Displaying 4 results from an estimated 4 matches for "vlentoken".
2018 Apr 11
5
RFC: Supporting the RISC-V vector extension in LLVM
...t also between
different calls to the same function. The exception is that functions with
the `inherits_vlen` attribute get the same dynamic vector length as their
caller (Note: this attribute is a straw man, target-specific calling
conventions may work better for this purpose).
A new instruction `vlentoken` is added, which has no operands and is of
type `token`. This token represents the dynamic vector length of the
function execution it is in. There can only be one such instruction per
function (this is inconsequential to the operational semantics, but it
simplifies IR passes).
A new kind of type i...
2018 Apr 12
0
RFC: Supporting the RISC-V vector extension in LLVM
...calls to the same function. The exception is that functions with
> the `inherits_vlen` attribute get the same dynamic vector length as their
> caller (Note: this attribute is a straw man, target-specific calling
> conventions may work better for this purpose).
>
> A new instruction `vlentoken` is added, which has no operands and is of
> type `token`. This token represents the dynamic vector length of the
> function execution it is in. There can only be one such instruction per
> function (this is inconsequential to the operational semantics, but it
> simplifies IR passes).
&...
2018 Apr 13
0
RFC: Supporting the RISC-V vector extension in LLVM
...legal, particularly outlining --
> details are in the RFC text. If loss of those optimizations is OK for SVE,
> we could talk about unifying the two vector types. Alternatively, I believe
> we could generalize the scalable vector type and my proposal, e.g., by
> adding my proposed "vlentoken" to instructions on scalable vectors and
> encode the "unknown, but fixed at program load time" vectorlength (i.e.,
> SVE) by using `token none` as vector length token.
>
> For the initial proposal, I did not want to impose either the loss in
> optimization power or th...
2018 Apr 16
1
RFC: Supporting the RISC-V vector extension in LLVM
...rfect sense are not legal, particularly outlining -- details are in the RFC text. If loss of those optimizations is OK for SVE, we could talk about unifying the two vector types. Alternatively, I believe we could generalize the scalable vector type and my proposal, e.g., by adding my proposed "vlentoken" to instructions on scalable vectors and encode the "unknown, but fixed at program load time" vectorlength (i.e., SVE) by using `token none` as vector length token.
I think your proposed function attribute 'inherits_vlen' would allow for outlining for SVE without causing pr...