Enable TensorIndexer with all C++ tests

[naoyam]

No description provided.

[naoyam]

!test

[github-actions]

Review updated until commit 0823fe7

Description

• Move circular buffer index variable allocation earlier in the logic flow

• Add support for Stream parallel type in loop index variable allocation

• Enable IdModel (TensorIndexer) by default in matmul scheduler tests

• Enable IdModel by default in stream tests and base test fixture

Changes walkthrough

	Relevant files
Bug fix	id_model.cpp Fix circular buffer allocation order and add stream support csrc/id_model/id_model.cpp Move circular buffer index variable allocation before thread index handling Add ParallelType::Stream support to thread index condition Add error checking for missing circular buffer index variables +23/-17
Tests	test_matmul_scheduler.cpp Enable IdModel in matmul scheduler test fixtures                  tests/cpp/test_matmul_scheduler.cpp Add SetUp() methods to enable IdModel with "all" option Apply to MatmulSchedulerTest, MatmulSchedulerPluginTest, AllocationDomainTest Apply to HopperPlusMatmulSchedulerTest +17/-0    test_stream.cpp Enable IdModel in stream tests                                                      tests/cpp/test_stream.cpp Enable IdModel with "all" option in StreamTest constructor +1/-0      utils.cpp Enable IdModel in base test fixture                                            tests/cpp/utils.cpp Enable IdModel with "all" option in NVFuserTest::SetUp() +1/-0

PR Reviewer Guide

Here are some key observations to aid the review process:

🧪 PR contains tests

⚡ Recommended focus areas for review

Code Movement Impact The circular buffer handling code has been moved earlier in the allocateLoopIndexVariables() function. While the logic appears preserved, this reordering could potentially change execution flow or introduce subtle bugs. The moved code handles allocation of circular buffer index variables for each stage, and any change in when this code executes relative to other logic could have unintended consequences. if (GpuLower::current()->circularBufferInfo().isCircularBufferedIterDomain( loop_group->front()->as<IterDomain>())) { // Allocate index variable for each stage of the circular // buffered loop. auto indices = std::make_unique<CircularBufferIndices>(); for (auto i : arange(static_cast<int>(CircularBufferLoopStage::EndOfStages))) { indices->emplace( static_cast<CircularBufferLoopStage>(i), IrBuilder::create<Val>(DataType::Index)); } circular_buffered_loop_index_variable_map_[loop_group] = std::move(indices); continue; } New Error Checking New error checking has been added for circular buffer index variables that will throw NVF_ERROR if the circular_buffered_loop_index_variable_map_ doesn't contain the expected loop_group. This is good for catching bugs but could cause test failures if the assumptions about when this function is called don't hold true in all test scenarios. NVF_ERROR( circular_buffered_loop_index_variable_map_.contains(loop_group), "Failed to find circular buffer index var for: ", nvfuser::toString(loop_group), ", ", loop_group->front()->toString()); Parallel Type Handling The condition for using NamedScalar::getParallelIndex has been expanded to include ParallelType::Stream in addition to thread parallel types. This change in indexing logic for stream types could potentially affect performance or correctness in stream-based operations and should be validated. } else if (isParallelTypeThread(ptype) || ptype == ParallelType::Stream) { loop_index = NamedScalar::getParallelIndex(ptype); }

Test failures

• (High, 189) CUDA driver version insufficient for runtime on dlcluster_h100 (nvFuser test suites)

  Test Name	H100	Source
  ArgsortParameterizedWithBlockAndBatch.SharedMemoryRequirement/128_1_0_0	❌	Link
  ArgsortParameterizedWithBlockAndBatch.SharedMemoryRequirement/4096_2_1_1	❌	Link
  ArgsortParameterizedWithBlockAndBatch.SharedMemoryRequirement/512_1_0_0	❌	Link
  ArgsortParameterizedWithBlockAndBatch.SharedMemoryRequirement/512_1_1_1	❌	Link
  ArgsortTest.ZeroDimensionalInput	❌	Link
  BlackwellMatmulTest.EpilogueSiluPersistentBroadcastInputs	❌	Link
  BlockSizeAndItemsPerThread/ArgSortComprehensiveTest.ComprehensiveValidation/BlockSize128_ItemsPerThread4	❌	Link
  ClusterReductionTest.SimpleFusionAllReduce/cluster_10_dtype_float	❌	Link
  ClusterReductionTest.SimpleFusionAllReduce/cluster_6_dtype_double	❌	Link
  ClusterReductionTest.SimpleFusionAllReduce/cluster_9_dtype___bfloat	❌	Link
  ... with 179 more test failures omitted. Check internal logs.

• (High, 44) NCCL NVLS multicast memory bind failures across multidevice/nvfuser test suites on dlcluster_viking_ci

  Test Name	H100 (dist.)	Source
  tests.python.multidevice.test_communication.test_allgather	❌
  tests.python.multidevice.test_communication.test_allgather_expanded_broadcast	❌
  tests.python.multidevice.test_communication.test_allreduce	❌
  tests.python.multidevice.test_communication.test_reduce_scatter	❌
  tests.python.multidevice.test_communication.test_reduce_scatter_noncontiguous	❌
  tests.python.multidevice.test_dtensor.test_column_parallel_linear	❌
  tests.python.multidevice.test_dtensor.test_plus_one	❌
  tests.python.multidevice.test_dtensor.test_row_parallel_linear	❌
  tests.python.multidevice.test_expert_parallel.test_dispatch_and_combine	❌
  tests.python.multidevice.test_matmul.test_column_parallel_grouped_mm	❌
  ... with 34 more test failures omitted. Check internal logs.

• (Medium, 1) NCCL invalid usage error in multidevice overlap tests (tests/python/multidevice/test_overlap.py)

  Test Name	H100 (dist.)	Source
  tests.python.multidevice.test_overlap.test_overlap_allgather_matmul_shard_outermost[backend_type=CommunicatorBackend.cuda]	❌

[naoyam]

!test