Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Allow ROW::CopyRangeFrom to be vectorized #15267

Merged
merged 1 commit into from
Jun 22, 2023
Merged

Allow ROW::CopyRangeFrom to be vectorized #15267

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions .github/actions/spelling/expect/expect.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2062,6 +2062,7 @@ vcpkg
vcprintf
vcxitems
vec
vectorize
vectorized
VERCTRL
VERTBAR
Expand Down
38 changes: 24 additions & 14 deletions src/buffer/out/Row.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -631,26 +631,36 @@ catch (...)
const auto baseOffset = til::at(charOffsets, 0);
const auto endOffset = til::at(charOffsets, colEndInput);
const auto inToOutOffset = gsl::narrow_cast<uint16_t>(chBeg - baseOffset);
#pragma warning(suppress : 26481) // Don't use pointer arithmetic. Use span instead (bounds.1).
const auto dst = row._charOffsets.data() + colEnd;

// Now with the `colEndInput` figured out, we can easily copy the `charOffsets` into the `_charOffsets`.
// It's possible to use SIMD for this loop for extra perf gains. Something like this for SSE2 (~8x faster):
// const auto in = _mm_loadu_si128(...);
// const auto off = _mm_and_epi32(in, _mm_set1_epi16(CharOffsetsMask));
// const auto trailer = _mm_and_epi32(in, _mm_set1_epi16(CharOffsetsTrailer));
// const auto out = _mm_or_epi32(_mm_add_epi16(off, _mm_set1_epi16(inToOutOffset)), trailer);
// _mm_store_si128(..., out);
for (uint16_t i = 0; i < colEndInput; ++i, ++colEnd)
{
const auto ch = til::at(charOffsets, i);
const auto off = ch & CharOffsetsMask;
const auto trailer = ch & CharOffsetsTrailer;
til::at(row._charOffsets, colEnd) = gsl::narrow_cast<uint16_t>((off + inToOutOffset) | trailer);
}
_copyOffsets(dst, charOffsets.data(), colEndInput, inToOutOffset);

colEnd += colEndInput;
colEndDirty = gsl::narrow_cast<uint16_t>(colBeg + colEndDirtyInput);
charsConsumed = endOffset - baseOffset;
}

#pragma warning(push)
#pragma warning(disable : 26481) // Don't use pointer arithmetic. Use span instead (bounds.1).
[[msvc::forceinline]] void ROW::WriteHelper::_copyOffsets(uint16_t* __restrict dst, const uint16_t* __restrict src, uint16_t size, uint16_t offset) noexcept
{
__assume(src != nullptr);
__assume(dst != nullptr);

// All tested compilers (including MSVC) will neatly unroll and vectorize
// this loop, which is why it's written in this particular way.
for (const auto end = src + size; src != end; ++src, ++dst)
{
const uint16_t ch = *src;
const uint16_t off = ch & CharOffsetsMask;
const uint16_t trailer = ch & CharOffsetsTrailer;
const uint16_t newOff = off + offset;
*dst = newOff | trailer;
}
}
#pragma warning(pop)

[[msvc::forceinline]] void ROW::WriteHelper::Finish()
{
colEndDirty = row._adjustForward(colEndDirty);
Expand Down
1 change: 1 addition & 0 deletions src/buffer/out/Row.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,6 +171,7 @@ class ROW final
void ReplaceCharacters(til::CoordType width) noexcept;
void ReplaceText() noexcept;
void CopyTextFrom(const std::span<const uint16_t>& charOffsets) noexcept;
static void _copyOffsets(uint16_t* dst, const uint16_t* src, uint16_t size, uint16_t offset) noexcept;
void Finish();

// Parent pointer.
Expand Down