Skip to content

Wire calculatePartialSums to native SIMD via Panama FFI downcall #651

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

Open
r-devulap wants to merge 5 commits into
base: main
Choose a base branch
from
Open

Wire calculatePartialSums to native SIMD via Panama FFI downcall #651

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a9a414f
Add optimal AVX2/SSE reduction and refactor native SIMD code
r-devulap Apr 7, 2026
a490cc0
Wire calculatePartialSums to native SIMD via Panama FFI downcall
r-devulap Apr 7, 2026
b530cb7
perf: Optimize PQ distance look up table calculation for subvector si…
r-devulap Apr 7, 2026
97a1ada
Optimize size == 2
r-devulap Apr 8, 2026
0276bf1
Remove calculate_partial_sums_f32_512 dispatcher, expose variants dir…
r-devulap May 4, 2026
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
4 changes: 2 additions & 2 deletions jvector-native/src/main/c/jextract_vector_simd.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ CURRENT_GCC_VERSION=$(gcc -dumpversion)
# Check if the current GCC version is greater than or equal to the minimum required version
if [ "$(printf '%s\n' "$MIN_GCC_VERSION" "$CURRENT_GCC_VERSION" | sort -V | head -n1)" = "$MIN_GCC_VERSION" ]; then
rm -rf ../resources/libjvector.so
gcc -fPIC -O3 -march=icelake-server -c jvector_simd.c -o jvector_simd.o
gcc -fPIC -O3 -march=skylake-avx512 -c jvector_simd.c -o jvector_simd.o
Copy link
Copy Markdown
Contributor

@ashkrisk ashkrisk Apr 27, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a strong reason to lower the target micro-architecture version?

Copy link
Copy Markdown
Contributor Author

@r-devulap r-devulap May 4, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This change addresses an actual bug in our build configuration. We currently compile targeting icelake-server, but at runtime we only check for skylake-avx512. This mismatch allows the compiler to emit Ice Lake–specific instructions that may be executed on a Skylake CPU, which can result in a SIGILL.

Targeting skylake-avx512 resolves this issue and is sufficient for the kernels we currently have; there’s no requirement for Ice Lake–specific features here.

gcc -fPIC -O3 -march=x86-64 -c jvector_simd_check.c -o jvector_simd_check.o
gcc -shared -o ../resources/libjvector.so jvector_simd_check.o jvector_simd.o

Expand Down Expand Up @@ -77,4 +77,4 @@ jextract \
jvector_simd.h

# Set critical linker option with heap-based segments for all generated methods
sed -i 's/DESC)/DESC, Linker.Option.critical(true))/g' ../java/io/github/jbellis/jvector/vector/cnative/NativeSimdOps.java
sed -i 's/DESC)/DESC, Linker.Option.critical(true))/g' ../java/io/github/jbellis/jvector/vector/cnative/NativeSimdOps.java
Loading
Loading