[llvm] [GISel][SDAG] Avoid push_back in loops for some shuffle mask handling. (PR #119434)

[Craig Topper via llvm-commits]

https://github.com/topperc created https://github.com/llvm/llvm-project/pull/119434

Each call to push_back contains a check to see if the vector needs to grow. Using resize or the giving the size to the constructor can reduce the number of checks for growing.

>From b8bc3d826bb7c489a720ff8630b4df9b26743e1c Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Tue, 10 Dec 2024 10:45:01 -0800
Subject: [PATCH] [GISel][SDAG] Avoid push_back in loops for some shuffle mask
 handling.

Each call to push_back contains a check to see if the vector needs
to grow. Using resize or the giving the size to the constructor can
reduce the number of checks for growing.
---
 llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp | 11 ++++-------
 .../SelectionDAG/LegalizeVectorTypes.cpp        | 17 ++++++-----------
 2 files changed, 10 insertions(+), 18 deletions(-)

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 2544f2479ddc68..6676cee8f618a8 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -6173,8 +6173,7 @@ LegalizerHelper::equalizeVectorShuffleLengths(MachineInstr &MI) {
     // Extend mask to match new destination vector size with
     // undef values.
     SmallVector<int, 16> NewMask(Mask);
-    for (unsigned I = MaskNumElts; I < SrcNumElts; ++I)
-      NewMask.push_back(-1);
+    NewMask.resize(SrcNumElts, -1);
 
     moreElementsVectorDst(MI, SrcTy, 0);
     MIRBuilder.setInstrAndDebugLoc(MI);
@@ -6254,16 +6253,14 @@ LegalizerHelper::moreElementsVectorShuffle(MachineInstr &MI,
   moreElementsVectorSrc(MI, MoreTy, 2);
 
   // Adjust mask based on new input vector length.
-  SmallVector<int, 16> NewMask;
+  SmallVector<int, 16> NewMask(WidenNumElts, -1);
   for (unsigned I = 0; I != NumElts; ++I) {
     int Idx = Mask[I];
     if (Idx < static_cast<int>(NumElts))
-      NewMask.push_back(Idx);
+      NewMask[I] = Idx;
     else
-      NewMask.push_back(Idx - NumElts + WidenNumElts);
+      NewMask[I] = Idx - NumElts + WidenNumElts;
   }
-  for (unsigned I = NumElts; I != WidenNumElts; ++I)
-    NewMask.push_back(-1);
   moreElementsVectorDst(MI, MoreTy, 0);
   MIRBuilder.setInstrAndDebugLoc(MI);
   MIRBuilder.buildShuffleVector(MI.getOperand(0).getReg(),
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 465128099f4447..205ad899e4c278 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -6421,16 +6421,14 @@ SDValue DAGTypeLegalizer::WidenVecRes_VECTOR_SHUFFLE(ShuffleVectorSDNode *N) {
   SDValue InOp2 = GetWidenedVector(N->getOperand(1));
 
   // Adjust mask based on new input vector length.
-  SmallVector<int, 16> NewMask;
+  SmallVector<int, 16> NewMask(WidenNumElts, -1);
   for (unsigned i = 0; i != NumElts; ++i) {
     int Idx = N->getMaskElt(i);
     if (Idx < (int)NumElts)
-      NewMask.push_back(Idx);
+      NewMask[i] = Idx;
     else
-      NewMask.push_back(Idx - NumElts + WidenNumElts);
+      NewMask[i] = Idx - NumElts + WidenNumElts;
   }
-  for (unsigned i = NumElts; i != WidenNumElts; ++i)
-    NewMask.push_back(-1);
   return DAG.getVectorShuffle(WidenVT, dl, InOp1, InOp2, NewMask);
 }
 
@@ -6478,12 +6476,9 @@ SDValue DAGTypeLegalizer::WidenVecRes_VECTOR_REVERSE(SDNode *N) {
 
   // Use VECTOR_SHUFFLE to combine new vector from 'ReverseVal' for
   // fixed-vectors.
-  SmallVector<int, 16> Mask;
-  for (unsigned i = 0; i != VTNumElts; ++i) {
-    Mask.push_back(IdxVal + i);
-  }
-  for (unsigned i = VTNumElts; i != WidenNumElts; ++i)
-    Mask.push_back(-1);
+  SmallVector<int, 16> Mask(VTNumElts, -1);
+  for (unsigned i = 0; i != VTNumElts; ++i)
+    Mask[i] = IdxVal + i;
 
   return DAG.getVectorShuffle(WidenVT, dl, ReverseVal, DAG.getUNDEF(WidenVT),
                               Mask);