In the current implementation the POW algorithm in dev pool section of a
currency will not be taken into account during the binary creation.
This PR changes the behavior and allow to create binaries for more than two POW algorihms.

Since #2080 bittube2 is broken.

- reintroduce special AES function for bittube2

NVIDIA is using clang as device compiler so the reciprocal optimizations was disabled with #2104.

- re-enable optimized reciprocal calculation

Add an option to brute force intensity settings and lock in at the intensity with the highest hashrate.

- update decumentation of the `interleave` option to mention the side effect with `auto-tune`
- disable `interleave` auto adjustment if `auto-tune` is enabled
- jconf: add `auto-tune` as optional option

- fix broken compile: change used `ULL` to `UL` because `UL` is defined as 64bit
- fix memory copy to shared memory via vload8 (somehow it create wrong access)

The auto config generates for AMD devices now by default two threads per GPU.

- remove the savety 128MiB memory now only from the max available GPU memory not from the avaialble memory for one alloc call
- extend the memory documentation in amd.txt

Due to a wrong implementation clamp was not working.

use for non clang (Rocm) OpenCL a optimized reciprocal calculation without lookup table.

Co-authored-by: SChernykh <sergey.v.chernykh@gmail.com>

Disable compatibility mode if intensity is a multiple of worksize. In that case enabled compaibility mode will only slow down the miner.

If two threads are using the same GPU device the start time of each hash round is optimized based on the average time needed to calculate a bunch of hashes.

This way to optimize the hash rate was first introduced by @SChernykh. This implementation based on the implementation in xmrig but differen in the details.

- introduce a new config option `interleave`
- implement thread interleaving

Use `mul24` to speedup the scratchpad index calculation.

Co-authored-by: SChernykh <sergey.v.chernykh@gmail.com>

Add new striding index where the memory is chunked by the size of the work group (worksize).

Co-authored-by: SChernykh <sergey.v.chernykh@gmail.com>

small optimization for non cryptonight_v8 algorithms

- optimize division

optimize cryptonight_heavy diff

- change a few 64bit variables into 32bit.
- provide defines type quallified

- remove useless `clFinish`
- avoid download num threads for skein&co and start always as much threads as in all other kernel (terminate useless threads)

Reduce local memory foot print to increase the occupancy.

Co-authored-by: SChernykh <sergey.v.chernykh@gmail.com>

define shared memory in the outer scope

x-ref: https://github.com/xmrig/xmrig-amd/pull/192

- optimize kernel cn0 and cn2
- optimize vast int math
- use more 32bit variables

Co-authored-by: psychocrypt <psychocryptHPC@gmail.com>

optimize the devision in cryptonight_heavy and cryptonight_haven

import of https://github.com/xmrig/xmrig-amd/pull/185/commits/5d9b9334654df25cea7707f667990fd1577ed290

Fix the fix from #1945. The initial fix produces invalid results.

The AMD compiler for OpenCL shipped with the driver 14XX is broken
and can not compile xmr-stak since the monero v8 changes are introduced.

- workaround a simple compare.
- add new device define `OPENCL_DRIVER_MAJOR`

In the current implementation the bit align is using signed integer which results in pulling in
ones in the case the sign bit is set.

- cast to unsigned integer before using bitshift

Strided index 1 is not allowed for cryptonight_v8 and monero.
In the case the dev pool is set to monero and the user tuned there settings for
an other currency the miner will crash if strided index or memChunk is not
fitting the requirement to mine monero.
This PR detects wrong configurations and will set strided index and memChunk to a valid
value but only for cryptonight_v8. The user pool settings will only be changed if monero or
cryptonight_v8 is selected.

With rocm we fighted very long with invalid shares. This is now solved with rocm 1.9 and
this tiny fix.
It is not fully clear where a memory optimization is kicking in and break the kernel `Groestl` if the variables `M` and `H` are not `volatile`.
The performance ill not change with this fix.

The fix is tested with rocm 1.9 with a VEGA64 and a RX570

add cpu implementation for the final monero POW

- remove unused host function (relict from old refactoring)
- remove unused OpenCL full div function

- fix code style issues
- fix spelling issue
- fix asm to support newer clang versions

add option `unroll` for OpenCL to allow better tuning the main POW kernel.

Create a special pass for NVIDIA GPUs to load memory chunks first into the shared memory.

Co-authored-by: SChernykh <sergey.v.chernykh@gmail.com>

- implement cryptonight_v8
- update auto adjust to fit the special requirements of `cryptonight_v8`
- add fast math integer implementation for `sqrt`, `reciprocal`  and `division`

Co-authored-by: SChernykh <sergey.v.chernykh@gmail.com>

During the initialization of the compile parameter for OpenCL it could be that the
fixed size buffer is to small. To avoid this we are now using `std::string`.
There is no problem by using `std::string` because this part of code is not perfromance critical.

If the first bit of the nonce is `1` (this is very often if we use a nicehash pool)
than it could be that some OpenCL implementations handle the 64bit representation of the 32bit
nonce on the device side as signed integer.
During a right bitshift we pull wrong ones from the wrong higher part of the 64bit
nonce representation into the 32bit part of the nonce.
The result will be that the computed share is invalid.

- explicit cast the nonce on the device to `uint` to avoid any side effects

Avoid that a OpenCL binary from the cache is used if the driver or xmr-stak version has changed.