SHA-3 Thumb2, ARM32 ASM: Add assembly implemention

[SparkiDev]

Description

Add SHA-3 assembly implementation for Thumb2 and ARM32.

Testing

QEMU with hosts: armv7m, armv8

Checklist

• added tests
• updated/added doxygen
• updated appropriate READMEs
• Updated manual and documentation

[bandi13]

retest this please

[dgarske]

STM32H7A Cortex-M7 @ 240MHz:

Could not build with WOLFSSL_ARMASM_INLINE because BlockSha3 is defined as static and called from sha3.c.

This is with "-Os".

SHA-3 Before:

SHA3-224                     1 MiB took 1.020 seconds,    1.125 MiB/s
SHA3-256                     1 MiB took 1.008 seconds,    1.066 MiB/s
SHA3-384                   850 KiB took 1.007 seconds,  844.091 KiB/s
SHA3-512                   600 KiB took 1.016 seconds,  590.551 KiB/s
SHAKE128                     1 MiB took 1.012 seconds,    1.303 MiB/s
SHAKE256                     1 MiB took 1.008 seconds,    1.066 MiB/s

SHA-3 With PR 7667:

SHA3-224                     1 MiB took 1.012 seconds,    1.254 MiB/s
SHA3-256                     1 MiB took 1.008 seconds,    1.187 MiB/s
SHA3-384                   950 KiB took 1.008 seconds,  942.460 KiB/s
SHA3-512                   675 KiB took 1.024 seconds,  659.180 KiB/s
SHAKE128                     1 MiB took 1.012 seconds,    1.447 MiB/s
SHAKE256                     1 MiB took 1.008 seconds,    1.187 MiB/s

For the Dilithium speed it improved to:

ML-DSA    44  key gen        56 ops took 1.028 sec, avg 18.357 ms, 54.475 ops/sec
ML-DSA    44     sign        16 ops took 1.133 sec, avg 70.812 ms, 14.122 ops/sec
ML-DSA    44   verify        52 ops took 1.031 sec, avg 19.827 ms, 50.436 ops/sec
ML-DSA    65  key gen        32 ops took 1.004 sec, avg 31.375 ms, 31.873 ops/sec
ML-DSA    65     sign         8 ops took 1.106 sec, avg 138.250 ms, 7.233 ops/sec
ML-DSA    65   verify        32 ops took 1.039 sec, avg 32.469 ms, 30.799 ops/sec
ML-DSA    87  key gen        20 ops took 1.051 sec, avg 52.550 ms, 19.029 ops/sec
ML-DSA    87     sign        10 ops took 1.039 sec, avg 103.900 ms, 9.625 ops/sec
ML-DSA    87   verify        20 ops took 1.082 sec, avg 54.100 ms, 18.484 ops/sec

[SparkiDev]

Fixed inline assembly code.

[dgarske]

Nice work!

STM32H7A Cortex M7 at 240MHz:

SHA3-224                     1 MiB took 1.004 seconds,    1.410 MiB/s
SHA3-256                     1 MiB took 1.008 seconds,    1.332 MiB/s
SHA3-384                     1 MiB took 1.016 seconds,    1.033 MiB/s
SHA3-512                   750 KiB took 1.011 seconds,  741.840 KiB/s
SHAKE128                     2 MiB took 1.012 seconds,    1.616 MiB/s
SHAKE256                     1 MiB took 1.008 seconds,    1.332 MiB/s
ML-DSA    44  key gen        60 ops took 1.031 sec, avg 17.183 ms, 58.196 ops/sec
ML-DSA    44     sign        18 ops took 1.078 sec, avg 59.889 ms, 16.698 ops/sec
ML-DSA    44   verify        54 ops took 1.008 sec, avg 18.667 ms, 53.571 ops/sec
ML-DSA    65  key gen        36 ops took 1.051 sec, avg 29.194 ms, 34.253 ops/sec
ML-DSA    65     sign        12 ops took 1.094 sec, avg 91.167 ms, 10.969 ops/sec
ML-DSA    65   verify        34 ops took 1.035 sec, avg 30.441 ms, 32.850 ops/sec
ML-DSA    87  key gen        22 ops took 1.070 sec, avg 48.636 ms, 20.561 ops/sec
ML-DSA    87     sign         8 ops took 1.259 sec, avg 157.375 ms, 6.354 ops/sec
ML-DSA    87   verify        20 ops took 1.008 sec, avg 50.400 ms, 19.841 ops/sec