openxla / iree

A retargetable MLIR-based machine learning compiler and runtime toolkit.

http://iree.dev/

openxla/iree Issues

Pending LLVM Dependent Patches
Updated 2 months ago2
LowerToUKernels should create a codegen fallback when the ukernel would be slow.
Updated 3 months ago10
Python 3.12 Wheels
Updated 3 months ago8
Turbine Llama2 requires `IREE_TASK_EXECUTOR_MAX_OUTSTANDING_WAITS >= 740`
Closed 3 months ago
[Epic] HIP HAL driver rewrite
Updated 3 months ago
[CPU] Missing ukernel for f32, i8 -> f32 mmt4d?
Updated 3 months ago11
[LLVMCPU] Potentially bad codegen for tensor.unpack op fused with linalg.generic op
Closed 3 months ago2
[CPU][DT] DT not triggering for i8,i8 -> i32 matmul?
Closed 4 months ago6
Embed executable sources in vmfb files for portable Tracy captures
Updated 4 months ago4
Numerical issue with chatglm2.vmfb model
Closed 5 months ago40
Update executable linking utilities/passes to support `hal.executable.condition`.
Closed 4 months ago
DecomposeSoftmax defines static methods not directly used in the translation unit
Closed 4 months ago1
Ensure LHS remains in L1 cache in 2 successive UKernel calls
Updated 4 months ago10
[DT][CPU] Some casting ops are not fused with mmt4d kernels
Closed 5 months ago3
Numerical issue with qwen7b.vmfb model
Closed 5 months ago4
[HIP] Support native executable and cache
Closed 5 months ago1
Unable to compile SHARK
Updated 5 months ago2
Regenerate `Falcon7bGptqPT` MLIR for benchmarking
Closed 5 months ago5
CUDA transform dialect reduction tests are flakey.
Updated 5 months ago1
[ROCM] Graph command buffer implementation
Closed 5 months ago
[ROCM] Direct command buffer update
Closed 5 months ago
The compilation time is hugely regressed after #15752
Updated 5 months ago12
Trying to run gpt2 results in `ValueError: unimplemented array format conversion from format: ?`
Updated 5 months ago2
coredump when use the iree-compile
Closed 5 months ago2
Disable `iree-input-demote-f64-to-f32` and `iree-input-demote-i64-to-i32` by default?
Updated 5 months ago5
[regression] Falcon7bGptqPT is 10x slower after #15766
Closed 5 months ago4
User build issue: `undefined symbol: mlir::torch::registerTorchConversionPasses()`
Updated 5 months ago1
when lowering to gpu.subgroup_mma for this datatype f16xf16+f32, the lowering is error
Updated 5 months ago
test_benchmark_suites workflow is not running any tests
Updated 5 months ago1
Support graph command buffer
Closed 5 months ago
[HIP] Support descriptor set and pipeline layout
Updated 5 months ago
[HIP] Support graph command buffer
Updated 5 months ago
IREE fails to build when it's source tree is not the root directory of the cmake call
Closed 5 months ago4
[CPU] Missing ukernel for f32, i8 -> f32 mmt4d
Closed 5 months ago2
Upload `iree-dump-*` to PIP for easy & cross-platform inspection of `.mlir` code.
Updated 5 months ago6
[spirv] Incorrect int4 store results
Updated 5 months ago
[CPU] Do not use mmt4d for non-DT on ARM
Closed 5 months ago5
[SDXL] CUDA A100 40GB Linux perf tracker
Updated 5 months ago
`iree-compile` crash after targeting multiple HAL backend
Updated 5 months ago
[data-tiling] UNet compilation fails with data tiling enabled
Updated 5 months ago3
[Flow] Fusion of `tensor.unpack` ops with reduction `linalg.generic` ops discussion
Updated 5 months ago7
Tracy sampling gets wrong symbol address for Turbine Llama2 mmt4d ukernel dispatch
Closed 5 months ago
Add revisioning (created and last update dates) to the website
Updated 5 months ago
[CPU][DT] Type propagation pass fails with DT enabled
Closed 5 months ago11
[rocm] Dumping intermediate executable results in crash in llvm backend isel
Closed 5 months ago1
Form cast ops and mmt4d ops into the same dispatch.
Closed 5 months ago
[Tracking] Upstreaming to llvm/mlir chores
Updated 5 months ago
Support i4 types in ConstEval
Updated 5 months ago
[Fusion] error codegen for fusing reduce_max and sub ops into a dispatch region
Updated 6 months ago
[CUDA] cuMemPrefetchAsync Error
Updated 6 months ago