Improve Compute Device Selection

[xperia64]

I have both a 1080 Ti and RX480 running in my system.
Due to this "metric" determining speed:

bfCL/ocl_util.c

Line 183 in 6730c32

cl_ulong cap = 1ull * devices[j].max_compute_units * devices[j].freq;

...the program selects my RX480 over the 1080 Ti.
A command line option to select the compute device number would probably work.

[Jimmy-Z]

You're right, but the command line "parser" is too weak, and I don't have much time on this.

[jason0597]

@Jimmy-Z i can try making a PR but i am having difficulty compiling this
are there any instructions somewhere on how to compile this or the prerequisites needed? does it only with with intel's openCL SDK? does it detect any other SDKs? does it work in linux, or windows only? if it's windows only, why does it use mingw and not vc++?

[Jimmy-Z]

@jason0597 Read Makefile, I prefer orange over apple.

[Jimmy-Z]

@xperia64 paste the output of bfcl and bfcl info, I might be able to add a quick fix for this.

[xperia64]

bfcl:
https://pastebin.com/WT6k6HaF
bfcl info:
https://pastebin.com/ahddR8Dr

[Jimmy-Z]

Thanks.
bfcl-test.zip
Would you try this one, and if it choose your 1080 Ti now, paste the output of bfcl, I'd like to see those numbers.

[xperia64]

Sorry about the delay:

selected device GeForce GTX 1080 Ti on platform NVIDIA CUDA
mbed TLS 2.7.0, AES-NI supported
self-test/benchmark mode
AES Key: 0d0b8bd02564dd0351d7e415e6f23f36
randomize source buffer using RDRAND
0.981 seconds for preparing test data, 68.40 MB/s
0.425 seconds for OpenCL compiling
0.017 seconds for data upload, 3844.24 MB/s
# sha1_16_test on 64 MB
0.004 seconds for OpenCL, 15040.09 MB/s
0.025 seconds for data download, 2655.25 MB/s
0.354 seconds for reference C(single thread), 189.38 MB/s
sha1_16_test: succeed
# aes_enc_128_test on 64 MB
0.033 seconds for OpenCL, 2056.60 MB/s
0.014 seconds for data download, 4861.90 MB/s
0.174 seconds for reference C(single thread), 386.39 MB/s
aes_enc_128_test: succeed
# aes_dec_128_test on 64 MB
0.036 seconds for OpenCL, 1884.92 MB/s
0.014 seconds for data download, 4873.20 MB/s
aes_dec_128_test: succeed
Press any key to continue . . .