Changes between Version 5 and Version 6 of Notes_on_CUDA_Semantics

Timestamp:

06/13/14 13:22:45 (12 years ago)

Author:

andrevm

Comment:

--

Notes_on_CUDA_Semantics

@@ -158 +158 @@
 Ideas:
 
+=== Thread Hierarchy Translation ===
+
 For each kernel `K`, create a function `__kernel_K` that takes the same arguments as K in addition to the execution configuration parameters.
 In the transformed program, `__kernel_K` is called wherever `K` was called in the original. 
@@ -184 +186 @@
 }
 }}}
+
+=== Memory Hierarchy Translation ===
 
 To help enforce the Cuda memory hierarchy, we introduce two root level variables, $host_scope and $device_scope. Here is the proposed layout.