intel/intel-xpu-backend-for-triton

Issues

• [Performance] Improve the flash attention performance on bottom-up optimization pipeline

  #2177 opened 2 months ago by chengjunlu
  4

• [PyTorch pin update] `951c21d6` - Inductor tests are starting to fail after pybind11 update to 2.13.6

  #2260 opened 2 months ago by pbchekin
  9

• Update auto-update-translator-cid to use upstream PyTorch

  #2214 opened 2 months ago by pbchekin
  0

• test-triton.sh does not work if git clone has different name

  #2253 opened 2 months ago by pbchekin
  0

• Compilation errors with Clang 17.0.6

  #2261 opened 2 months ago by pbchekin
  0

• Investigate failures from updating SPIRV-LLVM-Translator

  #2247 opened 2 months ago by whitneywhtsang
  1

• [E2E Accuracy] Enhancements

  #2165 opened 2 months ago by vlad-penkin
  0

• [E2E Accuracy] Baseline results per platform and driver

  #2164 opened 2 months ago by vlad-penkin
  0

• Remove outdated options from compile-pytorch-ipex.sh

  #2231 opened 2 months ago by Retribution98
  0

• Subset of E2E models for accuracy tests

  #2246 opened 2 months ago by pbchekin
  1

• Increase `warmup` and `rep` for FA benchmark

  #2255 opened 2 months ago by anmyachev
  0

• [FA] Identify the root causes for the recent Triton geomean performance degradation vs XeTLA

  #2257 opened 2 months ago by vlad-penkin
  0

• [Advanced Path] can not reproduce shader dump

  #2250 opened 2 months ago by Dewei-Wang-sh
  0

• [Feature Improvement] Change large GRF warnings to trigger by debug flag

  #2251 opened 2 months ago by Stonepia
  1

• Align the conditions under which benchmarks are run for different implementations

  #2235 opened 2 months ago by anmyachev
  0

• Rename branch llvm-target to main and make it default

  #2176 opened 2 months ago by pbchekin
  0

• Torch inductor tests failed for PyTorch main

  #2206 opened 2 months ago by pbchekin
  3

• Some triton kernels gernerated by inductors have low efficiency on PVC 1550 compare to A100

  #2229 opened 2 months ago by jianyizh
  2

• Merge OpenAI Triton till Sept 27th

  #2244 opened 2 months ago by whitneywhtsang
  0

• Decompose large `simdblockwrite` to smaller `simdblockwrite`s

  #2226 opened 2 months ago by whitneywhtsang
  0

• [test-triton.sh] Run micro benchmark tests without IPEX

  #2243 opened 2 months ago by whitneywhtsang
  0

• Lit tests executed twice in CI

  #2240 opened 2 months ago by pbchekin
  0

• CXX tests not running in CI

  #2238 opened 2 months ago by anmyachev
  4

• Decompose large simdblockread to smaller simdblockreads

  #2205 opened 2 months ago by vlad-penkin
  0

• [Benchdmark] [attention] core dumped

  #2228 opened 2 months ago by AshburnLee
  1

• [TritonGEN] Use type mangling utility to mangle the function name for 2D block reads calls

  #2225 opened 2 months ago by whitneywhtsang
  0

• Create presentation for Triton developer meeting

  #2169 opened 2 months ago by etiotto
  1

• Collect performance metrics for tutorials

  #2207 opened 2 months ago by pbchekin
  1

• [CI][Local runs] Unify test runners code in CI and in local env. Switch to `test-triton.sh` script in CI

  #2184 opened 2 months ago by vlad-penkin
  0

• Fix copy of `compile_commands.json` in `compile-triton.sh`

  #2222 opened 2 months ago by anmyachev
  0

• Re-organize build-test workflows

  #2218 opened 2 months ago by pbchekin
  0

• Make benchmarks compatible for both IPEX usage and upstream PyTorch usage

  #2201 opened 2 months ago by anmyachev
  0

• computed results are incorrect with blocked pointer matrix multiplication, in TRITON_INTEL_ADVANCED_PATH

  #2209 opened 2 months ago by arunjose696
  0

• [Local runs] Enhancements to `test-triton.sh` to support local test runs with nightly wheels mode

  #2187 opened 2 months ago by vlad-penkin
  0

• Clean up `test-triton.sh`

  #2208 opened 2 months ago by pbchekin
  0

• Provide the ability to run benchmarks with or without IPEX

  #2204 opened 2 months ago by anmyachev
  0

• [TritonGEN] Use OCL builtins for subgroup block read/write

  #2202 opened 2 months ago by vlad-penkin
  0

• ValueError: Pointer argument (at 0) doesn't reference XPU device memory (cpu tensor?)

  #2188 opened 2 months ago by dvrogozh
  1

• [sycl-free-inference-for-llms] Port and evaluate LLama3-8B and Granite-8B

  #2170 opened 2 months ago by etiotto
  1

• [sycl-free-inference-for-llms] flashattention kernel get up to ~100tflops

  #2195 opened 2 months ago by Dewei-Wang-sh
  1

• [sycl-free-inference-for-llms] gemm kernel perf get up to ~290tflops

  #2194 opened 2 months ago by Dewei-Wang-sh
  1

• [sycl-free-inference-for-llms] run llama3-8B with Pytorch for xpu and get the base line

  #2197 opened 2 months ago by Dewei-Wang-sh
  1

• [sycl-free-inference-for-llms] integrate triton gemm/attention in pytorch for xpu

  #2196 opened 2 months ago by Dewei-Wang-sh
  1

• Rewrite benchmarks to be more `elapsed_time` friendly

  #2198 opened 2 months ago by anmyachev
  0

• [CI] FlashAttension SLM path failure

  #2185 opened 2 months ago by vlad-penkin
  0

• [UT][Tutorials] Add tutorials pass rate summary subreport

  #2166 opened 2 months ago by vlad-penkin
  0

• Adding `XeTLA` FA implementation for all variants mentioned in the FA paper

  #2179 opened 2 months ago by ESI-SYD
  0

• [IPEX deprecation] Add Benchmarks job to the PyTorch pin update validation workflow

  #2172 opened 2 months ago by kwasd
  0

• [benchmarks] Add Prefix sums benchmark

  #2167 opened 2 months ago by vlad-penkin
  0

• [GEMM] Add autotune configs with different prefetch distances

  #2168 opened 2 months ago by vlad-penkin
  0