SimdUtilTest.gatherBits does not work for arm64(M1)

[icejoywoo]

Bug description

UT SimdUtilTest.gatherBits(added in #7726) is not passed in Mac M1 arm64, but can pass in Mac Intel. This also impact the PrestoVectorSerde, because only PrestoVectorSerializer use this function to handle nulls(see #7883).

I dig into this issue, and I tested on my 2 Macbook with intel chip and M1. Then I found that simd::gatherBits only works for xsimd::batch<int32_t>::size is 8 in Intel avx2, but it does NOT work for xsimd::batch<int32_t>::size is 4 in ARM neon.

I think the current implementation of simd::gatherBits is not correct for arm64 M1.

In #1505, function nonNullRowsFromSparse in DecoderUtil.cpp handle the different situtation with xsimd::batch<int32_t>::size = 4 or 8.

The code snippet in DecoderUtil.cpp follows:

constexpr int32_t kStep = xsimd::batch<int32_t>::size;
// ...
uint16_t flags = simd::gather8Bits(nulls, next8Rows, width);
if (outputNulls) {
  if constexpr (kStep == 8) {
    resultNullBytes[i / 8] = flags;
  } else {
    VELOX_DCHECK_EQ(kStep, 4);
    if (i % 8 == 0) {
      resultNullBytes[i / 8] = flags;
    } else {
      resultNullBytes[i / 8] |= flags << 4;
    }
  }
// ...
}

According to this snippet, I think this can fix this issue too.

System information

Velox System Info v0.0.2
Commit: ca8a0413aba800be0da7a0ece3d9c2ea26d6b7bf
CMake Version: 3.27.3
System: Darwin-22.5.0
Arch: arm64
C++ Compiler: /Library/Developer/CommandLineTools/usr/bin/c++
C++ Compiler Version: 14.0.3.14030022
C Compiler: /Library/Developer/CommandLineTools/usr/bin/cc
C Compiler Version: 14.0.3.14030022
CMake Prefix Path: /Library/Developer/CommandLineTools/SDKs/MacOSX13.3.sdk/usr;/opt/homebrew;/usr/local;/usr;/;/opt/homebrew/Cellar/cmake/3.27.3;/usr/local;/usr/X11R6;/usr/pkg;/opt;/sw;/opt/local

Relevant logs

No response

[icejoywoo]

@mbasmanova Could you please check this issue? I found an issue in simd::gatherBits. I already push a fix to this.

[mbasmanova]

CC: @oerling @Yuhta