Changes between Version 69 and Version 70 of Orio

Timestamp:

06/14/08 03:54:14 (15 years ago)

Author:

hartono

Comment:

--

Orio

@@ -277 +277 @@
 }}}
 
-Among all the class attributes, three of them are very important in generating optimized code: ''performance parameters'', ''module body code'', and ''annotation body code''. Performance parameters are information essential for performing code optimization and generation, such as unroll factor and tile size. These parameters are stored in a hashtable to facilitate quick accesses to parameter values. The code contained in the module body block normally outlines the applied optimization techniques and, possibly, the high level description of the computation itself. In order to extract this information, new language syntax and a corresponding parser component must be implemented for each transformation module. In addition, the annotation body code, currently expressed in C language, can electively be parsed and given as input to the transformation module.
-
+Among all the class attributes, three of them are very important in generating optimized code: ''performance parameters'', ''module body code'', and ''annotation body code''. Performance parameters are information essential for performing code optimization and generation, such as unroll factor and tile size. These parameters are stored in a hashtable to facilitate quick accesses to parameter values. The code contained in the module body block normally outlines the applied optimization techniques and, possibly, the high level description of the computation itself. In order to extract this information, new language syntax and a corresponding parser component must be implemented for each transformation module. In addition, the annotation body code, currently expressed in C language, can electively be parsed and given as input to the transformation module. 
+
+In the earlier AXPY-4 case, we can view the annotation body block as redundant, since the module body already contains the same or similar program. The motivation behind requiring Orio users to include the computation itself as part of the annotation is to avoid making use of a full-fledged C compiler infrastructure that potentially will compromise the simplicity, reliability, and portability of Orio. Furthermore, another reason is to allow the computation to be expressed by using domain-specific high-level languages, thus capturing the semantics without imposing tuning constraints resulting from the use of a general-purpose language.
 
 ''Currently under construction''