Search results

[TritonIntelGPU] Enable AnnotateCacheControl pass by default back

hw: bmg

performance

tests: e2e

windows

Status: Open.

Task

#7579 In intel/intel-xpu-backend-for-triton;

· anmyachev opened on Jul 24, 2026

[vLLM][MoE] Block fp8 should not select native grouped GEMM with only 1D scale support

vllm

Status: Open.

Bug

#7574 In intel/intel-xpu-backend-for-triton;

· sarah12121212 opened on Jul 24, 2026

[Pytorch] layernorm+softmax fusion from inductor has lower hardware efficiency than CUDA

performance

upstream: pytorch

Status: Open.

#7571 In intel/intel-xpu-backend-for-triton;

· LuFinch opened on Jul 24, 2026

[Inductor GEMM] Investigate unexpected MOV instructions between 2D Block Load and DPAS in FP8 scaled_mm triton kernel

performance

Status: Open.

#7561 In intel/intel-xpu-backend-for-triton;

· xiaowangintel opened on Jul 23, 2026

"Test with pip" is broken

ci

dependencies: pti

Status: Open.

Bug

#7554 In intel/intel-xpu-backend-for-triton;

· anmyachev opened on Jul 22, 2026

Track mean performance of E2E models from "performance" suite to avoid regressions.

performance

tests: e2e

Status: Open.

Task

#7552 In intel/intel-xpu-backend-for-triton;

· anmyachev opened on Jul 22, 2026

Triton FA FP8 fwd kernel benchmark runs too long

benchmarks

Status: Open.

Task

#7543 In intel/intel-xpu-backend-for-triton;

· anmyachev opened on Jul 21, 2026

[TritonIntelGPU] Enable CanonicalizePointers pass by default back

performance

tests: e2e

Status: Open.

Task

#7540 In intel/intel-xpu-backend-for-triton;

· anmyachev opened on Jul 21, 2026

[vLLM] Update vLLM pin

vllm

Status: Open.

#7531 In intel/intel-xpu-backend-for-triton;

· quinnlp opened on Jul 20, 2026

Enable "[LoadStoreOpToLLVM] Honor eviction_policy hints via LSC cache modes (#6775)" back

codegen: cache-controlCache policy annotation, eviction/cache hints, LSC cache modes.Cache policy annotation, eviction/cache hints, LSC cache modes.

performance

tests: e2e

Status: Open.

Task

#7530 In intel/intel-xpu-backend-for-triton;

· anmyachev opened on Jul 20, 2026

[Codegen] Auto-GRF / register-mode selection

codegen: compiler-infraUmbrella label for reusable compiler infrastructure that is not tied to a single feature.Umbrella label for reusable compiler infrastructure that is not tied to a single feature.

dependencies: igc

performance

Status: Open.

Epic

#7529 In intel/intel-xpu-backend-for-triton;

· vlad-penkin opened on Jul 20, 2026

·  

4. [Performance] Core

[Windows] test_compile_only_* fail on Intel-only CI: missing ptxas-blackwell.exe / AMD toolchain

tests: ut

upstream: triton

windows

Status: Open.

Bug

#7504 In intel/intel-xpu-backend-for-triton;

· kwasd opened on Jul 16, 2026

·  

3. [Triton] Language and Runtime