as word64 is not always available, introduce an abstract type and companion
operations. They use a word64 if available and fallback on word32[2] otherwise.
configure.ac: fix ed25519/sha512 dependency test to not misfire when ENABLED_32BIT;
wolfssl/wolfcrypt/curve{25519,448}.h: fix redundant typedefs of curve{25519,448}_key (fixes -Wpedantic warnings);
configure.ac: fix for "ISO C forbids an empty translation unit [-Werror=pedantic]", re wolfcrypt/src/sp_c{32,64}.c;
configure.ac: fixes for --enable-32bit versus pedantic "ISO C forbids an empty translation unit", including explicit exclusion of 32bit-incompatible algorithms from enable-all and enable-all-crypto sets;
tests/api.c: fixes for a couple inadequately gated SHA2 dependencies;
tests/api.c:test_wolfSSL_set_alpn_protos(): fix prototype missing (void);
wolfcrypt/src/misc.c and wolfssl/wolfcrypt/misc.h: fix ForceZero() definition and NO_INLINE prototype to not counterfactually constify the mem ptr, to avoid -Wmaybe-uninitialized from gcc11;
wolfcrypt/src/des3.c: drop obsolete register qualifier from declaration in DesSetKey(), for c++17 compatibility;
src/ssl.c:wolfSSL_BN_mod_word(): fix cast of arg2 to mp_mod_d().
For byte order reverse and rotation we have builtins
in the CC-RX compiler.
Especially when rotating registers with fixed amounts
this can be compiled efficiently into opcodes with
embedded values for shift (without needing other registers).
We separate out 64-bit desktop support based on the Intel check. With
the advent of the new Apple chip, ARM can also be a desktop processor.
Detect it like we do the Intel 64-bit, and treat it similarly with
respect to fast and normal math.
Right shift of a negative value is undefined.
Add an 'int' return version of constant time greater than equal.
Change equal and not equal to be constant time in more situations.
* Fixed some `tls_bench` build issues with various configure options.
* Moved the `WOLFSSL_PACK` and `WC_NORETURN` macros into types.h.
* Added support for `__builtin_bswap32` and `__builtin_bswap64`. Since the performance of the builtins varries by platform its off by default, but can be enabled by customer using `WOLF_ALLOW_BUILTIN`. Quick check on x86 showed the 32-bit swap performance matched, but 64-bit swap was slower.