[llvm-dev] [RFC] Vector Predication (original) (raw)

Simon Moll via llvm-dev llvm-dev at lists.llvm.org
Thu Feb 7 09:03:10 PST 2019

On 2/7/19 5:56 PM, David Greene wrote:

Simon Moll <moll at cs.uni-saarland.de> writes:

In unscaled IR types, this means VL masks each scalar result, in scaled types VL masks sub vectors. E.g. for %L == 1 the following call produces a pair of floats as the result:

<scalable 2 x float> evl.fsub(<scalable 2 x float> %x, <scalable 2 x float> %y, <scalable 2 x i1> %M, i32 %L) That seems wrong to me. In the SVE proposal, <scalable 2 x float> means a dynamic vector length guaranteed to be a multiple of 2 floats long. There is no notion of sub-vector. The vector length parameter should result in <scalable 2 x float> result, but where the second float value is undefined. It would be surprising to get two full results. If sub-vector types were notated as <scalable 1 x <2 x float>> then %L == 1 would result in two full float results. -David

We have agreed to interpret vlen elementwise by now. Let's keep discussing that in the RFC so everything is in one place.

I've summarized the current state there: https://reviews.llvm.org/D57504#1387733

Thanks

--

Simon Moll Researcher / PhD Student

Compiler Design Lab (Prof. Hack) Saarland University, Computer Science Building E1.3, Room 4.31

Tel. +49 (0)681 302-57521 : moll at cs.uni-saarland.de Fax. +49 (0)681 302-3065 : http://compilers.cs.uni-saarland.de/people/moll

More information about the llvm-dev mailing list