[PATCH] D75653: [mlir] Introduce an intrinsic for llvm.matrix.multiply

[Nicolas Vasilache via Phabricator via llvm-commits]

nicolasvasilache created this revision.
nicolasvasilache added a reviewer: ftynse.
Herald added subscribers: llvm-commits, Joonsoo, liufengdb, lucyrfox, mgester, arpith-jacob, antiagainst, shauheen, burmako, jpienaar, rriddle, mehdi_amini, dexonsmith.
Herald added a project: LLVM.

This revision adds the first intrinsic for llvm.matrix.multiply.
This uses the more general `LLVM_OneResultOp` for now since the goal is to use the
specific Matrix builders that @fhahn has created recently.

We can discuss in the future whether to provide special idiomatic tblgen support for this, but it is only 4 ops atm so unclear it is worth it yet.

When piped through:

  opt -O3 -enable-matrix | llc -O3 -march=x86-64 -mcpu=skylake-avx512

this has been verified to generate zmm instructions.

Benchmarking will be provided separately with the experimental metaprogramming  ModelBuilder <https://github.com/google/iree/tree/master/experimental/ModelBuilder> tool when ready.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D75653

Files:
  mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td
  mlir/include/mlir/Target/LLVMIR/ModuleTranslation.h
  mlir/test/Target/llvmir-intrinsics.mlir

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D75653.248344.patch
Type: text/x-patch
Size: 4618 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200304/2c05e2bd/attachment.bin>