[mlir][SCF] `scf.for`: Add support for unsigned integer comparisonusers/matthias-springer/scf_for_unsigned

19 files changed, 133 insertions, 41 deletions

diff --git a/mlir/include/mlir/Dialect/SCF/IR/SCFOps.td b/mlir/include/mlir/Dialect/SCF/IR/SCFOps.td
index 0c1c15b..88df541 100644
--- a/mlir/include/mlir/Dialect/SCF/IR/SCFOps.td
+++ b/mlir/include/mlir/Dialect/SCF/IR/SCFOps.td
@@ -169,9 +169,13 @@ def ForOp : SCF_Op<"for",
     region capturing the loop body. The induction variable is represented as an
     argument of this region. This SSA value is a signless integer or index.
     The step is a value of same type but required to be positive, the lower and
-    upper bounds can be also negative or zero. The lower and upper bounds specify
-    a half-open range: the iteration is executed iff the signed comparison of induction
-    variable value is less than the upper bound and bigger or equal to the lower bound.
+    upper bounds can be also negative or zero. The lower and upper bounds
+    specify a half-open range: the iteration is executed iff the comparison of
+    induction variable value is less than the upper bound and bigger or equal
+    to the lower bound.
+
+    By default, the integer comparison is signed. If the `unsignedCmp` unit
+    attribute is specified, the integer comparison is unsigned.
 
     The body region must contain exactly one block that terminates with
     `scf.yield`. Calling ForOp::build will create such a region and insert
@@ -184,8 +188,8 @@ def ForOp : SCF_Op<"for",
       ... // body
     }
     ...
-    // Integer case.
-    scf.for %iv_32 = %lb_32 to %ub_32 step %step_32 : i32 {
+    // Unsigned integer case.
+    scf.for unsigned %iv_32 = %lb_32 to %ub_32 step %step_32 : i32 {
       ... // body
     }
     ```
@@ -258,7 +262,8 @@ def ForOp : SCF_Op<"for",
   let arguments = (ins AnySignlessIntegerOrIndex:$lowerBound,
                        AnySignlessIntegerOrIndex:$upperBound,
                        AnySignlessIntegerOrIndex:$step,
-                       Variadic<AnyType>:$initArgs);
+                       Variadic<AnyType>:$initArgs,
+                       UnitAttr:$unsignedCmp);
   let results = (outs Variadic<AnyType>:$results);
   let regions = (region SizedRegion<1>:$region);
 
@@ -266,7 +271,8 @@ def ForOp : SCF_Op<"for",
   let builders = [OpBuilder<(ins "Value":$lowerBound, "Value":$upperBound,
       "Value":$step, CArg<"ValueRange", "{}">:$initArgs,
       CArg<"function_ref<void(OpBuilder &, Location, Value, ValueRange)>",
-           "nullptr">)>];
+           "nullptr">,
+      CArg<"bool", "false">:$unsignedCmp)>];
 
   let extraClassDeclaration = [{
     using BodyBuilderFn =
diff --git a/mlir/lib/Conversion/SCFToControlFlow/SCFToControlFlow.cpp b/mlir/lib/Conversion/SCFToControlFlow/SCFToControlFlow.cpp
index ba448e4..37cfc9f 100644
--- a/mlir/lib/Conversion/SCFToControlFlow/SCFToControlFlow.cpp
+++ b/mlir/lib/Conversion/SCFToControlFlow/SCFToControlFlow.cpp
@@ -382,8 +382,11 @@ LogicalResult ForLowering::matchAndRewrite(ForOp forOp,
 
   // With the body block done, we can fill in the condition block.
   rewriter.setInsertionPointToEnd(conditionBlock);
-  auto comparison = arith::CmpIOp::create(
-      rewriter, loc, arith::CmpIPredicate::slt, iv, upperBound);
+  arith::CmpIPredicate predicate = forOp.getUnsignedCmp()
+                                       ? arith::CmpIPredicate::ult
+                                       : arith::CmpIPredicate::slt;
+  auto comparison =
+      arith::CmpIOp::create(rewriter, loc, predicate, iv, upperBound);
 
   cf::CondBranchOp::create(rewriter, loc, comparison, firstBodyBlock,
                            ArrayRef<Value>(), endBlock, ArrayRef<Value>());
diff --git a/mlir/lib/Conversion/SCFToEmitC/SCFToEmitC.cpp b/mlir/lib/Conversion/SCFToEmitC/SCFToEmitC.cpp
index 84cbd86..1f239aa 100644
--- a/mlir/lib/Conversion/SCFToEmitC/SCFToEmitC.cpp
+++ b/mlir/lib/Conversion/SCFToEmitC/SCFToEmitC.cpp
@@ -154,6 +154,10 @@ ForLowering::matchAndRewrite(ForOp forOp, OpAdaptor adaptor,
                              ConversionPatternRewriter &rewriter) const {
   Location loc = forOp.getLoc();
 
+  if (forOp.getUnsignedCmp())
+    return rewriter.notifyMatchFailure(forOp,
+                                       "unsigned loops are not supported");
+
   // Create an emitc::variable op for each result. These variables will be
   // assigned to by emitc::assign ops within the loop body.
   SmallVector<Value> resultVariables;
diff --git a/mlir/lib/Conversion/SCFToSPIRV/SCFToSPIRV.cpp b/mlir/lib/Conversion/SCFToSPIRV/SCFToSPIRV.cpp
index dc92367..55ed31e 100644
--- a/mlir/lib/Conversion/SCFToSPIRV/SCFToSPIRV.cpp
+++ b/mlir/lib/Conversion/SCFToSPIRV/SCFToSPIRV.cpp
@@ -178,8 +178,14 @@ struct ForOpConversion final : SCFToSPIRVPattern<scf::ForOp> {
     // Generate the rest of the loop header.
     rewriter.setInsertionPointToEnd(header);
     auto *mergeBlock = loopOp.getMergeBlock();
-    auto cmpOp = spirv::SLessThanOp::create(rewriter, loc, rewriter.getI1Type(),
-                                            newIndVar, adaptor.getUpperBound());
+    Value cmpOp;
+    if (forOp.getUnsignedCmp()) {
+      cmpOp = spirv::ULessThanOp::create(rewriter, loc, rewriter.getI1Type(),
+                                         newIndVar, adaptor.getUpperBound());
+    } else {
+      cmpOp = spirv::SLessThanOp::create(rewriter, loc, rewriter.getI1Type(),
+                                         newIndVar, adaptor.getUpperBound());
+    }
 
     spirv::BranchConditionalOp::create(rewriter, loc, cmpOp, body,
                                        ArrayRef<Value>(), mergeBlock,
diff --git a/mlir/lib/Dialect/Linalg/Transforms/HoistPadding.cpp b/mlir/lib/Dialect/Linalg/Transforms/HoistPadding.cpp
index fd530f2..9436f1c 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/HoistPadding.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/HoistPadding.cpp
@@ -594,7 +594,8 @@ static FailureOr<PackingResult> buildPackingLoopNestImpl(
     auto clonedForOp = scf::ForOp::create(
         rewriter, loc, bvm.lookupOrDefault(forOp.getLowerBound()),
         bvm.lookupOrDefault(forOp.getUpperBound()),
-        bvm.lookupOrDefault(forOp.getStep()), hoistedPackedTensor);
+        bvm.lookupOrDefault(forOp.getStep()), hoistedPackedTensor,
+        /*bodyBuilder=*/nullptr, forOp.getUnsignedCmp());
 
     // Map the induction var, region args and results to the `clonedForOp`.
     bvm.map(forOp.getInductionVar(), clonedForOp.getInductionVar());
diff --git a/mlir/lib/Dialect/Linalg/Transforms/Hoisting.cpp b/mlir/lib/Dialect/Linalg/Transforms/Hoisting.cpp
index 58986a6..922b7d6 100644
--- a/mlir/lib/Dialect/Linalg/Transforms/Hoisting.cpp
+++ b/mlir/lib/Dialect/Linalg/Transforms/Hoisting.cpp
@@ -55,7 +55,8 @@ static scf::ForOp replaceWithDifferentYield(RewriterBase &rewriter,
 
   scf::ForOp newLoop = scf::ForOp::create(
       rewriter, loop.getLoc(), loop.getLowerBound(), loop.getUpperBound(),
-      loop.getStep(), inits, [](OpBuilder &, Location, Value, ValueRange) {});
+      loop.getStep(), inits, [](OpBuilder &, Location, Value, ValueRange) {},
+      loop.getUnsignedCmp());
 
   // Generate the new yield with the replaced operand.
   auto yieldOp = cast<scf::YieldOp>(loop.getBody()->getTerminator());
diff --git a/mlir/lib/Dialect/SCF/IR/SCF.cpp b/mlir/lib/Dialect/SCF/IR/SCF.cpp
index 0262a1b..89731de 100644
--- a/mlir/lib/Dialect/SCF/IR/SCF.cpp
+++ b/mlir/lib/Dialect/SCF/IR/SCF.cpp
@@ -318,9 +318,12 @@ void ConditionOp::getSuccessorRegions(
 
 void ForOp::build(OpBuilder &builder, OperationState &result, Value lb,
                   Value ub, Value step, ValueRange initArgs,
-                  BodyBuilderFn bodyBuilder) {
+                  BodyBuilderFn bodyBuilder, bool unsignedCmp) {
   OpBuilder::InsertionGuard guard(builder);
 
+  if (unsignedCmp)
+    result.addAttribute(getUnsignedCmpAttrName(result.name),
+                        builder.getUnitAttr());
   result.addOperands({lb, ub, step});
   result.addOperands(initArgs);
   for (Value v : initArgs)
@@ -450,6 +453,9 @@ static void printInitializationList(OpAsmPrinter &p,
 }
 
 void ForOp::print(OpAsmPrinter &p) {
+  if (getUnsignedCmp())
+    p << " unsigned";
+
   p << " " << getInductionVar() << " = " << getLowerBound() << " to "
     << getUpperBound() << " step " << getStep();
 
@@ -462,7 +468,8 @@ void ForOp::print(OpAsmPrinter &p) {
   p.printRegion(getRegion(),
                 /*printEntryBlockArgs=*/false,
                 /*printBlockTerminators=*/!getInitArgs().empty());
-  p.printOptionalAttrDict((*this)->getAttrs());
+  p.printOptionalAttrDict((*this)->getAttrs(),
+                          /*elidedAttrs=*/getUnsignedCmpAttrName().strref());
 }
 
 ParseResult ForOp::parse(OpAsmParser &parser, OperationState &result) {
@@ -472,6 +479,10 @@ ParseResult ForOp::parse(OpAsmParser &parser, OperationState &result) {
   OpAsmParser::Argument inductionVariable;
   OpAsmParser::UnresolvedOperand lb, ub, step;
 
+  if (succeeded(parser.parseOptionalKeyword("unsigned")))
+    result.addAttribute(getUnsignedCmpAttrName(result.name),
+                        builder.getUnitAttr());
+
   // Parse the induction variable followed by '='.
   if (parser.parseOperand(inductionVariable.ssaName) || parser.parseEqual() ||
       // Parse loop bounds.
@@ -562,7 +573,7 @@ ForOp::replaceWithAdditionalYields(RewriterBase &rewriter,
   inits.append(newInitOperands.begin(), newInitOperands.end());
   scf::ForOp newLoop = scf::ForOp::create(
       rewriter, getLoc(), getLowerBound(), getUpperBound(), getStep(), inits,
-      [](OpBuilder &, Location, Value, ValueRange) {});
+      [](OpBuilder &, Location, Value, ValueRange) {}, getUnsignedCmp());
   newLoop->setAttrs(getPrunedAttributeList(getOperation(), {}));
 
   // Generate the new yield values and append them to the scf.yield operation.
@@ -806,7 +817,8 @@ mlir::scf::replaceAndCastForOpIterArg(RewriterBase &rewriter, scf::ForOp forOp,
   // 2. Create the new forOp shell.
   scf::ForOp newForOp = scf::ForOp::create(
       rewriter, forOp.getLoc(), forOp.getLowerBound(), forOp.getUpperBound(),
-      forOp.getStep(), newIterOperands);
+      forOp.getStep(), newIterOperands, /*bodyBuilder=*/nullptr,
+      forOp.getUnsignedCmp());
   newForOp->setAttrs(forOp->getAttrs());
   Block &newBlock = newForOp.getRegion().front();
   SmallVector<Value, 4> newBlockTransferArgs(newBlock.getArguments().begin(),
@@ -931,7 +943,8 @@ struct ForOpIterArgsFolder : public OpRewritePattern<scf::ForOp> {
 
     scf::ForOp newForOp =
         scf::ForOp::create(rewriter, forOp.getLoc(), forOp.getLowerBound(),
-                           forOp.getUpperBound(), forOp.getStep(), newIterArgs);
+                           forOp.getUpperBound(), forOp.getStep(), newIterArgs,
+                           /*bodyBuilder=*/nullptr, forOp.getUnsignedCmp());
     newForOp->setAttrs(forOp->getAttrs());
     Block &newBlock = newForOp.getRegion().front();
 
@@ -989,12 +1002,12 @@ struct ForOpIterArgsFolder : public OpRewritePattern<scf::ForOp> {
 /// Util function that tries to compute a constant diff between u and l.
 /// Returns std::nullopt when the difference between two AffineValueMap is
 /// dynamic.
-static std::optional<int64_t> computeConstDiff(Value l, Value u) {
+static std::optional<APInt> computeConstDiff(Value l, Value u) {
   IntegerAttr clb, cub;
   if (matchPattern(l, m_Constant(&clb)) && matchPattern(u, m_Constant(&cub))) {
     llvm::APInt lbValue = clb.getValue();
     llvm::APInt ubValue = cub.getValue();
-    return (ubValue - lbValue).getSExtValue();
+    return ubValue - lbValue;
   }
 
   // Else a simple pattern match for x + c or c + x
@@ -1003,7 +1016,7 @@ static std::optional<int64_t> computeConstDiff(Value l, Value u) {
           u, m_Op<arith::AddIOp>(matchers::m_Val(l), m_ConstantInt(&diff))) ||
       matchPattern(
           u, m_Op<arith::AddIOp>(m_ConstantInt(&diff), matchers::m_Val(l))))
-    return diff.getSExtValue();
+    return diff;
   return std::nullopt;
 }
 
@@ -1022,13 +1035,15 @@ struct SimplifyTrivialLoops : public OpRewritePattern<ForOp> {
       return success();
     }
 
-    std::optional<int64_t> diff =
+    std::optional<APInt> diff =
         computeConstDiff(op.getLowerBound(), op.getUpperBound());
     if (!diff)
       return failure();
 
     // If the loop is known to have 0 iterations, remove it.
-    if (*diff <= 0) {
+    bool zeroOrLessIterations =
+        diff->isZero() || (!op.getUnsignedCmp() && diff->isNegative());
+    if (zeroOrLessIterations) {
       rewriter.replaceOp(op, op.getInitArgs());
       return success();
     }
@@ -3384,9 +3399,8 @@ ParseResult scf::WhileOp::parse(OpAsmParser &parser, OperationState &result) {
 
   if (functionType.getNumInputs() != operands.size()) {
     return parser.emitError(typeLoc)
-           << "expected as many input types as operands "
-           << "(expected " << operands.size() << " got "
-           << functionType.getNumInputs() << ")";
+           << "expected as many input types as operands " << "(expected "
+           << operands.size() << " got " << functionType.getNumInputs() << ")";
   }
 
   // Resolve input operands.
diff --git a/mlir/lib/Dialect/SCF/Transforms/BufferizableOpInterfaceImpl.cpp b/mlir/lib/Dialect/SCF/Transforms/BufferizableOpInterfaceImpl.cpp
index f8799c5..fb179e6 100644
--- a/mlir/lib/Dialect/SCF/Transforms/BufferizableOpInterfaceImpl.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/BufferizableOpInterfaceImpl.cpp
@@ -769,7 +769,8 @@ struct ForOpInterface
     // Construct a new scf.for op with memref instead of tensor values.
     auto newForOp = scf::ForOp::create(
         rewriter, forOp.getLoc(), forOp.getLowerBound(), forOp.getUpperBound(),
-        forOp.getStep(), castedInitArgs);
+        forOp.getStep(), castedInitArgs, /*bodyBuilder=*/nullptr,
+        forOp.getUnsignedCmp());
     newForOp->setAttrs(forOp->getAttrs());
     Block *loopBody = newForOp.getBody();
 
diff --git a/mlir/lib/Dialect/SCF/Transforms/ForToWhile.cpp b/mlir/lib/Dialect/SCF/Transforms/ForToWhile.cpp
index bee7780..ae52af5 100644
--- a/mlir/lib/Dialect/SCF/Transforms/ForToWhile.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/ForToWhile.cpp
@@ -58,9 +58,12 @@ struct ForLoopLoweringPattern : public OpRewritePattern<ForOp> {
     auto *beforeBlock = rewriter.createBlock(
         &whileOp.getBefore(), whileOp.getBefore().begin(), lcvTypes, lcvLocs);
     rewriter.setInsertionPointToStart(whileOp.getBeforeBody());
-    auto cmpOp = arith::CmpIOp::create(
-        rewriter, whileOp.getLoc(), arith::CmpIPredicate::slt,
-        beforeBlock->getArgument(0), forOp.getUpperBound());
+    arith::CmpIPredicate predicate = forOp.getUnsignedCmp()
+                                         ? arith::CmpIPredicate::ult
+                                         : arith::CmpIPredicate::slt;
+    auto cmpOp = arith::CmpIOp::create(rewriter, whileOp.getLoc(), predicate,
+                                       beforeBlock->getArgument(0),
+                                       forOp.getUpperBound());
     scf::ConditionOp::create(rewriter, whileOp.getLoc(), cmpOp.getResult(),
                              beforeBlock->getArguments());
 
diff --git a/mlir/lib/Dialect/SCF/Transforms/LoopPipelining.cpp b/mlir/lib/Dialect/SCF/Transforms/LoopPipelining.cpp
index 1130538..7e7fba4 100644
--- a/mlir/lib/Dialect/SCF/Transforms/LoopPipelining.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/LoopPipelining.cpp
@@ -791,6 +791,11 @@ FailureOr<ForOp> mlir::scf::pipelineForLoop(RewriterBase &rewriter, ForOp forOp,
                                             bool *modifiedIR) {
   if (modifiedIR)
     *modifiedIR = false;
+
+  // TODO: Add support for unsigned loops.
+  if (forOp.getUnsignedCmp())
+    return failure();
+
   LoopPipelinerInternal pipeliner;
   if (!pipeliner.initializeLoopInfo(forOp, options))
     return failure();
diff --git a/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp b/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp
index 4752c08..f1203b2 100644
--- a/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp
@@ -256,6 +256,10 @@ struct ForLoopPeelingPattern : public OpRewritePattern<ForOp> {
 
   LogicalResult matchAndRewrite(ForOp forOp,
                                 PatternRewriter &rewriter) const override {
+    if (forOp.getUnsignedCmp())
+      return rewriter.notifyMatchFailure(forOp,
+                                         "unsigned loops are not supported");
+
     // Do not peel already peeled loops.
     if (forOp->hasAttr(kPeeledLoopLabel))
       return failure();
diff --git a/mlir/lib/Dialect/SCF/Transforms/StructuralTypeConversions.cpp b/mlir/lib/Dialect/SCF/Transforms/StructuralTypeConversions.cpp
index 1b07b77..3b75970 100644
--- a/mlir/lib/Dialect/SCF/Transforms/StructuralTypeConversions.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/StructuralTypeConversions.cpp
@@ -116,7 +116,8 @@ public:
                                 llvm::getSingleElement(adaptor.getLowerBound()),
                                 llvm::getSingleElement(adaptor.getUpperBound()),
                                 llvm::getSingleElement(adaptor.getStep()),
-                                flattenValues(adaptor.getInitArgs()));
+                                flattenValues(adaptor.getInitArgs()),
+                                /*bodyBuilder=*/nullptr, op.getUnsignedCmp());
 
     // Reserve whatever attributes in the original op.
     newOp->setAttrs(op->getAttrs());
diff --git a/mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp b/mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp
index c0e47ee..250c413 100644
--- a/mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp
@@ -797,7 +797,8 @@ FailureOr<LoopLikeOpInterface> yieldTiledValuesAndReplaceLoop<scf::ForOp>(
   inits.append(newInitOperands.begin(), newInitOperands.end());
   auto newLoop = scf::ForOp::create(
       rewriter, loc, loopOp.getLowerBound(), loopOp.getUpperBound(),
-      loopOp.getStep(), inits, [](OpBuilder &, Location, Value, ValueRange) {});
+      loopOp.getStep(), inits, [](OpBuilder &, Location, Value, ValueRange) {},
+      loopOp.getUnsignedCmp());
 
   // Move the loop body to the new op.
   Block *loopBody = loopOp.getBody();
@@ -935,7 +936,8 @@ static LogicalResult addInitOperandsToLoopNest(
     auto newLoop = scf::ForOp::create(
         rewriter, forLoop.getLoc(), forLoop.getLowerBound(),
         forLoop.getUpperBound(), forLoop.getStep(), newInits,
-        [&](OpBuilder &b, Location loc, Value iv, ValueRange iterArgs) {});
+        [&](OpBuilder &b, Location loc, Value iv, ValueRange iterArgs) {},
+        forLoop.getUnsignedCmp());
 
     // Merge the body of the new loop with the body of the old loops.
     SmallVector<Value> sourceBlockArgs;
diff --git a/mlir/lib/Dialect/SCF/Utils/Utils.cpp b/mlir/lib/Dialect/SCF/Utils/Utils.cpp
index 5731795..4910258 100644
--- a/mlir/lib/Dialect/SCF/Utils/Utils.cpp
+++ b/mlir/lib/Dialect/SCF/Utils/Utils.cpp
@@ -1233,6 +1233,7 @@ static void getPerfectlyNestedLoopsImpl(
 
 static Loops stripmineSink(scf::ForOp forOp, Value factor,
                            ArrayRef<scf::ForOp> targets) {
+  assert(!forOp.getUnsignedCmp() && "unsigned loops are not supported");
   auto originalStep = forOp.getStep();
   auto iv = forOp.getInductionVar();
 
@@ -1241,6 +1242,8 @@ static Loops stripmineSink(scf::ForOp forOp, Value factor,
 
   Loops innerLoops;
   for (auto t : targets) {
+    assert(!t.getUnsignedCmp() && "unsigned loops are not supported");
+
     // Save information for splicing ops out of t when done
     auto begin = t.getBody()->begin();
     auto nOps = t.getBody()->getOperations().size();
@@ -1415,6 +1418,8 @@ scf::ForallOp mlir::fuseIndependentSiblingForallLoops(scf::ForallOp target,
 scf::ForOp mlir::fuseIndependentSiblingForLoops(scf::ForOp target,
                                                 scf::ForOp source,
                                                 RewriterBase &rewriter) {
+  assert(source.getUnsignedCmp() == target.getUnsignedCmp() &&
+         "incompatible signedness");
   unsigned numTargetOuts = target.getNumResults();
   unsigned numSourceOuts = source.getNumResults();
 
@@ -1428,7 +1433,8 @@ scf::ForOp mlir::fuseIndependentSiblingForLoops(scf::ForOp target,
   rewriter.setInsertionPointAfter(source);
   scf::ForOp fusedLoop = scf::ForOp::create(
       rewriter, source.getLoc(), source.getLowerBound(), source.getUpperBound(),
-      source.getStep(), fusedInitArgs);
+      source.getStep(), fusedInitArgs, /*bodyBuilder=*/nullptr,
+      source.getUnsignedCmp());
 
   // Map original induction variables and operands to those of the fused loop.
   IRMapping mapping;
diff --git a/mlir/lib/Dialect/SparseTensor/Transforms/SparseVectorization.cpp b/mlir/lib/Dialect/SparseTensor/Transforms/SparseVectorization.cpp
index 4464450..febec6d 100644
--- a/mlir/lib/Dialect/SparseTensor/Transforms/SparseVectorization.cpp
+++ b/mlir/lib/Dialect/SparseTensor/Transforms/SparseVectorization.cpp
@@ -533,8 +533,10 @@ static bool vectorizeStmt(PatternRewriter &rewriter, scf::ForOp forOp, VL vl,
       VectorType vtp = vectorType(vl, init.getType());
       Value vinit = genVectorReducInit(rewriter, loc, yield->getOperand(0),
                                        forOp.getRegionIterArg(0), init, vtp);
-      forOpNew = scf::ForOp::create(rewriter, loc, forOp.getLowerBound(),
-                                    forOp.getUpperBound(), step, vinit);
+      forOpNew =
+          scf::ForOp::create(rewriter, loc, forOp.getLowerBound(),
+                             forOp.getUpperBound(), step, vinit,
+                             /*bodyBuilder=*/nullptr, forOp.getUnsignedCmp());
       forOpNew->setAttr(
           LoopEmitter::getLoopEmitterLoopAttrName(),
           forOp->getAttr(LoopEmitter::getLoopEmitterLoopAttrName()));
@@ -605,8 +607,8 @@ public:
 
   ForOpRewriter(MLIRContext *context, unsigned vectorLength,
                 bool enableVLAVectorization, bool enableSIMDIndex32)
-      : OpRewritePattern(context), vl{vectorLength, enableVLAVectorization,
-                                      enableSIMDIndex32} {}
+      : OpRewritePattern(context),
+        vl{vectorLength, enableVLAVectorization, enableSIMDIndex32} {}
 
   LogicalResult matchAndRewrite(scf::ForOp op,
                                 PatternRewriter &rewriter) const override {
diff --git a/mlir/lib/Dialect/Vector/Transforms/VectorDistribute.cpp b/mlir/lib/Dialect/Vector/Transforms/VectorDistribute.cpp
index bb0f339..be0d28a 100644
--- a/mlir/lib/Dialect/Vector/Transforms/VectorDistribute.cpp
+++ b/mlir/lib/Dialect/Vector/Transforms/VectorDistribute.cpp
@@ -1826,7 +1826,8 @@ struct WarpOpScfForOp : public WarpDistributionPattern {
     rewriter.setInsertionPointAfter(newWarpOp);
     auto newForOp = scf::ForOp::create(
         rewriter, forOp.getLoc(), forOp.getLowerBound(), forOp.getUpperBound(),
-        forOp.getStep(), newForOpOperands);
+        forOp.getStep(), newForOpOperands, /*bodyBuilder=*/nullptr,
+        forOp.getUnsignedCmp());
     // Next, we insert a new `WarpOp` (called inner `WarpOp`) inside the
     // newly created `ForOp`. This `WarpOp` will contain all ops that were
     // contained within the original `ForOp` body.
diff --git a/mlir/test/Conversion/SCFToControlFlow/convert-to-cfg.mlir b/mlir/test/Conversion/SCFToControlFlow/convert-to-cfg.mlir
index ef0fa08..483c7b3 100644
--- a/mlir/test/Conversion/SCFToControlFlow/convert-to-cfg.mlir
+++ b/mlir/test/Conversion/SCFToControlFlow/convert-to-cfg.mlir
@@ -18,6 +18,24 @@ func.func @simple_std_for_loop(%arg0 : index, %arg1 : index, %arg2 : index) {
   return
 }
 
+// CHECK-LABEL: func @unsigned_loop(%{{.*}}: index, %{{.*}}: index, %{{.*}}: index) {
+//  CHECK-NEXT:  cf.br ^bb1(%{{.*}} : index)
+//  CHECK-NEXT:  ^bb1(%{{.*}}: index):    // 2 preds: ^bb0, ^bb2
+//  CHECK-NEXT:    %{{.*}} = arith.cmpi ult, %{{.*}}, %{{.*}} : index
+//  CHECK-NEXT:    cf.cond_br %{{.*}}, ^bb2, ^bb3
+//  CHECK-NEXT:  ^bb2:   // pred: ^bb1
+//  CHECK-NEXT:    %{{.*}} = arith.constant 1 : index
+//  CHECK-NEXT:    %[[iv:.*]] = arith.addi %{{.*}}, %{{.*}} : index
+//  CHECK-NEXT:    cf.br ^bb1(%[[iv]] : index)
+//  CHECK-NEXT:  ^bb3:   // pred: ^bb1
+//  CHECK-NEXT:    return
+func.func @unsigned_loop(%arg0 : index, %arg1 : index, %arg2 : index) {
+  scf.for unsigned %i0 = %arg0 to %arg1 step %arg2 {
+    %c1 = arith.constant 1 : index
+  }
+  return
+}
+
 // CHECK-LABEL: func @simple_std_2_for_loops(%{{.*}}: index, %{{.*}}: index, %{{.*}}: index) {
 //  CHECK-NEXT:    cf.br ^bb1(%{{.*}} : index)
 //  CHECK-NEXT:  ^bb1(%[[ub0:.*]]: index):    // 2 preds: ^bb0, ^bb5
diff --git a/mlir/test/Conversion/SCFToSPIRV/for.mlir b/mlir/test/Conversion/SCFToSPIRV/for.mlir
index 81661ec..9c55216 100644
--- a/mlir/test/Conversion/SCFToSPIRV/for.mlir
+++ b/mlir/test/Conversion/SCFToSPIRV/for.mlir
@@ -5,6 +5,7 @@ module attributes {
     #spirv.vce<v1.0, [Shader], [SPV_KHR_storage_buffer_storage_class]>, #spirv.resource_limits<>>
 } {
 
+// CHECK-LABEL: @loop_kernel
 func.func @loop_kernel(%arg2 : memref<10xf32, #spirv.storage_class<StorageBuffer>>, %arg3 : memref<10xf32, #spirv.storage_class<StorageBuffer>>) {
   // CHECK: %[[LB:.*]] = spirv.Constant 4 : i32
   %lb = arith.constant 4 : index
@@ -34,6 +35,19 @@ func.func @loop_kernel(%arg2 : memref<10xf32, #spirv.storage_class<StorageBuffer
   return
 }
 
+// CHECK-LABEL: @unsigned_loop
+//       CHECK:   spirv.ULessThan
+func.func @unsigned_loop(%arg2 : memref<10xf32, #spirv.storage_class<StorageBuffer>>, %arg3 : memref<10xf32, #spirv.storage_class<StorageBuffer>>) {
+  %lb = arith.constant 4 : index
+  %ub = arith.constant 42 : index
+  %step = arith.constant 2 : index
+  scf.for unsigned %arg4 = %lb to %ub step %step {
+    %1 = memref.load %arg2[%arg4] : memref<10xf32, #spirv.storage_class<StorageBuffer>>
+    memref.store %1, %arg3[%arg4] : memref<10xf32, #spirv.storage_class<StorageBuffer>>
+  }
+  return
+}
+
 // CHECK-LABEL: @loop_yield
 func.func @loop_yield(%arg2 : memref<10xf32, #spirv.storage_class<StorageBuffer>>, %arg3 : memref<10xf32, #spirv.storage_class<StorageBuffer>>) {
   // CHECK: %[[LB:.*]] = spirv.Constant 4 : i32
diff --git a/mlir/test/Dialect/SCF/ops.mlir b/mlir/test/Dialect/SCF/ops.mlir
index 7f457ef..5930a1d 100644
--- a/mlir/test/Dialect/SCF/ops.mlir
+++ b/mlir/test/Dialect/SCF/ops.mlir
@@ -28,14 +28,14 @@ func.func @std_for(%arg0 : index, %arg1 : index, %arg2 : index) {
 
 func.func @std_for_i32(%arg0 : i32, %arg1 : i32, %arg2 : i32) {
   scf.for %i0 = %arg0 to %arg1 step %arg2 : i32 {
-    scf.for %i1 = %arg0 to %arg1 step %arg2 : i32 {
+    scf.for unsigned %i1 = %arg0 to %arg1 step %arg2 : i32 {
     }
   }
   return
 }
 // CHECK-LABEL: func @std_for_i32(
 //  CHECK-NEXT:   scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} : i32 {
-//  CHECK-NEXT:     scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} : i32 {
+//  CHECK-NEXT:     scf.for unsigned %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} : i32 {
 
 func.func @scf_for_i64_iter(%arg1: i64, %arg2: i64) {
   %c1_i64 = arith.constant 1 : i64