Add macros to make the code simpler around allocating from dynamic
memory when WOLFSSL_SP_SMALL_STACK is defined.
Change over to using macros where it makes sense.
Don't use RIP relative with XMM/YMM instructions.
For MSVC asm, explicitly state type for pointer.
For MSVC asm, don't use vmodvqu for saving XMM registers unless this is
AVX2 code.
IAR doesn't parse register variable declarations with specified
registers. IAR doesn't even honor the register keyword.
Can use small negative but IAR doesn't like it.
Specify the positive value instead.
Add a small code size version of mont_reduce_full using umlal and umaal.
Make 'asm' usage in variables use keyword '__asm__'.
Explicitly don't inline some functions when compiling with IAR.
Thumb2/ARM32: use umaal instruction to speed up mulitplication, squaring
and reduction. umaal not always available so use umlal instead.
Implementations for architectures less than 4 still kept - no umull.
Cleanup point arithmetic to not take shortcuts.
Make masking for constant time faster.
Add alternate asm for different compilers.
Get ARMv4 and ARMv6 compiling.
Remove whitespace at end of lines.
For the CMake and autotools flows, enabling user_settings.h will define the
macro WOLFSSL_USER_SETTINGS_ASM. This will cause the generated
user_settings_asm.h to be included in the .S assembly files. If the user is
building without autotools or CMake, these files will include user_settings.h
directly, unless WOLFSSL_USER_SETTINGS_ASM is defined.
For user_settings.h builds, .S assembly files need to include user_settings.h
in order to get the defines used by the build. However, a user_settings.h may
contain code only understood by a C compiler and not the assembler (e.g. a
typedef). This commit makes it so our autotools and CMake builds produce a file
user_settings_asm.h when doing a user_settings.h build. This generated header
contains only the preprocessor directives from the user_settings.h. As a result,
it can be safely included by our assembly code files.
Curve25519: reduce to less than mod at end of curve25519, optimize to
not use imulq when andq can be used instead.
x86_64 Karatsuba Mul: Don't put add carry into overflow and then
continue adding zero. Overflow is lost.
EC point add is a double if affine ordinates are the same.
Fix AVX2 implementation of mod inv to do conversion from 26 bits in
32-bit registers to 64 bits in 64-bit registers. Also handle negative
value.
RSA OAEP: loop down to last byte. Must have a separator.
Div for x86_64: fix to have all 1s when high word equals divisor.
Change Karatsuba implementations for x86_64.
Fix ECC code to better handle corner cases.
Add 'lower' versions of functions wehn an input is known to be less than m.
Add mont_add/dbl/tpl/sub for P384.
Change ECC point add to be cache-attack resistant.
Change mod_exp to be cache-attack resistant.
Need to right shift number down when hash is more than 521 bits.
Previously handled at a byte level and now at bit level.
Always return err from sp_*_ecc_mulmod_add_only_*().
When ECC add and double points are public and only have SP
implementation, check that the point ordinates are the right size.
Faster small code and fast code.
Allow fixed 4096-bit FFDHE parameters in benchmark.
Convert [u]int[32|64|128]*_t types to sp_[u]int[32|64|128].
Add a div for when top bits are all 1
WOLFSSL_SP_FAST_LARGE_CODE added to make mul_add function faster on
non-embedded platforms.
Change mod_exp window sizes for same performance but less memory.
P256 with c32 now 9 words instead of 10.
modinv AVX2: do quick norm on result twice and convert 32-bit signed to
64-bit signed before adding
P-256 mont_add, mont_dbl, mont_tpl, mont_sub x64: handle mul/sqr result
being greater than modulus but not greater than 1<<256.
Sean Parkinson
JacobBarthelmeh
Daniel Pouzzner
Eric Blankenhorn
Hayden Roche
David Garske
toddouska
Chris Conlon
John Safranek