OpenCL Kernels: Use the kernel local buffer size as additional reference in order to limit the thread-count
This commit is contained in:
@@ -61724,7 +61724,8 @@ __kernel void gpu_atinit (__global pw_t *buf, const u64 gid_max)
|
||||
pw.i[62] = 0;
|
||||
pw.i[63] = 0; // yep that's faster
|
||||
|
||||
pw.pw_len = 1 + (l32 & 15);
|
||||
//pw.pw_len = 1 + (l32 & 15);
|
||||
pw.pw_len = 7; // some algorithms are very sensible on this (example: 12500)
|
||||
|
||||
buf[gid] = pw;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user