[Mlir-commits] [mlir] [mlir][SCF] Add support for peeling the first iteration out of the loop (PR #74015)

[llvmlistbot at llvm.org]

================
@@ -205,32 +222,92 @@ LogicalResult mlir::scf::peelForLoopAndSimplifyBounds(RewriterBase &rewriter,
   return success();
 }
 
+LogicalResult mlir::scf::peelFirstIterationForLoop(RewriterBase &b, ForOp forOp,
+                                                   ForOp &firstIteration) {
+  RewriterBase::InsertionGuard guard(b);
+  auto lbInt = getConstantIntValue(forOp.getLowerBound());
+  auto ubInt = getConstantIntValue(forOp.getUpperBound());
+  auto stepInt = getConstantIntValue(forOp.getStep());
+
+  // Peeling is not needed if there is one or less iteration.
+  if (lbInt && ubInt && stepInt && (*ubInt - *lbInt) / *stepInt <= 1)
+    return success();
+
+  // Slow path: Examine the ops that define lb, ub and step.
+  AffineExpr sym0, sym1, sym2;
+  bindSymbols(b.getContext(), sym0, sym1, sym2);
+  SmallVector<Value> operands{forOp.getLowerBound(), forOp.getUpperBound(),
+                              forOp.getStep()};
+  AffineMap map = AffineMap::get(0, 3, {(sym1 - sym0) % sym2});
----------------
MaheshRavishankar wrote:

I am not sure I understand this check... this is checking if (ub - lb) % step == 0 and bails, but that is true for step == 1 always... So what is the condition you are tyring to satisfy?

https://github.com/llvm/llvm-project/pull/74015