[FuncSpec] Handle ssa_copy intrinsic calls in InstCostVisitor

[hazzlim]

Look through ssa_copy intrinsic calls when computing codesize bonus for a specialization.

Also remove redundant logic to skip computing codesize bonus for ssa_copy intrinsics, now these are considered zero-cost by TTI (in PR #75294).

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-function-specialization

Author: Hari Limaye (hazzlim)

Changes

Look through ssa_copy intrinsic calls when computing codesize bonus for a specialization.

Also remove redundant logic to skip computing codesize bonus for ssa_copy intrinsics, now these are considered zero-cost by TTI (in PR#75294).

---

Full diff: https://github.com/llvm/llvm-project/pull/114247.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/IPO/FunctionSpecialization.cpp (+8-4)

diff --git a/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp b/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
index 919d3143a13f7e..1efec22624dd4f 100644
--- a/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
+++ b/llvm/lib/Transforms/IPO/FunctionSpecialization.cpp
@@ -120,10 +120,6 @@ Cost InstCostVisitor::estimateBasicBlocks(
       continue;
 
     for (Instruction &I : *BB) {
-      // Disregard SSA copies.
-      if (auto *II = dyn_cast<IntrinsicInst>(&I))
-        if (II->getIntrinsicID() == Intrinsic::ssa_copy)
-          continue;
       // If it's a known constant we have already accounted for it.
       if (KnownConstants.contains(&I))
         continue;
@@ -402,6 +398,14 @@ Constant *InstCostVisitor::visitFreezeInst(FreezeInst &I) {
 }
 
 Constant *InstCostVisitor::visitCallBase(CallBase &I) {
+  assert(LastVisited != KnownConstants.end() && "Invalid iterator!");
+
+  // Look through calls to ssa_copy intrinsics.
+  if (auto *II = dyn_cast<IntrinsicInst>(&I);
+      II && II->getIntrinsicID() == Intrinsic::ssa_copy) {
+    return LastVisited->second;
+  }
+
   Function *F = I.getCalledFunction();
   if (!F || !canConstantFoldCallTo(&I, F))
     return nullptr;

[labrinea]

Great thanks! Unit/regression test?

[hazzlim]

Great thanks! Unit/regression test?

Added a regression test, as a prior commit so it's more obvious how this change affects things :)

As additional context for this change, this improves the estimation of the CodeSize bonus for a specialization of the function digits_2 from the SPEC2k17 benchmark exchange2, increasing this from 43% to 68% (with -mcpu=neoverse-{n1|v2}).