Crash on CPU: no diagnostic output or terminate called after throwing an instance of 'std::length_error'
bugCPU
Running the attached StableHLO module crashes `run_hlo_module` on the CPU platform (SIGSEGV). Crash signature: no diagnostic output or `terminate called after throwing an instance of 'std::length_error'`.
The crash also reproduces on:
- `run_hlo_module` on GPU (StableHLO input) — failed (rc=-11, SIGSEGV)
- `run_hlo_module` on CPU (HLO proto input) — failed (rc=-11, SIGSEGV)
- `run_hlo_module` on GPU (HLO proto input) — failed (rc=-11, SIGSEGV)
- `hlo_runner_main` on CPU (HLO proto input, compile-only via `--run=false`) — crashed
- `hlo_runner_main` on GPU (HLO proto input, compile-only via `--run=false`) — failed (rc=-11, SIGSEGV)
The following backends do not terminate within the replay timeout of 30s:
- `run_hlo_module` on Interpreter (StableHLO input)
- `run_hlo_module` on Interpreter (HLO proto input)
The StableHLO→HLO-proto conversion:
- `hlo-translate` — succeeds
**Command:**
```
hlo-translate \
--mlir-to-hlo \
--emit-proto \
992013097850b5bf.mlir \
-o \
992013097850b5bf.hlo
sed -i '1{/^goo\.gle\//d}' 992013097850b5bf.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
```mlir
module @IrToHlo.54 attributes {mhlo.cross_program_prefetches = [], mhlo.input_output_alias = [], mhlo.is_dynamic = false, mhlo.use_auto_spmd_partitioning = false} {
func.func @main(%arg0: tensor<f64>, %arg1: tensor<1xf64>, %arg2: tensor<1xf64>, %arg3: tensor<46592xf32>) -> (tensor<f64>, tensor<1x1xf64>, tensor<i64>, tensor<f64>, tensor<1xf64>, tensor<46592xf32>, tensor<46592xf32>) {
%c = stablehlo.constant dense<0> : tensor<1xi32>
%c_0 = stablehlo.constant dense<0> : tensor<i32>
%cst = stablehlo.constant dense<0xFFF0000000000000> : tensor<f64>
%cst_1 = stablehlo.constant dense<0.000000e+00> : tensor<f64>
%0 = stablehlo.maximum %arg2, %arg1 : tensor<1xf64>
%1 = stablehlo.logistic %0 : tensor<1xf64>
%2 = stablehlo.tan %arg0 : tensor<f64>
%3 = stablehlo.reshape %2 : (tensor<f64>) -> tensor<1xf64>
%4 = stablehlo.maximum %1, %3 : tensor<1xf64>
%5 = stablehlo.reduce(%4 init: %cst_1) applies stablehlo.add across dimensions = [0] : (tensor<1xf64>, tensor<f64>) -> tensor<f64>
%6 = stablehlo.add %4, %4 : tensor<1xf64>
%7 = stablehlo.reshape %6 : (tensor<1xf64>) -> tensor<1x1xf64>
%8:2 = stablehlo.reduce(%6 init: %cst), (%c init: %c_0) across dimensions = [0] : (tensor<1xf64>, tensor<1xi32>, tensor<f64>, tensor<i32>) -> (tensor<f64>, tensor<i32>)
reducer(%arg4: tensor<f64>, %arg6: tensor<f64>) (%arg5: tensor<i32>, %arg7: tensor<i32>) {
%14 = stablehlo.compare GE, %arg4, %arg6 : (tensor<f64>, tensor<f64>) -> tensor<i1>
%15 = stablehlo.select %14, %arg4, %arg6 : tensor<i1>, tensor<f64>
%16 = stablehlo.compare EQ, %arg4, %arg6 : (tensor<f64>, tensor<f64>) -> tensor<i1>
%17 = stablehlo.minimum %arg5, %arg7 : tensor<i32>
%18 = stablehlo.select %14, %arg5, %arg7 : tensor<i1>, tensor<i32>
%19 = stablehlo.select %16, %17, %18 : tensor<i1>, tensor<i32>
stablehlo.return %15, %19 : tensor<f64>, tensor<i32>
}
%9 = stablehlo.convert %8#1 : (tensor<i32>) -> tensor<i64>
%10 = stablehlo.reshape %6 : (tensor<1xf64>) -> tensor<f64>
%11 = stablehlo.tan %6 : tensor<1xf64>
%cst_2 = stablehlo.constant dense<0xFF800000> : tensor<f32>
%cst_3 = stablehlo.constant dense<0x7F800000> : tensor<f32>
%12 = stablehlo.convert %5 : (tensor<f64>) -> tensor<f32>
%13:2 = "stablehlo.reduce_window"(%arg3, %arg3, %12, %cst_3) <{base_dilations = array<i64: 1>, padding = dense<[[46591, 0]]> : tensor<1x2xi64>, window_dilations = array<i64: 1>, window_dimensions = array<i64: 46592>, window_strides = array<i64: 1>}> ({
^bb0(%arg4: tensor<f32>, %arg5: tensor<f32>, %arg6: tensor<f32>, %arg7: tensor<f32>):
%14 = stablehlo.maximum %arg4, %arg6 : tensor<f32>
%15 = stablehlo.minimum %arg5, %arg7 : tensor<f32>
stablehlo.return %14, %15 : tensor<f32>, tensor<f32>
}) : (tensor<46592xf32>, tensor<46592xf32>, tensor<f32>, tensor<f32>) -> (tensor<46592xf32>, tensor<46592xf32>)
return %5, %7, %9, %10, %11, %13#0, %13#1 : tensor<f64>, tensor<1x1xf64>, tensor<i64>, tensor<f64>, tensor<1xf64>, tensor<46592xf32>, tensor<46592xf32>
}
func.func private @MaxMin_1285(%arg0: tensor<f32>, %arg1: tensor<f32>, %arg2: tensor<f32>, %arg3: tensor<f32>) -> (tensor<f32>, tensor<f32>) {
%0 = stablehlo.maximum %arg0, %arg2 : tensor<f32>
%1 = stablehlo.minimum %arg1, %arg3 : tensor<f32>
return %0, %1 : tensor<f32>, tensor<f32>
}
}
```
## Reproduction (CPU (StableHLO input) — crashed)
**Command:**
```bash
export TF_CPP_VMODULE=hlo_pass_pipeline=3
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 \
992013097850b5bf.mlir
```
**Output:**
```
** Running 992013097850b5bf.mlir**
Running HLO module with runner HloRunner...
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1786848746.938112 3845513 hlo_pass_pipeline.cc:303] Running HLO pass pipeline on module IrToHlo.54: async-collective
I0000 00:00:1786848746.938141 3845513 hlo_pass_pipeline.cc:181] HLO pass async-collective-replacer
I0000 00:00:1786848746.938155 3845513 hlo_pass_pipeline.cc:185] Module hash 11011067031028060739
I0000 00:00:1786848746.938167 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 57
I0000 00:00:1786848746.938175 3845513 hlo_pass_pipeline.cc:176] HLO pass: async-collective-replacer time: 34 us (34 us) (cumulative: 34 us, max: 34 us, #called: 1)
I0000 00:00:1786848746.938181 3845513 hlo_pass_pipeline.cc:303] Running HLO pass pipeline on module IrToHlo.54: pre-spmd-partitioner
I0000 00:00:1786848746.938185 3845513 hlo_pass_pipeline.cc:181] HLO pass strip-memory-placement-annotations
I0000 00:00:1786848746.938191 3845513 hlo_pass_pipeline.cc:185] Module hash 11011067031028060739
I0000 00:00:1786848746.938192 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 57
I0000 00:00:1786848746.938195 3845513 hlo_pass_pipeline.cc:176] HLO pass: strip-memory-placement-annotations time: 10 us (10 us) (cumulative: 44 us, max: 34 us, #called: 2)
I0000 00:00:1786848746.938199 3845513 hlo_pass_pipeline.cc:303] Running HLO pass pipeline on module IrToHlo.54: sharding-removal
I0000 00:00:1786848746.938203 3845513 hlo_pass_pipeline.cc:86] Invariant checker hlo-verifier
I0000 00:00:1786848746.938255 3845513 hlo_pass_pipeline.cc:89] Invariant checker done hlo-verifier
I0000 00:00:1786848746.938257 3845513 hlo_pass_pipeline.cc:181] HLO pass flatten-call-graph
I0000 00:00:1786848746.938263 3845513 hlo_pass_pipeline.cc:185] Module hash 11011067031028060739
I0000 00:00:1786848746.938265 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 57
I0000 00:00:1786848746.938267 3845513 hlo_pass_pipeline.cc:176] HLO pass: flatten-call-graph time: 10 us (10 us) (cumulative: 54 us, max: 34 us, #called: 3)
I0000 00:00:1786848746.938269 3845513 hlo_pass_pipeline.cc:181] HLO pass sharding-remover
I0000 00:00:1786848746.938275 3845513 hlo_pass_pipeline.cc:185] Module hash 11011067031028060739
I0000 00:00:1786848746.938276 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 57
I0000 00:00:1786848746.938280 3845513 hlo_pass_pipeline.cc:176] HLO pass: sharding-remover time: 11 us (11 us) (cumulative: 65 us, max: 34 us, #called: 4)
I0000 00:00:1786848746.938282 3845513 hlo_pass_pipeline.cc:181] HLO pass control-dep-rewriter
I0000 00:00:1786848746.938288 3845513 hlo_pass_pipeline.cc:185] Module hash 11011067031028060739
I0000 00:00:1786848746.938289 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 57
I0000 00:00:1786848746.938291 3845513 hlo_pass_pipeline.cc:176] HLO pass: control-dep-rewriter time: 9 us (9 us) (cumulative: 74 us, max: 34 us, #called: 5)
I0000 00:00:1786848746.938293 3845513 hlo_pass_pipeline.cc:181] HLO pass dce
I0000 00:00:1786848746.938298 3845513 hlo_pass_pipeline.cc:185] Module hash 11011067031028060739
I0000 00:00:1786848746.938299 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 57
I0000 00:00:1786848746.938313 3845513 hlo_pass_pipeline.cc:213] Pass caused changes dce
I0000 00:00:1786848746.938314 3845513 hlo_pass_pipeline.cc:86] Invariant checker hlo-verifier
I0000 00:00:1786848746.938344 3845513 hlo_pass_pipeline.cc:89] Invariant checker done hlo-verifier
I0000 00:00:1786848746.938346 3845513 hlo_pass_pipeline.cc:176] HLO pass: dce time: 53 us (53 us) (cumulative: 127 us, max: 53 us, #called: 6)
I0000 00:00:1786848746.938349 3845513 hlo_pass_pipeline.cc:303] Running HLO pass pipeline on module IrToHlo.54: SubbytePacker pipeline
I0000 00:00:1786848746.938353 3845513 hlo_pass_pipeline.cc:181] HLO pass sub-byte-size-setter
I0000 00:00:1786848746.938358 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938360 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938367 3845513 hlo_pass_pipeline.cc:176] HLO pass: sub-byte-size-setter time: 14 us (14 us) (cumulative: 141 us, max: 53 us, #called: 7)
I0000 00:00:1786848746.938388 3845513 hlo_pass_pipeline.cc:303] Running HLO pass pipeline on module IrToHlo.54: HLO passes through layout assignment
I0000 00:00:1786848746.938397 3845513 hlo_pass_pipeline.cc:86] Invariant checker hlo-verifier
I0000 00:00:1786848746.938426 3845513 hlo_pass_pipeline.cc:89] Invariant checker done hlo-verifier
I0000 00:00:1786848746.938428 3845513 hlo_pass_pipeline.cc:181] HLO pass batched_gather_scatter_normalizer
I0000 00:00:1786848746.938434 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938435 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938438 3845513 hlo_pass_pipeline.cc:176] HLO pass: batched_gather_scatter_normalizer time: 10 us (10 us) (cumulative: 151 us, max: 53 us, #called: 8)
I0000 00:00:1786848746.938440 3845513 hlo_pass_pipeline.cc:181] HLO pass result_caster
I0000 00:00:1786848746.938445 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938446 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938449 3845513 hlo_pass_pipeline.cc:176] HLO pass: result_caster time: 9 us (9 us) (cumulative: 160 us, max: 53 us, #called: 9)
I0000 00:00:1786848746.938451 3845513 hlo_pass_pipeline.cc:181] HLO pass dot_decomposer
I0000 00:00:1786848746.938456 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938457 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938459 3845513 hlo_pass_pipeline.cc:176] HLO pass: dot_decomposer time: 8 us (8 us) (cumulative: 168 us, max: 53 us, #called: 10)
I0000 00:00:1786848746.938460 3845513 hlo_pass_pipeline.cc:181] HLO pass operand_upcaster
I0000 00:00:1786848746.938466 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938467 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938470 3845513 hlo_pass_pipeline.cc:176] HLO pass: operand_upcaster time: 10 us (10 us) (cumulative: 178 us, max: 53 us, #called: 11)
I0000 00:00:1786848746.938473 3845513 hlo_pass_pipeline.cc:181] HLO pass rng-expander
I0000 00:00:1786848746.938477 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938478 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938481 3845513 hlo_pass_pipeline.cc:176] HLO pass: rng-expander time: 8 us (8 us) (cumulative: 186 us, max: 53 us, #called: 12)
I0000 00:00:1786848746.938483 3845513 hlo_pass_pipeline.cc:181] HLO pass rng-bit-generator-expander
I0000 00:00:1786848746.938487 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938489 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938491 3845513 hlo_pass_pipeline.cc:176] HLO pass: rng-bit-generator-expander time: 8 us (8 us) (cumulative: 194 us, max: 53 us, #called: 13)
I0000 00:00:1786848746.938493 3845513 hlo_pass_pipeline.cc:181] HLO pass zero_sized_hlo_elimination
I0000 00:00:1786848746.938498 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938499 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938502 3845513 hlo_pass_pipeline.cc:176] HLO pass: zero_sized_hlo_elimination time: 9 us (9 us) (cumulative: 203 us, max: 53 us, #called: 14)
I0000 00:00:1786848746.938504 3845513 hlo_pass_pipeline.cc:181] HLO pass dynamic-index-splitter
I0000 00:00:1786848746.938508 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938510 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938513 3845513 hlo_pass_pipeline.cc:176] HLO pass: dynamic-index-splitter time: 8 us (8 us) (cumulative: 211 us, max: 53 us, #called: 15)
I0000 00:00:1786848746.938515 3845513 hlo_pass_pipeline.cc:181] HLO pass conditional-to-select
I0000 00:00:1786848746.938519 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938520 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938528 3845513 hlo_pass_pipeline.cc:176] HLO pass: conditional-to-select time: 13 us (13 us) (cumulative: 224 us, max: 53 us, #called: 16)
I0000 00:00:1786848746.938530 3845513 hlo_pass_pipeline.cc:181] HLO pass map-inline
I0000 00:00:1786848746.938535 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938536 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938543 3845513 hlo_pass_pipeline.cc:176] HLO pass: map-inline time: 13 us (13 us) (cumulative: 237 us, max: 53 us, #called: 17)
I0000 00:00:1786848746.938545 3845513 hlo_pass_pipeline.cc:181] HLO pass topk-decomposer
I0000 00:00:1786848746.938549 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938551 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938556 3845513 hlo_pass_pipeline.cc:176] HLO pass: topk-decomposer time: 11 us (11 us) (cumulative: 248 us, max: 53 us, #called: 18)
I0000 00:00:1786848746.938558 3845513 hlo_pass_pipeline.cc:181] HLO pass comparison-expander
I0000 00:00:1786848746.938563 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938564 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938566 3845513 hlo_pass_pipeline.cc:176] HLO pass: comparison-expander time: 8 us (8 us) (cumulative: 256 us, max: 53 us, #called: 19)
I0000 00:00:1786848746.938569 3845513 hlo_pass_pipeline.cc:181] HLO pass cholesky_expander
I0000 00:00:1786848746.938573 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938575 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938577 3845513 hlo_pass_pipeline.cc:176] HLO pass: cholesky_expander time: 8 us (8 us) (cumulative: 264 us, max: 53 us, #called: 20)
I0000 00:00:1786848746.938580 3845513 hlo_pass_pipeline.cc:181] HLO pass qr_expander
I0000 00:00:1786848746.938584 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938585 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938588 3845513 hlo_pass_pipeline.cc:176] HLO pass: qr_expander time: 8 us (8 us) (cumulative: 272 us, max: 53 us, #called: 21)
I0000 00:00:1786848746.938590 3845513 hlo_pass_pipeline.cc:181] HLO pass eigh_expander
I0000 00:00:1786848746.938594 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938596 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938598 3845513 hlo_pass_pipeline.cc:176] HLO pass: eigh_expander time: 8 us (8 us) (cumulative: 280 us, max: 53 us, #called: 22)
I0000 00:00:1786848746.938600 3845513 hlo_pass_pipeline.cc:181] HLO pass triangular_solve_expander
I0000 00:00:1786848746.938604 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938606 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938608 3845513 hlo_pass_pipeline.cc:176] HLO pass: triangular_solve_expander time: 8 us (8 us) (cumulative: 288 us, max: 53 us, #called: 23)
I0000 00:00:1786848746.938610 3845513 hlo_pass_pipeline.cc:181] HLO pass all_to_all_decomposer
I0000 00:00:1786848746.938614 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938616 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938618 3845513 hlo_pass_pipeline.cc:176] HLO pass: all_to_all_decomposer time: 8 us (8 us) (cumulative: 296 us, max: 53 us, #called: 24)
I0000 00:00:1786848746.938620 3845513 hlo_pass_pipeline.cc:181] HLO pass stochastic_convert_decomposer
I0000 00:00:1786848746.938624 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938625 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938628 3845513 hlo_pass_pipeline.cc:176] HLO pass: stochastic_convert_decomposer time: 8 us (8 us) (cumulative: 304 us, max: 53 us, #called: 25)
I0000 00:00:1786848746.938630 3845513 hlo_pass_pipeline.cc:181] HLO pass call-inliner
I0000 00:00:1786848746.938635 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938636 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938642 3845513 hlo_pass_pipeline.cc:176] HLO pass: call-inliner time: 12 us (12 us) (cumulative: 316 us, max: 53 us, #called: 26)
I0000 00:00:1786848746.938644 3845513 hlo_pass_pipeline.cc:181] HLO pass batch-dot-simplification
I0000 00:00:1786848746.938649 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938650 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938652 3845513 hlo_pass_pipeline.cc:176] HLO pass: batch-dot-simplification time: 7 us (7 us) (cumulative: 323 us, max: 53 us, #called: 27)
I0000 00:00:1786848746.938653 3845513 hlo_pass_pipeline.cc:181] HLO pass dot_decomposer
I0000 00:00:1786848746.938658 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938659 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938660 3845513 hlo_pass_pipeline.cc:176] HLO pass: dot_decomposer time: 7 us (7 us) (cumulative: 330 us, max: 53 us, #called: 28)
I0000 00:00:1786848746.938662 3845513 hlo_pass_pipeline.cc:181] HLO pass all-reduce-promotion
I0000 00:00:1786848746.938667 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938668 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938671 3845513 hlo_pass_pipeline.cc:176] HLO pass: all-reduce-promotion time: 9 us (9 us) (cumulative: 339 us, max: 53 us, #called: 29)
I0000 00:00:1786848746.938673 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-bf16
I0000 00:00:1786848746.938677 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938679 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938692 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-bf16 time: 19 us (19 us) (cumulative: 358 us, max: 53 us, #called: 30)
I0000 00:00:1786848746.938695 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e5m2
I0000 00:00:1786848746.938699 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938701 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938709 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e5m2 time: 14 us (14 us) (cumulative: 372 us, max: 53 us, #called: 31)
I0000 00:00:1786848746.938711 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e4m3
I0000 00:00:1786848746.938716 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938717 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938725 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e4m3 time: 14 us (14 us) (cumulative: 386 us, max: 53 us, #called: 32)
I0000 00:00:1786848746.938727 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e4m3fn
I0000 00:00:1786848746.938731 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938732 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938740 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e4m3fn time: 13 us (13 us) (cumulative: 399 us, max: 53 us, #called: 33)
I0000 00:00:1786848746.938742 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e4m3b11fnuz
I0000 00:00:1786848746.938746 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938747 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938755 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e4m3b11fnuz time: 12 us (12 us) (cumulative: 411 us, max: 53 us, #called: 34)
I0000 00:00:1786848746.938757 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e5m2fnuz
I0000 00:00:1786848746.938761 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938762 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938770 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e5m2fnuz time: 13 us (13 us) (cumulative: 424 us, max: 53 us, #called: 35)
I0000 00:00:1786848746.938772 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e4m3fnuz
I0000 00:00:1786848746.938776 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938777 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938784 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e4m3fnuz time: 12 us (12 us) (cumulative: 436 us, max: 53 us, #called: 36)
I0000 00:00:1786848746.938786 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e3m4
I0000 00:00:1786848746.938791 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938792 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938799 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e3m4 time: 13 us (13 us) (cumulative: 449 us, max: 53 us, #called: 37)
I0000 00:00:1786848746.938801 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-s4
I0000 00:00:1786848746.938806 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938807 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938814 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-s4 time: 13 us (13 us) (cumulative: 462 us, max: 53 us, #called: 38)
I0000 00:00:1786848746.938816 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-u4
I0000 00:00:1786848746.938826 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938827 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938838 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-u4 time: 21 us (21 us) (cumulative: 483 us, max: 53 us, #called: 39)
I0000 00:00:1786848746.938840 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f4e2m1fn
I0000 00:00:1786848746.938844 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938845 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938852 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f4e2m1fn time: 12 us (12 us) (cumulative: 495 us, max: 53 us, #called: 40)
I0000 00:00:1786848746.938854 3845513 hlo_pass_pipeline.cc:181] HLO pass float-normalization-f8e8m0fnu
I0000 00:00:1786848746.938859 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938860 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938868 3845513 hlo_pass_pipeline.cc:176] HLO pass: float-normalization-f8e8m0fnu time: 14 us (14 us) (cumulative: 509 us, max: 53 us, #called: 41)
I0000 00:00:1786848746.938870 3845513 hlo_pass_pipeline.cc:181] HLO pass batch-dot-simplification
I0000 00:00:1786848746.938874 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938875 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938877 3845513 hlo_pass_pipeline.cc:176] HLO pass: batch-dot-simplification time: 7 us (7 us) (cumulative: 516 us, max: 53 us, #called: 42)
I0000 00:00:1786848746.938879 3845513 hlo_pass_pipeline.cc:181] HLO pass convolution-group-converter
I0000 00:00:1786848746.938883 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938884 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938891 3845513 hlo_pass_pipeline.cc:176] HLO pass: convolution-group-converter time: 12 us (12 us) (cumulative: 528 us, max: 53 us, #called: 43)
I0000 00:00:1786848746.938893 3845513 hlo_pass_pipeline.cc:181] HLO pass convolution-group-converter
I0000 00:00:1786848746.938897 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938898 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938903 3845513 hlo_pass_pipeline.cc:176] HLO pass: convolution-group-converter time: 10 us (10 us) (cumulative: 538 us, max: 53 us, #called: 44)
I0000 00:00:1786848746.938905 3845513 hlo_pass_pipeline.cc:181] HLO pass batchnorm_expander
I0000 00:00:1786848746.938909 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938910 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938916 3845513 hlo_pass_pipeline.cc:176] HLO pass: batchnorm_expander time: 11 us (11 us) (cumulative: 549 us, max: 53 us, #called: 45)
I0000 00:00:1786848746.938918 3845513 hlo_pass_pipeline.cc:181] HLO pass logistic-expander
I0000 00:00:1786848746.938922 3845513 hlo_pass_pipeline.cc:185] Module hash 13232447938432050832
I0000 00:00:1786848746.938924 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 48
I0000 00:00:1786848746.938944 3845513 hlo_pass_pipeline.cc:213] Pass caused changes logistic-expander
I0000 00:00:1786848746.938945 3845513 hlo_pass_pipeline.cc:86] Invariant checker hlo-verifier
I0000 00:00:1786848746.938979 3845513 hlo_pass_pipeline.cc:89] Invariant checker done hlo-verifier
I0000 00:00:1786848746.938980 3845513 hlo_pass_pipeline.cc:176] HLO pass: logistic-expander time: 62 us (62 us) (cumulative: 611 us, max: 62 us, #called: 46)
I0000 00:00:1786848746.938982 3845513 hlo_pass_pipeline.cc:181] HLO pass conditional-canonicalizer
I0000 00:00:1786848746.938988 3845513 hlo_pass_pipeline.cc:185] Module hash 7454174032573720876
I0000 00:00:1786848746.938989 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 53
I0000 00:00:1786848746.938992 3845513 hlo_pass_pipeline.cc:176] HLO pass: conditional-canonicalizer time: 10 us (10 us) (cumulative: 621 us, max: 62 us, #called: 47)
I0000 00:00:1786848746.938995 3845513 hlo_pass_pipeline.cc:181] HLO pass dynamic-dimension-simplifier
I0000 00:00:1786848746.938999 3845513 hlo_pass_pipeline.cc:185] Module hash 7454174032573720876
I0000 00:00:1786848746.939001 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 53
I0000 00:00:1786848746.939009 3845513 hlo_pass_pipeline.cc:176] HLO pass: dynamic-dimension-simplifier time: 14 us (14 us) (cumulative: 635 us, max: 62 us, #called: 48)
I0000 00:00:1786848746.939012 3845513 hlo_pass_pipeline.cc:181] HLO pass associative-scan-rewriter
I0000 00:00:1786848746.939016 3845513 hlo_pass_pipeline.cc:185] Module hash 7454174032573720876
I0000 00:00:1786848746.939018 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 53
I0000 00:00:1786848746.939022 3845513 hlo_pass_pipeline.cc:176] HLO pass: associative-scan-rewriter time: 10 us (10 us) (cumulative: 645 us, max: 62 us, #called: 49)
I0000 00:00:1786848746.939024 3845513 hlo_pass_pipeline.cc:181] HLO pass reduce-window-rewriter
I0000 00:00:1786848746.939029 3845513 hlo_pass_pipeline.cc:185] Module hash 7454174032573720876
I0000 00:00:1786848746.939030 3845513 hlo_pass_pipeline.cc:187] Number of instructions: 53
```
## Reference run (hlo-translate — succeeds)
**Command:**
```bash
hlo-translate \
--mlir-to-hlo \
--emit-proto \
992013097850b5bf.mlir \
-o \
992013097850b5bf.hlo
sed -i '1{/^goo\.gle\//d}' 992013097850b5bf.hlo
```
**Output:**
```
(no output)
```
## Contact
- **Email**: ``ch395@njit.edu, zhihao.yao@njit.edu, benquike@gmail.com``
关闭于 2 天前 1 条评论