similar to: Loop-vectorizer prototype for the EPI Project based on the RISC-V Vector Extension (Scalable vectors)

Hi all, It's been a while since we've given an update on scalable vector support in LLVM. Over the last 12 months a lot of work has been done to make LLVM cope with scalable vectors. This effort is now starting to bear fruit with LLVM gaining more capabilities, including an intrinsics interface for AArch64 SVE/SVE2, LLVM IR Codegen for scalable vectors, and several loop-vectorization

Hi all, On 11/5/20 10:32 AM, Roger Ferrer Ibáñez wrote: Hi Sjoerd, thanks for pointing us to this intrinsic. I see it returns a mask/predicate type. My understanding is that VPred intrinsics have both a vector length operand and a mask operand. It looks to me that a "popcount" of get.active.lane.mask would correspond to the vector length operand. Then additional "control

For RISC-V V and VE being explicit about %evl is important for performance & correctness and that is what VP does. The get.active.lane.mask intrinsic is used as a hint for the MVE, SVE backends to use hardware tail-predication (the backends reverse engineer that hint by pattern matching for get.active.lane.mask in the mask parameter of "some" masked intrinsics). IMHO, it's more

On 11/6/20 12:39 PM, Sjoerd Meijer wrote: Hello Simon, Thanks for your replies, very useful. And yes, thanks for the example and making the target differences clear: ; Some examples: ; RISC-V V & VE(*): ; %mask = (splat i1 1) ; %evl = min(256, %n - %i) ; MVE/SVE : ; %mask = get.active.lane.mask(%i, %n) ; %evl = call @llvm.vscale() ; AVX: ; %mask = icmp (%i + (seq

On 11/6/20 8:49 AM, Roger Ferrer Ibáñez wrote: Hi Sjoerd, Trying to remember how everything fits together here, but could get.active.lane.mask not create the %mask of the VP intrinsics? Or in other words, in the vectoriser, who's producing the %mask and %evl that is consumed by the VP intrinsics? I'm not sure what would be the best way here. I think about the Loop Vectorizer. I imagine

; RISC-V V & VE(*): ; %mask = get.active.lane.mask(%i, %i) ; %evl = min(256, %n - %i) ; MVE/SVE/AVX : ; %mask = get.active.lane.mask(%i, %n) ; %evl = call @llvm.vscale() For VE, we want to do as much predication as possible through %evl and as little as possible with %mask. This has performance implications on VE and RISC-V - VE does not generate a mask from %evl but %evl is

Hello Simon, Thanks for your replies, very useful. And yes, thanks for the example and making the target differences clear: ; Some examples: ; RISC-V V & VE(*): ; %mask = (splat i1 1) ; %evl = min(256, %n - %i) ; MVE/SVE : ; %mask = get.active.lane.mask(%i, %n) ; %evl = call @llvm.vscale() ; AVX: ; %mask = icmp (%i + (seq <8 x i32> 0,1,2,.,)), %n, ; %evl

Hi all, Several people have expressed interest in having regular sync-up calls to follow the progress of ARM SVE/SVE2 support in LLVM. With this effort picking up steam and multiple people contributing patches, regular sync-up calls would provide a platform to coordinate our efforts. Items to discuss in these sync-up calls would include: - SVE/SVE2 CodeGen support - Adding support for C/C++

Thanks to all who attended last week's SVE/SVE2 sync-up meeting! The next meeting will be on Thursday, January 9th, at 5pm GMT / 9am PST. I've created the agenda for the next meeting here: https://docs.google.com/document/d/17GqhBZaADtnDlDL3L4wWqcMd0qvfzB-VEe_ntodgYjA (If you want to attend, please add your name, email and organisation to the list referenced in the document above so I

Hi Sander, Awesome work from everyone involved. Thank you very much for your efforts! I know some people wanted it to go a lot faster than it did, but now we have an infrastructure that has reached consensus across different companies and industries. We're finally discussing high level vectorisation strategies without having to worry about the mechanics of scalable vector representation.

Hello, Just bringing to people's attention that we organise a bi-weekly sync-up call to discuss and collaborate on upstream support for scalable vectors and SVE/SVE2 CodeGen support in LLVM. The meetings are held every other Thursday, with the next meeting tomorrow (April 2nd) at 3pm GMT / 7am PST. The invite and agenda for tomorrow's meeting (including links to minutes from previous

On 31 July 2018 at 21:10, David A. Greene via llvm-dev <llvm-dev at lists.llvm.org> wrote: > Renato Golin via llvm-dev <llvm-dev at lists.llvm.org> writes: > >> Hi David, >> >> Let me put the last two comments up: >> >>> > But we're trying to represent slightly different techniques >>> > (predication, vscale change) which need

Hi Simon, Thanks for reposting the example, and looking at it more carefully, I think it is very similar to my first proposal. This was met with some resistance here because it dumps loop information in the vector preheader. Doing it this early, we want to emit this in the vectoriser, puts a restriction on (future) optimisations that transform vector loops to honour/update/support this intrinsic

Hi, I'd like to propose a change to our cost interfaces so that instead of returning an unsigned value from functions like getInstructionCost, getUserCost, etc., we instead return a wrapper class that encodes an integer cost along with extra state. The extra state can be used to express: 1. A cost as infinitely expensive in order to prevent certain optimisations taking place. For example,

Hello, We would like to propose making LoopVectorize aware of SLP operations, to improve the generated code for loops operating on struct fields or doing complex math. At the moment, LoopVectorize uses interleaving to vectorize loops that operate on values loaded/stored from consecutive addresses: vector loads/stores are generated to combine consecutive loads/stores and then shufflevector

Hello, I''m trying to get working infiniband device in Xen. I have Xen v3 with 2.6.18 kernel. I export PCI device from Dom0 into DomU. I''ve enabled permisive device in Dom0. However, inserting infiniband module results in the following messages (with little more verbosity which I have added): pcifront pci-0: Installing PCI frontend pcifront pci-0: Creating PCI Frontend Bus

When I pass an empty string for cpu and features to createTargetMachine, and then use LLVMGetTargetMachineCPU() and LLVMGetTargetMachineFeatureString() to get the strings back, they are still empty. Is there a way to have llvm compute the effective cpu/features string, and provide it so that I can inspect it? I'm trying to figure out how the cpu/features string that I am explicitly passing,

All, I am running xen3.0.2 and xen unstable on RHEL 4, and get this error when trying to boot Xen kernel(domain0) ...... Creating root device Mounting root filesystem mount: error 6 mounting ext3 mount: error 2 mounting none Switching to new root switchroot: mount failed: 22 umount /initrd/dev failed: 2 Kernel panic - not syncing: Attempting to kill init! Then system reboot.In grub.conf,I have

All, I am running xen3.0.2 and xen unstable on RHEL 4, and get this error when trying to boot Xen kernel(domain0) ...... Creating root device Mounting root filesystem mount: error 6 mounting ext3 mount: error 2 mounting none Switching to new root switchroot: mount failed: 22 umount /initrd/dev failed: 2 Kernel panic - not syncing: Attempting to kill init! Then system reboot.In grub.conf,I have

Hi David, Let me put the last two comments up: > > But we're trying to represent slightly different techniques > > (predication, vscale change) which need to be tied down to only > > exactly what they do. > > Wouldn't intrinsics to change vscale do exactly that? You're right. I've been using the same overloaded term and this is probably what caused the