synchronize for honest speed compare

This commit is contained in:
George Hotz
2023-03-24 10:24:27 -07:00
parent 1cb5b2d015
commit 23f88fb026
5 changed files with 26 additions and 11 deletions

View File

@@ -75,4 +75,4 @@ class CLCodegen(CStyleCodegen):
barrier = "barrier(CLK_LOCAL_MEM_FENCE);", float4 = "(float4)",
gid = [f'get_global_id({i})' for i in range(3)], lid = [f'get_local_id({i})' for i in range(3)], uses_vload=True)
GPUBuffer = Compiled(CLBuffer, CLCodegen, CLProgram)
GPUBuffer = Compiled(CLBuffer, CLCodegen, CLProgram, CL.cl_queue.finish)