Repository Issue Activity (beta)

#49123 [XLA:GPU] cuDNN int8 convolution with `feature_group_count=2` returns a wrong output channel

#49122 [XLA] SPMD partitioner clamps a partitioned `dynamic-slice` per shard instead of globally

#49121 [XLA] AlgebraicSimplifier folds nested dynamic-slice / dynamic-update-slice without clamping the outer start

#49120 [XLA:GPU] dynamic-slice copy path writes a constant-offset `dynamic-update-slice` at the unclamped offse

#49118 [XLA] SPMD partitioner drops a dynamic-update-slice whose start index needs clamping

#49119 [XLA:GPU] `IndexingMap` composition through `dynamic-update-slice` binds runtime variables to the wrong start operands

#46982 GPU ragged-all-to-all: zero-size peer slices cost ~12 us each, making cost O(n_devices) at fixed traffic

#49047 GPU compilation is not reproducible: GpuHloCostAnalysis sums float utilizations in pointer-hash order, PriorityFusion near-ties flip between processes

#49005 [SPMD] TopK produces incorrect local index shapes when returned directly

#48979 [GPU] WSL2 large-allocation regression after switching device memory to exportable VMM

#48717 [XLA] Constant unsigned gather / dynamic_slice index with the MSB set folds to index 0 (getSExtValue)

#48719 [XLA] RET_CHECK abort compiling reduce_window with an all-1 window and only negative padding

#37116 Multiplication by max fp64 is not elementwise

#47371 Crash on Interpreter: layout_assignment.cc:742 get_channel_constraints(…) Multi-module layout assignment requires…

#47372 Crash on GPU: llvm_loop.cc:76 !preheader_bb_->hasTerminator()

#47373 Crash on GPU: computation_partitioner.cc:269 operand_maps.size() == N

#44942 XLA jit_compile=True crashes with segmentation fault when executing tf.image.rot90 followed by tf.experimental.numpy.cumsum

#23110 Pmap slower with new CPU runtime

#20440 Copies supersede OptimizationBarrier

#15207 Involuntary Full Rematerialization

#44943 XLA jit_compile=True aborts with LLVM "Cannot select" during compilation of Igamma with DT_HALF

#45837 [XLA] compilation fails with InvalidArgumentError: Binary op multiply with different element types: f32[1] and c64[]. for tf.raw_ops.SqrtGrad

#45971 XLA compilation fails for tf.image.pad_to_bounding_box on SelfAdjointEigV2 eigenvalues with InvalidArgumentError

#47344 Lifting bug on CPU: hlo_input_output_alias_config.cc:220 ShapeUtil::IndexIsValid(…)

#47354 Lifting bug: hang/timeout on CPU after HLO-to-StableHLO conversion