motorhead/qemu
1
0
Fork 0
mirror of https://github.com/Motorhead1991/qemu.git synced 2025-08-06 09:13:55 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions

target/i386: implement SHA instructions

Browse source 
The implementation was validated with OpenSSL and with the test vectors in
https://github.com/rust-lang/stdarch/blob/master/crates/core_arch/src/x86/sha.rs.

The instructions provide a ~25% improvement on hashing a 64 MiB file:
runtime goes down from 1.8 seconds to 1.4 seconds; instruction count on
the host goes down from 5.8 billion to 4.8 billion with slightly better
IPC too.  Good job Intel. ;)

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This commit is contained in:
Paolo Bonzini 2023-10-10 10:31:17 +02:00
parent 45b5933f7a
commit e582b629f0
6 changed files with 209 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

128
target/i386/ops_sse.h
View file

@ -2527,6 +2527,134 @@ SSE_HELPER_FMAP(helper_fma4ps, ZMM_S, 2 << SHIFT, float32_muladd)
SSE_HELPER_FMAP(helper_fma4pd, ZMM_D, 1 << SHIFT, float64_muladd)
#endif
#if SHIFT == 1
#define SSE_HELPER_SHA1RNDS4(name, F, K) \
void name(Reg *d, Reg *a, Reg *b) \
{ \
uint32_t A, B, C, D, E, t, i; \
\
A = a->L(3); \
B = a->L(2); \
C = a->L(1); \
D = a->L(0); \
E = 0; \
\
for (i = 0; i <= 3; i++) { \
t = F(B, C, D) + rol32(A, 5) + b->L(3 - i) + E + K; \
E = D; \
D = C; \
C = rol32(B, 30); \
B = A; \
A = t; \
} \
\
d->L(3) = A; \
d->L(2) = B; \
d->L(1) = C; \
d->L(0) = D; \
}
#define SHA1_F0(b, c, d) (((b) & (c)) ^ (~(b) & (d)))
#define SHA1_F1(b, c, d) ((b) ^ (c) ^ (d))
#define SHA1_F2(b, c, d) (((b) & (c)) ^ ((b) & (d)) ^ ((c) & (d)))
SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f0, SHA1_F0, 0x5A827999)
SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f1, SHA1_F1, 0x6ED9EBA1)
SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f2, SHA1_F2, 0x8F1BBCDC)
SSE_HELPER_SHA1RNDS4(helper_sha1rnds4_f3, SHA1_F1, 0xCA62C1D6)
void helper_sha1nexte(Reg *d, Reg *a, Reg *b)
{
d->L(3) = b->L(3) + rol32(a->L(3), 30);
d->L(2) = b->L(2);
d->L(1) = b->L(1);
d->L(0) = b->L(0);
}
void helper_sha1msg1(Reg *d, Reg *a, Reg *b)
{
/* These could be overwritten by the first two assignments, save them. */
uint32_t b3 = b->L(3);
uint32_t b2 = b->L(2);
d->L(3) = a->L(3) ^ a->L(1);
d->L(2) = a->L(2) ^ a->L(0);
d->L(1) = a->L(1) ^ b3;
d->L(0) = a->L(0) ^ b2;
}
void helper_sha1msg2(Reg *d, Reg *a, Reg *b)
{
d->L(3) = rol32(a->L(3) ^ b->L(2), 1);
d->L(2) = rol32(a->L(2) ^ b->L(1), 1);
d->L(1) = rol32(a->L(1) ^ b->L(0), 1);
d->L(0) = rol32(a->L(0) ^ d->L(3), 1);
}
#define SHA256_CH(e, f, g) (((e) & (f)) ^ (~(e) & (g)))
#define SHA256_MAJ(a, b, c) (((a) & (b)) ^ ((a) & (c)) ^ ((b) & (c)))
#define SHA256_RNDS0(w) (ror32((w), 2) ^ ror32((w), 13) ^ ror32((w), 22))
#define SHA256_RNDS1(w) (ror32((w), 6) ^ ror32((w), 11) ^ ror32((w), 25))
#define SHA256_MSGS0(w) (ror32((w), 7) ^ ror32((w), 18) ^ ((w) >> 3))
#define SHA256_MSGS1(w) (ror32((w), 17) ^ ror32((w), 19) ^ ((w) >> 10))
void helper_sha256rnds2(Reg *d, Reg *a, Reg *b, uint32_t wk0, uint32_t wk1)
{
uint32_t t, AA, EE;
uint32_t A = b->L(3);
uint32_t B = b->L(2);
uint32_t C = a->L(3);
uint32_t D = a->L(2);
uint32_t E = b->L(1);
uint32_t F = b->L(0);
uint32_t G = a->L(1);
uint32_t H = a->L(0);
/* Even round */
t = SHA256_CH(E, F, G) + SHA256_RNDS1(E) + wk0 + H;
AA = t + SHA256_MAJ(A, B, C) + SHA256_RNDS0(A);
EE = t + D;
/* These will be B and F at the end of the odd round */
d->L(2) = AA;
d->L(0) = EE;
D = C, C = B, B = A, A = AA;
H = G, G = F, F = E, E = EE;
/* Odd round */
t = SHA256_CH(E, F, G) + SHA256_RNDS1(E) + wk1 + H;
AA = t + SHA256_MAJ(A, B, C) + SHA256_RNDS0(A);
EE = t + D;
d->L(3) = AA;
d->L(1) = EE;
}
void helper_sha256msg1(Reg *d, Reg *a, Reg *b)
{
/* b->L(0) could be overwritten by the first assignment, save it. */
uint32_t b0 = b->L(0);
d->L(0) = a->L(0) + SHA256_MSGS0(a->L(1));
d->L(1) = a->L(1) + SHA256_MSGS0(a->L(2));
d->L(2) = a->L(2) + SHA256_MSGS0(a->L(3));
d->L(3) = a->L(3) + SHA256_MSGS0(b0);
}
void helper_sha256msg2(Reg *d, Reg *a, Reg *b)
{
/* Earlier assignments cannot overwrite any of the two operands. */
d->L(0) = a->L(0) + SHA256_MSGS1(b->L(2));
d->L(1) = a->L(1) + SHA256_MSGS1(b->L(3));
/* Yes, this reuses the previously computed values. */
d->L(2) = a->L(2) + SHA256_MSGS1(d->L(0));
d->L(3) = a->L(3) + SHA256_MSGS1(d->L(1));
}
#endif
#undef SSE_HELPER_S
#undef LANE_WIDTH