AOTI fallback ops: remove ops that were never codegen'ed (pytorch#143421

)

Removes 4 fallback ops that are currently not possible to codegen, which does not break ABI-compatibility.

1. `_cudnn_rnn_backward` and `_histogramdd_bin_edges` both return `Tensor[]`, which we cannot codegen with the current design.
2. `_sparse_coo_tensor_with_dims_and_tensors` only supplies a Sparse operator, which we don't support.
3. `zeros.names` requires a `Dimname` input, which we can't currently codegen.

Removing these ops from the list will improve test performance, since the fallback op generation will use the Python proxy executor instead of calling non-existent C functions.

Pull Request resolved: pytorch#143421
Approved by: https://github.com/desertfire
ghstack dependencies: pytorch#141371, pytorch#143223

torchgen/aoti/fallback_ops.py

@@ -39,7 +39,6 @@
     "aten.cholesky_solve.default",
     "aten.convolution_backward.default",
     "aten._cudnn_rnn.default",
-    "aten._cudnn_rnn_backward.default",
     "aten.convolution.default",
     "aten.cummax.default",
     "aten.cummin.default",
@@ -70,7 +69,6 @@
     "aten.grid_sampler_2d_backward.default",
     "aten.histc.default",
     "aten.histogram.bin_ct",
-    "aten._histogramdd_bin_edges.default",
     "aten._histogramdd_from_bin_cts.default",
     "aten.index_put.default",
     "aten.index_reduce.default",
@@ -139,7 +137,6 @@
     "aten.soft_margin_loss_backward.default",
     "aten.sort.default",
     "aten.sort.stable",
-    "aten._sparse_coo_tensor_with_dims_and_tensors.default",
     "aten._thnn_fused_lstm_cell.default",
     "aten.topk.default",
     "aten._to_sparse.default",
@@ -154,5 +151,4 @@
     "aten.view_as_real.default",
     "aten.view.dtype",
     "aten._weight_int8pack_mm.default",
-    "aten.zeros.names",
 }