llvm
/
torch-mlir
mirror of https://github.com/llvm/torch-mlir
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Apply the trait NoMemoryEffect to most ReadOnly ops

pull/3891/head
zjgarvey 2024-11-22 16:55:03 -06:00
parent 0913b967ac
commit 7e30ef798b
3 changed files with 1211 additions and 604 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1806
include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td
View File

File diff suppressed because it is too large Load Diff

7
projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py
View File

@ -242,6 +242,7 @@ def emit_op(
has_folder: bool = False,
has_canonicalizer: bool = False,
has_verifier: bool = False,
has_memory_effects: bool = False,
):
"""Main entry point for op emission.
@ -257,6 +258,12 @@ def emit_op(
traits += ["HasValueSemantics"]
if operator.is_readonly():
traits += ["ReadOnly"]
# If a ReadOnly op has no returns, it is likely to have side effects.
# E.g. `prim.RaiseException` and `prim.Print`
# Besides ops with no returned values, there may be other ReadOnly ops with memory effects.
# In such cases, these ops can be emitted with `has_memory_effects=True` to avoid this trait.
if operator.is_readonly() and len(operator.returns) != 0 and not has_memory_effects:
traits += ["NoMemoryEffect"]
raw_emit_op(
operator,

2
test/Dialect/Torch/canonicalize.mlir
View File

@ -1557,7 +1557,6 @@ func.func @torch.prim.unchecked_cast$derefine(%arg0: !torch.list<int>) -> !torch
// CHECK-LABEL: func.func @torch.aten.Int.Tensor(
// CHECK-SAME: %[[NUM:.*]]: !torch.int) -> !torch.int {
// CHECK: %[[T:.*]] = torch.prim.NumToTensor.Scalar %[[NUM]] : !torch.int -> !torch.vtensor<[],si64>
// CHECK: return %[[NUM]] : !torch.int
func.func @torch.aten.Int.Tensor(%arg0: !torch.int) -> !torch.int {
%tensor = torch.prim.NumToTensor.Scalar %arg0: !torch.int -> !torch.vtensor<[],si64>
@ -1585,7 +1584,6 @@ func.func @torch.aten.Int.float() -> !torch.int {
// CHECK-LABEL: func.func @torch.aten.Float.Tensor(
// CHECK-SAME: %[[NUM:.*]]: !torch.float) -> !torch.float {
// CHECK: %[[T:.*]] = torch.prim.NumToTensor.Scalar %[[NUM]] : !torch.float -> !torch.vtensor<[],f64>
// CHECK: return %[[NUM]] : !torch.float
func.func @torch.aten.Float.Tensor(%arg0: !torch.float) -> !torch.float {
%tensor = torch.prim.NumToTensor.Scalar %arg0: !torch.float -> !torch.vtensor<[],f64>