Do we need an unsigned multiplyHigh? (original) (raw)
John Rose john.r.rose at oracle.com
Wed Sep 27 18:45:24 UTC 2017
- Previous message: Do we need an unsigned multiplyHigh?
- Next message: Do we need an unsigned multiplyHigh?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
On Sep 27, 2017, at 8:25 AM, Peter Lawrey <peter.lawrey at gmail.com> wrote:
Perhaps project Valhalla will be a way to return multiple values by having a composite value type, or panama with it's support for XMM instructions (or both)
That seems likely; we are aiming in both of those directions, to support direct programming with AVX-type registers (not just x86 specific, by the way) and general support for by-value return of structured objects (starting with "minimal value types").
For now, though, every method is limited to at most 64 bits of return value "payload", which means that 128-bit operations need to be split into two method calls, or else buffer their result into a temp object (e.g., array). The JIT knows how to combine two intrinsic calls into a single machine operation, in some very limited circumstances, notably the classic "div/rem" instructions. This technique would probably work for 64-to-128 multiplies. (Also AES-128, by the way.)
— John
- Previous message: Do we need an unsigned multiplyHigh?
- Next message: Do we need an unsigned multiplyHigh?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]