intel/llvm
1
0
Fork 0
mirror of https://github.com/intel/llvm.git synced 2026-01-26 12:26:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[mlir] Change vector.transfer_read/write "masked" attribute to "in_bounds".

Browse Source 
This is in preparation for adding a new "mask" operand. The existing "masked" attribute was used to specify dimensions that may be out-of-bounds. Such transfers can be lowered to masked load/stores. The new "in_bounds" attribute is used to specify dimensions that are guaranteed to be within bounds. (Semantics is inverted.)

Differential Revision: https://reviews.llvm.org/D99639
This commit is contained in:
Matthias Springer
2021-03-31 14:59:30 +09:00
parent b5995fced4
commit 95f8135043
26 changed files with 315 additions and 302 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp
View File

@@ -1248,14 +1248,15 @@ public:
Value vectorDataPtr =
castDataPtr(rewriter, loc, dataPtr, memRefType, toLLVMTy(vtp));
if (!xferOp.isMaskedDim(0))
if (xferOp.isDimInBounds(0))
return replaceTransferOpWithLoadOrStore(rewriter,
*this->getTypeConverter(), loc,
xferOp, operands, vectorDataPtr);
// 2. Create a vector with linear indices [ 0 .. vector_length - 1 ].
// 3. Create offsetVector = [ offset + 0 .. offset + vector_length - 1 ].
// 4. Let dim the memref dimension, compute the vector comparison mask:
// 4. Let dim the memref dimension, compute the vector comparison mask
// (in-bounds mask):
// [ offset + 0 .. offset + vector_length - 1 ] < [ dim .. dim ]
//
// TODO: when the leaf transfer rank is k > 1, we need the last `k`