[AsmPrinter] Fix bit pattern for i1 vectors.

[efriedma]

Summary

Vectors are defined to be tightly packed, regardless of the element type. The AsmPrinter didn't realize this, and was allocating padding between elements.

Fixes https://github.com/llvm/llvm-project/issues/55522 .

[dmgreen]

LGTM. I think I've seen this other issues from i1 constant vectors somewhere, but I can't find a reference to them now. 53246 maybe, but I thought there was another one.

llvm/test/CodeGen/AArch64/vector-global-i1.ll

1

This can remove the --check-prefixes=CHECK.

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

static void emitGlobalConstantArray(const DataLayout &DL,
   else {
     for (unsigned i = 0, e = CA->getNumOperands(); i != e; ++i) {
       emitGlobalConstantImpl(DL, CA->getOperand(i), AP, BaseCV, Offset);
       Offset += DL.getTypeAllocSize(CA->getOperand(i)->getType());
     }
   }
 }
 
+static void emitGlobalConstantLargeInt(const ConstantInt *CI, AsmPrinter &AP);
+
 static void emitGlobalConstantVector(const DataLayout &DL,
                                      const ConstantVector *CV, AsmPrinter &AP) {
+  Type *ElementType = CV->getType()->getElementType();
+  uint64_t ElementSizeInBits = DL.getTypeSizeInBits(ElementType);
+  uint64_t ElementAllocSizeInBits = DL.getTypeAllocSizeInBits(ElementType);
+  uint64_t EmittedSize;
+  if (ElementSizeInBits != ElementAllocSizeInBits) {
+    // If the allocation size of an element is different from the size in bits,
+    // printing each element separately will insert incorrect padding.
+    //
+    // The general algorithm here is complicated; instead of writing it out
+    // here, just use the existing code in ConstantFolding.
+    Type *IntT =
+        IntegerType::get(CV->getContext(), DL.getTypeSizeInBits(CV->getType()));
+    ConstantInt *CI = dyn_cast_or_null<ConstantInt>(ConstantFoldConstant(
+        ConstantExpr::getBitCast(const_cast<ConstantVector *>(CV), IntT), DL));
+    if (!CI) {
+      report_fatal_error(
+          "Cannot lower vector global with unusual element type");
+    }
+    emitGlobalConstantLargeInt(CI, AP);
+    EmittedSize = DL.getTypeStoreSize(CV->getType());
+  } else {
     for (unsigned i = 0, e = CV->getType()->getNumElements(); i != e; ++i)
       emitGlobalConstantImpl(DL, CV->getOperand(i), AP);
+    EmittedSize =
+        DL.getTypeAllocSize(ElementType) * CV->getType()->getNumElements();
+  }
 
   unsigned Size = DL.getTypeAllocSize(CV->getType());
-  unsigned EmittedSize = DL.getTypeAllocSize(CV->getType()->getElementType()) *
-                         CV->getType()->getNumElements();
   if (unsigned Padding = Size - EmittedSize)
     AP.OutStreamer->emitZeros(Padding);
 }
 
 static void emitGlobalConstantStruct(const DataLayout &DL,
                                      const ConstantStruct *CS, AsmPrinter &AP,
                                      const Constant *BaseCV, uint64_t Offset) {
   // Print the fields in successive locations. Pad to align if needed!

llvm/test/CodeGen/AArch64/vector-global-i1.ll

+; RUN: llc < %s -mtriple aarch64 | FileCheck --check-prefixes=CHECK %s
+
+; CHECK: a:
+; CHECK-NEXT: .zero 1
+@a = internal constant <4 x i1> <i1 false, i1 false, i1 false, i1 false>
+; CHECK: b:
+; CHECK-NEXT: .byte 5
+@b = internal constant <4 x i1> <i1 true, i1 false, i1 true, i1 false>
+; CHECK: c:
+; CHECK-NEXT: .hword 1
+; CHECK-NEXT: .byte 0
+; CHECK-NEXT: .zero 1
+@c = internal constant <24 x i1> <i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>
+