Failure on Interpreter: dynamic_dimension_inference.cc:1962 operand_shape.IsTuple() Only tuple typed inputs can have dynamic dimension…

[cxxhou]

Running the attached StableHLO module aborts run_hlo_module on Interpreter through an internal XLA check. Failure signature: dynamic_dimension_inference.cc:1962 operand_shape.IsTuple() Only tuple typed inputs can have dynamic dimension…. The same module runs successfully on CPU.

The following backends fail without a crash signal:

• run_hlo_module on Interpreter (HLO proto input) — failed (rc=255)

The same module runs successfully on:

• run_hlo_module on CPU (StableHLO input)
• run_hlo_module on GPU (StableHLO input)
• run_hlo_module on CPU (HLO proto input)
• run_hlo_module on GPU (HLO proto input)

The module also compiles cleanly on:

• hlo_runner_main on CPU (HLO proto input, compile-only via --run=false)
• hlo_runner_main on GPU (HLO proto input, compile-only via --run=false)

The StableHLO→HLO-proto conversion:

• hlo-translate — succeeds

Command:

hlo-translate \
  --mlir-to-hlo \
  --emit-proto \
  4385dedf54d6caf5.mlir \
  -o \
  4385dedf54d6caf5.hlo
sed -i '1{/^goo\.gle\//d}' 4385dedf54d6caf5.hlo

Environment

• CPU: Intel(R) Core(TM) i9-14900HX
• GPU: NVIDIA GeForce RTX 4060 Laptop GPU
• CUDA Driver: 580.126.09
• XLA commit: 6f1f2d916dcef0428bfa7ea66765be539126c4f2
• StableHLO commit: e6f81ebd06b3509f2c7fa6175430aadbd4d724ca

IR

module {
  func.func @main(%arg0: tensor<5xf32>, %arg1: tensor<5xf32>, %arg2: tensor<5xf32>, %arg3: tensor<3x2xi32>) -> (tensor<5xf32>, tensor<2xi32>) {
    %0 = stablehlo.add %arg0, %arg1 : tensor<5xf32>
    %1 = stablehlo.multiply %0, %arg2 : tensor<5xf32>
    %2 = stablehlo.subtract %1, %0 : tensor<5xf32>
    %3 = stablehlo.multiply %2, %1 : tensor<5xf32>
    %4 = call @wrapped_157(%arg3) : (tensor<3x2xi32>) -> tensor<2xi32>
    return %3, %4 : tensor<5xf32>, tensor<2xi32>
  }
  func.func private @wrapped_157(%arg0: tensor<3x2xi32>) -> tensor<2xi32> {
    %c = stablehlo.constant dense<false> : tensor<i1>
    %c_0 = stablehlo.constant dense<2> : tensor<i32>
    %0 = stablehlo.set_dimension_size %arg0, %c_0, dim = 0 : (tensor<3x2xi32>, tensor<i32>) -> tensor<?x2xi32, #stablehlo.bounds<3, ?>>
    %1:2 = "stablehlo.if"(%c) ({
      %3 = stablehlo.add %0, %0 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>
      stablehlo.return %3, %3 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>
    }, {
      %3 = stablehlo.add %0, %0 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>
      stablehlo.return %3, %3 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>
    }) : (tensor<i1>) -> (tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>)
    %c_1 = stablehlo.constant dense<0> : tensor<i32>
    %2 = stablehlo.reduce(%1#1 init: %c_1) applies stablehlo.add across dimensions = [0] : (tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<i32>) -> tensor<2xi32>
    return %2 : tensor<2xi32>
  }
  func.func private @true_branch_157(%arg0: tensor<?x2xi32, #stablehlo.bounds<3, ?>>) -> (tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>) {
    %0 = stablehlo.add %arg0, %arg0 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>
    return %0, %0 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>
  }
  func.func private @false_branch_157(%arg0: tensor<?x2xi32, #stablehlo.bounds<3, ?>>) -> (tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>) {
    %0 = stablehlo.add %arg0, %arg0 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>
    return %0, %0 : tensor<?x2xi32, #stablehlo.bounds<3, ?>>, tensor<?x2xi32, #stablehlo.bounds<3, ?>>
  }
  func.func private @update_s32_157(%arg0: tensor<i32>, %arg1: tensor<i32>) -> tensor<i32> {
    %0 = stablehlo.add %arg0, %arg1 : tensor<i32>
    return %0 : tensor<i32>
  }
}

Reproduction (Interpreter (StableHLO input) — failed (rc=255))

Command:

export XLA_FLAGS='--xla_backend_extra_options=xla_cpu_use_multi_output_fusion --xla_enable_enzyme_comms_opt=true --xla_gpu_async_dot=true --xla_gpu_enable_cudnn_layer_norm=true --xla_gpu_enable_dus_accumulator_zero_init_elimination=true --xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_scatter_determinism_expander=true --xla_gpu_experimental_stream_annotation=true --xla_gpu_experimental_use_ragged_dot_fusion --xla_gpu_gemm_rewrite_size_threshold=0 --xla_gpu_multi_streamed_windowed_einsum=true --xla_recognize_reduction_optimization_level=1 --xla_reduce_window_rewrite_base_length=1'
run_hlo_module \
  --platform=Interpreter \
  --reference_platform= \
  --input_format=stablehlo \
  4385dedf54d6caf5.mlir

Output:

 ** Running 4385dedf54d6caf5.mlir**
Running HLO module with runner HloRunner...
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1786848417.278327 3828318 status_macros.cc:58] INTERNAL: RET_CHECK failure (xla/service/dynamic_dimension_inference.cc:1962) operand_shape.IsTuple() Only tuple typed inputs can have dynamic dimension. Please file a bug against XLA team.
*** Begin stack trace ***


























	__libc_start_main

*** End stack trace ***

... compiled and ran in 0.000665344s.
INTERNAL: RET_CHECK failure (xla/service/dynamic_dimension_inference.cc:1962) operand_shape.IsTuple() Only tuple typed inputs can have dynamic dimension. Please file a bug against XLA team.


Suspected Python Code Location:
  -:14
	Failed to execute on HloRunner [xla::PythonStackTrace='']

Reference run (CPU (StableHLO input) — succeeds)

Command:

export XLA_FLAGS='--xla_backend_extra_options=xla_cpu_use_multi_output_fusion --xla_enable_enzyme_comms_opt=true --xla_gpu_async_dot=true --xla_gpu_enable_cudnn_layer_norm=true --xla_gpu_enable_dus_accumulator_zero_init_elimination=true --xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_scatter_determinism_expander=true --xla_gpu_experimental_stream_annotation=true --xla_gpu_experimental_use_ragged_dot_fusion --xla_gpu_gemm_rewrite_size_threshold=0 --xla_gpu_multi_streamed_windowed_einsum=true --xla_recognize_reduction_optimization_level=1 --xla_reduce_window_rewrite_base_length=1'
run_hlo_module \
  --platform=CPU \
  --reference_platform= \
  --input_format=stablehlo \
  4385dedf54d6caf5.mlir

Output:

 ** Running 4385dedf54d6caf5.mlir**
Running HLO module with runner HloRunner...
... compiled and ran in 0.0283194s.
Skipping reference runner

Contact

• Email: ch395@njit.edu, zhihao.yao@njit.edu, benquike@gmail.com