fixup! [MachinePipeliner] Improve loop carried dependence analysis

Author: ytmukai

llvm/lib/CodeGen/MachinePipeliner.cpp

@@ -2824,6 +2824,10 @@ bool SwingSchedulerDAG::mayOverlapInLaterIter(
            << ", Len: " << AccessSizeO.getValue() << "\n";
   });
 
+  // Excessive overlap may be detected in strided patterns.
+  // For example, the memory addresses of the store and the load in
+  //   for (i=0; i<n; i+=2) a[i+1] = a[i];
+  // are assumed to overlap.
   if (Delta < 0) {
     int64_t BaseMinAddr = OffsetB;
     int64_t OhterNextIterMaxAddr = OffsetO + Delta + AccessSizeO.getValue() - 1;

llvm/test/CodeGen/Hexagon/swp-carried-dep2.mir

@@ -1,16 +1,27 @@
 # RUN: llc -mtriple=hexagon -run-pass pipeliner -debug-only=pipeliner %s -o /dev/null 2>&1 -pipeliner-experimental-cg=true | FileCheck %s
 # REQUIRES: asserts
 
-# Test that it correctly recognizes that there is no loop carried dependence
+# Test that the loop carried dependence check correctly identifies dependences
 # when the loop variable decreases and the array index offset is negative.
 
+# No dependence from the store to the load.
 # CHECK: Overlap check:
 # CHECK-NEXT:   BaseMI:   S2_storeri_io %{{[0-9]+}}:intregs, 0, %{{[0-9]+}}:intregs :: (store (s32) into %ir.lsr.iv1)
 # CHECK-NEXT:     Base + 0 + I * -4, Len: 4
 # CHECK-NEXT:   OtherMI:   %{{[0-9]+}}:intregs = L2_loadri_io %{{[0-9]+}}:intregs, -8 :: (load (s32) from %ir.cgep)
 # CHECK-NEXT:     Base + -8 + I * -4, Len: 4
 # CHECK-NEXT:   Result: No overlap
 
+# TODO: There is a loop carried dependence from the load to the store but it
+#   is not recognised. addLoopCarriedDependences() should be modified to
+#   recognise the dependence and enable the following checks.
+# CHECK-AFTER-FIX: Overlap check:
+# CHECK-AFTER-FIX-NEXT:   BaseMI:   %{{[0-9]+}}:intregs = L2_loadri_io %{{[0-9]+}}:intregs, -8 :: (load (s32) from %ir.cgep)
+# CHECK-AFTER-FIX-NEXT:     Base + -8 + I * -4, Len: 4
+# CHECK-AFTER-FIX-NEXT:   OtherMI:   S2_storeri_io %{{[0-9]+}}:intregs, 0, %{{[0-9]+}}:intregs :: (store (s32) into %ir.lsr.iv1)
+# CHECK-AFTER-FIX-NEXT:     Base + 0 + I * -4, Len: 4
+# CHECK-AFTER-FIX-NEXT:   Result: Overlap!
+
 --- |
 
   define void @test() {