[llvm] [AArch64] Skip if then else diamonds (PR #82135)

[via llvm-commits]

https://github.com/AtariDreams updated https://github.com/llvm/llvm-project/pull/82135

>From 63118ed96177841ccc486ccd69572c5c11a13780 Mon Sep 17 00:00:00 2001
From: Rose <83477269+AtariDreams at users.noreply.github.com>
Date: Sat, 17 Feb 2024 17:14:49 -0500
Subject: [PATCH] [AArch64] Skip if then else diamonds

---
 .../AArch64/AArch64TargetTransformInfo.cpp    | 28 +++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index 010e569809e276..548fca17b158b8 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -3324,6 +3324,29 @@ unsigned AArch64TTIImpl::getMaxInterleaveFactor(ElementCount VF) {
   return ST->getMaxInterleaveFactor();
 }
 
+static bool isPartOfIfThenElseDiamond(const BasicBlock *BB) {
+  if (!BB)
+    return false;
+  auto *BI = dyn_cast<BranchInst>(BB->getTerminator());
+  if (!BI || !BI->isConditional())
+    return false;
+
+  BasicBlock *Succ0 = BI->getSuccessor(0);
+  BasicBlock *Succ1 = BI->getSuccessor(1);
+
+  if (!Succ0->getSinglePredecessor())
+    return false;
+  if (!Succ1->getSinglePredecessor())
+    return false;
+
+  BasicBlock *Succ0Succ = Succ0->getSingleSuccessor();
+  BasicBlock *Succ1Succ = Succ1->getSingleSuccessor();
+  // Ignore triangles.
+  if (!Succ0Succ || !Succ1Succ || Succ0Succ != Succ1Succ)
+    return false;
+  return true;
+}
+
 // For Falkor, we want to avoid having too many strided loads in a loop since
 // that can exhaust the HW prefetcher resources.  We adjust the unroller
 // MaxCount preference below to attempt to ensure unrolling doesn't create too
@@ -3334,9 +3357,10 @@ getFalkorUnrollingPreferences(Loop *L, ScalarEvolution &SE,
   enum { MaxStridedLoads = 7 };
   auto countStridedLoads = [](Loop *L, ScalarEvolution &SE) {
     int StridedLoads = 0;
-    // FIXME? We could make this more precise by looking at the CFG and
-    // e.g. not counting loads in each side of an if-then-else diamond.
     for (const auto BB : L->blocks()) {
+      // Skip blocks that are part of an if-then-else diamond
+      if (isPartOfIfThenElseDiamond(BB))
+        continue;
       for (auto &I : *BB) {
         LoadInst *LMemI = dyn_cast<LoadInst>(&I);
         if (!LMemI)