archive/jdk
1
0
Fork 0
mirror of https://github.com/openjdk/jdk.git synced 2025-08-26 14:24:46 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

8288047: Accelerate Poly1305 on x86_64 using AVX512 instructions

Browse source 
Reviewed-by: sviswanathan, vlivanov
This commit is contained in:
Volodymyr Paprotski 2022-11-21 21:01:25 +00:00 • committed by Sandhya Viswanathan
parent cd6a203a3e
commit f12710e938
32 changed files with 1857 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

63
src/java.base/share/classes/com/sun/crypto/provider/Poly1305.java
View file

@ -34,6 +34,8 @@ import java.util.Objects;
import sun.security.util.math.*;
import sun.security.util.math.intpoly.*;
import jdk.internal.vm.annotation.IntrinsicCandidate;
import jdk.internal.vm.annotation.ForceInline;
/**
* This class represents the Poly1305 function defined in RFC 7539.
@ -59,8 +61,10 @@ final class Poly1305 {
private IntegerModuloP s;
private MutableIntegerModuloP a;
private final MutableIntegerModuloP n = ipl1305.get1().mutable();
private final boolean checkWeakKey;
Poly1305() { }
Poly1305() { this(true); }
Poly1305(boolean checkKey) { checkWeakKey = checkKey; }
/**
* Initialize the Poly1305 object
@ -165,11 +169,15 @@ final class Poly1305 {
blockOffset = 0;
}
}
while (len >= BLOCK_LENGTH) {
processBlock(input, offset, BLOCK_LENGTH);
offset += BLOCK_LENGTH;
len -= BLOCK_LENGTH;
}
int blockMultipleLength = len & (~(BLOCK_LENGTH-1));
long[] aLimbs = a.getLimbs();
long[] rLimbs = r.getLimbs();
processMultipleBlocksCheck(input, offset, blockMultipleLength, aLimbs, rLimbs);
processMultipleBlocks(input, offset, blockMultipleLength, aLimbs, rLimbs);
offset += blockMultipleLength;
len -= blockMultipleLength;
if (len > 0) { // and len < BLOCK_LENGTH
System.arraycopy(input, offset, block, 0, len);
blockOffset = len;
@ -235,12 +243,35 @@ final class Poly1305 {
a.setProduct(r); // a = (a * r) % p
}
// This is an intrinsified method. The unused parameters aLimbs and rLimbs are used by the intrinsic.
// They correspond to this.a and this.r respectively
@ForceInline
@IntrinsicCandidate
private void processMultipleBlocks(byte[] input, int offset, int length, long[] aLimbs, long[] rLimbs) {
while (length >= BLOCK_LENGTH) {
processBlock(input, offset, BLOCK_LENGTH);
offset += BLOCK_LENGTH;
length -= BLOCK_LENGTH;
}
}
private static void processMultipleBlocksCheck(byte[] input, int offset, int length, long[] aLimbs, long[] rLimbs) {
Objects.checkFromIndexSize(offset, length, input.length);
final int numLimbs = 5; // Intrinsic expects exactly 5 limbs
if (aLimbs.length != numLimbs) {
throw new RuntimeException("invalid accumulator length: " + aLimbs.length);
}
if (rLimbs.length != numLimbs) {
throw new RuntimeException("invalid R length: " + rLimbs.length);
}
}
/**
* Partition the authentication key into the R and S components, clamp
* the R value, and instantiate IntegerModuloP objects to R and S's
* numeric values.
*/
private void setRSVals() {
private void setRSVals() throws InvalidKeyException {
// Clamp the bytes in the "r" half of the key.
keyBytes[3] &= 15;
keyBytes[7] &= 15;
@ -250,6 +281,24 @@ final class Poly1305 {
keyBytes[8] &= (byte)252;
keyBytes[12] &= (byte)252;
if (checkWeakKey) {
byte keyIsZero = 0;
for (int i = 0; i < RS_LENGTH; i++) {
keyIsZero |= keyBytes[i];
}
if (keyIsZero == 0) {
throw new InvalidKeyException("R is set to zero");
}
keyIsZero = 0;
for (int i = RS_LENGTH; i < 2*RS_LENGTH; i++) {
keyIsZero |= keyBytes[i];
}
if (keyIsZero == 0) {
throw new InvalidKeyException("S is set to zero");
}
}
// Create IntegerModuloP elements from the r and s values
r = ipl1305.getElement(keyBytes, 0, RS_LENGTH, (byte)0);
s = ipl1305.getElement(keyBytes, RS_LENGTH, RS_LENGTH, (byte)0);