Skip to content

8351034: Add AVX-512 intrinsics for ML-DSA #23860

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

Closed
wants to merge 15 commits into from
Closed

8351034: Add AVX-512 intrinsics for ML-DSA #23860

Changes from 1 commit
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
1ff5851
JDK-8351034 Add AVX-512 intrinsics for ML-DSA
ferakocz Jan 10, 2025
fe50e0d
Added comments, removed debugging printfs
ferakocz Mar 3, 2025
331f1ec
Merged master.
ferakocz Mar 5, 2025
3aaa106
Added alignment to loop entries.
ferakocz Mar 5, 2025
64135f2
Accepted review comments.
ferakocz Mar 6, 2025
f65ef7c
Added validity test for the intrinsics.
ferakocz Mar 12, 2025
aa2fdf2
Made the intrinsics test separate from the pure java test.
ferakocz Mar 12, 2025
2438fb5
responding to review comments
ferakocz Mar 20, 2025
1cfab77
More beautification
ferakocz Mar 20, 2025
e9db09e
Fix windows build
ferakocz Mar 20, 2025
9597b53
Added asserts for array sizes
ferakocz Mar 21, 2025
5665689
Further readability improvements.
ferakocz Mar 22, 2025
7a9f664
Reacting to comments by Volodymyr.
ferakocz Mar 31, 2025
e4ab10b
Reacting to comment by Sandhya.
ferakocz Apr 2, 2025
0b0d096
Reacting to mor comments from Sandhya.
ferakocz Apr 8, 2025
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
18 changes: 7 additions & 11 deletions src/hotspot/cpu/x86/stubGenerator_x86_64_dilithium.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -341,8 +341,6 @@ static address generate_dilithiumAlmostNtt_avx512(StubGenerator *stubgen,
// memory after level 7, then loading back the coefficients that we saved after
// level 1 and do the same computation with those

store4Xmms(coeffs, 0, xmm0_3, _masm);
store4Xmms(coeffs, 4 * XMMBYTES, xmm4_7, _masm);
store4Xmms(coeffs, 8 * XMMBYTES, xmm8_11, _masm);
store4Xmms(coeffs, 12 * XMMBYTES, xmm12_15, _masm);

Expand All @@ -367,7 +365,7 @@ static address generate_dilithiumAlmostNtt_avx512(StubGenerator *stubgen,
// level 4
loadPerm(xmm16_19, perms, nttL4PermsIdx, _masm);
loadPerm(xmm12_15, perms, nttL4PermsIdx + 64, _masm);
load4Xmms(xmm24_27, zetas, 4 * 512, _masm); // for level 3
load4Xmms(xmm24_27, zetas, 4 * 512, _masm);

for (int i = 0; i < 8; i += 2) {
__ evpermi2d(xmm(i/2 + 16), xmm(i), xmm(i + 1), Assembler::AVX_512bit);
Expand Down Expand Up @@ -796,16 +794,15 @@ static address generate_dilithiumDecomposePoly_avx512(StubGenerator *stubgen,
const XMMRegister barrettMultiplier = xmm29;
const XMMRegister barrettAddend = xmm30;

__ xorl(scratch, scratch);
__ evpbroadcastd(zero, scratch, Assembler::AVX_512bit); // 0
__ addl(scratch, 1);
__ evpbroadcastd(one, scratch, Assembler::AVX_512bit); // 1
__ vpxor(zero, zero, zero, Assembler::AVX_512bit); // 0
__ vpternlogd(xmm0, 0xff, xmm0, xmm0, Assembler::AVX_512bit); // -1
__ vpsubd(one, zero, xmm0, Assembler::AVX_512bit); // 1
__ vpbroadcastd(dilithium_q,
ExternalAddress(dilithiumAvx512ConstsAddr(dilithium_qIdx)),
Assembler::AVX_512bit, scratch); // q^-1 mod 2^32
Assembler::AVX_512bit, scratch); // q
__ vpbroadcastd(barrettAddend,
ExternalAddress(dilithiumAvx512ConstsAddr(barrettAddendIdx)),
Assembler::AVX_512bit, scratch); // q
Assembler::AVX_512bit, scratch); // addend for Barrett reduction

__ evpbroadcastd(twoGamma2, rTwoGamma2, Assembler::AVX_512bit); // 2 * gamma2

Expand Down Expand Up @@ -1008,10 +1005,9 @@ static address generate_dilithiumDecomposePoly_avx512(StubGenerator *stubgen,
store4Xmms(highPart, 0, xmm0_3, _masm);
store4Xmms(lowPart, 0, xmm8_11, _masm);

__ subl(len, 4 * XMMBYTES);
__ addptr(highPart, 4 * XMMBYTES);
__ addptr(lowPart, 4 * XMMBYTES);
__ cmpl(len, 0);
__ subl(len, 4 * XMMBYTES);
__ jcc(Assembler::notEqual, L_loop);

__ leave(); // required for proper stackwalking of RuntimeStub frame
Expand Down