() {
    float V[4] = {1.xxxx};
    fn((float4)V); // valid
    fn(V); // invalid
}

see: https://godbolt.org/z/c3YcGWPex

\texttt{U[]} or a prvalue of type \texttt{T} can be converted to a
prvalue of type \texttt{U[]}, if there are valid conversions from
\texttt{T} to each \texttt{V} in flattened \texttt{U}. The destination value
is the source value \texttt{T} replicated into each element of the flattened
\texttt{U[]}.
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

U[] is not the same type as U[N], so we should be clear we're talking about sized arrays here.


\p A glvalue of type \texttt{vector<T,1>} can be converted to a cxvalue of type
\texttt{U[]} or a prvalue of type \texttt{vector<T,1>} can be converted to a
prvalue of type \texttt{U[]}, if there are valid conversions from
\texttt{T} to each \texttt{V} in flattened \texttt{U}. The destination value
is the value in the source vector replicated into each element of the flattened
\texttt{U[]}.

\Sec{Vector and matrix truncation conversion}{Conv.vtrunc}

\p A prvalue of type \texttt{vector<T,x>} can be converted to a prvalue of type:
Expand Down
Loading