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

[DOCS] Matmul and vecadd working examples

Browse Source
This commit is contained in:
jack-willturner
2020-05-04 16:25:17 +01:00
committed by Philippe Tillet
parent ce4a4728f5
commit 32819dea51
6 changed files with 159 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
docs/tutorials/putting-it-all-together.rst
View File

@@ -110,9 +110,9 @@ However, in practice only A, B are provided by the user, and all the other :code
'TYPE' : dtype,
'AT' : transpose_a,
'BT' : transpose_b,
'TM' : [32, 64, 128]
'TN' : [32, 64, 128]
'TK' : [8]
'TM' : [32, 64, 128],
'TN' : [32, 64, 128],
'TK' : [8],
# handle A transposition
'USE_A' : '^a' if transpose_a else 'a',
'STRIDE_AK' : 'lda' if transpose_a else '1',
@@ -177,4 +177,4 @@ A callable operation can be created using the :code:`apply` method of the :code:
dot = _dot.apply
And that's it! In just ~100 lines of pure python, we have written a fully functional matrix multiplication that will not only work with automatic differentiation but also provide performance very close to cuBLAS. And it's all open-source~
And that's it! In just ~100 lines of pure python, we have written a fully functional matrix multiplication that will not only work with automatic differentiation but also provide performance very close to cuBLAS. And it's all open-source~