intel/mlir-extensions

Issues

• RFC - introducing tile attribute to support cooperative prefetch and load at SG level XeTile dialect IR

  #678 opened 3 months ago by Jianhui-Li
  1

• TEST 'IMEX :: Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir' FAILED

  #682 opened 4 months ago by chsasank
  3

• Build failure on main branch.

  #663 opened 8 months ago by Menooker
  4

• Unsupported operations in ConvertGPUtoSPIRV

  #665 opened 8 months ago by fschlimb
  0

• Auto-inserted deallocation of GPU memory leads to crash

  #664 opened 8 months ago by fschlimb
  0

• Level-Zero system installs don't work

  #652 opened 8 months ago by fschlimb
  0

• [Triton] To inline the VC intrinsic in the SIMT kernel.

  #658 opened a year ago by chengjunlu
  9

• To support user to define a global string in SPIRV dialect.

  #642 opened a year ago by chengjunlu
  2

• Checking Test Suite Gives error: failed to legalize operation 'xegpu.load_2d'

  #661 opened 9 months ago by Scarlet1ssimo
  0

• Add dynamic shape handling for GPU to SPIRV lowering.

  #653 opened a year ago by silee2
  3

• [Triton] Triton generated kernel cannot be load correctly thru the L0 API.

  #659 opened a year ago by chengjunlu
  3

• Spirv.ConstantOp does not support lowering from arith.ConstantOp with bf16

  #657 opened a year ago by Stonepia
  0

• Support atomic operations for low precision

  #654 opened a year ago by tye1
  0

• The SPIRV constant translation has issue in translating the signless value

  #638 opened a year ago by chengjunlu
  6

• Add function decoration FuncParamAttr support in SPIRV dialect.

  #645 opened a year ago by chengjunlu
  1

• Add more arith operation optimization in SPIRV dialect

  #637 opened a year ago by chengjunlu
  6

• It seems the SPIRV serialization has issue in translating the block parameter (SSA in MLIR) into the Phi node (SSA in SPIRV IR)

  #646 opened a year ago by chengjunlu
  2

• Triton reduce kernel get incorrect results on PVC while it is correct on Gen9. Need IGC help to narrow down the root cause.

  #647 opened a year ago by chengjunlu
  4

• Failure in GPU pass pipeline

  #651 opened a year ago by fschlimb
  1

• To support printf in SPIRV dialect which could be landing to the OCL printf.

  #643 opened a year ago by chengjunlu
  1

• Enhance the CompositeExtractOp fold like LLVM ExtractValueOp.

  #648 opened a year ago by chengjunlu
  2

• To support assert

  #644 opened a year ago by chengjunlu
  1

• Add general int2ptr and ptr2int operation in SPIRV dialect.

  #640 opened a year ago by chengjunlu
  2

• Add external call capability in SPIRV dialect

  #636 opened a year ago by chengjunlu
  5

• MathToSPIRV missed math::AbsIop to spirv::CLSAbsOp conversion.

  #650 opened a year ago by chengjunlu
  1

• The GPUBarrierConversion is not align to the SPIRV memory model

  #649 opened a year ago by chengjunlu
  4

• Segmentation Fault in IGC when generating the GPU binary.

  #639 opened a year ago by chengjunlu
  2

• Support the x1 vector data type support for easy to use

  #641 opened a year ago by chengjunlu
  0

• IR print seems corrupted.

  #589 opened a year ago by chencha3
  1

• Rename main to numba-dev and refactor to main

  #551 opened 2 years ago by silee2
  2

• cmake config needs fixing

  #289 opened 2 years ago by fschlimb
  1

• Print functions for I8 and I1 are not supported by upstream

  #479 opened 2 years ago by nbpatel
  3

• Add documentation for imex-runner.py and pass pipeline file syntax

  #539 opened 2 years ago by silee2
  0

• Index lowering failure in upstream ArithToSpirv

  #534 opened 2 years ago by mdvoretc-intel
  0

• Sycl vs level zero runtime pros & cons

  #508 opened 2 years ago by nbpatel
  0

• Convert linalg.generic to linalg.matmul

  #489 opened 2 years ago by mshahneo
  0

• runtime error: symbols not found [memrefcopy]

  #486 opened 2 years ago by Dewei-Wang-sh
  0

• Some math opes (e.g., math.tan) can't be lowered to spirv

  #478 opened 2 years ago by nbpatel
  2

• Complex Type is not supported on GPU

  #468 opened 2 years ago by nbpatel
  1

• Lowering of arith.cmpi "ult" op is not supported by upstream ArithToSpirv pass.

  #469 opened 2 years ago by nbpatel
  1

• GPU Kernel Fusion

  #481 opened 2 years ago by chencha3
  0

• InsertGpuAlloc Pass cannot handle memref's produced from call op.

  #459 opened 2 years ago by nbpatel
  0

• ExplicitPadding Test case fails on GPU pipeline

  #456 opened 2 years ago by nbpatel
  0

• Jax test case failure on GPU pipeline

  #464 opened 2 years ago by nbpatel
  0

• Enable Mobilenet and Resnet Models on CPU and GPU

  #470 opened 2 years ago by nbpatel
  0

• [ptensor] Better IR for elementwise ops

  #380 opened 2 years ago by fschlimb
  0

• [ptensor] Add elementwise unary ops

  #317 opened 2 years ago by fschlimb
  0

• [ptensor] add missing elementwise binary operations

  #316 opened 2 years ago by fschlimb
  0

• New execution engine for Python E2E pipleine

  #313 opened 2 years ago by Hardcode84
  1

• Debugging support in Python E2E pipeline

  #312 opened 2 years ago by Hardcode84
  1