mapleFU commented on code in PR #40335:
URL: https://github.com/apache/arrow/pull/40335#discussion_r1516607762
##########
cpp/src/arrow/util/byte_stream_split_internal.h:
##########
@@ -139,40 +146,72 @@ void ByteStreamSplitEncodeSse2(const uint8_t* raw_values,
const int64_t num_valu
for (int stage_lvl = 0; stage_lvl < 2; ++stage_lvl) {
for (int i = 0; i < kNumStreams / 2; ++i) {
stage[stage_lvl + 1][i * 2] =
- _mm_unpacklo_epi8(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2
+ 1]);
+ xsimd::zip_lo(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2 +
1]);
stage[stage_lvl + 1][i * 2 + 1] =
- _mm_unpackhi_epi8(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2
+ 1]);
+ xsimd::zip_hi(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2 +
1]);
}
}
if constexpr (kNumStreams == 8) {
- // This is the path for double.
- __m128i tmp[8];
+ // This is the path for 64bits data.
+ simd_batch tmp[8];
+ using int32_batch = xsimd::batch<int32_t, simd_arch>;
+ // This is a workaround, see:
https://github.com/xtensor-stack/xsimd/issues/735
Review Comment:
Using memcpy as a workaround.
##########
cpp/src/arrow/util/byte_stream_split_internal.h:
##########
@@ -139,40 +146,72 @@ void ByteStreamSplitEncodeSse2(const uint8_t* raw_values,
const int64_t num_valu
for (int stage_lvl = 0; stage_lvl < 2; ++stage_lvl) {
for (int i = 0; i < kNumStreams / 2; ++i) {
stage[stage_lvl + 1][i * 2] =
- _mm_unpacklo_epi8(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2
+ 1]);
+ xsimd::zip_lo(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2 +
1]);
stage[stage_lvl + 1][i * 2 + 1] =
- _mm_unpackhi_epi8(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2
+ 1]);
+ xsimd::zip_hi(stage[stage_lvl][i * 2], stage[stage_lvl][i * 2 +
1]);
}
}
if constexpr (kNumStreams == 8) {
- // This is the path for double.
- __m128i tmp[8];
+ // This is the path for 64bits data.
+ simd_batch tmp[8];
+ using int32_batch = xsimd::batch<int32_t, simd_arch>;
+ // This is a workaround, see:
https://github.com/xtensor-stack/xsimd/issues/735
+ auto from_int32_batch = [](int32_batch from) -> simd_batch {
+ simd_batch dest;
+ memcpy(&dest, &from, sizeof(simd_batch));
+ return dest;
+ };
+ auto to_int32_batch = [](simd_batch from) -> int32_batch {
+ int32_batch dest;
+ memcpy(&dest, &from, sizeof(simd_batch));
+ return dest;
+ };
for (int i = 0; i < 4; ++i) {
- tmp[i * 2] = _mm_unpacklo_epi32(stage[2][i], stage[2][i + 4]);
- tmp[i * 2 + 1] = _mm_unpackhi_epi32(stage[2][i], stage[2][i + 4]);
+ tmp[i * 2] = from_int32_batch(
+ xsimd::zip_lo(to_int32_batch(stage[2][i]),
to_int32_batch(stage[2][i + 4])));
+ tmp[i * 2 + 1] = from_int32_batch(
+ xsimd::zip_hi(to_int32_batch(stage[2][i]),
to_int32_batch(stage[2][i + 4])));
}
for (int i = 0; i < 4; ++i) {
- final_result[i * 2] = _mm_unpacklo_epi32(tmp[i], tmp[i + 4]);
- final_result[i * 2 + 1] = _mm_unpackhi_epi32(tmp[i], tmp[i + 4]);
+ final_result[i * 2] = from_int32_batch(
+ xsimd::zip_lo(to_int32_batch(tmp[i]), to_int32_batch(tmp[i + 4])));
+ final_result[i * 2 + 1] = from_int32_batch(
+ xsimd::zip_hi(to_int32_batch(tmp[i]), to_int32_batch(tmp[i + 4])));
}
} else {
- // this is the path for float.
- __m128i tmp[4];
+ // This is the path for 32bits data.
+ using int64_batch = xsimd::batch<int64_t, simd_arch>;
+ // This is a workaround, see:
https://github.com/xtensor-stack/xsimd/issues/735
+ auto from_int64_batch = [](int64_batch from) -> simd_batch {
Review Comment:
ditto
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
