|
33 | 33 | function matmuladd!(C::AbstractMatrix, ::LoopedArrayOp, A::AbstractMatrix, |
34 | 34 | B::AbstractMatrix, bias::AbstractVector) |
35 | 35 | if unrolled_any(≤(256), (size(C, 1), size(A, 2), size(B, 2))) && |
| 36 | + unrolled_all(≤(1024), (size(C, 1), size(A, 2), size(B, 2))) && |
36 | 37 | LoopVectorization.check_args(C, A, B) |
37 | 38 | __matmuladd_loopvec!(C, A, B, bias) |
38 | 39 | return |
@@ -92,6 +93,7 @@ function matmul!(C::AbstractMatrix, ::AbstractInternalArrayOpMode, |
92 | 93 | end |
93 | 94 | function matmul!(C::AbstractMatrix, ::LoopedArrayOp, A::AbstractMatrix, B::AbstractMatrix) |
94 | 95 | if unrolled_any(≤(256), (size(C, 1), size(A, 2), size(B, 2))) && |
| 96 | + unrolled_all(≤(1024), (size(C, 1), size(A, 2), size(B, 2))) && |
95 | 97 | LoopVectorization.check_args(C, A, B) |
96 | 98 | __matmul_loopvec!(C, A, B) |
97 | 99 | return |
|
0 commit comments