This is the first time for me to ask a question here.
I'm facing the same problem as the following discussion with Radeon R7 260X.
The details are as follows;
Offline compile failed with Radeon R7 260X (and R9 290X)
Steps to Reproduce:
1. Create a context by clCreateContext() with CL_CONTEXT_OFFLINE_DEVICES_AMD flag.
2. Create multiple binaries of OpenCL kernel program for each devices by clBuildProgram(), and save them as files.
3. Get the device name of the execution environment by calling clGetDeviceInfo() with CL_DEVICE_NAME flag.
4. Choose the binary for the device name obtained in Step 3 from the binaries created in Step 2, and execute the kernel programs.
When I execute Step 1-4 with each GPU as the following matrix,
OK: the kernel programs work correctly.
NG: the kernel programs generate incorrect results but no error is returned.
|↓Step 1&2 Step 3&4→||R9 285 (Tonga)||Other GPUs||R9 290X (Hawaii)||R7 260X (Bonaire)|
|R9 285 (Tonga)||OK||OK||NG||NG|
|R9 290X (Hawaii)||NG||NG||OK||NG|
|R7 260X (Bonaire)||NG||NG||NG||OK|
All machines are running Windows 8.1, Catalyst Omega 14.12, OpenCL 2.0 AMD-APP(1642.5).
Wii this problem be solved in the next version of Catalyst?