cpuminer-opt-3.19.2

Fixed log displaying incorrect memory usage for scrypt, broken in v3.19.1.

Reduce log noise when replies to submitted shares are lost due to stratum errors.

Fugue prehash optimization for X16r family AVX2 & AVX512.

Small speed improvement for Hamsi AVX2 & AVX512.

Win: With CPU groups enabled the number of CPUs displayed in the ASCII art
affinity map is the number of CPUs in a CPU group, was number of CPUs up to 64.

cpuminer-opt-3.19.1

Changes to Windows binaries package:

• builds for CPUs with AVX or lower have CPU groups disabled,
• zen3 build renamed to avx2-sha-vaes to support Alderlake as well as Zen3,
• zen build renamed to avx2-sha, supports Zen1 & Zen2,
• avx512-sha build removed, Rocketlake CPUs can use avx512-sha-vaes,
• see README.txt for compatibility details.

Fixed a few compiler warnings that are new in GCC 11.
Other minor fixes.

cpuminer-opt-3.19.0

Windows binaries for CPUs with AVX2 and newer are now built with support for CPU groups, requires Windows 7.

Changes to cpu-affinity:

• PR#346: Fixed incorrect CPU affinity on Windows built for CPU groups,
• added support for CPU affinity for up to 256 threads or CPUs,
• streamlined code for more efficient initialization of miner threads,
• precise affining of each miner thread to a specific CPU,
• added an option to disable CPU affinity with "--cpu-affinity 0"

Faster sha256t with AVX512 & AVX2.

Added stratum error count to stats log, reported only when non-zero.

Note: Windows binaries have been rebuilt for CPUs with AVX, SSE* due to problems with CPU groups.
cpuminer-opt-3.19.0-windows2.zip contains rebuilt binaries without support for CPU groups for those builds.
There are no source code changes and no changes to other builds.

cpuminer-opt-3.18.2

Issue #342, fixed Groestl AES on Windows, broken in v3.18.0.

AVX512 for sha256d.

SSE42 and AVX may now be displayed as mining features at startup.
This is hard coded for each algo, and is only implemented for scrypt
at this time as it is the only algo with significant performance differences
with those features.

Fixed an issue where a high hashrate algo could cause excessive invalid hash
rate log reports when starting up in benchmark mode.

cpuminer-opt-3.18.1

More speed for scrypt:

• additional scryptn2 optimizations for all CPU architectures,
• AVX2 is now used by default on CPUS with SHA but not AVX512,
• scrypt:1024 performance lost in v3.18.0 is restored,
• AVX512 & AVX2 improvements to scrypt:1024.

Big speedup for SwiFFTx AVX2 & SSE4.1: x22i +55%, x25x +22%.

Issue #337: fixed a problem that could display negative stats values in the first summary report if the
report was forced prematurely due to a stratum diff change. The stats will still be invalid but should
display zeros.

cpuminer-opt-3.18.0

Complete rewrite of Scrypt code, optimized for large N factor (scryptn2):

• AVX512 & SHA support for SHA256, AVX512 has priority,
• up to 50% increase in hashrate,
• memory requirements reduced 30-60% depending on CPU architecture,
• memory usage displayed at startup,
• scrypt, default N=1024 (LTC), will likely perform slower.

Improved stale share detection and handling for Scrypt with large N factor:

• abort and discard partially computed hash when new work is detected,
• quicker response to new job, less time wasted mining stale job.

Improved stale share handling for all algorithms:

• report possible stale share when new work received with a previously submitted share still pending,
• when new work is detected report the submission of an already completed, otherwise valid, but likely stale, share,
• fixed incorrect block height in stale share log.

Small performance improvements to sha, bmw, cube & hamsi for AVX512 & AVX2.

When stratum disconnects miner threads go to idle until reconnected.

Colour changes to some logs.

Some low level function name changes for clarity and consistency.

The reference hashrate in the summary log and the benchmark total hashrate are now the mean hashrate for the session.

Update errata:

The changes to scrypt have reduced scryptn2 performance on AVX CPUs significantly, SSE2 also likely. This occurs with the
same number of threads. Memory usage reductions in v3.18.1 may allow more cores to run mitigating the slow down. Owners
with these CPUs who haven't previously been able to run all cores may want to compare performance and may want to continue
using v3.17.1 until the problem is resolved. CPUs with AVX512 or AVX2 are not affected. The problem is under investigation.

Ryzen users will find better performance on scryptn2 using the AVX2 build instead of the native zen or zen3 build with SHA.
The default will be changed in the next release to use AVX2 for scryptn2 when SHA is also available. AVX512 is already the
default when it is available.

cpuminer-opt-3.17.1

Fixed Windows build for AES+SSE4.2 (Westmere), was missing AES.

More ternary logic optimizations for AVX512, AVX512+VAES, and AVX512+AES.

Fixed my-gr algo for VAES.

cpuminer-opt-3.17.0

AVX512 optimized using ternary logic instructions.
Faster sha256t on all CPU architectures: AVX512 +30%, SHA +30%, AVX2 +9%.
Use SHA on supported CPUs to produce merkle hash.
Fixed byte order in Extranonce2 log & replaced Block height with Job ID.

cpuminer-opt-3.16.5

#329: Fixed GBT incorrect target diff in stats, second attempt.

Fixed formatting error in share result log when --no-color option is used.

cpuminer-opt-3.16.4

Faster sha512 and sha256 when not using SHA CPU extension.

Issue #329: Fixed GBT incorrect target diff in stats.