mirror of
https://github.com/henrydcase/pqc.git
synced 2024-11-24 00:11:27 +00:00
a655ec8a9d
* Add state destroy to SHA2 API * Include optimized SPHINCS+ implementations I've generated new implementations from the sphincsplus repository. * Don't destroy sha256ctx after finalize * Attempt to shut up MSVC * Make sure to drop errors in rmtree
129 lines
5.4 KiB
C
129 lines
5.4 KiB
C
#include <string.h>
|
|
|
|
#include "sha256.h"
|
|
#include "sha256avx.h"
|
|
#include "sha256x8.h"
|
|
#include "utils.h"
|
|
|
|
void PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_seed_statex8(sha256ctxx8 *ctx, const unsigned char *pub_seed) {
|
|
uint8_t block[PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_BLOCK_BYTES];
|
|
size_t i;
|
|
|
|
for (i = 0; i < PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_N; ++i) {
|
|
block[i] = pub_seed[i];
|
|
}
|
|
for (i = PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_N; i < PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_BLOCK_BYTES; ++i) {
|
|
block[i] = 0;
|
|
}
|
|
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256_init8x(ctx);
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256_update8x(ctx, block, block, block, block, block, block, block, block, PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_BLOCK_BYTES);
|
|
|
|
}
|
|
|
|
/* This provides a wrapper around the internals of 8x parallel SHA256 */
|
|
void PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256x8(unsigned char *out0,
|
|
unsigned char *out1,
|
|
unsigned char *out2,
|
|
unsigned char *out3,
|
|
unsigned char *out4,
|
|
unsigned char *out5,
|
|
unsigned char *out6,
|
|
unsigned char *out7,
|
|
const unsigned char *in0,
|
|
const unsigned char *in1,
|
|
const unsigned char *in2,
|
|
const unsigned char *in3,
|
|
const unsigned char *in4,
|
|
const unsigned char *in5,
|
|
const unsigned char *in6,
|
|
const unsigned char *in7, unsigned long long inlen) {
|
|
sha256ctxx8 ctx;
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256_init8x(&ctx);
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256_update8x(&ctx, in0, in1, in2, in3, in4, in5, in6, in7, inlen);
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256_final8x(&ctx, out0, out1, out2, out3, out4, out5, out6, out7);
|
|
}
|
|
|
|
/**
|
|
* Note that inlen should be sufficiently small that it still allows for
|
|
* an array to be allocated on the stack. Typically 'in' is merely a seed.
|
|
* Outputs outlen number of bytes
|
|
*/
|
|
void PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_mgf1x8(
|
|
unsigned char *outx8,
|
|
unsigned long outlen,
|
|
const unsigned char *in0,
|
|
const unsigned char *in1,
|
|
const unsigned char *in2,
|
|
const unsigned char *in3,
|
|
const unsigned char *in4,
|
|
const unsigned char *in5,
|
|
const unsigned char *in6,
|
|
const unsigned char *in7,
|
|
unsigned long inlen) {
|
|
unsigned char inbufx8[8 * ((PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_N + PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_ADDR_BYTES) + 4)];
|
|
unsigned char outbufx8[8 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES];
|
|
unsigned long i;
|
|
unsigned int j;
|
|
|
|
memcpy(inbufx8 + 0 * (inlen + 4), in0, inlen);
|
|
memcpy(inbufx8 + 1 * (inlen + 4), in1, inlen);
|
|
memcpy(inbufx8 + 2 * (inlen + 4), in2, inlen);
|
|
memcpy(inbufx8 + 3 * (inlen + 4), in3, inlen);
|
|
memcpy(inbufx8 + 4 * (inlen + 4), in4, inlen);
|
|
memcpy(inbufx8 + 5 * (inlen + 4), in5, inlen);
|
|
memcpy(inbufx8 + 6 * (inlen + 4), in6, inlen);
|
|
memcpy(inbufx8 + 7 * (inlen + 4), in7, inlen);
|
|
|
|
/* While we can fit in at least another full block of SHA256 output.. */
|
|
for (i = 0; (i + 1)*PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES <= outlen; i++) {
|
|
for (j = 0; j < 8; j++) {
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_ull_to_bytes(inbufx8 + inlen + j * (inlen + 4), 4, i);
|
|
}
|
|
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256x8(outx8 + 0 * outlen,
|
|
outx8 + 1 * outlen,
|
|
outx8 + 2 * outlen,
|
|
outx8 + 3 * outlen,
|
|
outx8 + 4 * outlen,
|
|
outx8 + 5 * outlen,
|
|
outx8 + 6 * outlen,
|
|
outx8 + 7 * outlen,
|
|
inbufx8 + 0 * (inlen + 4),
|
|
inbufx8 + 1 * (inlen + 4),
|
|
inbufx8 + 2 * (inlen + 4),
|
|
inbufx8 + 3 * (inlen + 4),
|
|
inbufx8 + 4 * (inlen + 4),
|
|
inbufx8 + 5 * (inlen + 4),
|
|
inbufx8 + 6 * (inlen + 4),
|
|
inbufx8 + 7 * (inlen + 4), inlen + 4);
|
|
outx8 += PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES;
|
|
}
|
|
/* Until we cannot anymore, and we fill the remainder. */
|
|
for (j = 0; j < 8; j++) {
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_ull_to_bytes(inbufx8 + inlen + j * (inlen + 4), 4, i);
|
|
}
|
|
PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_sha256x8(outbufx8 + 0 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 1 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 2 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 3 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 4 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 5 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 6 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outbufx8 + 7 * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
inbufx8 + 0 * (inlen + 4),
|
|
inbufx8 + 1 * (inlen + 4),
|
|
inbufx8 + 2 * (inlen + 4),
|
|
inbufx8 + 3 * (inlen + 4),
|
|
inbufx8 + 4 * (inlen + 4),
|
|
inbufx8 + 5 * (inlen + 4),
|
|
inbufx8 + 6 * (inlen + 4),
|
|
inbufx8 + 7 * (inlen + 4), inlen + 4);
|
|
|
|
for (j = 0; j < 8; j++) {
|
|
memcpy(outx8 + j * outlen,
|
|
outbufx8 + j * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES,
|
|
outlen - i * PQCLEAN_SPHINCSSHA256256SSIMPLE_AVX2_SHA256_OUTPUT_BYTES);
|
|
}
|
|
}
|