gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Revert "[BACKEND] Various bug fixes; making reductions faster (#533)".

Browse Source 
This is a more stable commit that produce bitwise identical code to earlier
versions. Using commits after this one may lead to slightly different numerics
This commit is contained in:
Philippe Tillet
2022-06-03 11:36:06 -07:00
parent efa04cac1f
commit a60374a597
11 changed files with 65 additions and 173 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/triton/codegen/analysis/layout.h
View File

@@ -224,7 +224,7 @@ struct scanline_layout: public distributed_layout {
int nts(size_t k) { return nts_.at(k); }
int contig_per_thread(size_t k) { return nts_.at(k); }
int per_thread(size_t k) { return contig_per_thread(k) * shape_[k] / shape_per_cta(k);}
int per_thread(size_t k) { return nts(k) * shape_[k] / shape_per_cta(k);}
public:
// micro tile size. The size of a tile held by a thread block.
std::vector<int> mts_;