pytorch

mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00

Files

IvanKobzarev 8ec01f34e9 [bucketing] custom_ops mode to hide inductor copies overhead (#161499 )

Adding "_custom_ops" bucketing to temporary fallback to eager execution of for_each,
to workaround too many generated kernels on inductor side.

This PR also reverts parts of bucketing changes for cycles detection that resulted in accuracy problems.

Differential Revision: [D81152293](https://our.internmc.facebook.com/intern/diff/D81152293)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/161499
Approved by: https://github.com/eellison

2025-09-08 20:03:08 +00:00

analysis

Add profiler analysis flag to combine multiple profiles into one (#161145 )

2025-08-21 21:36:58 +00:00

autoheuristic

[BE][Ez]: Optimize unnecessary lambda with operator (#154722 )

2025-05-30 23:47:10 +00:00

codegen

Revert "[inductor] pdl inductor option (disabled by default) (#160928 )"

2025-09-07 07:37:37 +00:00

compile_worker

Allow for using a dedicated binary for the torch subproc pool. (#162093 )

2025-09-05 01:43:46 +00:00

fx_passes

[bucketing] custom_ops mode to hide inductor copies overhead (#161499 )

2025-09-08 20:03:08 +00:00

kernel

Revert "Add return-max-scores to flex-attention (#161667 )"

2025-09-07 06:00:55 +00:00

package

Grab bag of (mostly) typing improvements (#158075 )

2025-07-21 19:17:01 +00:00

runtime

[inductor][triton] support static cuda launcher after triton # 7866 (#162309 )

2025-09-08 07:57:48 +00:00

template_heuristics

[inductor][ez] return choicecallers directly (#161345 )

2025-09-05 18:02:38 +00:00

__autotune_main__.py

Improve subproc autotuning implementation (#149700 )

2025-03-28 01:06:39 +00:00

__init__.py

[aoti-fx] Initial AOTInductor FX (#160765 )

2025-08-18 18:14:08 +00:00

analyze_preserves_zero_mask.py

Revert two recent prologue prs (#151013 )

2025-04-10 23:48:41 +00:00

aoti_eager.py

PEP585 update - torch/_inductor/[_-i]* (#145137 )

2025-01-19 01:22:47 +00:00

async_compile.py

[inductor][triton] more JITCallable._hash_lock support (#162244 )

2025-09-08 07:57:48 +00:00

autotune_process.py

Add cutedsl template support to compile (#160108 )

2025-08-18 04:37:15 +00:00

await_utils.py

integrate kernacle into inductor (#160121 )

2025-08-08 02:14:44 +00:00

bounds.py

[inductor] Refactor op handlers part 5 (#146257 )

2025-02-08 18:00:30 +00:00

choices.py

Revert "[inductor] add kernel template choice (ktc) (#161347 )"

2025-09-07 20:39:39 +00:00

codecache.py

allow user to pass in custom partitioner function (#157580 )

2025-09-05 22:49:39 +00:00

comm_analysis.py

[inductor] Runtime estimations: use nccl estimator; mm only benchmark mode (#161405 )

2025-09-08 14:33:19 +00:00

comm_lowering.py

[AOTI] Fix memory leak from all_reduce (#159818 )

2025-08-06 18:11:14 +00:00

comms_debug.py

[inductor] Estimate peak memory allocfree and applying to reordering collectives (#160113 )

2025-08-22 14:19:57 +00:00

comms.py

[inductor] Runtime estimations: use nccl estimator; mm only benchmark mode (#161405 )

2025-09-08 14:33:19 +00:00

compile_fx_async.py

[pc] introduce ProgressiveCompilationState and clear callback (#157619 )

2025-07-05 07:55:11 +00:00

compile_fx_ext.py

Migrate from lru_cache to cache (#155613 )

2025-06-11 19:44:18 +00:00

compile_fx_subproc.py

Rename inductor cache (#156128 )

2025-06-17 03:57:18 +00:00

compile_fx.py

allow user to pass in custom partitioner function (#157580 )

2025-09-05 22:49:39 +00:00

compiler_bisector.py

Migrate from lru_cache to cache (#155613 )

2025-06-11 19:44:18 +00:00

config_comms.py

[inductor] Runtime estimations: use nccl estimator; mm only benchmark mode (#161405 )

2025-09-08 14:33:19 +00:00

config.py

[inductor] Runtime estimations: use nccl estimator; mm only benchmark mode (#161405 )

2025-09-08 14:33:19 +00:00

constant_folding.py

Add dont constant fold flag (#154945 )

2025-06-10 14:52:26 +00:00

cpp_builder.py

Revert "[inductor] Windows inductor use intel-openmp. (#160258 )"

2025-08-25 12:57:47 +00:00

cpu_vec_isa.py

[CPU][Inductor] Improve performance of A16W8 GEMM template (#161148 )

2025-08-31 09:56:29 +00:00

cudagraph_trees.py

Fix types in graphs.py (#158192 )

2025-07-15 19:49:38 +00:00

cudagraph_utils.py

[CUDAGraph] add config to error on skipping cudagraph (#161862 )

2025-09-04 15:52:39 +00:00

custom_graph_pass.py

allow user to pass in custom partitioner function (#157580 )

2025-09-05 22:49:39 +00:00

debug.py

[reland] Add inductor provenance mapping for cpp extern kernel (#161656 ) (#162069 )

2025-09-04 04:18:43 +00:00

decomposition.py

[Inductor] Add DeviceAssert op to enable device-side assertion in torch.compile (#160677 )

2025-08-28 18:57:34 +00:00

dependencies.py

DDE-Free select with unbacked index. (#157605 )

2025-07-24 20:08:05 +00:00

dtype_propagation.py

[Inductor] Add DeviceAssert op to enable device-side assertion in torch.compile (#160677 )

2025-08-28 18:57:34 +00:00

exc.py

[BE] Save attributes for CppCompileError for pickleing (#160294 )

2025-08-12 22:03:36 +00:00

extern_node_serializer.py

[BE][AOTI] Remove duplicate schema for ExternKernelNode (#155867 )

2025-06-14 02:03:27 +00:00

freezing_utils.py

PEP585: More UP006 fixes (#146392 )

2025-02-20 06:18:13 +00:00

freezing.py

mypy 1.16.0 (#155821 )

2025-06-14 18:18:43 +00:00

fuzzer.py

allow user to pass in custom partitioner function (#157580 )

2025-09-05 22:49:39 +00:00

fx_utils.py

Inductor logging + analysis of torch.profile (#149697 )

2025-07-07 22:13:34 +00:00

graph.py

[invoke_subgraph][inductor] Thread graphsafe rng input states for hops (#160713 )

2025-08-21 20:41:29 +00:00

hooks.py

PEP585 update - torch/_inductor/[_-i]* (#145137 )

2025-01-19 01:22:47 +00:00

index_propagation.py

[inductor] Follow integer overflow rules in TypedExpr (#161922 )

2025-09-03 18:33:18 +00:00

inductor_prims.py

[inductor] lowering for fractional_max_pool3d (#148630 )

2025-05-22 16:06:29 +00:00

ir.py

[while_loop][autograd] add hop while_loop_stack_output (#160467 )

2025-09-06 21:26:33 +00:00

jagged_lowerings.py

Fix XPU CI UT test_circular_dependencies (#158189 )

2025-07-13 09:30:57 +00:00

kernel_inputs.py

[inductor] move scaled_mm template args into heuristics (#161126 )

2025-09-03 01:03:57 +00:00

loop_body.py

[Inductor] Improve RoPE (#161420 )

2025-09-05 20:55:20 +00:00

lowering.py

[while_loop][autograd] add hop while_loop_stack_output (#160467 )

2025-09-06 21:26:33 +00:00

memory.py

[inductor] Estimate peak memory allocfree and applying to reordering collectives (#160113 )

2025-08-22 14:19:57 +00:00

metrics.py

Replace runtime type parameterization (#155221 )

2025-06-05 21:43:54 +00:00

mkldnn_ir.py

[AOTI][Intel GPU] Add XPU quantization ops to AOT Inductor. (#156572 )

2025-08-29 09:19:44 +00:00

mkldnn_lowerings.py

[Inductor][Float8] Add float8_e4m3fn into assertion dtype list. (#157684 )

2025-07-15 06:02:01 +00:00

mock_cache.py

PEP585 update - torch/_inductor (#145198 )

2025-01-21 21:04:33 +00:00

ops_handler.py

[Inductor] Add DeviceAssert op to enable device-side assertion in torch.compile (#160677 )

2025-08-28 18:57:34 +00:00

optimize_indexing.py

PEP585 update - torch/_inductor (#145198 )

2025-01-21 21:04:33 +00:00

output_code.py

[Graph Partition] interface for custom cg wrapper (#162207 )

2025-09-06 03:13:01 +00:00

pattern_matcher.py

Make pattern matcher resilient to ddes (#161843 )

2025-09-02 05:16:13 +00:00

quantized_lowerings.py

[BE][1/X] Phase out usage of use_max_autotune() (#155847 )

2025-06-14 03:16:20 +00:00

remote_cache.py

pt2_remote_cache: Log sample for failures, and log the explicit reason we're faling. (#156874 )

2025-07-18 20:28:27 +00:00

remote_gemm_autotune_cache.py

integrate kernacle into inductor (#160121 )

2025-08-08 02:14:44 +00:00

scheduler.py

[inductor] fuse for scalar shared data (#162311 )

2025-09-08 17:20:46 +00:00

script.ld

Place .lrodata later in the binary (#117575 )

2024-01-18 17:58:18 +00:00

select_algorithm.py

[inductor][ez] return choicecallers directly (#161345 )

2025-09-05 18:02:38 +00:00

shape_propagation.py

[Inductor] Add DeviceAssert op to enable device-side assertion in torch.compile (#160677 )

2025-08-28 18:57:34 +00:00

sizevars.py

multi-kernel matmuls based on varying hint sizes (#156628 )

2025-07-12 15:08:21 +00:00

standalone_compile.py

[inductor] fix CompiledArtifact.load path on Windows. (#160268 )

2025-08-10 14:22:52 +00:00

subgraph_lowering.py

[FlexAttention] fixing learnable bias assertion error in inductor (#161170 )

2025-08-23 06:24:22 +00:00

test_case.py

Rename inductor cache (#156128 )

2025-06-17 03:57:18 +00:00

test_operators.py

[BE] remove torch deploy - conditionals (#158288 )

2025-07-29 17:40:49 +00:00

tiling_utils.py

[BE][3/16] fix typos in torch/ (torch/_inductor/) (#156313 )

2025-06-23 02:57:12 +00:00

triton_bundler.py

[inductor][triton] more JITCallable._hash_lock support (#162244 )

2025-09-08 07:57:48 +00:00

utils.py

[inductor] Runtime estimations: use nccl estimator; mm only benchmark mode (#161405 )

2025-09-08 14:33:19 +00:00

virtualized.py

[export] Fix custom ops in subgraphs (#160004 )

2025-08-18 15:42:19 +00:00

wrapper_benchmark.py

[inductor] Make times and repeat parameters command line args (#158590 )

2025-07-18 20:07:55 +00:00