Doesn't work for MTT AIBOOK (Linux on ARM aarch64)

[Oslomayor]

.-----------------------------------------------------------------------------.
|----------------.------------------------------------------------------------|
| Device ID    0 | Vivante OpenCL Device VIP9000L-Q+I.9202.0000               |
|----------------'------------------------------------------------------------|
|----------------.------------------------------------------------------------|
| Device ID      | 0                                                          |
| Device Name    | Vivante OpenCL Device VIP9000L-Q+I.9202.0000               |
| Device Vendor  | Vivante Corporation                                        |
| Device Driver  | OpenCL 3.0 V6.4.16.4.741359 (Linux)                        |
| OpenCL Version | OpenCL C 3.0                                               |
| Compute Units  | 2 at 3 MHz (2 cores, 0.000 TFLOPs/s)                       |
| Memory, Cache  | 256 MB RAM, 32 KB global / 32 KB local                     |
| Buffer Limits  | 128 MB global, 64 KB constant                              |
|----------------'------------------------------------------------------------|
| Warning: (2:0) : warning : #undef a undefined id. (3:0) : warning : #undef  |
|          a undefined id. (39:0) : error : Not expected token('(') in        |
|          expression. (39:0) : error : Not expected                          |
|          token('__builtin_amdgcn_sdot4') in expression. (39:0) : error :    |
|          Not expected token(')') in expression. (45:0) : error : Not        |
|          expected token('(') in expression. (45:0) : error : Not expected   |
|          token('__builtin_amdgcn_sudot4') in expression. (45:0) : error :   |
|          Not expected token(')') in expression.                             |
| Error: OpenCL C code compilation failed with error code -11. Make sure      |
|        there are no errors in kernel.cpp.                                   |
'-----------------------------------------------------------------------------'

[ProjectPhysX]

Hi @Oslomayor,

first MTT device I've seen in OpenCL, cool!

You can do 2 small modifications to get the benchmark running (hopefully):

1. Replace the dp4a wrapper function with this:

   int dp4a(const char4 a, const char4 b, const int c) { // 4-wide byte dot product and accumulate
   	return c+a.x*b.x+a.y*b.y+a.z*b.z+a.w*b.w;
   }
   

2. Reduce memory footprint for the benchmark from 1GB to 96MB: here set const uint N = 4096u*384u;

If you could upload the OpenCL specs of your MTT AIBOOK to https://opencl.gpuinfo.org/ that would help a lot!

Thanks and kind regards,
Moritz