colin
/
mlir-hlo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[XLA/GPU] Add XLA HLO -> LMHLO conversion to several ops, and implement them in XLA/GPU. 

PiperOrigin-RevId: 353158172
This commit is contained in:
Tim Shen 2021-01-21 19:56:11 -08:00 committed by TensorFlow MLIR Team
parent ae10640d78
commit d1c785381d
3 changed files with 25 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
include/mlir-hlo/Dialect/mhlo/IR/hlo_ops.td
View File

@ -713,7 +713,7 @@ def HLO_SliceOp: HLO_Op<
}
def HLO_DynamicSliceOp: HLO_Op<"dynamic-slice",
[NoSideEffect, AllElementTypesMatch<["operand", "result"]>]> {
[NoSideEffect, AllElementTypesMatch<["operand", "result"]>]>, BASE_HLO_DynamicSliceOp {
let arguments = (ins
HLO_Tensor:$operand,
Variadic<HLO_ScalarIntTensor>:$start_indices,
@ -726,7 +726,7 @@ def HLO_DynamicSliceOp: HLO_Op<"dynamic-slice",
def HLO_DynamicUpdateSliceOp: HLO_Op<"dynamic-update-slice",
[NoSideEffect, AllElementTypesMatch<["operand", "update", "result"]>,
AllShapesMatch<["operand", "result"]>]> {
AllShapesMatch<["operand", "result"]>]>, BASE_HLO_DynamicUpdateSliceOp {
let arguments = (ins
HLO_Tensor:$operand,
HLO_Tensor:$update,

20
include/mlir-hlo/Dialect/mhlo/IR/lhlo_ops.td
View File

@ -214,9 +214,7 @@ def LHLO_ReduceOp: LHLO_Op<"reduce", [SameVariadicOperandSize]>, BASE_HLO_Reduce
let regions = (region SizedRegion<1>:$body);
}
def LHLO_ReduceWindowOp: LHLO_Op<"reduce_window", [
SingleBlockImplicitTerminator<"TerminatorOp">
]>, BASE_HLO_ReduceWindowOp {
def LHLO_ReduceWindowOp: LHLO_Op<"reduce_window", []>, BASE_HLO_ReduceWindowOp {
let arguments = (ins
Arg<LHLO_Buffer, "", [MemRead]>:$operand,
@ -309,12 +307,22 @@ def LHLO_SliceOp: LHLO_Op<
);
}
def HLO_DynamicUpdateSliceOp: LHLO_Op<"dynamic-update-slice", []> {
def LHLO_DynamicSliceOp: LHLO_Op<"dynamic_slice",
[AllElementTypesMatch<["operand", "output"]>]>, BASE_HLO_DynamicSliceOp {
let arguments = (ins
Arg<LHLO_Buffer, "", [MemRead]>:$operand,
Arg<Variadic<LHLO_Buffer>, "", [MemRead]>:$start_indices,
Arg<LHLO_Buffer, "", [MemWrite]>:$output,
I64ElementsAttr:$slice_sizes
);
}
def LHLO_DynamicUpdateSliceOp: LHLO_Op<"dynamic-update-slice", []>, BASE_HLO_DynamicUpdateSliceOp {
let arguments = (ins
Arg<LHLO_Buffer, "", [MemRead]>:$operand,
Arg<LHLO_Buffer, "", [MemRead]>:$update,
Arg<LHLO_Buffer, "", [MemWrite]>:$output,
Arg<Variadic<LHLO_Buffer>, "", [MemRead]>:$start_indices
Arg<Variadic<LHLO_Buffer>, "", [MemRead]>:$start_indices,
Arg<LHLO_Buffer, "", [MemWrite]>:$output
);
}

9
lib/utils/hlo_utils.cc
View File

@ -135,6 +135,15 @@ DenseElementsAttr GetScalarLimitOfType(Type ty, ScalarLimit limit) {
std::string LmhloToMhloOpName(llvm::StringRef op_name,
mlir::MLIRContext *context) {
assert(op_name.startswith("lmhlo.") && "Expected an LMHLO op");
if (op_name == "lmhlo.dot") {
return "mhlo.dot_general";
}
if (op_name == "lmhlo.dynamic_slice") {
return "mhlo.dynamic-slice";
}
std::string mhlo_op_name(op_name.drop_front(1));
if (context->isOperationRegistered(mhlo_op_name)) return mhlo_op_name;
return "";