From 44cdc9e5e20b9ca122bfb932f8e5d0308a37cb36 Mon Sep 17 00:00:00 2001
From: icejoywoo <icejoywoo@gmail.com>
Date: Thu, 18 Jan 2024 12:39:13 -0800
Subject: [PATCH] Fix simd::gatherBits for Mac M1 or when AVX2 is disabled
 (#8415)

Summary:
Fix https://github.com/facebookincubator/velox/issues/8377

When avx2 is disabled or run on Mac M1(arm64), simd::gatherBits works incorrectly.

This fix comes from DecoderUtil::nonNullRowsFromSparse.

Pull Request resolved: https://github.com/facebookincubator/velox/pull/8415

Reviewed By: xiaoxmeng

Differential Revision: D52873543

Pulled By: Yuhta

fbshipit-source-id: ce8cbeb2069a809410b7a259e05285be2e1a70b5
---
 velox/common/base/SimdUtil.cpp | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/velox/common/base/SimdUtil.cpp b/velox/common/base/SimdUtil.cpp
index 6b84435dd893..03576ac31ec4 100644
--- a/velox/common/base/SimdUtil.cpp
+++ b/velox/common/base/SimdUtil.cpp
@@ -23,6 +23,7 @@ void gatherBits(
     const uint64_t* bits,
     folly::Range<const int32_t*> indexRange,
     uint64_t* result) {
+  constexpr int32_t kStep = xsimd::batch<int32_t>::size;
   const auto size = indexRange.size();
   auto indices = indexRange.data();
   uint8_t* resultPtr = reinterpret_cast<uint8_t*>(result);
@@ -37,14 +38,16 @@ void gatherBits(
   }
 
   int32_t i = 0;
-  for (; i + 8 < size; i += 8) {
-    *(resultPtr++) =
-        simd::gather8Bits(bits, xsimd::load_unaligned(indices + i), 8);
+  for (; i + kStep < size; i += kStep) {
+    uint16_t flags =
+        simd::gather8Bits(bits, xsimd::load_unaligned(indices + i), kStep);
+    bits::storeBitsToByte<kStep>(flags, resultPtr, i);
   }
   const auto bitsLeft = size - i;
   if (bitsLeft > 0) {
-    *resultPtr =
+    uint16_t flags =
         simd::gather8Bits(bits, xsimd::load_unaligned(indices + i), bitsLeft);
+    bits::storeBitsToByte<kStep>(flags, resultPtr, i);
   }
 }