[AArch64] Use tbl for truncating vector FPtoUI conversions.

On AArch64, doing the vector truncate separately after the fptoui
conversion can be lowered more efficiently using tbl.4, building on
D133495.

https://alive2.llvm.org/ce/z/T538CC

Depends on D133495

Reviewed By: t.p.northover

Differential Revision: https://reviews.llvm.org/D133496

Author: fhahn

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -8047,7 +8047,7 @@ bool CodeGenPrepare::optimizeInst(Instruction *I, ModifyDT &ModifiedDT) {
     if (OptimizeNoopCopyExpression(CI, *TLI, *DL))
       return true;
 
-    if ((isa<UIToFPInst>(I) || isa<TruncInst>(I)) &&
+    if ((isa<UIToFPInst>(I) || isa<FPToUIInst>(I) || isa<TruncInst>(I)) &&
         TLI->optimizeExtendOrTruncateConversion(I,
                                                 LI->getLoopFor(I->getParent())))
       return true;

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -13289,6 +13289,23 @@ bool AArch64TargetLowering::optimizeExtendOrTruncateConversion(Instruction *I,
     return true;
   }
 
+  // Convert 'fptoui <(8|16) x float> to <(8|16) x i8>' to a wide fptoui
+  // followed by a truncate lowered to using tbl.4.
+  auto *FPToUI = dyn_cast<FPToUIInst>(I);
+  if (FPToUI &&
+      (SrcTy->getNumElements() == 8 || SrcTy->getNumElements() == 16) &&
+      SrcTy->getElementType()->isFloatTy() &&
+      DstTy->getElementType()->isIntegerTy(8)) {
+    IRBuilder<> Builder(I);
+    auto *WideConv = Builder.CreateFPToUI(FPToUI->getOperand(0),
+                                          VectorType::getInteger(SrcTy));
+    auto *TruncI = Builder.CreateTrunc(WideConv, DstTy);
+    I->replaceAllUsesWith(TruncI);
+    I->eraseFromParent();
+    createTblForTrunc(cast<TruncInst>(TruncI), Subtarget->isLittleEndian());
+    return true;
+  }
+
   // Convert 'trunc <(8|16) x i32> %x to <(8|16) x i8>' to a single tbl.4
   // instruction selecting the lowest 8 bits per lane of the input interpreted
   // as 2 or 4 <4 x i32> vectors.