search for: fcmla

...re-start the discussion on the topic. Given that this original thread was started a while ago apologies if I missed anything already discussed earlier on the list or the round-table. The original mail is quoted below. > > In particular, I’m interested in the AArch64 side of things, like using FCMLA [1] for complex multiplications to start with. > > To get the discussion going, I’d like to share an alternative pitch. Instead of starting with adding complex types, we could start with adding a set of intrinsics that operate on complex values packed into vectors instead. > > Starting...

Krzysztof Parzyszek via llvm-dev <llvm-dev at lists.llvm.org> writes: > Examples of complex instructions? Sorry, I was referring specifically to this statement: >> Without intrinsics it may be hard to use such instructions especially >> because of the arithmetic simplifications. I was asking the question in the context of intrinsics vs. a first-class complex type.

...re-start the discussion on the topic. Given that this original thread was started a while ago apologies if I missed anything already discussed earlier on the list or the round-table. The original mail is quoted below. > > In particular, I’m interested in the AArch64 side of things, like using FCMLA [1] for complex multiplications to start with. > > To get the discussion going, I’d like to share an alternative pitch. Instead of starting with adding complex types, we could start with adding a set of intrinsics that operate on complex values packed into vectors instead. > > Starting...

...the discussion on the topic. Given that this original thread was started a while ago apologies if I missed anything already discussed earlier on the list or the round-table. The original mail is quoted below. >> >> In particular, I’m interested in the AArch64 side of things, like using FCMLA [1] for complex multiplications to start with. >> >> To get the discussion going, I’d like to share an alternative pitch. Instead of starting with adding complex types, we could start with adding a set of intrinsics that operate on complex values packed into vectors instead. >>...

...like to re-start the discussion on the topic. Given that this original thread was started a while ago apologies if I missed anything already discussed earlier on the list or the round-table. The original mail is quoted below. In particular, I’m interested in the AArch64 side of things, like using FCMLA [1] for complex multiplications to start with. To get the discussion going, I’d like to share an alternative pitch. Instead of starting with adding complex types, we could start with adding a set of intrinsics that operate on complex values packed into vectors instead. Starting with intrinsics...

Ahead of the Wednesday’s roundtable at the developers’ conference, here is version three of the proposal for first-class complex types in LLVM. I was not able to add Krzysztof Parzyszek’s suggestion of a “cunzip” intrinsic returning two vectors as I could not find examples of intrinsics that return two values at the IR level. The Hexagon intrinsics declared to return two values do not actually

...y), or where one compound group needs to be reordered, like A[i].x + B[i].y, A[i].y + B[i].x * loops where only parts are applicable to SLP-style vectorization * loops with complex operations that can be mapped to specialized HW instructions, like complex multiply and accumulate using FCMLA (Arm v8.3-a) We do not necessarily have to do 3. in LoopVectorize. If we detect SLP opportunities, instead of choosing to interleave, we could decide to just unroll the loop so we can make optimal use of the vector registers with compound operations and let the SLP vectorizer generate code....

...547.html), summarising the current state of support, regulary sync-up meeting timings, and what's next. * Florian Hahn brought life back to an old discussion on [better support for complex numbers](http://lists.llvm.org/pipermail/llvm-dev/2020-November/146568.html), e.g. to support lowering to FCMLA on AArch64. * ORC JIT Weekly #26 [is out](http://lists.llvm.org/pipermail/llvm-dev/2020-November/146633.html), covering he Orc library break-up, remote TargetProcessControl, and the beginnings of a runtime. * Alexander Yermolovich started a discussion on [supporting DWARF64 with `debug_info` sort...

...re-start the discussion on the topic. Given that this original thread was started a while ago apologies if I missed anything already discussed earlier on the list or the round-table. The original mail is quoted below. > > In particular, I’m interested in the AArch64 side of things, like using FCMLA [1] for complex multiplications to start with. > > To get the discussion going, I’d like to share an alternative pitch. Instead of starting with adding complex types, we could start with adding a set of intrinsics that operate on complex values packed into vectors instead. > > Starting...

...y), or where one compound group needs to be reordered, like A[i].x + B[i].y, A[i].y + B[i].x * loops where only parts are applicable to SLP-style vectorization * loops with complex operations that can be mapped to specialized HW instructions, like complex multiply and accumulate using FCMLA (Arm v8.3-a) We do not necessarily have to do 3. in LoopVectorize. If we detect SLP opportunities, instead of choosing to interleave, we could decide to just unroll the loop so we can make optimal use of the vector registers with compound operations and let the SLP vectorizer generate code. Bu...