Merge pull request #30239 from benbarsdell:amp-add-batchmatmulv2
PiperOrigin-RevId: 255699129
This commit is contained in:
commit
dcbf7f7bc0
@ -62,7 +62,7 @@ class AutoMixedPrecisionLists {
|
||||
|
||||
auto list = gtl::FlatSet<string> {
|
||||
#if CUDA_VERSION >= 9010 // Fp16 BatchMatMul is slow before CUDA 9.1.
|
||||
"BatchMatMul",
|
||||
"BatchMatMul", "BatchMatMulV2",
|
||||
#endif
|
||||
"BlockLSTM", "BlockLSTMGrad", "Conv2D", "Conv2DBackpropFilter",
|
||||
"Conv2DBackpropInput",
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user