Bump llvm - get updated Python bindings for linalg.elementwise

build_tools/llvm_version.txt

@@ -1 +1 @@
-a99e055030f0da61651e808cbb208bb39594cdc0
+faf5d747f174cc9d714839f0d3bce1a783eac2ac

lib/TPP/Runner/MLIRBench.cpp

@@ -388,7 +388,7 @@ LogicalResult MLIRBench::printResult(Operation *kernelCall) {
 
     if (isa<TensorType>(result.getType())) {
       result =
-          builder.create<bufferization::ToMemrefOp>(unkLoc, memrefType, result);
+          builder.create<bufferization::ToBufferOp>(unkLoc, memrefType, result);
     }
 
     auto outBuf = builder.create<memref::AllocOp>(unkLoc, memrefType);

lib/TPP/Transforms/BrgemmLinalgTiling.cpp

@@ -202,7 +202,7 @@ struct BrgemmLinalgTiling
     RewritePatternSet patterns(&getContext());
     populateBrgemmLinalgTilingPatterns(patterns, options);
     GreedyRewriteConfig config;
-    config.strictMode = GreedyRewriteStrictness::ExistingOps;
+    config.setStrictness(GreedyRewriteStrictness::ExistingOps);
 
     (void)applyPatternsGreedily(getOperation(), std::move(patterns), config);
   }

lib/TPP/Transforms/HoistVectorTransfers.cpp

@@ -250,7 +250,7 @@ struct HoistVectorTransfers
     RewritePatternSet patterns(&getContext());
     populateHoistVectorTransferPatterns(patterns);
     GreedyRewriteConfig config;
-    config.strictMode = GreedyRewriteStrictness::ExistingOps;
+    config.setStrictness(GreedyRewriteStrictness::ExistingOps);
     (void)applyPatternsGreedily(getOperation(), std::move(patterns), config);
   }
 };

lib/TPP/Transforms/SplitReductionDim.cpp

@@ -103,7 +103,7 @@ struct SplitReductionDim
     RewritePatternSet patterns(ctx);
     patterns.add<SplitContractionReduction>(ctx, options);
     GreedyRewriteConfig config;
-    config.strictMode = GreedyRewriteStrictness::ExistingOps;
+    config.setStrictness(GreedyRewriteStrictness::ExistingOps);
     (void)applyPatternsGreedily(getOperation(), std::move(patterns), config);
   }
 };

lib/TPP/Transforms/VectorContractToAMX.cpp

@@ -392,7 +392,7 @@ struct VectorContractToAMXPattern
     int64_t M = accType.getDimSize(0);
     int64_t N = accType.getDimSize(1);
 
-    auto accSubview = accDefiningOp.getSource();
+    auto accSubview = accDefiningOp.getBase();
     Location loc = op.getLoc();
     scf::ForOp insertAt =
         getOutermostLoopWithIterargAccumulator(ctx.innerForOp, acc);
@@ -472,14 +472,14 @@ struct VectorContractToAMXPattern
       // Update index of LHS matrix subview for batch dimension if corresponding
       // loop is needed.
       if (iv)
-        mapping.map(lhsDefiningOp.getSource().getDefiningOp()->getOperand(1),
+        mapping.map(lhsDefiningOp.getBase().getDefiningOp()->getOperand(1),
                     iv);
       // Update index of LHS matrix subview for K dimension.
       mapping.map(
-          lhsDefiningOp.getSource().getDefiningOp()->getOperand(iv ? 3 : 1),
+          lhsDefiningOp.getBase().getDefiningOp()->getOperand(iv ? 3 : 1),
           innerIv);
       auto lhsClone = innerBuilder.clone(
-          *lhsDefiningOp.getSource().getDefiningOp(), mapping);
+          *lhsDefiningOp.getBase().getDefiningOp(), mapping);
       // Load matrix A tile
       SmallVector<Value, 4> aLoadTiles =
           createTileLoads(innerBuilder, loc, amxInputTilesOf16x32xBf16Ty,
@@ -489,14 +489,14 @@ struct VectorContractToAMXPattern
       // Update index of LHS matrix subview for batch dimension if corresponding
       // loop is needed.
       if (iv)
-        rhsMapping.map(rhsDefiningOp.getSource().getDefiningOp()->getOperand(1),
+        rhsMapping.map(rhsDefiningOp.getBase().getDefiningOp()->getOperand(1),
                        iv);
       // Update index of LHS matrix subview for K dimension.
       rhsMapping.map(
-          rhsDefiningOp.getSource().getDefiningOp()->getOperand(iv ? 2 : 1),
+          rhsDefiningOp.getBase().getDefiningOp()->getOperand(iv ? 2 : 1),
           innerIv);
       auto rhsClone = innerBuilder.clone(
-          *rhsDefiningOp.getSource().getDefiningOp(), rhsMapping);
+          *rhsDefiningOp.getBase().getDefiningOp(), rhsMapping);
       // Load matrix B tile, vnni factor and N tile size will be collapsed as
       // effective tilse size.
       SmallVector<Value, 4> bLoadTiles =

lib/TPP/Transforms/VectorContractToBF16DotProduct.cpp

@@ -402,13 +402,13 @@ struct BF16DotProductOp : OpRewritePattern<vector::ContractionOp> {
                   Value ivNewKForOp, ValueRange iterArgsNewKForOp) {
                 IRMapping mapping;
                 mapping.map(
-                    vectorReadOpLhs.getSource().getDefiningOp()->getOperand(1),
+                    vectorReadOpLhs.getBase().getDefiningOp()->getOperand(1),
                     ivNewReductionForOp);
                 mapping.map(
-                    vectorReadOpLhs.getSource().getDefiningOp()->getOperand(3),
+                    vectorReadOpLhs.getBase().getDefiningOp()->getOperand(3),
                     ivNewKForOp);
                 auto lhsClone = rewriterNewKForOp.clone(
-                    *vectorReadOpLhs.getSource().getDefiningOp(), mapping);
+                    *vectorReadOpLhs.getBase().getDefiningOp(), mapping);
 
                 // Memory access for A Matrix into <32xbf16>
                 llvm::SmallVector<Value, 8> vectorA;
@@ -440,13 +440,13 @@ struct BF16DotProductOp : OpRewritePattern<vector::ContractionOp> {
 
                 IRMapping rhsMapping;
                 rhsMapping.map(
-                    vectorReadOpRhs.getSource().getDefiningOp()->getOperand(1),
-                    ivNewReductionForOp);
+                  vectorReadOpRhs.getBase().getDefiningOp()->getOperand(1),
+                  ivNewReductionForOp);
                 rhsMapping.map(
-                    vectorReadOpRhs.getSource().getDefiningOp()->getOperand(2),
+                    vectorReadOpRhs.getBase().getDefiningOp()->getOperand(2),
                     ivNewKForOp);
                 auto rhsClone = rewriterNewKForOp.clone(
-                    *vectorReadOpRhs.getSource().getDefiningOp(), rhsMapping);
+                    *vectorReadOpRhs.getBase().getDefiningOp(), rhsMapping);
 
                 // Memory access for B Matrix into <32xbf16>
                 llvm::SmallVector<Value, 8> vectorB;
@@ -516,7 +516,7 @@ struct BF16DotProduct : public impl::BF16DotProductBase<BF16DotProduct> {
     RewritePatternSet patterns(&getContext());
     populateBF16DotProductPatterns(patterns);
     GreedyRewriteConfig config;
-    config.strictMode = GreedyRewriteStrictness::ExistingOps;
+    config.setStrictness(GreedyRewriteStrictness::ExistingOps);
     (void)applyPatternsGreedily(getOperation(), std::move(patterns), config);
   }
 };

lib/TPP/Transforms/VectorContractToFMA.cpp

@@ -245,7 +245,7 @@ struct VectorContractToFMAPattern
     if (K != 1)
       return failure();
 
-    auto accSubview = accDefiningOp.getSource();
+    auto accSubview = accDefiningOp.getBase();
     Location loc = op.getLoc();
 
     // Create M different <1xN> subviews.
@@ -295,13 +295,13 @@ struct VectorContractToFMAPattern
                   ValueRange innerIterArgs) {
                 IRMapping mapping;
                 mapping.map(
-                    lhsDefiningOp.getSource().getDefiningOp()->getOperand(1),
+                    lhsDefiningOp.getBase().getDefiningOp()->getOperand(1),
                     iv);
                 mapping.map(
-                    lhsDefiningOp.getSource().getDefiningOp()->getOperand(3),
+                    lhsDefiningOp.getBase().getDefiningOp()->getOperand(3),
                     innerIv);
                 auto lhsClone = innerBuilder.clone(
-                    *lhsDefiningOp.getSource().getDefiningOp(), mapping);
+                    *lhsDefiningOp.getBase().getDefiningOp(), mapping);
 
                 // Load and broadcast individual elements
                 SmallVector<Value, 4> broadcasts;
@@ -319,13 +319,13 @@ struct VectorContractToFMAPattern
 
                 IRMapping rhsMapping;
                 rhsMapping.map(
-                    rhsDefiningOp.getSource().getDefiningOp()->getOperand(1),
+                    rhsDefiningOp.getBase().getDefiningOp()->getOperand(1),
                     iv);
                 rhsMapping.map(
-                    rhsDefiningOp.getSource().getDefiningOp()->getOperand(2),
+                    rhsDefiningOp.getBase().getDefiningOp()->getOperand(2),
                     innerIv);
                 auto rhsClone = innerBuilder.clone(
-                    *rhsDefiningOp.getSource().getDefiningOp(), rhsMapping);
+                    *rhsDefiningOp.getBase().getDefiningOp(), rhsMapping);
                 auto rowVec = innerBuilder.create<vector::LoadOp>(
                     loc, VectorType::get({N}, elementType),
                     rhsClone->getResult(0), ValueRange{c0, c0, c0});

lib/TPP/Transforms/VectorContractToOuterproduct.cpp

@@ -230,8 +230,8 @@ struct VectorContractToOuterproductPattern
                                     nestedBuilder.getContext());
           }
 
-          Value lhsTensor = lhsDefiningOp.getSource();
-          Value rhsTensor = rhsDefiningOp.getSource();
+          Value lhsTensor = lhsDefiningOp.getBase();
+          Value rhsTensor = rhsDefiningOp.getBase();
           // Read vector slices using TransferReadOp
           auto lhsSlice = nestedBuilder.create<vector::TransferReadOp>(
               nestedLoc, VectorType::get({M}, lhsType.getElementType()),