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

[OPTIMIZER] Made layout simplification pass efficient for fused attention kernels (#790)

Browse Source
This commit is contained in:
Philippe Tillet
2022-10-21 16:52:15 -07:00
committed by GitHub
parent c4726333bf
commit bb0f9235d1
26 changed files with 683 additions and 229 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
test/Conversion/triton_to_tritongpu.mlir
View File

@@ -5,7 +5,7 @@ func @ops() {
%a = arith.constant dense<1.00e+00> : tensor<128x32xf16>
%b = arith.constant dense<2.00e+00> : tensor<32x128xf16>
%c = arith.constant dense<3.00e+00> : tensor<128x128xf32>
%0 = tt.dot %a, %b, %c {allowTF32 = true} : tensor<128x32xf16> * tensor<32x128xf16> -> tensor<128x128xf32>
%0 = tt.dot %a, %b, %c {allowTF32 = true, transA = false, transB = false} : tensor<128x32xf16> * tensor<32x128xf16> -> tensor<128x128xf32>
return
}