search

cplusplus / papers

ISO/IEC JTC1 SC22 WG21 paper scheduling and management
625 stars 18 forks source link

P0350 Integrating simd with parallel algorithms #544

Open wg21bot opened 5 years ago

wg21bot commented 5 years ago

P0350R2 Integrating simd with parallel algorithms (Matthias Kretz)

tituswinters commented 5 years ago

Prefer expressing simd execution via policy tag (not newly named algorithms, std::simd_transform, etc). SF F N A SA 4 9 3 0 0

What do we do with predicates? (Roughly option 2 from the paper, be mindful of the amount of wording that is thrown out of conformance.)

Require that this goes first into std::ranges algorithms (which haven’t been parallelized yet). SF F N A SA 0 2 7 5 2

Revise and come back.

wg21bot commented 4 years ago

P0350R3 Integrating simd with parallel algorithms (Matthias Kretz)

mattkretz commented 4 years ago

P0350R3 was discussed in Cologne already (as D0350R3). No need to discuss it again until R4. Sorry for the confusion.

wg21bot commented 4 years ago

P0350R4 Integrating simd with parallel algorithms (Matthias Kretz)

mattkretz commented 4 years ago

I had to cancel my plans to attend Prague. Please postpone discussion of R4 to the next meeting.

cor3ntin commented 3 years ago

Mailing list review: 2021-01-25 Few participants (5 people, including the author). Some concerns over the fact that the simd interface would use different types in the same interface (when invoking functions and predicates), but not enough to warrant a design change without more objections. If anything, the author is encouraged to make that more clear in the paper introduction! We noted that this design is not unusual and used in similar 3rd party libraries Some wording suggestions Guidance: Clarify more explicitly the types of the predicate parameters, Improve the wording, go to LEWG meeting next

brycelelbach commented 1 year ago

Closing due to more than 1 year of inactivity. Please re-open if further work is expected.

mattkretz commented 1 year ago

I will work on this again as soon as the workload of P1928 and related papers doesn't take up all of my time.