trunk/652012a9e9691c0ecb931df78f17f323f6fe3cde: Disable shape padding in test_slice_scatter_reinplace (#197997)
- PyTorch: 1319 events in the last 90 days
- PyTorch: 1304th Release in the last 90 days
- Previous: earlier the same day · trunk/dc0133edcfb1aded1e2e054b31e643ed459472b6: Fix meta function for `dyn_quant_pack_4bit_weight` on ARM (#197518)
What happened
test_slice_scatter_reinplace asserts that inductor generates exactly one kernel, to show the slice_scatter into the cache was reinplaced. The model also runs a matmul whose bmm has a 33-column operand, and pad_mm decides whether to pad it to 36 columns by timing the padded and unpadded bmm at compile time and padding when the padded one looks more than 1.1x faster. At these sizes each timed call takes well under 0.1…
Summary assembled by rule from the sources below