CLBlast: All sgemm tests fail on Intel HD530 gpu

All [edit: sgemm] tests fail on Intel HD530 gpu.

Output of running ./clblast_test_xgemm -device 1 > clblast_test_xgemm_out.txt 2>&1: https://gist.github.com/hughperkins/711741a770949889fbe8122a0c0fd288

git log:

$ git status
On branch cedric-master
Your branch is up-to-date with 'cedric/master'.
Untracked files:
  (use "git add <file>..." to include in what will be committed)

	./

nothing added to commit but untracked files present (use "git add" to track)
carrot:mac hugh2$ git log -n 3
commit e52f9a9ff23f73325e5a8bc58765554d5f5eedc4
Merge: 115af8c 2cf7d84
Author: Cedric Nugteren <web@cedricnugteren.nl>
Date:   Sun Nov 27 15:59:21 2016 +0100

    Merge pull request #127 from CNugteren/development
    
    Update to version 0.10.0

commit 2cf7d8429a58ac09842815d9859c90acbc38e8b7
Author: Cedric Nugteren <web@cedricnugteren.nl>
Date:   Sun Nov 27 13:34:18 2016 +0100

    Updated to version 0.10.0

commit 39c49bf4f977427de42fdfe27e8a2ed41ae4923e
Author: Cedric Nugteren <web@cedricnugteren.nl>
Date:   Sun Nov 27 11:00:29 2016 +0100

    Made it possible to use the command-line environmental variables for each executable and without re-running CMake

Environment:

• Mac Sierra
• HD530 is -device 1. I havent figured out how to install clinfo yet, but EasyCL’s gpuinfo gives:

$ ./gpuinfo 
num platforms: 1

platform index: 0:
platform id: 0x7fff0000
platform vendor: Apple
platform name: Apple
platform num devices: 3

   device index: 0
   device id: 0xffffffff
   device type: 2
   global memory size: 16384MB
   local memory size: 32KB
   global cache size: 0KB
   global cacheline size: 6291456
   max memory alloc size: 4096MB
   max compute units: 8
   max workgroup size: 1024
   max workitem dimensions: 3
   max workitem sizes: 1024 1 1
   device name: Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz
   opencl c version: OpenCL C 1.2 
   opencl device version: OpenCL 1.2 
   frequency MHz: 2600

   device index: 1
   device id: 0x1024500
   device type: 4
   global memory size: 1536MB
   local memory size: 64KB
   global cache size: 0KB
   global cacheline size: 0
   max memory alloc size: 384MB
   max compute units: 24
   max workgroup size: 256
   max workitem dimensions: 3
   max workitem sizes: 256 256 256
   device name: Intel(R) HD Graphics 530
   opencl c version: OpenCL C 1.2 
   opencl device version: OpenCL 1.2 
   frequency MHz: 1050

   device index: 2
   device id: 0x1021c00
   device type: 4
   global memory size: 2048MB
   local memory size: 32KB
   global cache size: 0KB
   global cacheline size: 0
   max memory alloc size: 512MB
   max compute units: 10
   max workgroup size: 256
   max workitem dimensions: 3
   max workitem sizes: 256 256 256
   device name: AMD Radeon Pro 450 Compute Engine
   opencl c version: OpenCL C 1.2 
   opencl device version: OpenCL 1.2 
   frequency MHz: 800

Edit: note, for the record, all the easycl unit tests pass ok:

• gist of the output: https://gist.github.com/hughperkins/741e95ec3975a05d9efc46c289f33781
• example of a test that uses workgroup size > 1, and local memory:
  • cpp hostside code: https://github.com/hughperkins/EasyCL/blob/master/test/testlocal.cpp
  • corresponding opencl code: https://github.com/hughperkins/EasyCL/blob/master/test/testlocal.cl