Displaying 3 results from an estimated 3 matches for "isl_ast_build_expr_from_pw_aff".
2013 Sep 08
0
[LLVMdev] [Polly] Compile-time of Polly's code generation
...distinct base pointers that may alias and start from there. The
idea would be to collect for each of the base pointers all accesses that
use it, and to create an isl_pw_aff that is 'one' if the pointers do
overlap and 'zero' otherwise. You can use the isl code ast generator<
(isl_ast_build_expr_from_pw_aff()) to create LLVM IR that performs
exactly this check at run-time and you can use the result of this check
in executeScopConditionally() to only execute the modified SCoP, if we
found it safe to do so.
Cheers,
Tobias
2013 Sep 09
1
[LLVMdev] [Polly] Compile-time of Polly's code generation
...ointers that may alias and start from there. The
>idea would be to collect for each of the base pointers all accesses that
>use it, and to create an isl_pw_aff that is 'one' if the pointers do
>overlap and 'zero' otherwise. You can use the isl code ast generator<
>(isl_ast_build_expr_from_pw_aff()) to create LLVM IR that performs
>exactly this check at run-time and you can use the result of this check
>in executeScopConditionally() to only execute the modified SCoP, if we
>found it safe to do so.
I see, you mean we can generate LLVM code for runtime alias checking to allow mor...
2013 Sep 08
2
[LLVMdev] [Polly] Compile-time of Polly's code generation
At 2013-09-02 17:05:52,"Tobias Grosser" <tobias at grosser.es> wrote:
>On 09/01/2013 08:02 PM, Star Tan wrote:
>> Hi all,
>>
>>
>> It seems that Polly's code generation can leads to high compile-time overhead, especially for PolyBench applications such as 2mm, 3mm, gemm, syrk, etc. Some basic evaluation and analysis for Polly's code generation