nnicolosi/Raptor
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

reimplement pool lowering

Browse Source 
add pool validation
align PIM ops/codegen/parser with the ISA
move constant materialization to MLIR
rename the PIM verification/materialization passes
better folded-constant handling
This commit is contained in:
NiccoloN
2026-03-23 19:14:50 +01:00
parent 461bdd808d
commit 661170a9aa
30 changed files with 912 additions and 512 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
src/PIM/Conversion/SpatialToPim/SpatialToPim.td
View File

@@ -27,9 +27,21 @@ def spatToPimMVMOp : Pat<
(NativeCodeCall<"onnx_mlir::getBestOutputTensorFromOperandsOrAllocate($_builder, $0.getDefiningOp())"> $srcOpRes))
>;
def spatToPimVAddOp : Pat<
def spatToPimVVAddOp : Pat<
(SpatVAddOp:$srcOpRes $a, $b),
(PimVAddOp $a, $b,
(PimVVAddOp $a, $b,
(NativeCodeCall<"onnx_mlir::getBestOutputTensorFromOperandsOrAllocate($_builder, $0.getDefiningOp())"> $srcOpRes))
>;
def spatToPimVVMulOp : Pat<
(SpatVMulOp:$srcOpRes $a, $b),
(PimVVMulOp $a, $b,
(NativeCodeCall<"onnx_mlir::getBestOutputTensorFromOperandsOrAllocate($_builder, $0.getDefiningOp())"> $srcOpRes))
>;
def spatToPimVVMaxOp : Pat<
(SpatVMaxOp:$srcOpRes $a, $b),
(PimVVMaxOp $a, $b,
(NativeCodeCall<"onnx_mlir::getBestOutputTensorFromOperandsOrAllocate($_builder, $0.getDefiningOp())"> $srcOpRes))
>;