Here are some key observations to aid the review process:

std::cout <<"reference_tv:" << reference_tv->toString() << std::endl;reference_tv->printTransforms();

at::Tensor t1 = at::randn({512,1024,2}, options);// NOTE force pointwise scheduler here just for testing purpose

// bool use_tma = mayUseTma(prop, runtime_info) &&// isOptionEnabled(EnableOption::TmaPointwise); for CI testing, use tma always// if possible

• (Medium, 90)NVFuser internal assert (Unknown tensor map data type) in test_direct_ops opinfo suite

  Test Name	GB200	H100	Source
  tests.python.direct.test_repro.test_issue1277	❌	❌
  tests.python.opinfo.test_direct_ops.test_correctness_abs_complex128		❌
  tests.python.opinfo.test_direct_ops.test_correctness_abs_complex64	❌	❌
  tests.python.opinfo.test_direct_ops.test_correctness_acos_complex128		❌
  tests.python.opinfo.test_direct_ops.test_correctness_acos_complex64	❌	❌
  tests.python.opinfo.test_direct_ops.test_correctness_acosh_complex128	❌	❌
  tests.python.opinfo.test_direct_ops.test_correctness_acosh_complex64	❌	❌
  tests.python.opinfo.test_direct_ops.test_correctness_add_complex128		❌
  tests.python.opinfo.test_direct_ops.test_correctness_add_complex64		❌
  tests.python.opinfo.test_direct_ops.test_correctness_asin_complex128		❌
  ... with 57 more test failures omitted. Check internal logs.

• (Medium, 46)NVFuser internal assert: Unknown tensor map data type on complex dtype ops (opinfo direct & UnaryTests)

  Test Name	GB200	H100	Source
  UnaryTests/UnaryTest.Neg/std__complex_float_	❌	❌	Link
  tests.python.opinfo.test_direct_ops.test_correctness_abs_complex128	❌
  tests.python.opinfo.test_direct_ops.test_correctness_acos_complex128	❌
  tests.python.opinfo.test_direct_ops.test_correctness_add_complex128	❌
  tests.python.opinfo.test_direct_ops.test_correctness_add_complex64	❌
  tests.python.opinfo.test_direct_ops.test_correctness_asin_complex128	❌
  tests.python.opinfo.test_direct_ops.test_correctness_asin_complex64	❌
  tests.python.opinfo.test_direct_ops.test_correctness_asinh_complex128	❌
  tests.python.opinfo.test_direct_ops.test_correctness_asinh_complex64	❌
  tests.python.opinfo.test_direct_ops.test_correctness_atan_complex128	❌
  ... with 35 more test failures omitted. Check internal logs.

• (Medium, 12)NVFuser internal assertion failures in BlockQuantizationSchedulingTestSuite and MatmulSchedulerTest

  Test Name	GB200	Source
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/__bfloat_1024x1024_WithGlobalScale_NoSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/__bfloat_128x64_NoGlobalScale_WithSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/__bfloat_2048x128_NoGlobalScale_NoSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/__bfloat_2048x128_WithGlobalScale_WithSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/__bfloat_2048x2048_WithGlobalScale_NoSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/float_1024x1024_NoGlobalScale_NoSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/float_1024x1024_WithGlobalScale_WithSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/float_128x64_WithGlobalScale_NoSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/float_2048x128_NoGlobalScale_WithSwizzle	❌	Link
  BlockQuantizationSchedulingTestSuite/BlockQuantizationSchedulingTest.AutoScheduleSingleOp/float_2048x2048_NoGlobalScale_NoSwizzle	❌	Link
  ... with 2 more test failures omitted. Check internal logs.

• (Medium, 9)Multiple NVFuser internal assertion failures across grouped_mm, multidevice matmul/transformer, and thunderfx MoE tests

  Test Name	GB200	GB200 (dist.)	H100	H100 (dist.)	Source
  tests.python.direct.test_with_id_model_indexer.test_layout_op_and_cutlass_nvfp4_grouped_mm[out_dtype=torch.bfloat16-tokens_per_expert_neg_one=[115, 144, 8]-config=[1024, 128, 256]]	❌
  tests.python.multidevice.test_matmul.test_linear_reduce_scatter		❌		❌
  tests.python.multidevice.test_matmul.test_sequence_parallel_linear		❌		❌
  tests.python.multidevice.test_transformer.test_grouped_mlp		❌		❌
  tests.python.test_moe.test_llama4_moe_thunderfx	❌		❌

• (Medium, 8)NVFuser TMA analysis internal asserts (merge-discontiguous / extent divisibility) in PointwiseTest, ResizeTest, matmul_stride, and issue1953 suites

  Test Name	GB200	H100	Source
  PointwiseTest.VIssue1567ectorizationFactorAnalysisCase3	❌	❌	Link
  ResizeTest.PadAndCacheUses	❌	❌	Link
  tests.python.direct.test_matmul.test_matmul_stride	❌	❌
  tests.python.direct.test_repro.test_issue1953	❌	❌

• (Medium, 2)nvFuser internal input-size assert in test_schedule_ops::TestScheduleOps.test_concretize_reshape_pointwise

  Test Name	GB200	H100	Source
  tests.python.test_schedule_ops.TestScheduleOps.test_concretize_reshape_pointwise	❌	❌

• (Medium, 2)nvFuser split-after-parallelization assertion in multidevice transformer tests

  Test Name	GB200	H100	Source
  tests.python.multidevice.test_transformer.test_grouped_mlp	❌	❌

• (Medium, 2)Heuristic string mismatch in test_tutorial_compute_heuristics_and_schedule

  Test Name	GB200	H100	Source
  tests.python.direct.test_tutorial.test_tutorial_compute_heuristics_and_schedule	❌	❌

• (Medium, 1)nvFuser pointwise heuristic unroll factor mismatch in PointwiseTest

  Test Name	GB200	Source
  PointwiseTest.Heuristicst1Compute2Unroll4	❌	Link