shadps4-emu/ext-cryptopp
1
0
Fork 0
mirror of https://github.com/shadps4-emu/ext-cryptopp.git synced 2024-11-23 18:09:48 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
776a2195bd
ext-cryptopp/adv-simd.h

2382 lines
101 KiB
C
Raw Normal View History

Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// adv-simd.h - written and placed in the public domain by Jeffrey Walton
Update documentation
2018-07-02 02:25:07 +00:00
/// \file adv-simd.h
/// \brief Template for AdvancedProcessBlocks and SIMD processing
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// The SIMD based implementations for ciphers that use SSE, NEON and Power7
// have a commom pattern. Namely, they have a specialized implementation of
// AdvancedProcessBlocks which processes multiple block using hardware
// acceleration. After several implementations we noticed a lot of copy and
// paste occuring. adv-simd.h provides a template to avoid the copy and paste.
//
Update comments
2018-08-14 09:15:32 +00:00
// There are 11 templates provided in this file. The number following the
// function name, 64 or 128, is the block size. The name following the block
// size is the arrangement and acceleration. For example 4x1_SSE means Intel
// SSE using two encrypt (or decrypt) functions: one that operates on 4 SIMD
// words, and one that operates on 1 SIMD words.
//
// The distinction between SIMD words versus cipher blocks is important
Update comments
2018-08-15 00:49:26 +00:00
// because 64-bit ciphers use one SIMD word for two cipher blocks. For
Update comments
2018-08-14 09:15:32 +00:00
// example, AdvancedProcessBlocks64_6x2_ALTIVEC operates on 6 and 2 SIMD
// words, which is 12 and 4 cipher blocks. The function will do the right
// thing even if there is only one 64-bit block to encrypt.
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
//
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
// * AdvancedProcessBlocks64_2x1_SSE
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
// * AdvancedProcessBlocks64_4x1_SSE
// * AdvancedProcessBlocks128_4x1_SSE
// * AdvancedProcessBlocks64_6x2_SSE
// * AdvancedProcessBlocks128_6x2_SSE
// * AdvancedProcessBlocks64_6x2_NEON
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
// * AdvancedProcessBlocks128_4x1_NEON
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
// * AdvancedProcessBlocks128_6x2_NEON
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
// * AdvancedProcessBlocks64_6x2_ALTIVEC
Add AdvancedProcessBlocks128_4x1_ALTIVEC template
2018-08-13 10:38:30 +00:00
// * AdvancedProcessBlocks128_4x1_ALTIVEC
Update comments
2018-08-12 05:12:00 +00:00
// * AdvancedProcessBlocks128_6x1_ALTIVEC
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
//
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
// If an arrangement ends in 2, like 6x2, then the template will handle the
Update comments
2018-08-15 00:49:26 +00:00
// single block case by padding with 0's and using the two SIMD word
// function. This happens at most one time when processing multiple blocks.
// The extra processing of a zero block is trivial and worth the tradeoff.
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
//
// The MAYBE_CONST macro present on x86 is a SunCC workaround. Some versions
// of SunCC lose/drop the const-ness in the F1 and F4 functions. It eventually
// results in a failed link due to the const/non-const mismatch.
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#ifndef CRYPTOPP_ADVANCED_SIMD_TEMPLATES
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#define CRYPTOPP_ADVANCED_SIMD_TEMPLATES
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#include "config.h"
#include "misc.h"
Fix Clang 5.0 "runtime error: addition of unsigned offset to 0xXXXX overflowed to 0xYYYY" (GH #549)
2017-12-16 23:18:53 +00:00
#include "stdcpp.h"
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#if (CRYPTOPP_ARM_NEON_AVAILABLE)
# include <arm_neon.h>
#endif
Sync CRYPTOPP_ARM_ACLE_AVAILABLE with Autotools
2018-10-25 18:08:09 +00:00
#if (CRYPTOPP_ARM_ACLE_AVAILABLE)
Add ARMv8.4 cpu feature detection support (GH #685) (#687) This PR adds ARMv8.4 cpu feature detection support. Previously we only needed ARMv8.1 and things were much easier. For example, ARMv8.1 `__ARM_FEATURE_CRYPTO` meant PMULL, AES, SHA-1 and SHA-256 were available. ARMv8.4 `__ARM_FEATURE_CRYPTO` means PMULL, AES, SHA-1, SHA-256, SHA-512, SHA-3, SM3 and SM4 are available. We still use the same pattern as before. We make something available based on compiler version and/or preprocessor macros. But this time around we had to tighten things up a bit to ensure ARMv8.4 did not cross-pollinate down into ARMv8.1. ARMv8.4 is largely untested at the moment. There is no hardware in the field and CI lacks QEMU with the relevant patches/support. We will probably have to revisit some of this stuff in the future. Since this update applies to ARM gadgets we took the time to expand Android and iOS testing on Travis. Travis now tests more platforms, and includes Autotools and CMake builds, too.
2018-07-15 12:35:14 +00:00
# include <stdint.h>
# include <arm_acle.h>
#endif
Fix build with Embarcadero C++Builder 10.2.3 (#696) Fix two compilation errors encountered with C++Builder (Starter Edition): - In `cpu.cpp`, 0ccdc197b introduced a dependency on `_xgetbv()` from `<immintrin.h>` that doesn't exist on C++Builder. Enlist it for the workaround, similar to SunCC in 692ed2a2b. - In `adv-simd.h`, `<pmmintrin.h>` is being #included under the `CRYPTOPP_SSE2_INTRIN_AVAILABLE` macro. This header, [which apparently provides SSE3 intrinsics](https://stackoverflow.com/a/11228864/1433768), is not shipped with C++Builder. (This section of code was recently downgraded from a SSSE3 to a SSE2 block in 09c8ae28, followed by moving away from `<immintrin.h>` in bc8da71a, followed by reintroducing the SSSE3 check in d1e646a5.) Split the SSE2 and SSSE3 cases such that `<pmmintrin.h>` is not #included for SSE2. This seems safe to do, because some `git grep` analysis shows that: - `adv-simd.h` is not #included by any other header, but only directly #included by some `.cpp` files. - Among those `.cpp` files, only `sm4-simd.cpp` has a `CRYPTOPP_SSE2_INTRIN_AVAILABLE` preprocessor block, and there it again includes the other two headers (`<emmintrin.h>` and `<xmmintrin.h>`). NOTE: I was compiling via the IDE after [setting up a project file](https://github.com/tanzislam/cryptopals/wiki/Importing-into-Embarcadero-C%E2%94%BC%E2%94%BCBuilder-Starter-10.2#using-the-crypto-library). My compilation command was effectively: ``` bcc32c.exe -DCRYPTOPP_NO_CXX11 -DCRYPTOPP_DISABLE_SSSE3 -D__SSE2__ -D__SSE__ -D__MMX__ ```
2018-08-05 02:54:36 +00:00
#if (CRYPTOPP_SSE2_INTRIN_AVAILABLE)
# include <emmintrin.h>
# include <xmmintrin.h>
#endif
Fix SunStudio 12.6 compile on i386
2018-07-16 13:37:08 +00:00
// SunCC needs CRYPTOPP_SSSE3_AVAILABLE, too
Fix build with Embarcadero C++Builder 10.2.3 (#696) Fix two compilation errors encountered with C++Builder (Starter Edition): - In `cpu.cpp`, 0ccdc197b introduced a dependency on `_xgetbv()` from `<immintrin.h>` that doesn't exist on C++Builder. Enlist it for the workaround, similar to SunCC in 692ed2a2b. - In `adv-simd.h`, `<pmmintrin.h>` is being #included under the `CRYPTOPP_SSE2_INTRIN_AVAILABLE` macro. This header, [which apparently provides SSE3 intrinsics](https://stackoverflow.com/a/11228864/1433768), is not shipped with C++Builder. (This section of code was recently downgraded from a SSSE3 to a SSE2 block in 09c8ae28, followed by moving away from `<immintrin.h>` in bc8da71a, followed by reintroducing the SSSE3 check in d1e646a5.) Split the SSE2 and SSSE3 cases such that `<pmmintrin.h>` is not #included for SSE2. This seems safe to do, because some `git grep` analysis shows that: - `adv-simd.h` is not #included by any other header, but only directly #included by some `.cpp` files. - Among those `.cpp` files, only `sm4-simd.cpp` has a `CRYPTOPP_SSE2_INTRIN_AVAILABLE` preprocessor block, and there it again includes the other two headers (`<emmintrin.h>` and `<xmmintrin.h>`). NOTE: I was compiling via the IDE after [setting up a project file](https://github.com/tanzislam/cryptopals/wiki/Importing-into-Embarcadero-C%E2%94%BC%E2%94%BCBuilder-Starter-10.2#using-the-crypto-library). My compilation command was effectively: ``` bcc32c.exe -DCRYPTOPP_NO_CXX11 -DCRYPTOPP_DISABLE_SSSE3 -D__SSE2__ -D__SSE__ -D__MMX__ ```
2018-08-05 02:54:36 +00:00
#if (CRYPTOPP_SSSE3_AVAILABLE)
Fix early Fedora compiles
2018-07-06 05:14:28 +00:00
# include <emmintrin.h>
# include <pmmintrin.h>
Use inline for LEA_Encryption and LEA_Decryption
2018-06-23 16:58:55 +00:00
# include <xmmintrin.h>
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#endif
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
#if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
# include "ppc-simd.h"
#endif
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// ************************ All block ciphers *********************** //
ANONYMOUS_NAMESPACE_BEGIN
using CryptoPP::BlockTransformation;
CRYPTOPP_CONSTANT(BT_XorInput = BlockTransformation::BT_XorInput)
CRYPTOPP_CONSTANT(BT_AllowParallel = BlockTransformation::BT_AllowParallel)
CRYPTOPP_CONSTANT(BT_InBlockIsCounter = BlockTransformation::BT_InBlockIsCounter)
CRYPTOPP_CONSTANT(BT_ReverseDirection = BlockTransformation::BT_ReverseDirection)
CRYPTOPP_CONSTANT(BT_DontIncrementInOutPointers = BlockTransformation::BT_DontIncrementInOutPointers)
ANONYMOUS_NAMESPACE_END
// *************************** ARM NEON ************************** //
Allow user to set -DCRYPTOPP_ARM_NEON_AVAILABLE=0 (#595) Disable neon through -DCRYPTOPP_ARM_NEON_AVAILABLE=0, replace "if defined(CRYPTOPP_ARM_NEON_AVAILABLE)" by "if (CRYPTOPP_ARM_NEON_AVAILABLE)" Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
2018-03-05 23:49:10 +00:00
#if (CRYPTOPP_ARM_NEON_AVAILABLE)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
NAMESPACE_BEGIN(CryptoPP)
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 2 and 6 blocks
/// \tparam F2 function to process 2 64-bit blocks
/// \tparam F6 function to process 6 64-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks64_6x2_NEON processes 6 and 2 NEON SIMD words
/// at a time. For a single block the template uses F2 with a zero block.
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type.
template <typename F2, typename F6, typename W>
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
inline size_t AdvancedProcessBlocks64_6x2_NEON(F2 func2, F6 func6,
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
const W *subKeys, size_t rounds, const byte *inBlocks,
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 8);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint32x4_t s_one = {0, 0, 0, 1<<24};
const uint32x4_t s_two = {0, 2<<24, 0, 2<<24};
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#else
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
// TODO: verify these constants on ARM-BE
const uint32x4_t s_one = {0, 0, 0, 1};
const uint32x4_t s_two = {0, 2, 0, 2};
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#endif
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 8;
const size_t neonBlockSize = 16;
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : neonBlockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? neonBlockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : neonBlockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - neonBlockSize);
xorBlocks = PtrAdd(xorBlocks, length - neonBlockSize);
outBlocks = PtrAdd(outBlocks, length - neonBlockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 6*neonBlockSize)
{
uint32x4_t block0, block1, block2, block3, block4, block5;
if (flags & BT_InBlockIsCounter)
{
// For 64-bit block ciphers we need to load the CTR block, which is 8 bytes.
// After the dup load we have two counters in the NEON word. Then we need
// to increment the low ctr by 0 and the high ctr by 1.
const uint8x8_t ctr = vld1_u8(inBlocks);
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block0 = vaddq_u32(s_one, vreinterpretq_u32_u8(vcombine_u8(ctr,ctr)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// After initial increment of {0,1} remaining counters increment by {2,2}.
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block1 = vaddq_u32(s_two, block0);
block2 = vaddq_u32(s_two, block1);
block3 = vaddq_u32(s_two, block2);
block4 = vaddq_u32(s_two, block3);
block5 = vaddq_u32(s_two, block4);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1_u8(const_cast<byte*>(inBlocks), vget_low_u8(
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
vreinterpretq_u8_u32(vaddq_u32(s_two, block5))));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u32(block0, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u32(block1, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = veorq_u32(block2, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = veorq_u32(block3, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = veorq_u32(block4, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = veorq_u32(block5, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func6(block0, block1, block2, block3, block4, block5, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u32(block0, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u32(block1, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = veorq_u32(block2, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = veorq_u32(block3, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = veorq_u32(block4, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = veorq_u32(block5, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block0));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block1));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block2));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block3));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block4));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block5));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 6*neonBlockSize;
}
while (length >= 2*neonBlockSize)
{
uint32x4_t block0, block1;
if (flags & BT_InBlockIsCounter)
{
// For 64-bit block ciphers we need to load the CTR block, which is 8 bytes.
// After the dup load we have two counters in the NEON word. Then we need
// to increment the low ctr by 0 and the high ctr by 1.
const uint8x8_t ctr = vld1_u8(inBlocks);
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block0 = vaddq_u32(s_one, vreinterpretq_u32_u8(vcombine_u8(ctr,ctr)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// After initial increment of {0,1} remaining counters increment by {2,2}.
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block1 = vaddq_u32(s_two, block0);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1_u8(const_cast<byte*>(inBlocks), vget_low_u8(
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
vreinterpretq_u8_u32(vaddq_u32(s_two, block1))));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = vreinterpretq_u32_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u32(block0, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u32(block1, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func2(block0, block1, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u32(block0, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u32(block1, vreinterpretq_u32_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block0));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u32(block1));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 2*neonBlockSize;
}
}
if (length)
{
// Adjust to real block size
if (flags & BT_ReverseDirection)
{
inIncrement += inIncrement ? blockSize : 0;
xorIncrement += xorIncrement ? blockSize : 0;
outIncrement += outIncrement ? blockSize : 0;
Fix typo in AdvancedProcessBlocks64_6x2_NEON AdvancedProcessBlocks64_6x2_NEON was only used by Simon and Speck. It was causing failures in wide-block tests for CBC decryption only.
2018-10-13 23:20:08 +00:00
inBlocks = PtrSub(inBlocks, inIncrement);
Use PtrSub to subtract offset from pointer
2018-08-15 07:16:08 +00:00
xorBlocks = PtrSub(xorBlocks, xorIncrement);
outBlocks = PtrSub(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
inIncrement -= inIncrement ? blockSize : 0;
xorIncrement -= xorIncrement ? blockSize : 0;
outIncrement -= outIncrement ? blockSize : 0;
}
while (length >= blockSize)
{
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
uint32x4_t block, zero = {0};
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
const uint8x8_t v = vld1_u8(inBlocks);
block = vreinterpretq_u32_u8(vcombine_u8(v,v));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
const uint8x8_t x = vld1_u8(xorBlocks);
block = veorq_u32(block, vreinterpretq_u32_u8(vcombine_u8(x,x)));
}
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[7]++;
func2(block, zero, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
const uint8x8_t x = vld1_u8(xorBlocks);
block = veorq_u32(block, vreinterpretq_u32_u8(vcombine_u8(x,x)));
}
vst1_u8(const_cast<byte*>(outBlocks),
vget_low_u8(vreinterpretq_u8_u32(block)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= blockSize;
}
}
return length;
}
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 1 and 6 blocks
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
/// \tparam F1 function to process 1 128-bit block
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \tparam F6 function to process 6 128-bit blocks
/// \tparam W word type of the subkey table
Fix AdvancedProcessBlocks128_6x1_NEON template name
2018-06-23 16:35:06 +00:00
/// \details AdvancedProcessBlocks128_6x1_NEON processes 6 and 2 NEON SIMD words
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// at a time.
/// \details The subkey type is usually word32 or word64. F1 and F6 must use the
/// same word type.
template <typename F1, typename F6, typename W>
Fix AdvancedProcessBlocks128_6x1_NEON template name
2018-06-23 16:35:06 +00:00
inline size_t AdvancedProcessBlocks128_6x1_NEON(F1 func1, F6 func6,
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
const W *subKeys, size_t rounds, const byte *inBlocks,
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint32x4_t s_one = {0, 0, 0, 1<<24};
const uint32x4_t s_two = {0, 2<<24, 0, 2<<24};
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#else
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
// TODO: verify these constants on ARM-BE
const uint32x4_t s_one = {0, 0, 0, 1};
const uint32x4_t s_two = {0, 2, 0, 2};
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#endif
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 16;
// const size_t neonBlockSize = 16;
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 6*blockSize)
{
uint64x2_t block0, block1, block2, block3, block4, block5;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint64x2_t one = vreinterpretq_u64_u32(s_one);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block1 = vaddq_u64(block0, one);
block2 = vaddq_u64(block1, one);
block3 = vaddq_u64(block2, one);
block4 = vaddq_u64(block3, one);
block5 = vaddq_u64(block4, one);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
vst1q_u8(const_cast<byte*>(inBlocks),
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
vreinterpretq_u8_u64(vaddq_u64(block5, one)));
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
else
{
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block1 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block2 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block3 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block4 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block5 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block2 = veorq_u64(block2, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block3 = veorq_u64(block3, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block4 = veorq_u64(block4, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block5 = veorq_u64(block5, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
func6(block0, block1, block2, block3, block4, block5, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block2 = veorq_u64(block2, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block3 = veorq_u64(block3, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block4 = veorq_u64(block4, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block5 = veorq_u64(block5, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block0));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block1));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block2));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block3));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block4));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block5));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
length -= 6*blockSize;
}
}
while (length >= blockSize)
{
uint64x2_t block;
block = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block = veorq_u64(block, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func1(block, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block = veorq_u64(block, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
length -= blockSize;
}
return length;
}
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
/// \brief AdvancedProcessBlocks for 1 and 4 blocks
/// \tparam F1 function to process 1 128-bit block
/// \tparam F4 function to process 4 128-bit blocks
/// \tparam W word type of the subkey table
Update documentation
2018-06-23 16:27:25 +00:00
/// \tparam V vector type of the NEON datatype
/// \details AdvancedProcessBlocks128_4x1_NEON processes 4 and 1 NEON SIMD words
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
/// at a time.
/// \details The subkey type is usually word32 or word64. V is the vector type and it is
/// usually uint32x4_t or uint64x2_t. F1, F4, W and V must use the same word and
Update documentation
2018-06-23 16:27:25 +00:00
/// vector type. The V parameter is used to avoid template argument
/// deduction/substitution failures.
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
template <typename F1, typename F4, typename W, typename V>
inline size_t AdvancedProcessBlocks128_4x1_NEON(F1 func1, F4 func4,
const V& unused, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
CRYPTOPP_UNUSED(unused);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint32x4_t s_one = {0, 0, 0, 1<<24};
const uint32x4_t s_two = {0, 2<<24, 0, 2<<24};
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
#else
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
// TODO: verify these constants on ARM-BE
const uint32x4_t s_one = {0, 0, 0, 1};
const uint32x4_t s_two = {0, 2, 0, 2};
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
#endif
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 16;
// const size_t neonBlockSize = 16;
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 4*blockSize)
{
Clear unused variable warning
2018-07-14 16:59:42 +00:00
uint64x2_t block0, block1, block2, block3;
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint64x2_t one = vreinterpretq_u64_u32(s_one);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block1 = vaddq_u64(block0, one);
block2 = vaddq_u64(block1, one);
block3 = vaddq_u64(block2, one);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
vst1q_u8(const_cast<byte*>(inBlocks),
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
vreinterpretq_u8_u64(vaddq_u64(block3, one)));
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
}
else
{
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block1 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block2 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block3 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
}
if (xorInput)
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block2 = veorq_u64(block2, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block3 = veorq_u64(block3, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
}
func4((V&)block0, (V&)block1, (V&)block2, (V&)block3, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block2 = veorq_u64(block2, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
block3 = veorq_u64(block3, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
}
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block0));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block1));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block2));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block3));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
length -= 4*blockSize;
}
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
}
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
while (length >= blockSize)
{
uint64x2_t block = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
if (xorInput)
block = veorq_u64(block, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func1( (V&)block, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
block = veorq_u64(block, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add LEA-128 NEON and ARMv8 implementation (GH #669) LEA-128(128) from 35.6 cpb to 14.11 cpb on a LeMaker HiKey dev-board. LEA-128 from 12.60 cpb to 11.89 cpb on AMD Opteron 1100.
2018-06-23 07:54:51 +00:00
length -= blockSize;
}
return length;
}
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 2 and 6 blocks
/// \tparam F2 function to process 2 128-bit blocks
/// \tparam F6 function to process 6 128-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks128_6x2_NEON processes 6 and 2 NEON SIMD words
/// at a time. For a single block the template uses F2 with a zero block.
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type.
template <typename F2, typename F6, typename W>
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
inline size_t AdvancedProcessBlocks128_6x2_NEON(F2 func2, F6 func6,
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
const W *subKeys, size_t rounds, const byte *inBlocks,
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint32x4_t s_one = {0, 0, 0, 1<<24};
const uint32x4_t s_two = {0, 2<<24, 0, 2<<24};
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#else
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
// TODO: verify these constants on ARM-BE
const uint32x4_t s_one = {0, 0, 0, 1};
const uint32x4_t s_two = {0, 2, 0, 2};
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
#endif
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 16;
// const size_t neonBlockSize = 16;
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 6*blockSize)
{
uint64x2_t block0, block1, block2, block3, block4, block5;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint64x2_t one = vreinterpretq_u64_u32(s_one);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block1 = vaddq_u64(block0, one);
block2 = vaddq_u64(block1, one);
block3 = vaddq_u64(block2, one);
block4 = vaddq_u64(block3, one);
block5 = vaddq_u64(block4, one);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(const_cast<byte*>(inBlocks),
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
vreinterpretq_u8_u64(vaddq_u64(block5, one)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = veorq_u64(block2, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = veorq_u64(block3, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = veorq_u64(block4, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = veorq_u64(block5, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func6(block0, block1, block2, block3, block4, block5, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = veorq_u64(block2, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = veorq_u64(block3, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = veorq_u64(block4, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = veorq_u64(block5, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block0));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block1));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block2));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block3));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block4));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block5));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 6*blockSize;
}
while (length >= 2*blockSize)
{
uint64x2_t block0, block1;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
const uint64x2_t one = vreinterpretq_u64_u32(s_one);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
block1 = vaddq_u64(block0, one);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(const_cast<byte*>(inBlocks),
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
vreinterpretq_u8_u64(vaddq_u64(block1, one)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func2(block0, block1, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = veorq_u64(block0, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = veorq_u64(block1, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block0));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block1));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 2*blockSize;
}
}
while (length >= blockSize)
{
uint64x2_t block, zero = {0,0};
block = vreinterpretq_u64_u8(vld1q_u8(inBlocks));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block = veorq_u64(block, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func2(block, zero, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block = veorq_u64(block, vreinterpretq_u64_u8(vld1q_u8(xorBlocks)));
vst1q_u8(outBlocks, vreinterpretq_u8_u64(block));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= blockSize;
}
return length;
}
Fix unaligned load for _mm_loaddup_pd with GCC and UBsan This function was missed earlier. Unfortunately, it does not squash all of the unaligned load findings. I'm pretty sure it is a GCC problem
2017-12-28 06:16:17 +00:00
NAMESPACE_END // CryptoPP
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#endif // CRYPTOPP_ARM_NEON_AVAILABLE
// *************************** Intel SSE ************************** //
#if defined(CRYPTOPP_SSSE3_AVAILABLE)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
// Hack for SunCC, http://github.com/weidai11/cryptopp/issues/224
#if (__SUNPRO_CC >= 0x5130)
# define MAYBE_CONST
# define MAYBE_UNCONST_CAST(T, x) const_cast<MAYBE_CONST T>(x)
#else
# define MAYBE_CONST const
# define MAYBE_UNCONST_CAST(T, x) (x)
#endif
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// Clang __m128i casts, http://bugs.llvm.org/show_bug.cgi?id=20670
#ifndef M128_CAST
# define M128_CAST(x) ((__m128i *)(void *)(x))
#endif
#ifndef CONST_M128_CAST
# define CONST_M128_CAST(x) ((const __m128i *)(const void *)(x))
#endif
NAMESPACE_BEGIN(CryptoPP)
Update documentation
2018-07-01 07:29:12 +00:00
/// \brief AdvancedProcessBlocks for 1 and 2 blocks
/// \tparam F1 function to process 1 64-bit block
Update documentation
2018-07-01 07:42:17 +00:00
/// \tparam F2 function to process 2 64-bit blocks
Update documentation
2018-07-01 07:29:12 +00:00
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks64_2x1_SSE processes 2 and 1 SSE SIMD words
/// at a time.
/// \details The subkey type is usually word32 or word64. F1 and F2 must use the
/// same word type.
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
template <typename F1, typename F2, typename W>
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
inline size_t AdvancedProcessBlocks64_2x1_SSE(F1 func1, F2 func2,
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 8);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 8;
const size_t xmmBlockSize = 16;
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : xmmBlockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? xmmBlockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : xmmBlockSize;
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - xmmBlockSize);
xorBlocks = PtrAdd(xorBlocks, length - xmmBlockSize);
outBlocks = PtrAdd(outBlocks, length - xmmBlockSize);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
double temp[2];
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
while (length >= 2*xmmBlockSize)
{
__m128i block0, block1;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 and 2 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
const __m128i s_two = _mm_set_epi32(2<<24, 0, 2<<24, 0);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
// For 64-bit block ciphers we need to load the CTR block, which is 8 bytes.
// After the dup load we have two counters in the XMM word. Then we need
// to increment the low ctr by 0 and the high ctr by 1.
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, inBlocks, blockSize);
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block0 = _mm_add_epi32(s_one, _mm_castpd_si128(_mm_loaddup_pd(temp)));
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
// After initial increment of {0,1} remaining counters increment by {2,2}.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(s_two, block0);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
Update comments
2018-07-01 08:03:30 +00:00
// Store the next counter. When BT_InBlockIsCounter is set then
// inBlocks is backed by m_counterArray which is non-const.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
_mm_store_sd(temp, _mm_castsi128_pd(_mm_add_epi64(s_two, block1)));
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(const_cast<byte*>(inBlocks), temp, blockSize);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
if (xorInput)
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
func2(block0, block1, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
length -= 2*xmmBlockSize;
}
}
if (length)
{
// Adjust to real block size
if (flags & BT_ReverseDirection)
{
inIncrement += inIncrement ? blockSize : 0;
xorIncrement += xorIncrement ? blockSize : 0;
outIncrement += outIncrement ? blockSize : 0;
Use PtrSub to subtract offset from pointer
2018-08-15 07:16:08 +00:00
inBlocks = PtrSub(inBlocks, inIncrement);
xorBlocks = PtrSub(xorBlocks, xorIncrement);
outBlocks = PtrSub(outBlocks, outIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
else
{
inIncrement -= inIncrement ? blockSize : 0;
xorIncrement -= xorIncrement ? blockSize : 0;
outIncrement -= outIncrement ? blockSize : 0;
}
while (length >= blockSize)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
double temp[2];
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, inBlocks, blockSize);
__m128i block = _mm_castpd_si128(_mm_load_sd(temp));
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
if (xorInput)
{
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, xorBlocks, blockSize);
block = _mm_xor_si128(block, _mm_castpd_si128(_mm_load_sd(temp)));
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[7]++;
func1(block, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, xorBlocks, blockSize);
block = _mm_xor_si128(block, _mm_castpd_si128(_mm_load_sd(temp)));
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
}
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
_mm_store_sd(temp, _mm_castsi128_pd(block));
std::memcpy(outBlocks, temp, blockSize);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Add CHAM64 SSSE3 implementation (PR #670) CHAM64 from 20 cpb to 14 cpb on modern iCore. CHAM64 from 90 cpb to 18 cpb antique Core2 Duo
2018-06-21 04:37:10 +00:00
length -= blockSize;
}
}
return length;
}
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 2 and 6 blocks
/// \tparam F2 function to process 2 64-bit blocks
/// \tparam F6 function to process 6 64-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks64_6x2_SSE processes 6 and 2 SSE SIMD words
/// at a time. For a single block the template uses F2 with a zero block.
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type.
template <typename F2, typename F6, typename W>
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
inline size_t AdvancedProcessBlocks64_6x2_SSE(F2 func2, F6 func6,
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 8);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 8;
const size_t xmmBlockSize = 16;
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : xmmBlockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? xmmBlockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : xmmBlockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - xmmBlockSize);
xorBlocks = PtrAdd(xorBlocks, length - xmmBlockSize);
outBlocks = PtrAdd(outBlocks, length - xmmBlockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
double temp[2];
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
while (length >= 6*xmmBlockSize)
{
__m128i block0, block1, block2, block3, block4, block5;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 and 2 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
const __m128i s_two = _mm_set_epi32(2<<24, 0, 2<<24, 0);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// For 64-bit block ciphers we need to load the CTR block, which is 8 bytes.
// After the dup load we have two counters in the XMM word. Then we need
// to increment the low ctr by 0 and the high ctr by 1.
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, inBlocks, blockSize);
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block0 = _mm_add_epi32(s_one, _mm_castpd_si128(_mm_loaddup_pd(temp)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// After initial increment of {0,1} remaining counters increment by {2,2}.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(s_two, block0);
block2 = _mm_add_epi32(s_two, block1);
block3 = _mm_add_epi32(s_two, block2);
block4 = _mm_add_epi32(s_two, block3);
block5 = _mm_add_epi32(s_two, block4);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Update comments
2018-07-01 08:03:30 +00:00
// Store the next counter. When BT_InBlockIsCounter is set then
// inBlocks is backed by m_counterArray which is non-const.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
_mm_store_sd(temp, _mm_castsi128_pd(_mm_add_epi32(s_two, block5)));
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(const_cast<byte*>(inBlocks), temp, blockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = _mm_xor_si128(block4, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = _mm_xor_si128(block5, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func6(block0, block1, block2, block3, block4, block5, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = _mm_xor_si128(block4, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = _mm_xor_si128(block5, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block2);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block3);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block4);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block5);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 6*xmmBlockSize;
}
while (length >= 2*xmmBlockSize)
{
__m128i block0, block1;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 and 2 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
const __m128i s_two = _mm_set_epi32(2<<24, 0, 2<<24, 0);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// For 64-bit block ciphers we need to load the CTR block, which is 8 bytes.
// After the dup load we have two counters in the XMM word. Then we need
// to increment the low ctr by 0 and the high ctr by 1.
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, inBlocks, blockSize);
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block0 = _mm_add_epi32(s_one, _mm_castpd_si128(_mm_loaddup_pd(temp)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
// After initial increment of {0,1} remaining counters increment by {2,2}.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(s_two, block0);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Update comments
2018-07-01 08:03:30 +00:00
// Store the next counter. When BT_InBlockIsCounter is set then
// inBlocks is backed by m_counterArray which is non-const.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
_mm_store_sd(temp, _mm_castsi128_pd(_mm_add_epi64(s_two, block1)));
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(const_cast<byte*>(inBlocks), temp, blockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func2(block0, block1, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 2*xmmBlockSize;
}
}
if (length)
{
// Adjust to real block size
if (flags & BT_ReverseDirection)
{
inIncrement += inIncrement ? blockSize : 0;
xorIncrement += xorIncrement ? blockSize : 0;
outIncrement += outIncrement ? blockSize : 0;
Use PtrSub to subtract offset from pointer
2018-08-15 07:16:08 +00:00
inBlocks = PtrSub(inBlocks, inIncrement);
xorBlocks = PtrSub(xorBlocks, xorIncrement);
outBlocks = PtrSub(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
inIncrement -= inIncrement ? blockSize : 0;
xorIncrement -= xorIncrement ? blockSize : 0;
outIncrement -= outIncrement ? blockSize : 0;
}
while (length >= blockSize)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
double temp[2];
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
__m128i block, zero = _mm_setzero_si128();
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, inBlocks, blockSize);
block = _mm_castpd_si128(_mm_load_sd(temp));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, xorBlocks, blockSize);
block = _mm_xor_si128(block,
_mm_castpd_si128(_mm_load_sd(temp)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[7]++;
func2(block, zero, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, xorBlocks, blockSize);
block = _mm_xor_si128(block,
_mm_castpd_si128(_mm_load_sd(temp)));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
_mm_store_sd(temp, _mm_castsi128_pd(block));
std::memcpy(outBlocks, temp, blockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= blockSize;
}
}
return length;
}
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 2 and 6 blocks
/// \tparam F2 function to process 2 128-bit blocks
/// \tparam F6 function to process 6 128-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks128_6x2_SSE processes 6 and 2 SSE SIMD words
/// at a time. For a single block the template uses F2 with a zero block.
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type.
template <typename F2, typename F6, typename W>
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
inline size_t AdvancedProcessBlocks128_6x2_SSE(F2 func2, F6 func6,
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 16;
// const size_t xmmBlockSize = 16;
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 6*blockSize)
{
__m128i block0, block1, block2, block3, block4, block5;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(block0, s_one);
block2 = _mm_add_epi32(block1, s_one);
block3 = _mm_add_epi32(block2, s_one);
block4 = _mm_add_epi32(block3, s_one);
block5 = _mm_add_epi32(block4, s_one);
_mm_storeu_si128(M128_CAST(inBlocks), _mm_add_epi32(block5, s_one));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = _mm_xor_si128(block4, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = _mm_xor_si128(block5, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func6(block0, block1, block2, block3, block4, block5, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block4 = _mm_xor_si128(block4, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block5 = _mm_xor_si128(block5, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block2);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block3);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block4);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block5);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 6*blockSize;
}
while (length >= 2*blockSize)
{
__m128i block0, block1;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(block0, s_one);
_mm_storeu_si128(M128_CAST(inBlocks), _mm_add_epi32(block1, s_one));
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
func2(block0, block1, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= 2*blockSize;
}
}
while (length >= blockSize)
{
__m128i block, zero = _mm_setzero_si128();
block = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block = _mm_xor_si128(block, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func2(block, zero, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
block = _mm_xor_si128(block, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
_mm_storeu_si128(M128_CAST(outBlocks), block);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
length -= blockSize;
}
return length;
}
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 1 and 4 blocks
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
/// \tparam F1 function to process 1 128-bit block
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \tparam F4 function to process 4 128-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks128_4x1_SSE processes 4 and 1 SSE SIMD words
/// at a time.
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
/// \details The subkey type is usually word32 or word64. F1 and F4 must use the
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// same word type.
template <typename F1, typename F4, typename W>
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
inline size_t AdvancedProcessBlocks128_4x1_SSE(F1 func1, F4 func4,
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 16;
// const size_t xmmBlockSize = 16;
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 4*blockSize)
{
__m128i block0, block1, block2, block3;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(block0, s_one);
block2 = _mm_add_epi32(block1, s_one);
block3 = _mm_add_epi32(block2, s_one);
_mm_storeu_si128(M128_CAST(inBlocks), _mm_add_epi32(block3, s_one));
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block2 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block3 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
func4(block0, block1, block2, block3, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block2);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block3);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
length -= 4*blockSize;
}
}
while (length >= blockSize)
{
__m128i block = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block = _mm_xor_si128(block, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func1(block, subKeys, static_cast<unsigned int>(rounds));
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
block = _mm_xor_si128(block, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
_mm_storeu_si128(M128_CAST(outBlocks), block);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Refactor rijndael-simd.cpp and simon.simd.cpp to use adv-simd.h
2017-12-10 16:09:50 +00:00
length -= blockSize;
}
return length;
}
Update documentation
2018-07-01 07:29:12 +00:00
/// \brief AdvancedProcessBlocks for 1 and 4 blocks
/// \tparam F1 function to process 1 64-bit block
/// \tparam F4 function to process 6 64-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks64_4x1_SSE processes 4 and 1 SSE SIMD words
/// at a time.
/// \details The subkey type is usually word32 or word64. F1 and F4 must use the
/// same word type.
template <typename F1, typename F4, typename W>
inline size_t AdvancedProcessBlocks64_4x1_SSE(F1 func1, F4 func4,
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 8);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 8;
const size_t xmmBlockSize = 16;
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter | BT_DontIncrementInOutPointers)) ? 0 : xmmBlockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? xmmBlockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : xmmBlockSize;
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - xmmBlockSize);
xorBlocks = PtrAdd(xorBlocks, length - xmmBlockSize);
outBlocks = PtrAdd(outBlocks, length - xmmBlockSize);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
inIncrement = 0 - inIncrement;
xorIncrement = 0 - xorIncrement;
outIncrement = 0 - outIncrement;
}
if (flags & BT_AllowParallel)
{
Update comments
2018-08-14 09:15:32 +00:00
while (length >= 4*xmmBlockSize)
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
{
__m128i block0, block1, block2, block3;
if (flags & BT_InBlockIsCounter)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
// Increment of 1 and 2 in big-endian compatible with the ctr byte array.
const __m128i s_one = _mm_set_epi32(1<<24, 0, 0, 0);
const __m128i s_two = _mm_set_epi32(2<<24, 0, 2<<24, 0);
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
double temp[2];
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
// For 64-bit block ciphers we need to load the CTR block, which is 8 bytes.
// After the dup load we have two counters in the XMM word. Then we need
// to increment the low ctr by 0 and the high ctr by 1.
std::memcpy(temp, inBlocks, blockSize);
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block0 = _mm_add_epi32(s_one, _mm_castpd_si128(_mm_loaddup_pd(temp)));
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
// After initial increment of {0,1} remaining counters increment by {2,2}.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
block1 = _mm_add_epi32(s_two, block0);
block2 = _mm_add_epi32(s_two, block1);
block3 = _mm_add_epi32(s_two, block2);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
Update comments
2018-07-01 08:03:30 +00:00
// Store the next counter. When BT_InBlockIsCounter is set then
// inBlocks is backed by m_counterArray which is non-const.
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
_mm_store_sd(temp, _mm_castsi128_pd(_mm_add_epi64(s_two, block3)));
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(const_cast<byte*>(inBlocks), temp, blockSize);
}
else
{
block0 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block1 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block2 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block3 = _mm_loadu_si128(CONST_M128_CAST(inBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
}
if (xorInput)
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
}
Update documentation
2018-07-01 07:29:12 +00:00
func4(block0, block1, block2, block3, subKeys, static_cast<unsigned int>(rounds));
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
if (xorOutput)
{
block0 = _mm_xor_si128(block0, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block1 = _mm_xor_si128(block1, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block2 = _mm_xor_si128(block2, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
block3 = _mm_xor_si128(block3, _mm_loadu_si128(CONST_M128_CAST(xorBlocks)));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
}
_mm_storeu_si128(M128_CAST(outBlocks), block0);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block1);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block2);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
_mm_storeu_si128(M128_CAST(outBlocks), block3);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
Update comments
2018-08-14 09:15:32 +00:00
length -= 4*xmmBlockSize;
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
}
}
if (length)
{
// Adjust to real block size
if (flags & BT_ReverseDirection)
{
inIncrement += inIncrement ? blockSize : 0;
xorIncrement += xorIncrement ? blockSize : 0;
outIncrement += outIncrement ? blockSize : 0;
Use PtrSub to subtract offset from pointer
2018-08-15 07:16:08 +00:00
inBlocks = PtrSub(inBlocks, inIncrement);
xorBlocks = PtrSub(xorBlocks, xorIncrement);
outBlocks = PtrSub(outBlocks, outIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
}
else
{
inIncrement -= inIncrement ? blockSize : 0;
xorIncrement -= xorIncrement ? blockSize : 0;
outIncrement -= outIncrement ? blockSize : 0;
}
while (length >= blockSize)
{
Cleanup adv-simd.h for x86
2018-08-12 23:04:14 +00:00
double temp[2];
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
std::memcpy(temp, inBlocks, blockSize);
__m128i block = _mm_castpd_si128(_mm_load_sd(temp));
if (xorInput)
{
std::memcpy(temp, xorBlocks, blockSize);
block = _mm_xor_si128(block, _mm_castpd_si128(_mm_load_sd(temp)));
}
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[7]++;
func1(block, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
std::memcpy(temp, xorBlocks, blockSize);
block = _mm_xor_si128(block, _mm_castpd_si128(_mm_load_sd(temp)));
}
_mm_store_sd(temp, _mm_castsi128_pd(block));
std::memcpy(outBlocks, temp, blockSize);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Remove GCC_NO_UBSAN and double casts
2018-07-01 05:23:35 +00:00
length -= blockSize;
}
}
return length;
}
Fix unaligned load for _mm_loaddup_pd with GCC and UBsan This function was missed earlier. Unfortunately, it does not squash all of the unaligned load findings. I'm pretty sure it is a GCC problem
2017-12-28 06:16:17 +00:00
NAMESPACE_END // CryptoPP
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#endif // CRYPTOPP_SSSE3_AVAILABLE
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
// *********************** Altivec/Power 4 ********************** //
#if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
NAMESPACE_BEGIN(CryptoPP)
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
/// \brief AdvancedProcessBlocks for 2 and 6 blocks
/// \tparam F2 function to process 2 128-bit blocks
/// \tparam F6 function to process 6 128-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks64_6x2_Altivec processes 6 and 2 Altivec SIMD words
/// at a time. For a single block the template uses F2 with a zero block.
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type.
template <typename F2, typename F6, typename W>
Make AdvancedProcessBlocks64_6x2_ALTIVEC inline
2018-08-15 09:18:38 +00:00
inline size_t AdvancedProcessBlocks64_6x2_ALTIVEC(F2 func2, F6 func6,
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 8);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
enum {LowOffset=8, HighOffset=0};
const uint32x4_p s_one = {1,0,0,0};
const uint32x4_p s_two = {2,0,2,0};
#else
Make SPECK-64 more symmetrical
2018-08-14 10:19:34 +00:00
enum {LowOffset=8, HighOffset=0};
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
const uint32x4_p s_one = {0,0,0,1};
const uint32x4_p s_two = {0,2,0,2};
#endif
const size_t blockSize = 8;
const size_t vsxBlockSize = 16;
CRYPTOPP_ALIGN_DATA(16) uint8_t temp[16];
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : vsxBlockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? vsxBlockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : vsxBlockSize;
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
if (flags & BT_ReverseDirection)
{
inBlocks = PtrAdd(inBlocks, length - vsxBlockSize);
xorBlocks = PtrAdd(xorBlocks, length - vsxBlockSize);
outBlocks = PtrAdd(outBlocks, length - vsxBlockSize);
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 6*vsxBlockSize)
{
uint32x4_p block0, block1, block2, block3, block4, block5;
if (flags & BT_InBlockIsCounter)
{
// There is no easy way to load 8-bytes into a vector. It is
// even harder without POWER8 due to lack of 64-bit elements.
std::memcpy(temp+LowOffset, inBlocks, 8);
std::memcpy(temp+HighOffset, inBlocks, 8);
uint32x4_p ctr = (uint32x4_p)VectorLoadBE(temp);
// For 64-bit block ciphers we need to load the CTR block,
// which is 8 bytes. After the dup load we have two counters
// in the Altivec word. Then we need to increment the low ctr
// by 0 and the high ctr by 1.
block0 = VectorAdd(s_one, ctr);
// After initial increment of {0,1} remaining counters
// increment by {2,2}.
block1 = VectorAdd(s_two, block0);
block2 = VectorAdd(s_two, block1);
block3 = VectorAdd(s_two, block2);
block4 = VectorAdd(s_two, block3);
block5 = VectorAdd(s_two, block4);
Update comments
2018-08-14 09:15:32 +00:00
// Update the counter in the caller.
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
const_cast<byte*>(inBlocks)[7] += 12;
}
else
{
block0 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block1 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block2 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block3 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block4 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block5 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
}
if (xorInput)
{
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block2 = VectorXor(block2, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block3 = VectorXor(block3, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block4 = VectorXor(block4, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block5 = VectorXor(block5, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
}
func6(block0, block1, block2, block3, block4, block5, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block2 = VectorXor(block2, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block3 = VectorXor(block3, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block4 = VectorXor(block4, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block5 = VectorXor(block5, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
}
VectorStoreBE(block0, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block1, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block2, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block3, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block4, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block5, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
length -= 6*vsxBlockSize;
}
while (length >= 2*vsxBlockSize)
{
uint32x4_p block0, block1;
if (flags & BT_InBlockIsCounter)
{
// There is no easy way to load 8-bytes into a vector. It is
// even harder without POWER8 due to lack of 64-bit elements.
std::memcpy(temp+LowOffset, inBlocks, 8);
std::memcpy(temp+HighOffset, inBlocks, 8);
uint32x4_p ctr = (uint32x4_p)VectorLoadBE(temp);
// For 64-bit block ciphers we need to load the CTR block,
// which is 8 bytes. After the dup load we have two counters
// in the Altivec word. Then we need to increment the low ctr
// by 0 and the high ctr by 1.
block0 = VectorAdd(s_one, ctr);
// After initial increment of {0,1} remaining counters
// increment by {2,2}.
block1 = VectorAdd(s_two, block0);
Update comments
2018-08-14 09:15:32 +00:00
// Update the counter in the caller.
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
const_cast<byte*>(inBlocks)[7] += 4;
}
else
{
block0 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block1 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
}
if (xorInput)
{
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
}
func2(block0, block1, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
}
VectorStoreBE(block0, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block1, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
length -= 2*vsxBlockSize;
}
}
if (length)
{
// Adjust to real block size
if (flags & BT_ReverseDirection)
{
inIncrement += inIncrement ? blockSize : 0;
xorIncrement += xorIncrement ? blockSize : 0;
outIncrement += outIncrement ? blockSize : 0;
Use PtrSub to subtract offset from pointer
2018-08-15 07:16:08 +00:00
inBlocks = PtrSub(inBlocks, inIncrement);
xorBlocks = PtrSub(xorBlocks, xorIncrement);
outBlocks = PtrSub(outBlocks, outIncrement);
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
}
else
{
inIncrement -= inIncrement ? blockSize : 0;
xorIncrement -= xorIncrement ? blockSize : 0;
outIncrement -= outIncrement ? blockSize : 0;
}
while (length >= blockSize)
{
uint32x4_p block, zero = {0};
// There is no easy way to load 8-bytes into a vector. It is
// even harder without POWER8 due to lack of 64-bit elements.
Update comments
2018-08-14 10:07:19 +00:00
// The high 8 bytes are "don't care" but it if we don't
// initialize the block then it generates warnings.
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
std::memcpy(temp+LowOffset, inBlocks, 8);
Update comments
2018-08-14 10:07:19 +00:00
std::memcpy(temp+HighOffset, inBlocks, 8); // don't care
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
block = (uint32x4_p)VectorLoadBE(temp);
if (xorInput)
{
std::memcpy(temp+LowOffset, xorBlocks, 8);
Update comments
2018-08-14 10:07:19 +00:00
std::memcpy(temp+HighOffset, xorBlocks, 8); // don't care
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
uint32x4_p x = (uint32x4_p)VectorLoadBE(temp);
block = VectorXor(block, x);
}
Update comments
2018-08-14 09:15:32 +00:00
// Update the counter in the caller.
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[7]++;
func2(block, zero, subKeys, static_cast<unsigned int>(rounds));
if (xorOutput)
{
std::memcpy(temp+LowOffset, xorBlocks, 8);
Update comments
2018-08-14 10:07:19 +00:00
std::memcpy(temp+HighOffset, xorBlocks, 8); // don't care
Add AdvancedProcessBlocks64_6x2_ALTIVEC template
2018-08-14 07:32:45 +00:00
uint32x4_p x = (uint32x4_p)VectorLoadBE(temp);
block = VectorXor(block, x);
}
VectorStoreBE(block, temp);
std::memcpy(outBlocks, temp+LowOffset, 8);
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
length -= blockSize;
}
}
return length;
}
Add AdvancedProcessBlocks128_4x1_ALTIVEC template
2018-08-13 10:38:30 +00:00
/// \brief AdvancedProcessBlocks for 1 and 4 blocks
/// \tparam F1 function to process 1 128-bit block
/// \tparam F4 function to process 4 128-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks128_4x1_ALTIVEC processes 4 and 1 Altivec SIMD words
/// at a time.
/// \details The subkey type is usually word32 or word64. F1 and F4 must use the
/// same word type.
template <typename F1, typename F4, typename W>
inline size_t AdvancedProcessBlocks128_4x1_ALTIVEC(F1 func1, F4 func4,
const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Add AdvancedProcessBlocks128_4x1_ALTIVEC template
2018-08-13 10:38:30 +00:00
const uint32x4_p s_one = {1,0,0,0};
#else
const uint32x4_p s_one = {0,0,0,1};
#endif
const size_t blockSize = 16;
// const size_t vsxBlockSize = 16;
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
if (flags & BT_ReverseDirection)
{
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 4*blockSize)
{
uint32x4_p block0, block1, block2, block3;
if (flags & BT_InBlockIsCounter)
{
block0 = VectorLoadBE(inBlocks);
block1 = VectorAdd(block0, s_one);
block2 = VectorAdd(block1, s_one);
block3 = VectorAdd(block2, s_one);
// Hack due to big-endian loads used by POWER8 (and maybe ARM-BE).
// CTR_ModePolicy::OperateKeystream is wired such that after
// returning from this function CTR_ModePolicy will detect wrap on
// on the last counter byte and increment the next to last byte.
// The problem is, with a big-endian load, inBlocks[15] is really
// located at index 15. The vector addition using a 32-bit element
// generates a carry into inBlocks[14] and then CTR_ModePolicy
// increments inBlocks[14] too.
Use PtrSub to subtract offset from pointer
2018-08-15 07:16:08 +00:00
const_cast<byte*>(inBlocks)[15] += 6;
Add AdvancedProcessBlocks128_4x1_ALTIVEC template
2018-08-13 10:38:30 +00:00
}
else
{
block0 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block1 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block2 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
block3 = VectorLoadBE(inBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
}
if (xorInput)
{
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block2 = VectorXor(block2, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block3 = VectorXor(block3, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
}
func4(block0, block1, block2, block3, subKeys, rounds);
if (xorOutput)
{
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block2 = VectorXor(block2, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
block3 = VectorXor(block3, VectorLoadBE(xorBlocks));
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
}
VectorStoreBE(block0, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block1, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block2, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
VectorStoreBE(block3, outBlocks);
outBlocks = PtrAdd(outBlocks, outIncrement);
length -= 4*blockSize;
}
}
while (length >= blockSize)
{
uint32x4_p block = VectorLoadBE(inBlocks);
if (xorInput)
block = VectorXor(block, VectorLoadBE(xorBlocks));
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func1(block, subKeys, rounds);
if (xorOutput)
block = VectorXor(block, VectorLoadBE(xorBlocks));
VectorStoreBE(block, outBlocks);
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
length -= blockSize;
}
return length;
}
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \brief AdvancedProcessBlocks for 1 and 6 blocks
Add LEA-128 SSSE3 implementation (GH #669) LEA-128(128) from 6.73 cpb to 2.84 cpb on modern Core-i5 6400. LEA-128 from 10.12 cpb to 7.84 cpb antique Core2 Duo.
2018-06-22 20:26:27 +00:00
/// \tparam F1 function to process 1 128-bit block
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
/// \tparam F6 function to process 6 128-bit blocks
/// \tparam W word type of the subkey table
/// \details AdvancedProcessBlocks128_6x1_ALTIVEC processes 6 and 1 Altivec SIMD words
/// at a time.
/// \details The subkey type is usually word32 or word64. F1 and F6 must use the
/// same word type.
template <typename F1, typename F6, typename W>
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
inline size_t AdvancedProcessBlocks128_6x1_ALTIVEC(F1 func1, F6 func6,
Parameterize word type for subkeys in AdvancedProcessBlocks templates This was needed a while ago but we mostly side-stepped the issues with casts. CHAM64 uses a word16 type for subkeys and a cast won't fix it because we favor word32 for 64-bit block sizes.
2018-06-20 23:25:52 +00:00
const W *subKeys, size_t rounds, const byte *inBlocks,
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
{
CRYPTOPP_ASSERT(subKeys);
CRYPTOPP_ASSERT(inBlocks);
CRYPTOPP_ASSERT(outBlocks);
CRYPTOPP_ASSERT(length >= 16);
Sync CRYPTOPP_{BIG|LITTLE}_ENDIAN with Autotools Autotools sets up its config.h file with the '#define XXX 0' or '#define XXX 1' pattern. This check-in makes the sources Autotools aware. We need to verify CMake does the same
2018-10-28 08:24:22 +00:00
#if (CRYPTOPP_LITTLE_ENDIAN)
Fix ODR violation in AdvancedProcessBlocks_{ARCH} (GH #585) The ALTIVEC function required an inline declaration. Lack of inline caused the self test failure. Two NEON functions needed the same. We also cleaned up constants in unnamed namespaces
2018-02-20 18:17:05 +00:00
const uint32x4_p s_one = {1,0,0,0};
#else
const uint32x4_p s_one = {0,0,0,1};
#endif
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
const size_t blockSize = 16;
Cleanup adv-simd.h for ARM
2018-08-12 23:51:50 +00:00
// const size_t vsxBlockSize = 16;
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
size_t inIncrement = (flags & (BT_InBlockIsCounter|BT_DontIncrementInOutPointers)) ? 0 : blockSize;
size_t xorIncrement = (xorBlocks != NULLPTR) ? blockSize : 0;
size_t outIncrement = (flags & BT_DontIncrementInOutPointers) ? 0 : blockSize;
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
// Clang and Coverity are generating findings using xorBlocks as a flag.
const bool xorInput = (xorBlocks != NULLPTR) && (flags & BT_XorInput);
const bool xorOutput = (xorBlocks != NULLPTR) && !(flags & BT_XorInput);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
if (flags & BT_ReverseDirection)
{
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, length - blockSize);
xorBlocks = PtrAdd(xorBlocks, length - blockSize);
outBlocks = PtrAdd(outBlocks, length - blockSize);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
inIncrement = 0-inIncrement;
xorIncrement = 0-xorIncrement;
outIncrement = 0-outIncrement;
}
if (flags & BT_AllowParallel)
{
while (length >= 6*blockSize)
{
Fix carry bug in AdvancedProcessBlocks128_6x1_ALTIVEC
2018-08-13 05:44:23 +00:00
uint32x4_p block0, block1, block2, block3, block4, block5;
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
if (flags & BT_InBlockIsCounter)
{
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block0 = VectorLoadBE(inBlocks);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
block1 = VectorAdd(block0, s_one);
block2 = VectorAdd(block1, s_one);
block3 = VectorAdd(block2, s_one);
block4 = VectorAdd(block3, s_one);
block5 = VectorAdd(block4, s_one);
Fix carry bug in AdvancedProcessBlocks128_6x1_ALTIVEC
2018-08-13 05:44:23 +00:00
// Hack due to big-endian loads used by POWER8 (and maybe ARM-BE).
// CTR_ModePolicy::OperateKeystream is wired such that after
Update comments
2018-08-13 05:51:01 +00:00
// returning from this function CTR_ModePolicy will detect wrap on
// on the last counter byte and increment the next to last byte.
// The problem is, with a big-endian load, inBlocks[15] is really
// located at index 15. The vector addition using a 32-bit element
// generates a carry into inBlocks[14] and then CTR_ModePolicy
// increments inBlocks[14] too.
Fix carry bug in AdvancedProcessBlocks128_6x1_ALTIVEC
2018-08-13 05:44:23 +00:00
//
// To find this bug we needed a test case with a ctr of 0xNN...FA.
// The last octet is 0xFA and adding 6 creates the wrap to trigger
// the issue. If the last octet was 0xFC then 4 would trigger it.
// We dumb-lucked into the test with SPECK-128. The test case of
// interest is the one with IV 348ECA9766C09F04 826520DE47A212FA.
uint8x16_p temp = VectorAdd((uint8x16_p)block5, (uint8x16_p)s_one);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(temp, const_cast<byte*>(inBlocks));
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
}
else
{
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block0 = VectorLoadBE(inBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block1 = VectorLoadBE(inBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block2 = VectorLoadBE(inBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block3 = VectorLoadBE(inBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block4 = VectorLoadBE(inBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block5 = VectorLoadBE(inBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
}
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
{
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block2 = VectorXor(block2, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block3 = VectorXor(block3, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block4 = VectorXor(block4, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block5 = VectorXor(block5, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
}
func6(block0, block1, block2, block3, block4, block5, subKeys, rounds);
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
{
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block0 = VectorXor(block0, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block1 = VectorXor(block1, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block2 = VectorXor(block2, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block3 = VectorXor(block3, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block4 = VectorXor(block4, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block5 = VectorXor(block5, VectorLoadBE(xorBlocks));
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
}
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block0, outBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block1, outBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block2, outBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block3, outBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block4, outBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block5, outBlocks);
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
outBlocks = PtrAdd(outBlocks, outIncrement);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
length -= 6*blockSize;
}
}
while (length >= blockSize)
{
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
uint32x4_p block = VectorLoadBE(inBlocks);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorInput)
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block = VectorXor(block, VectorLoadBE(xorBlocks));
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
if (flags & BT_InBlockIsCounter)
const_cast<byte *>(inBlocks)[15]++;
func1(block, subKeys, rounds);
Add xorInput and xorOutput flags to adv-simd classes Analysis tools are generating findings when the pointer xorBlocks is used as the flag. The other missing piece is, xorBlocks is never NULL when either BT_XorInput or BT_XorOuput. But we don't know how to train the analyzers with the information, so we make it explicit with the boolean flags xorInput and xorOutput. Switching to the explicit flags costs us about 0.01 cpb on a modern Intel Core processor. In the typical case 0.01 is negligible.
2018-01-24 17:06:15 +00:00
if (xorOutput)
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
block = VectorXor(block, VectorLoadBE(xorBlocks));
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
Cleanup PPC vector functions The Crypto++ functions follow IBM's lead and provide VectorLoad, VectorLoadBE, VectorStore, and VectorStoreBE. Additionally, VectorLoadKey was removed in favor of vanilla VectorLoad.
2018-08-06 09:15:12 +00:00
VectorStoreBE(block, outBlocks);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
Add PtrAdd and PtrSub helper functions This helps contain UB on pointer subtraction by ensuring a ptrdiff_t is used. The code is a little uglier but it is also more portable.
2018-07-10 09:00:02 +00:00
inBlocks = PtrAdd(inBlocks, inIncrement);
outBlocks = PtrAdd(outBlocks, outIncrement);
xorBlocks = PtrAdd(xorBlocks, xorIncrement);
Move Altivec AdvancedProcessBlocks into adv-simd.h
2018-01-02 12:08:13 +00:00
length -= blockSize;
}
return length;
}
NAMESPACE_END // CryptoPP
#endif // CRYPTOPP_ALTIVEC_AVAILABLE
Refactor <cipher>_AdvancedProcessBlocks_<arch> into adv-simd.h This also fixes the SPECK64 bug where CTR mode self tests fail. It was an odd failure because it only affected 64-bit SPECK. SIMON was fine and it used nearly the same code. We tracked it down through trial and error to the table based rotates.
2017-12-10 02:04:25 +00:00
#endif // CRYPTOPP_ADVANCED_SIMD_TEMPLATES
Reference in New Issue Copy Permalink