CPU: remove AVX code, switch to __builtin for AES detection (#1959)

* [cpu] remove avx detect and code blocks, try to switch to __builtin
* [cpu] use __builtin_* only on x86 systems
* [cpu] perform check in separate function
* [cpu] set AES definition on MSVC
* update x86 and aes support checks at compile time
* [cmake] update comment about AES on MSVC
This commit is contained in:
R4SAS 2023-08-31 16:52:51 +00:00 committed by GitHub
parent 32c5ff23a6
commit 7b6aa41ca8
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
11 changed files with 222 additions and 232 deletions

View file

@ -1,5 +1,5 @@
/*
* Copyright (c) 2013-2022, The PurpleI2P Project
* Copyright (c) 2013-2023, The PurpleI2P Project
*
* This file is part of Purple i2pd project and licensed under BSD3
*
@ -28,6 +28,12 @@
#include "I2PEndian.h"
#include "Log.h"
#if defined(__AES__) && !defined(_MSC_VER) && ((defined(_M_AMD64) || defined(__x86_64__)) || (defined(_M_IX86) || defined(__i386__)))
#define SUPPORTS_AES 1
#else
#define SUPPORTS_AES 0
#endif
namespace i2p
{
namespace crypto
@ -361,7 +367,7 @@ namespace crypto
BIGNUM * b1 = BN_CTX_get (ctx);
BIGNUM * b = BN_CTX_get (ctx);
// select random k
#if defined(__x86_64__)
#if (defined(_M_AMD64) || defined(__x86_64__))
BN_rand (k, ELGAMAL_FULL_EXPONENT_NUM_BITS, -1, 1); // full exponent for x64
#else
BN_rand (k, ELGAMAL_SHORT_EXPONENT_NUM_BITS, -1, 1); // short exponent of 226 bits
@ -428,7 +434,7 @@ namespace crypto
void GenerateElGamalKeyPair (uint8_t * priv, uint8_t * pub)
{
#if defined(__x86_64__) || defined(__i386__) || defined(_MSC_VER)
#if (defined(_M_AMD64) || defined(__x86_64__)) || (defined(_M_IX86) || defined(__i386__)) || defined(_MSC_VER)
RAND_bytes (priv, 256);
#else
// lower 226 bits (28 bytes and 2 bits) only. short exponent
@ -555,7 +561,7 @@ namespace crypto
}
// AES
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
#define KeyExpansion256(round0,round1) \
"pshufd $0xff, %%xmm2, %%xmm2 \n" \
"movaps %%xmm1, %%xmm4 \n" \
@ -580,7 +586,7 @@ namespace crypto
"movaps %%xmm3, "#round1"(%[sched]) \n"
#endif
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
void ECBCryptoAESNI::ExpandKey (const AESKey& key)
{
__asm__
@ -621,7 +627,7 @@ namespace crypto
#endif
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
#define EncryptAES256(sched) \
"pxor (%["#sched"]), %%xmm0 \n" \
"aesenc 16(%["#sched"]), %%xmm0 \n" \
@ -642,16 +648,18 @@ namespace crypto
void ECBEncryption::Encrypt (const ChipherBlock * in, ChipherBlock * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
"movups (%[in]), %%xmm0 \n"
EncryptAES256(sched)
"movups %%xmm0, (%[out]) \n"
: : [sched]"r"(GetKeySchedule ()), [in]"r"(in), [out]"r"(out) : "%xmm0", "memory"
);
(
"movups (%[in]), %%xmm0 \n"
EncryptAES256(sched)
"movups %%xmm0, (%[out]) \n"
:
: [sched]"r"(GetKeySchedule ()), [in]"r"(in), [out]"r"(out)
: "%xmm0", "memory"
);
}
else
#endif
@ -660,7 +668,7 @@ namespace crypto
}
}
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
#define DecryptAES256(sched) \
"pxor 224(%["#sched"]), %%xmm0 \n" \
"aesdec 208(%["#sched"]), %%xmm0 \n" \
@ -681,16 +689,18 @@ namespace crypto
void ECBDecryption::Decrypt (const ChipherBlock * in, ChipherBlock * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
"movups (%[in]), %%xmm0 \n"
DecryptAES256(sched)
"movups %%xmm0, (%[out]) \n"
: : [sched]"r"(GetKeySchedule ()), [in]"r"(in), [out]"r"(out) : "%xmm0", "memory"
);
(
"movups (%[in]), %%xmm0 \n"
DecryptAES256(sched)
"movups %%xmm0, (%[out]) \n"
:
: [sched]"r"(GetKeySchedule ()), [in]"r"(in), [out]"r"(out)
: "%xmm0", "memory"
);
}
else
#endif
@ -699,7 +709,7 @@ namespace crypto
}
}
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
#define CallAESIMC(offset) \
"movaps "#offset"(%[shed]), %%xmm0 \n" \
"aesimc %%xmm0, %%xmm0 \n" \
@ -708,7 +718,7 @@ namespace crypto
void ECBEncryption::SetKey (const AESKey& key)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
ExpandKey (key);
@ -722,28 +732,30 @@ namespace crypto
void ECBDecryption::SetKey (const AESKey& key)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
ExpandKey (key); // expand encryption key first
// then invert it using aesimc
__asm__
(
CallAESIMC(16)
CallAESIMC(32)
CallAESIMC(48)
CallAESIMC(64)
CallAESIMC(80)
CallAESIMC(96)
CallAESIMC(112)
CallAESIMC(128)
CallAESIMC(144)
CallAESIMC(160)
CallAESIMC(176)
CallAESIMC(192)
CallAESIMC(208)
: : [shed]"r"(GetKeySchedule ()) : "%xmm0", "memory"
);
(
CallAESIMC(16)
CallAESIMC(32)
CallAESIMC(48)
CallAESIMC(64)
CallAESIMC(80)
CallAESIMC(96)
CallAESIMC(112)
CallAESIMC(128)
CallAESIMC(144)
CallAESIMC(160)
CallAESIMC(176)
CallAESIMC(192)
CallAESIMC(208)
:
: [shed]"r"(GetKeySchedule ())
: "%xmm0", "memory"
);
}
else
#endif
@ -754,28 +766,28 @@ namespace crypto
void CBCEncryption::Encrypt (int numBlocks, const ChipherBlock * in, ChipherBlock * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
"movups (%[iv]), %%xmm1 \n"
"1: \n"
"movups (%[in]), %%xmm0 \n"
"pxor %%xmm1, %%xmm0 \n"
EncryptAES256(sched)
"movaps %%xmm0, %%xmm1 \n"
"movups %%xmm0, (%[out]) \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"dec %[num] \n"
"jnz 1b \n"
"movups %%xmm1, (%[iv]) \n"
:
: [iv]"r"((uint8_t *)m_LastBlock), [sched]"r"(m_ECBEncryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(numBlocks)
: "%xmm0", "%xmm1", "cc", "memory"
);
(
"movups (%[iv]), %%xmm1 \n"
"1: \n"
"movups (%[in]), %%xmm0 \n"
"pxor %%xmm1, %%xmm0 \n"
EncryptAES256(sched)
"movaps %%xmm0, %%xmm1 \n"
"movups %%xmm0, (%[out]) \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"dec %[num] \n"
"jnz 1b \n"
"movups %%xmm1, (%[iv]) \n"
:
: [iv]"r"((uint8_t *)m_LastBlock), [sched]"r"(m_ECBEncryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(numBlocks)
: "%xmm0", "%xmm1", "cc", "memory"
);
}
else
#endif
@ -799,22 +811,22 @@ namespace crypto
void CBCEncryption::Encrypt (const uint8_t * in, uint8_t * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
"movups (%[iv]), %%xmm1 \n"
"movups (%[in]), %%xmm0 \n"
"pxor %%xmm1, %%xmm0 \n"
EncryptAES256(sched)
"movups %%xmm0, (%[out]) \n"
"movups %%xmm0, (%[iv]) \n"
:
: [iv]"r"((uint8_t *)m_LastBlock), [sched]"r"(m_ECBEncryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out)
: "%xmm0", "%xmm1", "memory"
);
(
"movups (%[iv]), %%xmm1 \n"
"movups (%[in]), %%xmm0 \n"
"pxor %%xmm1, %%xmm0 \n"
EncryptAES256(sched)
"movups %%xmm0, (%[out]) \n"
"movups %%xmm0, (%[iv]) \n"
:
: [iv]"r"((uint8_t *)m_LastBlock), [sched]"r"(m_ECBEncryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out)
: "%xmm0", "%xmm1", "memory"
);
}
else
#endif
@ -823,29 +835,29 @@ namespace crypto
void CBCDecryption::Decrypt (int numBlocks, const ChipherBlock * in, ChipherBlock * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
"movups (%[iv]), %%xmm1 \n"
"1: \n"
"movups (%[in]), %%xmm0 \n"
"movaps %%xmm0, %%xmm2 \n"
DecryptAES256(sched)
"pxor %%xmm1, %%xmm0 \n"
"movups %%xmm0, (%[out]) \n"
"movaps %%xmm2, %%xmm1 \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"dec %[num] \n"
"jnz 1b \n"
"movups %%xmm1, (%[iv]) \n"
:
: [iv]"r"((uint8_t *)m_IV), [sched]"r"(m_ECBDecryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(numBlocks)
: "%xmm0", "%xmm1", "%xmm2", "cc", "memory"
);
(
"movups (%[iv]), %%xmm1 \n"
"1: \n"
"movups (%[in]), %%xmm0 \n"
"movaps %%xmm0, %%xmm2 \n"
DecryptAES256(sched)
"pxor %%xmm1, %%xmm0 \n"
"movups %%xmm0, (%[out]) \n"
"movaps %%xmm2, %%xmm1 \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"dec %[num] \n"
"jnz 1b \n"
"movups %%xmm1, (%[iv]) \n"
:
: [iv]"r"((uint8_t *)m_IV), [sched]"r"(m_ECBDecryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(numBlocks)
: "%xmm0", "%xmm1", "%xmm2", "cc", "memory"
);
}
else
#endif
@ -869,22 +881,22 @@ namespace crypto
void CBCDecryption::Decrypt (const uint8_t * in, uint8_t * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
"movups (%[iv]), %%xmm1 \n"
"movups (%[in]), %%xmm0 \n"
"movups %%xmm0, (%[iv]) \n"
DecryptAES256(sched)
"pxor %%xmm1, %%xmm0 \n"
"movups %%xmm0, (%[out]) \n"
:
: [iv]"r"((uint8_t *)m_IV), [sched]"r"(m_ECBDecryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out)
: "%xmm0", "%xmm1", "memory"
);
(
"movups (%[iv]), %%xmm1 \n"
"movups (%[in]), %%xmm0 \n"
"movups %%xmm0, (%[iv]) \n"
DecryptAES256(sched)
"pxor %%xmm1, %%xmm0 \n"
"movups %%xmm0, (%[out]) \n"
:
: [iv]"r"((uint8_t *)m_IV), [sched]"r"(m_ECBDecryption.GetKeySchedule ()),
[in]"r"(in), [out]"r"(out)
: "%xmm0", "%xmm1", "memory"
);
}
else
#endif
@ -893,34 +905,34 @@ namespace crypto
void TunnelEncryption::Encrypt (const uint8_t * in, uint8_t * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
// encrypt IV
"movups (%[in]), %%xmm0 \n"
EncryptAES256(sched_iv)
"movaps %%xmm0, %%xmm1 \n"
// double IV encryption
EncryptAES256(sched_iv)
"movups %%xmm0, (%[out]) \n"
// encrypt data, IV is xmm1
"1: \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"movups (%[in]), %%xmm0 \n"
"pxor %%xmm1, %%xmm0 \n"
EncryptAES256(sched_l)
"movaps %%xmm0, %%xmm1 \n"
"movups %%xmm0, (%[out]) \n"
"dec %[num] \n"
"jnz 1b \n"
:
: [sched_iv]"r"(m_IVEncryption.GetKeySchedule ()), [sched_l]"r"(m_LayerEncryption.ECB().GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(63) // 63 blocks = 1008 bytes
: "%xmm0", "%xmm1", "cc", "memory"
);
(
// encrypt IV
"movups (%[in]), %%xmm0 \n"
EncryptAES256(sched_iv)
"movaps %%xmm0, %%xmm1 \n"
// double IV encryption
EncryptAES256(sched_iv)
"movups %%xmm0, (%[out]) \n"
// encrypt data, IV is xmm1
"1: \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"movups (%[in]), %%xmm0 \n"
"pxor %%xmm1, %%xmm0 \n"
EncryptAES256(sched_l)
"movaps %%xmm0, %%xmm1 \n"
"movups %%xmm0, (%[out]) \n"
"dec %[num] \n"
"jnz 1b \n"
:
: [sched_iv]"r"(m_IVEncryption.GetKeySchedule ()), [sched_l]"r"(m_LayerEncryption.ECB().GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(63) // 63 blocks = 1008 bytes
: "%xmm0", "%xmm1", "cc", "memory"
);
}
else
#endif
@ -934,35 +946,35 @@ namespace crypto
void TunnelDecryption::Decrypt (const uint8_t * in, uint8_t * out)
{
#if defined(__AES__) && (defined(__x86_64__) || defined(__i386__))
#if SUPPORTS_AES
if(i2p::cpu::aesni)
{
__asm__
(
// decrypt IV
"movups (%[in]), %%xmm0 \n"
DecryptAES256(sched_iv)
"movaps %%xmm0, %%xmm1 \n"
// double IV encryption
DecryptAES256(sched_iv)
"movups %%xmm0, (%[out]) \n"
// decrypt data, IV is xmm1
"1: \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"movups (%[in]), %%xmm0 \n"
"movaps %%xmm0, %%xmm2 \n"
DecryptAES256(sched_l)
"pxor %%xmm1, %%xmm0 \n"
"movups %%xmm0, (%[out]) \n"
"movaps %%xmm2, %%xmm1 \n"
"dec %[num] \n"
"jnz 1b \n"
:
: [sched_iv]"r"(m_IVDecryption.GetKeySchedule ()), [sched_l]"r"(m_LayerDecryption.ECB().GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(63) // 63 blocks = 1008 bytes
: "%xmm0", "%xmm1", "%xmm2", "cc", "memory"
);
(
// decrypt IV
"movups (%[in]), %%xmm0 \n"
DecryptAES256(sched_iv)
"movaps %%xmm0, %%xmm1 \n"
// double IV encryption
DecryptAES256(sched_iv)
"movups %%xmm0, (%[out]) \n"
// decrypt data, IV is xmm1
"1: \n"
"add $16, %[in] \n"
"add $16, %[out] \n"
"movups (%[in]), %%xmm0 \n"
"movaps %%xmm0, %%xmm2 \n"
DecryptAES256(sched_l)
"pxor %%xmm1, %%xmm0 \n"
"movups %%xmm0, (%[out]) \n"
"movaps %%xmm2, %%xmm1 \n"
"dec %[num] \n"
"jnz 1b \n"
:
: [sched_iv]"r"(m_IVDecryption.GetKeySchedule ()), [sched_l]"r"(m_LayerDecryption.ECB().GetKeySchedule ()),
[in]"r"(in), [out]"r"(out), [num]"r"(63) // 63 blocks = 1008 bytes
: "%xmm0", "%xmm1", "%xmm2", "cc", "memory"
);
}
else
#endif
@ -1285,9 +1297,9 @@ namespace crypto
}
}*/
void InitCrypto (bool precomputation, bool aesni, bool avx, bool force)
void InitCrypto (bool precomputation, bool aesni, bool force)
{
i2p::cpu::Detect (aesni, avx, force);
i2p::cpu::Detect (aesni, force);
#if LEGACY_OPENSSL
SSL_library_init ();
#endif
@ -1297,7 +1309,7 @@ namespace crypto
CRYPTO_set_locking_callback (OpensslLockingCallback);*/
if (precomputation)
{
#if defined(__x86_64__)
#if (defined(_M_AMD64) || defined(__x86_64__))
g_ElggTable = new BIGNUM * [ELGAMAL_FULL_EXPONENT_NUM_BYTES][255];
PrecalculateElggTable (g_ElggTable, ELGAMAL_FULL_EXPONENT_NUM_BYTES);
#else
@ -1312,7 +1324,7 @@ namespace crypto
if (g_ElggTable)
{
DestroyElggTable (g_ElggTable,
#if defined(__x86_64__)
#if (defined(_M_AMD64) || defined(__x86_64__))
ELGAMAL_FULL_EXPONENT_NUM_BYTES
#else
ELGAMAL_SHORT_EXPONENT_NUM_BYTES