[mlir] [llvm] [libc] [libcxx] [clang-tools-extra] [clang] [compiler-rt] [lldb] [lld] [MLIR][LLVM] Add Continuous Loop Peeling transform to SCF (PR #71555)

[Matthias Springer via cfe-commits]

================
@@ -0,0 +1,46 @@
+// RUN: mlir-opt %s -scf-for-loop-continuous-peeling=convert-single-iter-loops-to-if=true -split-input-file | FileCheck %s
+
+#map = affine_map<(d0, d1)[s0] -> (s0, d0 - d1)>
+func.func @foo(%ub: index) -> index {
+  %c0 = arith.constant 0 : index
+  %step = arith.constant 8 : index
+  %0 = scf.for %iv = %c0 to %ub step %step iter_args(%arg = %c0) -> (index) {
+    %1 = affine.min #map(%ub, %iv)[%step]
+    %2 = index.add %1, %arg
+    scf.yield %2 : index
+  }
+  return %0 : index
+}
+
+// CHECK: #[[MAP:.*]] = affine_map<()[s0, s1, s2] -> (s1 - s1 mod s2)>
+// CHECK: func.func @foo(%[[UB:.*]]: index) -> index {
+// CHECK: %[[STEP8:.*]] = arith.constant 8 : index
+// CHECK: %[[STEP4:.*]] = arith.constant 4 : index
+// CHECK: %[[STEP2:.*]] = arith.constant 2 : index
+// CHECK: %[[STEP1:.*]] = arith.constant 1 : index
+// CHECK: %[[LB:.*]] = arith.constant 0 : index
+// CHECK: %[[I0:.*]] = affine.apply #[[MAP]]()[%[[LB]], %[[UB]], %[[STEP8]]]
+// CHECK: %[[I1:.*]] = scf.for %{{.*}} = %[[LB]] to %[[I0]] step %[[STEP8]] iter_args(%[[ALB:.*]] = %[[LB]]) -> (index) {
+// CHECK: %[[SUM:.*]] = index.add %[[ALB]], %[[STEP8]]
+// CHECK: scf.yield %[[SUM]] : index
+// CHECK: %[[I2:.*]] = affine.apply #[[MAP]]()[%[[I0]], %[[UB]], %[[STEP4]]]
----------------
matthias-springer wrote:

Please indent for better readability.


https://github.com/llvm/llvm-project/pull/71555