[llvm] [AArch64] Refactor creation of a shuffle mask for TBL (NFC) (PR #92529)

[Momchil Velikov via llvm-commits]

https://github.com/momchil-velikov updated https://github.com/llvm/llvm-project/pull/92529

>From 6fe74a46fa45701c0039b2d5c31f01732fc6c883 Mon Sep 17 00:00:00 2001
From: Momchil Velikov <momchil.velikov at arm.com>
Date: Fri, 17 May 2024 11:50:31 +0100
Subject: [PATCH 1/2] [AArch64] Refactor creation of a shuffle mask for TBL
 (NFC)

---
 .../Target/AArch64/AArch64ISelLowering.cpp    | 88 +++++++++++--------
 1 file changed, 50 insertions(+), 38 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index c4f819f5fcdd2..e209bbcc478ce 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -15854,48 +15854,51 @@ bool AArch64TargetLowering::shouldSinkOperands(
   return false;
 }
 
-static bool createTblShuffleForZExt(ZExtInst *ZExt, FixedVectorType *DstTy,
-                                    bool IsLittleEndian) {
-  Value *Op = ZExt->getOperand(0);
-  auto *SrcTy = cast<FixedVectorType>(Op->getType());
-  auto SrcWidth = cast<IntegerType>(SrcTy->getElementType())->getBitWidth();
-  auto DstWidth = cast<IntegerType>(DstTy->getElementType())->getBitWidth();
+static bool createTblShuffleMask(unsigned SrcWidth, unsigned DstWidth,
+                                 unsigned NumElts, bool IsLittleEndian,
+                                 SmallVectorImpl<int> &Mask) {
   if (DstWidth % 8 != 0 || DstWidth <= 16 || DstWidth >= 64)
     return false;
 
-  assert(DstWidth % SrcWidth == 0 &&
-         "TBL lowering is not supported for a ZExt instruction with this "
-         "source & destination element type.");
-  unsigned ZExtFactor = DstWidth / SrcWidth;
+  if (DstWidth % SrcWidth != 0)
+    return false;
+
+  unsigned Factor = DstWidth / SrcWidth;
+  unsigned MaskLen = NumElts * Factor;
+
+  Mask.clear();
+  Mask.resize(MaskLen, NumElts);
+
+  unsigned SrcIndex = 0;
+  for (unsigned I = 0; I < MaskLen; I += Factor)
+    Mask[I] = SrcIndex++;
+
+  if (!IsLittleEndian)
+    std::rotate(Mask.rbegin(), Mask.rbegin() + Factor - 1, Mask.rend());
+
+  return true;
+}
+
+static Value *createTblShuffleForZExt(IRBuilderBase &Builder, Value *Op,
+                                      FixedVectorType *ZExtTy,
+                                      FixedVectorType *DstTy,
+                                      bool IsLittleEndian) {
+  auto *SrcTy = cast<FixedVectorType>(Op->getType());
   unsigned NumElts = SrcTy->getNumElements();
-  IRBuilder<> Builder(ZExt);
+  auto SrcWidth = cast<IntegerType>(SrcTy->getElementType())->getBitWidth();
+  auto DstWidth = cast<IntegerType>(DstTy->getElementType())->getBitWidth();
+
   SmallVector<int> Mask;
-  // Create a mask that selects <0,...,Op[i]> for each lane of the destination
-  // vector to replace the original ZExt. This can later be lowered to a set of
-  // tbl instructions.
-  for (unsigned i = 0; i < NumElts * ZExtFactor; i++) {
-    if (IsLittleEndian) {
-      if (i % ZExtFactor == 0)
-        Mask.push_back(i / ZExtFactor);
-      else
-        Mask.push_back(NumElts);
-    } else {
-      if ((i + 1) % ZExtFactor == 0)
-        Mask.push_back((i - ZExtFactor + 1) / ZExtFactor);
-      else
-        Mask.push_back(NumElts);
-    }
-  }
+  if (!createTblShuffleMask(SrcWidth, DstWidth, NumElts, IsLittleEndian, Mask))
+    return nullptr;
 
   auto *FirstEltZero = Builder.CreateInsertElement(
       PoisonValue::get(SrcTy), Builder.getInt8(0), uint64_t(0));
   Value *Result = Builder.CreateShuffleVector(Op, FirstEltZero, Mask);
   Result = Builder.CreateBitCast(Result, DstTy);
-  if (DstTy != ZExt->getType())
-    Result = Builder.CreateZExt(Result, ZExt->getType());
-  ZExt->replaceAllUsesWith(Result);
-  ZExt->eraseFromParent();
-  return true;
+  if (DstTy != ZExtTy)
+    Result = Builder.CreateZExt(Result, ZExtTy);
+  return Result;
 }
 
 static void createTblForTrunc(TruncInst *TI, bool IsLittleEndian) {
@@ -16060,21 +16063,30 @@ bool AArch64TargetLowering::optimizeExtendOrTruncateConversion(
 
       DstTy = TruncDstType;
     }
-
-    return createTblShuffleForZExt(ZExt, DstTy, Subtarget->isLittleEndian());
+    IRBuilder<> Builder(ZExt);
+    Value *Result = createTblShuffleForZExt(
+        Builder, ZExt->getOperand(0), cast<FixedVectorType>(ZExt->getType()),
+        DstTy, Subtarget->isLittleEndian());
+    if (!Result)
+      return false;
+    ZExt->replaceAllUsesWith(Result);
+    ZExt->eraseFromParent();
+    return true;
   }
 
   auto *UIToFP = dyn_cast<UIToFPInst>(I);
   if (UIToFP && SrcTy->getElementType()->isIntegerTy(8) &&
       DstTy->getElementType()->isFloatTy()) {
     IRBuilder<> Builder(I);
-    auto *ZExt = cast<ZExtInst>(
-        Builder.CreateZExt(I->getOperand(0), VectorType::getInteger(DstTy)));
+    Value *ZExt = createTblShuffleForZExt(
+        Builder, I->getOperand(0), FixedVectorType::getInteger(DstTy),
+        FixedVectorType::getInteger(DstTy), Subtarget->isLittleEndian());
+    if (!ZExt)
+      return false;
     auto *UI = Builder.CreateUIToFP(ZExt, DstTy);
     I->replaceAllUsesWith(UI);
     I->eraseFromParent();
-    return createTblShuffleForZExt(ZExt, cast<FixedVectorType>(ZExt->getType()),
-                                   Subtarget->isLittleEndian());
+    return true;
   }
 
   // Convert 'fptoui <(8|16) x float> to <(8|16) x i8>' to a wide fptoui

>From 1fc66ab684a2e95456968e0788feab8315c06762 Mon Sep 17 00:00:00 2001
From: Momchil Velikov <momchil.velikov at arm.com>
Date: Tue, 28 May 2024 14:35:33 +0100
Subject: [PATCH 2/2] [fixup] Simplify code

Change-Id: I0e51f00113d0692d5d3751c89eac8167511d1d9e
---
 llvm/lib/Target/AArch64/AArch64ISelLowering.cpp | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index e209bbcc478ce..c78b8c8a6fbbf 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -15870,12 +15870,9 @@ static bool createTblShuffleMask(unsigned SrcWidth, unsigned DstWidth,
   Mask.resize(MaskLen, NumElts);
 
   unsigned SrcIndex = 0;
-  for (unsigned I = 0; I < MaskLen; I += Factor)
+  for (unsigned I = IsLittleEndian ? 0 : Factor - 1; I < MaskLen; I += Factor)
     Mask[I] = SrcIndex++;
 
-  if (!IsLittleEndian)
-    std::rotate(Mask.rbegin(), Mask.rbegin() + Factor - 1, Mask.rend());
-
   return true;
 }