diff options
author | Benjamin Kramer <kramerb@google.com> | 2018-07-26 13:41:53 -0700 |
---|---|---|
committer | TensorFlower Gardener <gardener@tensorflow.org> | 2018-07-26 13:45:39 -0700 |
commit | 6f4e330eb8dc18a44fa496467c4a5caa19a64e14 (patch) | |
tree | 650a8dc79ac4c39562086301d916c01923707031 | |
parent | 58841d73a4877398fe678d38c56de1ab9c30be68 (diff) |
[XLA:GPU] Don't fuse random things into fusions that become a cublas call
Only transpose and broadcast are valid. I think this used to work because we
didn't emit cublas calls for fused dots until recently.
PiperOrigin-RevId: 206213730
-rw-r--r-- | tensorflow/compiler/xla/service/gpu/instruction_fusion.cc | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc b/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc index af6259ae83..0f2c83aeb2 100644 --- a/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc +++ b/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc @@ -202,6 +202,7 @@ bool GpuInstructionFusion::ShouldFuse(HloInstruction* consumer, IsIEEEFloatingPointScalarConstant(producer->operand(0)) && fused_parameter_users[0]->opcode() == HloOpcode::kMultiply; } + return false; } // Other output fusions are not currently supported on GPUs. |