Revert "[PyTorch] Make tls_local_dispatch_key_set inlineable (reapply) (#49412)"

[ezyang]

Stack from ghstack:

• Revert "[PyTorch] Make tls_local_dispatch_key_set inlineable (reapply) (#49412)" #49548 Revert "[PyTorch] Make tls_local_dispatch_key_set inlineable (reapply) ([PyTorch] Make tls_local_dispatch_key_set inlineable (reapply) #49412)"

This reverts commit 6f928a4.

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit c9f51f7 (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

pytorch_linux_backward_compatibility_check_test (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

Dec 17 19:20:17 The PR is introducing backward incompatible changes to the operator library. Please contact PyTorch team to confirm whether this change is wanted or not.

Dec 17 19:20:17 processing existing schema:  __setstate__(__torch__.torch.classes.quantized.EmbeddingPackedParamsBase _0, (int, Tensor[], float[], int[]) _1) -> (None _0)
Dec 17 19:20:17 processing existing schema:  bit_rate(__torch__.torch.classes.quantized.EmbeddingPackedParamsBase _0) -> (int _0)
Dec 17 19:20:17 processing existing schema:  version(__torch__.torch.classes.quantized.EmbeddingPackedParamsBase _0) -> (int _0)
Dec 17 19:20:17 processing existing schema:  __getstate__(__torch__.torch.classes.xnnpack.LinearOpContext _0) -> ((Tensor, Tensor?, Scalar?, Scalar?) _0)
Dec 17 19:20:17 processing existing schema:  __setstate__(__torch__.torch.classes.xnnpack.LinearOpContext _0, (Tensor, Tensor?, Scalar?, Scalar?) _1) -> (None _0)
Dec 17 19:20:17 processing existing schema:  __getstate__(__torch__.torch.classes.xnnpack.Conv2dOpContext _0) -> ((Tensor, Tensor?, int[], int[], int[], int, Scalar?, Scalar?) _0)
Dec 17 19:20:17 processing existing schema:  __setstate__(__torch__.torch.classes.xnnpack.Conv2dOpContext _0, (Tensor, Tensor?, int[], int[], int[], int, Scalar?, Scalar?) _1) -> (None _0)
Dec 17 19:20:17 processing existing schema:  __getstate__(__torch__.torch.classes.xnnpack.TransposeConv2dOpContext _0) -> ((Tensor, Tensor?, int[], int[], int[], int[], int, Scalar?, Scalar?) _0)
Dec 17 19:20:17 processing existing schema:  __setstate__(__torch__.torch.classes.xnnpack.TransposeConv2dOpContext _0, (Tensor, Tensor?, int[], int[], int[], int[], int, Scalar?, Scalar?) _1) -> (None _0)
Dec 17 19:20:17 processing existing schema:  __init__(__torch__.torch.classes.dist_rpc.WorkerInfo _0, str _1, int _2) -> (None _0)
Dec 17 19:20:17 The PR is introducing backward incompatible changes to the operator library. Please contact PyTorch team to confirm whether this change is wanted or not. 
Dec 17 19:20:17 
Dec 17 19:20:17 Broken ops: [
Dec 17 19:20:17 	aten::hash.str(str t) -> (int)
Dec 17 19:20:17 	aten::hash.int(int t) -> (int)
Dec 17 19:20:17 	aten::hash.float(float t) -> (int)
Dec 17 19:20:17 	__caffe2::BoxWithNMSLimit(Tensor scores, Tensor boxes, Tensor batch_splits, float score_thresh, float nms, int detections_per_im, bool soft_nms_enabled, str soft_nms_method, float soft_nms_sigma, float soft_nms_min_score_thres, bool rotated, bool cls_agnostic_bbox_reg, bool input_boxes_include_bg_cls, bool output_classes_include_bg_cls, bool legacy_plus_one, Tensor[]? _caffe2_preallocated_outputs=None) -> (Tensor scores, Tensor boxes, Tensor classes, Tensor batch_splits, Tensor keeps, Tensor keeps_size)
Dec 17 19:20:17 	__caffe2::RoIAlignRotated(Tensor features, Tensor rois, str order, float spatial_scale, int pooled_h, int pooled_w, int sampling_ratio, bool aligned, Tensor[]? _caffe2_preallocated_outputs=None) -> (Tensor)
Dec 17 19:20:17 	aten::choose_qparams_optimized(Tensor input, int numel, int n_bins, float ratio, int bit_width) -> (float, float)
Dec 17 19:20:17 	__caffe2::HeatmapMaxKeypoint(Tensor heatmaps, Tensor bboxes_in, bool should_output_softmax=True, Tensor[]? _caffe2_preallocated_outputs=None) -> (Tensor keypoints)
Dec 17 19:20:17 	__caffe2::GenerateProposals(Tensor scores, Tensor bbox_deltas, Tensor im_info, Tensor anchors, float spatial_scale, int pre_nms_topN, int post_nms_topN, float nms_thresh, float min_size, bool angle_bound_on, int angle_bound_lo, int angle_bound_hi, float clip_angle_thresh, bool legacy_plus_one, Tensor[]? _caffe2_preallocated_outputs=None) -> (Tensor output_0, Tensor output_1)

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

This comment has been revised 7 times.