[AArch64] Fix vectorToScalarBitmask BE (#156314)

Closes #156312

Author: GiuseppeCesarano

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -24329,14 +24329,18 @@ static SDValue vectorToScalarBitmask(SDNode *N, SelectionDAG &DAG) {
   // Ensure that all elements' bits are either 0s or 1s.
   ComparisonResult = DAG.getSExtOrTrunc(ComparisonResult, DL, VecVT);
 
+  bool IsLE = DAG.getDataLayout().isLittleEndian();
   SmallVector<SDValue, 16> MaskConstants;
   if (DAG.getSubtarget<AArch64Subtarget>().isNeonAvailable() &&
       VecVT == MVT::v16i8) {
     // v16i8 is a special case, as we have 16 entries but only 8 positional bits
     // per entry. We split it into two halves, apply the mask, zip the halves to
     // create 8x 16-bit values, and the perform the vector reduce.
     for (unsigned Half = 0; Half < 2; ++Half) {
-      for (unsigned MaskBit = 1; MaskBit <= 128; MaskBit *= 2) {
+      for (unsigned I = 0; I < 8; ++I) {
+        // On big-endian targets, the lane order in sub-byte vector elements
+        // gets reversed, so we need to flip the bit index.
+        unsigned MaskBit = IsLE ? (1u << I) : (1u << (7 - I));
         MaskConstants.push_back(DAG.getConstant(MaskBit, DL, MVT::i32));
       }
     }
@@ -24354,8 +24358,9 @@ static SDValue vectorToScalarBitmask(SDNode *N, SelectionDAG &DAG) {
   }
 
   // All other vector sizes.
-  unsigned MaxBitMask = 1u << (VecVT.getVectorNumElements() - 1);
-  for (unsigned MaskBit = 1; MaskBit <= MaxBitMask; MaskBit *= 2) {
+  unsigned NumEl = VecVT.getVectorNumElements();
+  for (unsigned I = 0; I < NumEl; ++I) {
+    unsigned MaskBit = IsLE ? (1u << I) : (1u << (NumEl - 1 - I));
     MaskConstants.push_back(DAG.getConstant(MaskBit, DL, MVT::i64));
   }
 

llvm/test/CodeGen/AArch64/vector-to-scalar-bitmask.ll

@@ -0,0 +1,75 @@
+; RUN: llc -O3 -mtriple=aarch64-linux-gnu %s -o - | FileCheck %s --check-prefix=CHECK-LE
+; RUN: llc -O3 -mtriple=aarch64_be-linux-gnu %s -o - | FileCheck %s --check-prefix=CHECK-BE
+
+define i16 @convert_to_bitmask16(<16 x i8> %vec) {
+  %cmp_result = icmp ne <16 x i8> %vec, zeroinitializer
+  %bitmask = bitcast <16 x i1> %cmp_result to i16
+  ret i16 %bitmask
+}
+
+define i16 @convert_to_bitmask8(<8 x i16> %vec) {
+  %cmp_result = icmp ne <8 x i16> %vec, zeroinitializer
+  %bitmask = bitcast <8 x i1> %cmp_result to i8
+  %extended_bitmask = zext i8 %bitmask to i16
+  ret i16 %extended_bitmask
+}
+
+; Little endian
+
+; CHECK-LE-LABEL: .LCPI0_0:
+; CHECK-LE-NEXT: .byte 1
+; CHECK-LE-NEXT: .byte 2
+; CHECK-LE-NEXT: .byte 4
+; CHECK-LE-NEXT: .byte 8
+; CHECK-LE-NEXT: .byte 16
+; CHECK-LE-NEXT: .byte 32
+; CHECK-LE-NEXT: .byte 64
+; CHECK-LE-NEXT: .byte 128
+; CHECK-LE-NEXT: .byte 1
+; CHECK-LE-NEXT: .byte 2
+; CHECK-LE-NEXT: .byte 4
+; CHECK-LE-NEXT: .byte 8
+; CHECK-LE-NEXT: .byte 16
+; CHECK-LE-NEXT: .byte 32
+; CHECK-LE-NEXT: .byte 64
+; CHECK-LE-NEXT: .byte 128
+
+; CHECK-LE-LABEL: .LCPI1_0:
+; CHECK-LE-NEXT: .hword 1
+; CHECK-LE-NEXT: .hword 2
+; CHECK-LE-NEXT: .hword 4
+; CHECK-LE-NEXT: .hword 8
+; CHECK-LE-NEXT: .hword 16
+; CHECK-LE-NEXT: .hword 32
+; CHECK-LE-NEXT: .hword 64
+; CHECK-LE-NEXT: .hword 128
+
+; Big endian
+
+; CHECK-BE-LABEL: .LCPI0_0:
+; CHECK-BE-NEXT: .byte 128
+; CHECK-BE-NEXT: .byte 64
+; CHECK-BE-NEXT: .byte 32
+; CHECK-BE-NEXT: .byte 16
+; CHECK-BE-NEXT: .byte 8
+; CHECK-BE-NEXT: .byte 4
+; CHECK-BE-NEXT: .byte 2
+; CHECK-BE-NEXT: .byte 1
+; CHECK-BE-NEXT: .byte 128
+; CHECK-BE-NEXT: .byte 64
+; CHECK-BE-NEXT: .byte 32
+; CHECK-BE-NEXT: .byte 16
+; CHECK-BE-NEXT: .byte 8
+; CHECK-BE-NEXT: .byte 4
+; CHECK-BE-NEXT: .byte 2
+; CHECK-BE-NEXT: .byte 1
+
+; CHECK-BE-LABEL: .LCPI1_0:
+; CHECK-BE-NEXT: .hword 128
+; CHECK-BE-NEXT: .hword 64
+; CHECK-BE-NEXT: .hword 32
+; CHECK-BE-NEXT: .hword 16
+; CHECK-BE-NEXT: .hword 8
+; CHECK-BE-NEXT: .hword 4
+; CHECK-BE-NEXT: .hword 2
+; CHECK-BE-NEXT: .hword 1