Specialize SIMD SHA-512 for half-length input, iterated hashing#5425

Merged

solardiz merged 24 commits intoopenwall:bleeding-jumbofrom solardiz:sha512opt

Jan 17, 2024

+476-165

Commits on Jan 16, 2024

SIMDSHA512body(): Optimize for elements 9 to 14 being all zeroes
solardiz
committed
SIMDSHA512body(): Build 3 specialized versions of the function
solardiz
committed
SIMDSHA512body() calls: Claim "elements 9 to 14 are 0" where we can
solardiz
committed
SIMDSHA512body() and calls: Implement and use iterated hashing
solardiz
committed
SIMDSHA512body(): Reset impossible flags to help the compiler optimize
solardiz
committed
SIMDSHA512body(): Add compile-time toggle for all-zeroes optimization
solardiz
committed
SIMDSHA512body(): Add redundant initialization of w[] with SSEi_HALF_IN
solardiz
committed
SIMDSHA512body() calls: With SSEi_HALF_IN, do not initialize second half
solardiz
committed
SIMDSHA512body() calls: With SSEi_HALF_IN, do not allocate second half
solardiz
committed
SIMDSHA512body() and calls: Fix SIMD_PARA_SHA512 > 1 support
solardiz
committed
SIMD*body(): Optimize the case of SIMD_PARA_* == 1
solardiz
committed
SIMDSHA512body(): Optimize SSEi_FLAT_OUT
solardiz
committed
SIMDSHA512body() and Drupal7 format: Implement and use iterated hashing
solardiz
committed
SIMDSHA512body(): Add redundant initialization of w[] with SSEi_LOOP
solardiz
committed
SIMDSHA512body(): Exclude SSEi_LOOP from the generic SSEi_HALF_IN code
solardiz
committed
SIMDSHA*body(): Exclude SSEi_FLAT_RELOAD_SWAPLAST support except for SHA-256
solardiz
committed
SIMDSHA512body(): Specialized version for sha512crypt and some dynamics
solardiz
committed
simd-intrinsics.c: Copyright years update
solardiz
committed
sha256crypt, sha512crypt CPU formats: Enable most ex-DEBUG test vectors
solardiz
committed
bench.c: Print multi-test/benchmark summary line even when aborted
solardiz
committed
bench.c: Print two fractional digits for numbers below 10
solardiz
committed
relbench: Update for current messages, handle more special cases
solardiz
committed
Armory format: Fetch a part of each hash first, then the rest of each
solardiz
committed
Keplr format: Benchmark only Raw, print algorithm detail
solardiz
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Specialize SIMD SHA-512 for half-length input, iterated hashing#5425

Commits on Jan 16, 2024

SIMDSHA512body(): Optimize for elements 9 to 14 being all zeroes

SIMDSHA512body(): Build 3 specialized versions of the function

SIMDSHA512body() calls: Claim "elements 9 to 14 are 0" where we can

SIMDSHA512body() and calls: Implement and use iterated hashing

SIMDSHA512body(): Reset impossible flags to help the compiler optimize

SIMDSHA512body(): Add compile-time toggle for all-zeroes optimization

SIMDSHA512body(): Add redundant initialization of w[] with SSEi_HALF_IN

SIMDSHA512body() calls: With SSEi_HALF_IN, do not initialize second half

SIMDSHA512body() calls: With SSEi_HALF_IN, do not allocate second half

SIMDSHA512body() and calls: Fix SIMD_PARA_SHA512 > 1 support

SIMDbody(): Optimize the case of SIMD_PARA_ == 1

SIMDSHA512body(): Optimize SSEi_FLAT_OUT

SIMDSHA512body() and Drupal7 format: Implement and use iterated hashing

SIMDSHA512body(): Add redundant initialization of w[] with SSEi_LOOP

SIMDSHA512body(): Exclude SSEi_LOOP from the generic SSEi_HALF_IN code

SIMDSHA*body(): Exclude SSEi_FLAT_RELOAD_SWAPLAST support except for SHA-256

SIMDSHA512body(): Specialized version for sha512crypt and some dynamics

simd-intrinsics.c: Copyright years update

sha256crypt, sha512crypt CPU formats: Enable most ex-DEBUG test vectors

bench.c: Print multi-test/benchmark summary line even when aborted

bench.c: Print two fractional digits for numbers below 10

relbench: Update for current messages, handle more special cases

Armory format: Fetch a part of each hash first, then the rest of each

Keplr format: Benchmark only Raw, print algorithm detail