Magic-Mirror/llvm-capstone
1
0
Fork 0
mirror of https://github.com/capstone-engine/llvm-capstone.git synced 2025-02-07 18:10:24 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[mlir][tensor] Do not use affine ops on non-perfect-tiling unpack cases.

Browse Source 
Do not create an Affine ops for expanded size because the affine op is
too complicated which would hit an assertion in affine ops
simplification.

Reviewed By: mravishankar

Differential Revision: https://reviews.llvm.org/D144151
This commit is contained in:
Hanhan Wang 2023-02-15 17:36:21 -08:00
parent 2b2b8409e6
commit 5fa9933c13
2 changed files with 5 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
mlir/lib/Dialect/Tensor/IR/TensorTilingInterfaceImpl.cpp
View File

@ -318,8 +318,11 @@ static UnpackTileDimInfo getUnpackTileDimInfo(OpBuilder &b, UnPackOp unpackOp,
ab.add(AV(dim0).bind(lengthMinusOne), AV(dim1).bind(oneAttr));
info.sourceOffset = firstCoord.quotient;
info.resultOffset = firstCoord.remainder;
info.destExpandedSize =
ab.mul(AV(dim0).bind(info.sourceSize), AV(sym0).bind(innerTileSize));
// Do not create an Affine ops for expanded size because the affine op is too
// complicated which would trigger an issue in affine ops simplification.
info.destExpandedSize = b.createOrFold<arith::MulIOp>(
loc, getValueOrCreateConstantIndexOp(b, loc, info.sourceSize),
getValueOrCreateConstantIndexOp(b, loc, innerTileSize));
return info;
}

4
mlir/test/Dialect/Tensor/tiling.mlir
View File

@ -391,11 +391,9 @@ transform.sequence failures(propagate) {
// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0) -> (d0 floordiv 32)>
// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0) -> (d0 mod 32)>
// CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0) -> ((d0 + 1) floordiv 32 - d0 floordiv 32 + 1)>
// CHECK-DAG: #[[MAP3:.+]] = affine_map<(d0) -> (((d0 + 1) floordiv 32) * 32 - (d0 floordiv 32) * 32 + 32)>
// CHECK-DAG: #[[MAP4:.+]] = affine_map<(d0) -> (d0 floordiv 16)>
// CHECK-DAG: #[[MAP5:.+]] = affine_map<(d0) -> (d0 mod 16)>
// CHECK-DAG: #[[MAP6:.+]] = affine_map<(d0) -> ((d0 + 3) floordiv 16 - d0 floordiv 16 + 1)>
// CHECK-DAG: #[[MAP7:.+]] = affine_map<(d0) -> (((d0 + 3) floordiv 16) * 16 - (d0 floordiv 16) * 16 + 16)>
// CHECK: func.func @NCnc_to_NC
// CHECK-SAME: %[[IN:[A-Za-z0-9]+]]:
// CHECK-SAME: %[[OUT:[A-Za-z0-9]+]]:
@ -440,11 +438,9 @@ transform.sequence failures(propagate) {
// CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0) -> (d0 floordiv 32)>
// CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0) -> (d0 mod 32)>
// CHECK-DAG: #[[MAP2:.+]] = affine_map<(d0) -> ((d0 + 1) floordiv 32 - d0 floordiv 32 + 1)>
// CHECK-DAG: #[[MAP3:.+]] = affine_map<(d0) -> (((d0 + 1) floordiv 32) * 32 - (d0 floordiv 32) * 32 + 32)>
// CHECK-DAG: #[[MAP4:.+]] = affine_map<(d0) -> (d0 floordiv 8)>
// CHECK-DAG: #[[MAP5:.+]] = affine_map<(d0) -> (d0 mod 8)>
// CHECK-DAG: #[[MAP6:.+]] = affine_map<(d0) -> ((d0 + 3) floordiv 8 - d0 floordiv 8 + 1)>
// CHECK-DAG: #[[MAP7:.+]] = affine_map<(d0) -> (((d0 + 3) floordiv 8) * 8 - (d0 floordiv 8) * 8 + 8)>
// CHECK: func.func @CKkc_to_KC
// CHECK-SAME: %[[IN:[A-Za-z0-9]+]]:
// CHECK-SAME: %[[OUT:[A-Za-z0-9]+]]: