Skip to content

8351412: Add AVX-512 intrinsics for ML-KEM #24953

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 7 commits into from
Closed

8351412: Add AVX-512 intrinsics for ML-KEM #24953

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
4ef7c73
8351412: Add AVX-512 intrinsics for ML-KEM
ferakocz Jan 10, 2025
c5c6449
Removed debugging changes.
ferakocz Apr 30, 2025
43455de
Eliminating some instructions from generate_kyber12To16_avx512() + fi…
ferakocz May 12, 2025
215b346
Restoring copyright notice on ML_KEM.java
ferakocz May 12, 2025
32571f3
Responding to comments by Sandhya.
ferakocz May 14, 2025
e4f3264
Response to review comment + loading constants with broadcast op.
ferakocz May 15, 2025
ea2152d
Added some comments.
ferakocz May 20, 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
136 changes: 68 additions & 68 deletions src/hotspot/cpu/x86/stubGenerator_x86_64_kyber.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,41 +43,27 @@
// Constants
//
ATTRIBUTE_ALIGNED(64) static const uint16_t kyberAvx512Consts[] = {
0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301,
0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301,
0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301,
0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301, 0xF301,

0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01,
0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01,
0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01,
0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01, 0x0D01,

0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF,
0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF,
0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF,
0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF,

0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200,
0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200,
0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200,
0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200, 0x0200,

0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549,
0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549,
0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549,
0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549, 0x0549,

0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00,
0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00,
0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00,
0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00, 0x0F00
0xF301, 0xF301, 0xF301, 0xF301, // q^-1 mod montR
0x0D01, 0x0D01, 0x0D01, 0x0D01, // q
0x4EBF, 0x4EBF, 0x4EBF, 0x4EBF, // Barrett multiplier
0x0200, 0x0200, 0x0200, 0x0200, //(dim/2)^-1 mod q
0x0549, 0x0549, 0x0549, 0x0549, // montR^2 mod q
0x0F00, 0x0F00, 0x0F00, 0x0F00 // mask for kyber12to16
};

static address kyberAvx512ConstsAddr() {
return (address) kyberAvx512Consts;
static int qInvModROffset = 0;
static int qOffset = 8;
static int barretMultiplierOffset = 16;
static int dimHalfInverseOffset = 24;
static int montRSquareModqOffset = 32;
static int f00Offset = 40;

static address kyberAvx512ConstsAddr(int offset) {
return ((address) kyberAvx512Consts) + offset;
}

const Register scratch = r10;

ATTRIBUTE_ALIGNED(64) static const uint16_t kyberAvx512NttPerms[] = {
// 0
0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
Expand Down Expand Up @@ -378,17 +364,19 @@ address generate_kyberNtt_avx512(StubGenerator *stubgen,
const Register coeffs = c_rarg0;
const Register zetas = c_rarg1;

const Register kyberConsts = r10;
const Register perms = r11;

__ lea(perms, ExternalAddress(kyberAvx512NttPermsAddr()));
__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));

load4regs(xmm4_7, coeffs, 256, _masm);
load4regs(xmm20_23, zetas, 0, _masm);

__ evmovdquw(xmm30, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ evmovdquw(xmm31, Address(kyberConsts, 0), Assembler::AVX_512bit);
__ vpbroadcastq(xmm30,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q
__ vpbroadcastq(xmm31,
ExternalAddress(kyberAvx512ConstsAddr(qInvModROffset)),
Assembler::AVX_512bit, scratch); // q^-1 mod montR

load4regs(xmm0_3, coeffs, 0, _masm);

Expand Down Expand Up @@ -472,18 +460,21 @@ address generate_kyberInverseNtt_avx512(StubGenerator *stubgen,
const Register coeffs = c_rarg0;
const Register zetas = c_rarg1;

const Register kyberConsts = r10;
const Register perms = r11;

__ lea(perms, ExternalAddress(kyberAvx512InverseNttPermsAddr()));
__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));

__ evmovdquw(xmm12, Address(perms, 0), Assembler::AVX_512bit);
__ evmovdquw(xmm16, Address(perms, 64), Assembler::AVX_512bit);

__ evmovdquw(xmm31, Address(kyberConsts, 0), Assembler::AVX_512bit);
__ evmovdquw(xmm30, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ evmovdquw(xmm29, Address(kyberConsts, 192), Assembler::AVX_512bit);
__ vpbroadcastq(xmm31,
ExternalAddress(kyberAvx512ConstsAddr(qInvModROffset)),
Assembler::AVX_512bit, scratch); // q^-1 mod montR
__ vpbroadcastq(xmm30,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q
__ vpbroadcastq(xmm29,
ExternalAddress(kyberAvx512ConstsAddr(dimHalfInverseOffset)),
Assembler::AVX_512bit, scratch); // (dim/2)^-1 mod q

load4regs(xmm0_3, coeffs, 0, _masm);
load4regs(xmm4_7, coeffs, 256, _masm);
Expand Down Expand Up @@ -518,8 +509,12 @@ address generate_kyberInverseNtt_avx512(StubGenerator *stubgen,

montmul(xmm12_15, xmm16_19, xmm8_11, xmm12_15, xmm8_11, _masm);

__ evmovdquw(xmm16, Address(kyberConsts, 128), Assembler::AVX_512bit);
__ evmovdquw(xmm17, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ vpbroadcastq(xmm16,
ExternalAddress(kyberAvx512ConstsAddr(barretMultiplierOffset)),
Assembler::AVX_512bit, scratch); // Barrett multiplier
__ vpbroadcastq(xmm17,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q

permute(xmm1357, xmm0246, xmm12_15, 20, _masm);
barrettReduce(_masm);
Expand All @@ -537,8 +532,12 @@ address generate_kyberInverseNtt_avx512(StubGenerator *stubgen,
// level 4
load4regs(xmm8_11, zetas, 1024, _masm);

__ evmovdquw(xmm16, Address(kyberConsts, 128), Assembler::AVX_512bit);
__ evmovdquw(xmm17, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ vpbroadcastq(xmm16,
ExternalAddress(kyberAvx512ConstsAddr(barretMultiplierOffset)),
Assembler::AVX_512bit, scratch); // Barrett multiplier
__ vpbroadcastq(xmm17,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q

sub_add(xmm12_15, xmm0246, xmm0246, xmm1357, _masm);
montmul(xmm1357, xmm12_15, xmm8_11, xmm1357, xmm8_11, _masm);
Expand Down Expand Up @@ -587,7 +586,6 @@ address generate_kyberNttMult_avx512(StubGenerator *stubgen,
const Register nttb = c_rarg2;
const Register zetas = c_rarg3;

const Register kyberConsts = r10;
const Register perms = r11;
const Register loopCnt = r12;

Expand All @@ -597,14 +595,18 @@ address generate_kyberNttMult_avx512(StubGenerator *stubgen,
Label Loop;

__ lea(perms, ExternalAddress(kyberAvx512_nttMultPermsAddr()));
__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));


load4regs(xmm26_29, perms, 0, _masm);
__ evmovdquw(xmm31, Address(kyberConsts, 0), Assembler::AVX_512bit);
__ evmovdquw(xmm30, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ evmovdquw(xmm23, Address(kyberConsts, 256), Assembler::AVX_512bit);

__ vpbroadcastq(xmm31,
ExternalAddress(kyberAvx512ConstsAddr(qInvModROffset)),
Assembler::AVX_512bit, scratch); // q^-1 mod montR
__ vpbroadcastq(xmm30,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q
__ vpbroadcastq(xmm23,
ExternalAddress(kyberAvx512ConstsAddr(montRSquareModqOffset)),
Assembler::AVX_512bit, scratch); // montR^2 mod q

__ BIND(Loop);

Expand Down Expand Up @@ -703,11 +705,9 @@ address generate_kyberAddPoly_2_avx512(StubGenerator *stubgen,
const Register a = c_rarg1;
const Register b = c_rarg2;

const Register kyberConsts = r11;

__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));

__ evmovdquw(xmm31, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ vpbroadcastq(xmm31,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q

for (int i = 0; i < 8; i++) {
__ evmovdquw(xmm(i), Address(a, 64 * i), Assembler::AVX_512bit);
Expand Down Expand Up @@ -752,10 +752,9 @@ address generate_kyberAddPoly_3_avx512(StubGenerator *stubgen,
const Register b = c_rarg2;
const Register c = c_rarg3;

const Register kyberConsts = r11;

__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));
__ evmovdquw(xmm31, Address(kyberConsts, 64), Assembler::AVX_512bit);
__ vpbroadcastq(xmm31,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q

for (int i = 0; i < 8; i++) {
__ evmovdquw(xmm(i), Address(a, 64 * i), Assembler::AVX_512bit);
Expand Down Expand Up @@ -807,19 +806,19 @@ address generate_kyber12To16_avx512(StubGenerator *stubgen,
const Register parsed = c_rarg2;
const Register parsedLength = c_rarg3;

const Register kyberConsts = r10;
const Register perms = r11;

Label Loop;

__ addptr(condensed, condensedOffs);

__ lea(perms, ExternalAddress(kyberAvx512_12To16PermsAddr()));
__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));

load4regs(xmm24_27, perms, 0, _masm);
load4regs(xmm28_31, perms, 256, _masm);
__ evmovdquw(xmm23, Address(kyberConsts, 320), Assembler::AVX_512bit);
__ vpbroadcastq(xmm23,
ExternalAddress(kyberAvx512ConstsAddr(f00Offset)),
Assembler::AVX_512bit, scratch); // 0xF00

__ BIND(Loop);
__ evmovdqub(xmm0, Address(condensed, 0),Assembler::AVX_256bit);
Expand Down Expand Up @@ -890,7 +889,6 @@ address generate_kyber12To16_avx512(StubGenerator *stubgen,
// Kyber barrett reduce function.
//
// coeffs (short[256]) = c_rarg0
// kyberConsts (short[40]) = c_rarg1
address generate_kyberBarrettReduce_avx512(StubGenerator *stubgen,
MacroAssembler *_masm) {

Expand All @@ -901,14 +899,16 @@ address generate_kyberBarrettReduce_avx512(StubGenerator *stubgen,
__ enter();

const Register coeffs = c_rarg0;
const Register kyberConsts = c_rarg1;

__ lea(kyberConsts, ExternalAddress(kyberAvx512ConstsAddr()));
__ vpbroadcastq(xmm16,
ExternalAddress(kyberAvx512ConstsAddr(barretMultiplierOffset)),
Assembler::AVX_512bit, scratch); // Barrett multiplier
__ vpbroadcastq(xmm17,
ExternalAddress(kyberAvx512ConstsAddr(qOffset)),
Assembler::AVX_512bit, scratch); // q

load4regs(xmm0_3, coeffs, 0, _masm);
load4regs(xmm4_7, coeffs, 256, _masm);
__ evmovdquw(xmm16, Address(kyberConsts, 128), Assembler::AVX_512bit);
__ evmovdquw(xmm17, Address(kyberConsts, 64), Assembler::AVX_512bit);

barrettReduce(_masm);

Expand Down