Issues
- 1
Make a CMake project importable via find_package
#946 opened by mryzhov - 0
- 0
Next steps on target descriptor
#934 opened by rengolin - 2
- 3
Investigate re-enabling `mlir-doc` target
#940 opened by rolfmorel - 0
Replicate MLP examples in Linalg Named Ops
#933 opened by rengolin - 3
Linking CXX executable bin/tpp-opt failed
#935 opened by rolfmorel - 1
Handle `linalg.max`
#671 opened by chelini - 0
Canonicalize affine.min map when tensor are dynamics
#676 opened by chelini - 5
MHA benchmarks are taking too long
#805 opened by rengolin - 1
- 6
tpp-run does not support ml_program dialect
#704 opened by nhasabni - 7
PyTorch MLP doesn't fuse XSMM
#881 opened by rengolin - 3
- 0
IntelAMX TileConfig Hoisting should be anchored on IntelAMXTileConfigDispatchOp
#896 opened by alheinecke - 1
SPR scalability issue
#895 opened by rengolin - 0
Make 2D parallelization a run time choice
#894 opened by rengolin - 1
SPR XSMM fusion does not trigger on packed models
#892 opened by rengolin - 0
Graviton 3 packing not working
#891 opened by rengolin - 0
Add Zen/Arm allocation for benchmark job
#798 opened by rengolin - 1
Update libxsmm-dnn with new argument for VNNI^T
#829 opened by rengolin - 0
Invalid constant in `WmmaOpsToNvvm` lowering
#870 opened by adam-smnk - 1
XSMM fusion at memref level
#659 opened by rengolin - 0
BF16 BRGEMM is not matched on mlir-gen input when using prepacked weights and tilesize and tensor size are identical
#806 opened by alheinecke - 0
VNNI pack: use linalg.generic instead of tpp.brgemm
#822 opened by chelini - 1
Post-bufferization buffer reuse pass
#658 opened by rengolin - 2
Check Buildkite's Test Analytics
#792 opened by rengolin - 2
Verifier for GEMM and BRGEMM in XSMM
#779 opened by chelini - 2
Lower pack to `linalg.copy` instead of `memref.copy`
#813 opened by rengolin - 0
Some warnings in the build
#815 opened by rengolin - 0
- 3
Add support in the `perf` dialect to start/stop perf counters when we start/stop the timer
#788 opened by rengolin - 0
PyTorch GEMM/MLP into benchmarks
#776 opened by rengolin - 6
`transform-convolutions.mlir` passes on my machine
#789 opened by rengolin - 10
- 0
Detect cmpf and select as Relu
#773 opened by nhasabni - 2
libxsmm pointer cast warnings
#707 opened by rengolin - 4
Move our dep to stable libxsmm
#649 opened by rengolin - 1
tpp-run crashes on TorchDynamo generated MLP
#770 opened by nhasabni - 2
Robust Performance Benchmarks
#614 opened by rengolin - 0
tpp.zero error: 'tpp.zero' op operand #0 must be 1D/2D memref of floating-point values or 1D/2D tensor of floating-point values or floating-point, but got 'tensor<>'
#743 opened by nhasabni - 7
- 4
- 2
Update CUDA driver on the CI.
#669 opened by chelini - 0
Constant-fold-pack pass error
#657 opened by KavithaTipturMadhu - 1
TPP-performance tests are not running
#639 opened by hfp - 0
- 0
Deprecate TPP conversion on memrefs
#600 opened by adam-smnk - 0
TPP bufferization does not deallocate
#586 opened by adam-smnk - 3
Enable running softmax with TPPs
#584 opened by chelini