8000 Tags · clayne/pospop · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

Tags: clayne/pospop

Tags

v1.4.2

Toggle v1.4.2's commit message 
pospop 1.4.2

Fix a copy-editing error preventing the tests from building on arm64.

v1.4.1

Toggle v1.4.1's commit message 
pospop 1.4.1

Bump golang.org/x/sys dependency to 0.7.0 for riscv64 support
Add dummy benchmark function that just measures memory bandwidth

v1.4.0

Toggle v1.4.0's commit message 
pospop 1.4.0

Add CountString() function; like Count8() but for strings.
Add unit test for CountString().
Require go 1.20.  Use v1.3.6 if you want to stay on go 1.14.

v1.3.6

Toggle v1.3.6's commit message 
pospop 1.3.6

Fix accumulator overflow issue in arm64 NEON kernel.
Add new unit test for accumulator overflow.
Flush accumulators as late as possible.

v1.3.5

Toggle v1.3.5's commit message 
pospop 1.3.5

Improve tail handling in amd64 avx2 kernel.
Explicitly check for BMI2 before using AVX2 kernel.

v1.3.4

Toggle v1.3.4's commit message 
pospop 1.3.4

Improved tail and runt processing for avx512 kernel; now up to 25%
faster for short inputs.
New unit tests checking for overreads on access to the input.
Fix overread conditions in various kernels.

v1.3.3

Toggle v1.3.3's commit message 
pospop 1.3.3

Slight optimisations to amd64 kernels.

v1.3.2

Toggle v1.3.2's commit message 
pospop v1.3.2

Improve countavx512_amd64 performance.
Suffix benchmark names with B for better usability.

v1.3.1

Toggle v1.3.1's commit message 
pospop 1.3.1

Fix possible overflow issues.
Fix clobbered registers for long input arrays.
Fix test case minimisation.
Add a test case to detect this kind of problem.
Put some prefetches back in.

v1.3.0

Toggle v1.3.0's commit message 
pospop 1.3.0

All new kernels for amd64 avx512 avx2 sse2 and arm64 neon.
General housekeeping.
Test suite can now generate minimal examples for failed test cases.
Expanded bench marks.
0