import and keep my code for older archs, like skein 64
reduce the gap between our versions...
+150kH x11 GTX 960 / +30kH 750Ti
+900kH quark GTX 960 / +230kH 750Ti
Reduce a bit the 750Ti speed but improve a lot the 9xx speed.
Keep compat for SM 3/3.5 in a second file..
Note: With this code and Cuda 7.5, the speed won is the reverse...
May be "reverted" soon
Should give same or better than SP and klaus versions
Keep old code for older devices and skein2 compat
Linux perf: 750Ti 78MH/s and GTX 970 260MH/s
Signed-off-by: Tanguy Pruvot <tanguy.pruvot@gmail.com>
Seems to be djm34 work, i recognize the code style ;)
Code was cleaned/indented and adapted to my fork...
Only usable on the test pool until 16 december 2014!
The core problem was the cuda hefty Thread per block set to high
but took me several hours to find that...
btw... +25% in heavy 12500 with 256 threads per block... vs 128 & 512
if max reg count is set to 80...
Small echo rewrite. +10KHASH on the 650(compute 3.0)
tpruvot: add Linux Makefile - Force to 80 registers (else -30KH/s)
Note : the hashrate seems more constant with this change
nvml.dll doesnt exists for 32bit binaries! use nvapi to get infos
seems to have more/different features than NVML... like pstate etc..
This is nvapi r343 : https://developer.nvidia.com/nvapi
Signed-off-by: Tanguy Pruvot <tanguy.pruvot@gmail.com>
There was a different behavior on linux and visual studio
That was making it hard to link functions correctly
That remove some ifdef / extern "C" requirements
note about x86 releases, x86 nvml.dll is not installed on Windows x64!
Based on mwhite73 <marvin.white@gmail.com> implementation
Linked to the api system
Also fix Makefile to support standard c++ files
This prevent nvcc use without device code
Signed-off-by: Tanguy Pruvot <tanguy.pruvot@gmail.com>
Displayed data is the average of the last 50 scans in the 5 last minutes
Also move cuda common functions in a new file (cuda.cu)
Signed-off-by: Tanguy Pruvot <tanguy.pruvot@gmail.com>