Start specifying standard conversions #58
Conversation
This change adds some basic language to describe values then moves into describing HLSL's standard conversions and standard conversion sequences.
specs/language/conversions.tex
Outdated
| chapter enumerates the full set of conversions. A \textit{standard conversion | ||
| sequence} is a sequence of standard conversions in the following order: | ||
| \begin{enumerate} | ||
| \item Zero or one conversion of either lvalue-to-rvalue, array-to-pinter or |
|
|
||
| \Sec{Floating point-integral conversion}{Conv.fpint} | ||
|
|
||
| \p A glvalue of floating point type can be converted to a cxvalue of integer |
There was a problem hiding this comment.
I think this sentence should be separated into two.
| discarding the fractional value. The behavior is undefined if the truncated | ||
| value cannot be represented in the destination type. | ||
|
|
||
| \p A glvalue of integer type can be converted to a cxvalue of floating point |
specs/language/conversions.tex
Outdated
| \p A glvalue of type \texttt{matrix<T,x,y>} can be converted to a cxvalue of type | ||
| \texttt{matrix<T,z,w>}, or a prvalue of type \texttt{matrix<T,x,y>} can be | ||
| converted to a prvalue of type \texttt{matrix<T,z,w>} only if \texttt{x} is less | ||
| than or equal to \texttt{z} and \texttt{y} is less than or equal to \texttt{w}. |
There was a problem hiding this comment.
Just curious, why not use the latex /leq operator?
specs/language/conversions.tex
Outdated
| \Sec{Qualification conversion}{Conv.qual} | ||
|
|
||
| A prvalue of type "\textit{cv1} \texttt{T}" can be converted to a prvalue of type | ||
| "\textit{cv2} \texttt{T}" type "\textit{cv2} \texttt{T}" is more cv-qualified |
There was a problem hiding this comment.
Should there be a sentence break here?
| \p A glvalue of floating point type can be converted to a cxvalue of integer | ||
| type. A prvalue of floating point type can be converted to a prvalue of | ||
| integer type. Conversion of floating point values to integer values truncates by | ||
| discarding the fractional value. The behavior is undefined if the truncated |
There was a problem hiding this comment.
Is "undefined" here different to "implementation defined" on line 53?
There was a problem hiding this comment.
Yea, "undefined" and "implementation defined" are defined terms earlier in the full document:
https://github.com/microsoft/hlsl-specs/blob/main/specs/language/introduction.tex#L148
Basically "implementation defined" behavior needs to be documented by the relevant part of the implementation (compiler, runtime, ISA, DXIL, SPIR-V, etc).
| point type. If the destination type can exactly represent the source value, the | ||
| result is the exact value. If the destination type cannot exactly represent the | ||
| source value, the conversion to a best-approximation of the source value is | ||
| implementation defined. |
There was a problem hiding this comment.
Do we need to talk about rounding modes here? I worry that "best-approximation" is very ambiguous otherwise. If we're just going to kick the can and worry about those details later I'd just say "approximation".
There was a problem hiding this comment.
I think the rounding modes falls under the "implementation defined" aspect and I would expect us to document the DXIL & SPIR-V behaviors separately since they may be different.
| \p A glvalue of type \texttt{matrix<T,x,y>} can be converted to a cxvalue of type | ||
| \texttt{matrix<T,z,w>}, or a prvalue of type \texttt{matrix<T,x,y>} can be | ||
| converted to a prvalue of type \texttt{matrix<T,z,w>} only if \( x \leq z \) | ||
| and \(y \leq w \) |
There was a problem hiding this comment.
It might be helpful to have a brief code listing with examples for each section of conversions. Is that something that would be practical without it taking up too much space?
This change adds some basic language to describe values then moves into describing HLSL's standard conversions and standard conversion sequences.