[Inductor] Fix Diode / exhaustive autotune crash on AMD#169225
Closed
JChunX wants to merge 1 commit intopytorch:mainfrom
Closed
[Inductor] Fix Diode / exhaustive autotune crash on AMD#169225JChunX wants to merge 1 commit intopytorch:mainfrom
JChunX wants to merge 1 commit intopytorch:mainfrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/169225
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New Failure, 7 Unrelated FailuresAs of commit a137b1a with merge base 6f7dcf5 ( NEW FAILURE - The following job has failed:
UNSTABLE - The following jobs are marked as unstable, possibly due to flakiness on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
JChunX
force-pushed
the
export-D87963125
branch
from
2132616 to
836994e
Compare
JChunX
added a commit
to JChunX/pytorch
that referenced
this pull request
Dec 1, 2025
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
1. matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
2. GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Differential Revision: D87963125
JChunX
force-pushed
the
export-D87963125
branch
from
836994e to
203e751
Compare
JChunX
added a commit
to JChunX/pytorch
that referenced
this pull request
Dec 1, 2025
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
1. matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
2. GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Differential Revision: D87963125
JChunX
force-pushed
the
export-D87963125
branch
from
203e751 to
facd747
Compare
JChunX
added a commit
to JChunX/pytorch
that referenced
this pull request
Dec 2, 2025
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
1. matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
2. GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Differential Revision: D87963125
JChunX
force-pushed
the
export-D87963125
branch
from
facd747 to
62bbc0f
Compare
JChunX
added a commit
to JChunX/pytorch
that referenced
this pull request
Dec 2, 2025
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
1. matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
2. GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Differential Revision: D87963125
coconutruben
approved these changes
Dec 2, 2025
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
1. matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
2. GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Reviewed By: coconutruben
Differential Revision: D87963125
JChunX
force-pushed
the
export-D87963125
branch
from
62bbc0f to
a137b1a
Compare
Contributor
|
@pytorchbot merge (Initiating merge automatically since Phabricator Diff has merged) |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Collaborator
Merge failedReason: 1 jobs have failed, first few of them are: trunk / linux-jammy-cuda12.8-py3.10-gcc11 / test (default, 3, 5, linux.g6.4xlarge.experimental.nvidia.gpu) Details for Dev Infra teamRaised by workflow job |
Contributor
|
@pytorchbot merge -i |
Collaborator
Merge startedYour change will be merged while ignoring the following 8 checks: trunk / linux-jammy-cuda12.8-py3.10-gcc11 / test (default, 3, 5, linux.g6.4xlarge.experimental.nvidia.gpu), trunk / linux-jammy-rocm-py3.10 / test (default, 4, 6, linux.rocm.gpu.gfx942.1), trunk / linux-jammy-rocm-py3.10 / test (default, 5, 6, linux.rocm.gpu.gfx942.1), trunk / linux-jammy-rocm-py3.10 / test (default, 3, 6, linux.rocm.gpu.gfx942.1), trunk / linux-jammy-rocm-py3.10 / test (default, 6, 6, linux.rocm.gpu.gfx942.1), trunk / linux-jammy-rocm-py3.10 / test (default, 2, 6, linux.rocm.gpu.gfx942.1), trunk / linux-jammy-rocm-py3.10 / test (distributed, 1, 3, linux.rocm.gpu.gfx942.4), trunk / linux-jammy-rocm-py3.10 / test (default, 1, 6, linux.rocm.gpu.gfx942.1) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
JacobSzwejbka
pushed a commit
that referenced
this pull request
Dec 8, 2025
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
1. matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
2. GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Differential Revision: D87963125
Pull Request resolved: #169225
Approved by: https://github.com/coconutruben
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
Two issues prevent using Diode w/ expanded search space on AMD:
matrix_instr_nonkdim=2 and kpack=2 causes triton compile to fail
GROUP_M=0 crashes AMD GPU (but not NV)
repro: P2057901593
Test Plan:
MODEL=822608598; SNAPSHOT=0
TORCHINDUCTOR_COMPILE_THREADS=1 TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 TORCHINDUCTOR_MAX_AUTOTUNE=1 TORCH_COMPILE_DEBUG=1 HIP_VISIBLE_DEVICES=7 buck2 run -m rocm640 mode/opt-split-dwarf mode/inplace mode/amd-gpu -c fbcode.triton_backend=amd -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=mi300 caffe2/torch/fb/model_transform/fx2trt/packaging:generate_merge_net_file -- --action=generate --max-batch-size=3072 --preset_lowerer='relu_nan_to_num;disable_new_lowering_weights' --input-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/input.predictor.disagg.gpu.merge --output-file=/home/${USER}/models/${MODEL}/${SNAPSHOT}/fp8_amd_output_diode.predictor.disagg.gpu.merge --diode-config="{'top_k': 100, 'expand_search_space': True, 'discard_unpredicted': False}" --lower-backend aot_inductor --add_passes="use_matmul_lce_replace_normal_LCE,use_triton_dot_compress,use_matmul_fuse_lce_replace_first_LCE,use_contiguous_linear_reduction_replace_linear_reduction" --aot_inductor_config="{'max_autotune': True, 'comprehensive_padding': False, 'aot_inductor.use_runtime_constant_folding': True}" --hardware-type GFX942_X86 --node_replacement_dict="{'torch.nn.Linear':{'(3000+, 3000+)':'fp8_float_model_dynamic_quantization_rowwise_triton'}}" 2>&1 | tee ~/logs/lower_diode.log
Differential Revision: D87963125
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben @jataylo