mirrors/tantivy
2
0
Fork 1
mirror of https://github.com/quickwit-oss/tantivy.git synced 2026-06-19 09:16:45 +00:00
Code Issues Projects Releases Packages Wiki Activity

CR comments from https://github.com/quickwit-oss/tantivy/pull/2940 (#2952)

Browse source 
Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
This commit is contained in:
Paul Masurel 2026-06-08 10:47:58 +02:00 • committed by GitHub
commit abcf6754a2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 34 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

16
bitpacker/src/filter_vec/mod.rs
View file

@ -92,7 +92,10 @@ impl FilterImplPerInstructionSet {
#[cfg(target_arch = "x86_64")] #[cfg(target_arch = "x86_64")]
FilterImplPerInstructionSet::AVX2 => avx2::filter_vec_in_place(range, offset, output), FilterImplPerInstructionSet::AVX2 => avx2::filter_vec_in_place(range, offset, output),
#[cfg(all(target_arch = "aarch64", not(target_vendor = "apple")))] #[cfg(all(target_arch = "aarch64", not(target_vendor = "apple")))]
FilterImplPerInstructionSet::SVE => sve::filter_vec_in_place(range, offset, output), // SAFETY: SVE availability was verified by is_available() before selecting this impl.
FilterImplPerInstructionSet::SVE => unsafe {
sve::filter_vec_in_place(range, offset, output)
},
#[cfg(target_arch = "aarch64")] #[cfg(target_arch = "aarch64")]
FilterImplPerInstructionSet::Neon => neon::filter_vec_in_place(range, offset, output), FilterImplPerInstructionSet::Neon => neon::filter_vec_in_place(range, offset, output),
FilterImplPerInstructionSet::Scalar => { FilterImplPerInstructionSet::Scalar => {
@ -266,15 +269,16 @@ mod tests {
start in 0u32..400u32, start in 0u32..400u32,
end in 0u32..400u32, end in 0u32..400u32,
offset in 0u32..2u32, offset in 0u32..2u32,
mut vals in vals_strategy()) { vals in vals_strategy()) {
for implementation in available_impls() { for implementation in available_impls() {
if implementation == FilterImplPerInstructionSet::Scalar { if implementation == FilterImplPerInstructionSet::Scalar {
continue; continue;
} }
let mut vals_clone = vals.clone(); let mut impl_output = vals.clone();
implementation.filter_vec_in_place(start..=end, offset, &mut vals); let mut scalar_output = vals.clone();
FilterImplPerInstructionSet::Scalar.filter_vec_in_place(start..=end, offset, &mut vals_clone); implementation.filter_vec_in_place(start..=end, offset, &mut impl_output);
assert_eq!(&vals, &vals_clone); FilterImplPerInstructionSet::Scalar.filter_vec_in_place(start..=end, offset, &mut scalar_output);
assert_eq!(&impl_output, &scalar_output);
} }
} }
} }

37
bitpacker/src/filter_vec/neon.rs
View file

@ -17,6 +17,7 @@ unsafe fn compact(data: uint32x4_t, mask: u8) -> uint32x4_t {
} }
} }
// Safe (not unsafe) because NEON is mandatory on aarch64: no runtime feature check needed.
#[inline(never)] #[inline(never)]
pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) { pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
let num_words = output.len() / NUM_LANES; let num_words = output.len() / NUM_LANES;
@ -94,20 +95,24 @@ unsafe fn filter_vec_neon_aux(
// Index is a 4-bit mask: bit k=1 means lane k (bytes 4k..4k+3) is in-range. // Index is a 4-bit mask: bit k=1 means lane k (bytes 4k..4k+3) is in-range.
// The j-th set bit determines which input lane goes to output position j. // The j-th set bit determines which input lane goes to output position j.
const BYTE_SHUFFLE_TABLE: [[u8; 16]; 16] = [ const BYTE_SHUFFLE_TABLE: [[u8; 16]; 16] = [
[0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0000: none [
[0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0001: lane 0 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16,
[4, 5, 6, 7, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0010: lane 1 ], // 0b0000: none
[0, 1, 2, 3, 4, 5, 6, 7, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0011: lanes 0,1 [0, 1, 2, 3, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16], // 0b0001: lane 0
[8, 9, 10, 11, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0100: lane 2 [4, 5, 6, 7, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16], // 0b0010: lane 1
[0, 1, 2, 3, 8, 9, 10, 11, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0101: lanes 0,2 [0, 1, 2, 3, 4, 5, 6, 7, 16, 16, 16, 16, 16, 16, 16, 16], // 0b0011: lanes 0,1
[4, 5, 6, 7, 8, 9, 10, 11, 0, 1, 2, 3, 0, 1, 2, 3], // 0b0110: lanes 1,2 [8, 9, 10, 11, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16], // 0b0100: lane 2
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 0, 1, 2, 3], // 0b0111: lanes 0,1,2 [0, 1, 2, 3, 8, 9, 10, 11, 16, 16, 16, 16, 16, 16, 16, 16], // 0b0101: lanes 0,2
[12, 13, 14, 15, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3], // 0b1000: lane 3 [4, 5, 6, 7, 8, 9, 10, 11, 16, 16, 16, 16, 16, 16, 16, 16], // 0b0110: lanes 1,2
[0, 1, 2, 3, 12, 13, 14, 15, 0, 1, 2, 3, 0, 1, 2, 3], // 0b1001: lanes 0,3 [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 16, 16, 16, 16], // 0b0111: lanes 0,1,2
[4, 5, 6, 7, 12, 13, 14, 15, 0, 1, 2, 3, 0, 1, 2, 3], // 0b1010: lanes 1,3 [
[0, 1, 2, 3, 4, 5, 6, 7, 12, 13, 14, 15, 0, 1, 2, 3], // 0b1011: lanes 0,1,3 12, 13, 14, 15, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16,
[8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 0, 1, 2, 3], // 0b1100: lanes 2,3 ], // 0b1000: lane 3
[0, 1, 2, 3, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3], // 0b1101: lanes 0,2,3 [0, 1, 2, 3, 12, 13, 14, 15, 16, 16, 16, 16, 16, 16, 16, 16], // 0b1001: lanes 0,3
[4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3], // 0b1110: lanes 1,2,3 [4, 5, 6, 7, 12, 13, 14, 15, 16, 16, 16, 16, 16, 16, 16, 16], // 0b1010: lanes 1,3
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15], // 0b1111: all lanes [0, 1, 2, 3, 4, 5, 6, 7, 12, 13, 14, 15, 16, 16, 16, 16], // 0b1011: lanes 0,1,3
[8, 9, 10, 11, 12, 13, 14, 15, 16, 16, 16, 16, 16, 16, 16, 16], // 0b1100: lanes 2,3
[0, 1, 2, 3, 8, 9, 10, 11, 12, 13, 14, 15, 16, 16, 16, 16], // 0b1101: lanes 0,2,3
[4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 16, 16, 16], // 0b1110: lanes 1,2,3
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15], // 0b1111: all lanes
]; ];

4
bitpacker/src/filter_vec/sve.rs
View file

@ -15,7 +15,9 @@ unsafe fn num_lanes() -> usize {
vl vl
} }
pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) { // SAFETY: caller must ensure SVE is available (checked via is_aarch64_feature_detected!("sve")).
// Unlike NEON, SVE is optional on aarch64 and not guaranteed by the target architecture.
pub unsafe fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
if range.start() > range.end() { if range.start() > range.end() {
output.clear(); output.clear();
return; return;