diff options
Diffstat (limited to 'tensorflow/compiler/xla/service/gpu/instruction_fusion.cc')
| -rw-r--r-- | tensorflow/compiler/xla/service/gpu/instruction_fusion.cc | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc b/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc index 0f2c83aeb2..de8d021321 100644 --- a/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc +++ b/tensorflow/compiler/xla/service/gpu/instruction_fusion.cc @@ -26,7 +26,7 @@ namespace gpu { namespace { -bool IsFusile(const HloInstruction& hlo) { +bool IsFusible(const HloInstruction& hlo) { // Don't fuse get-tuple-element on GPU: We can, but it's slower than not // fusing. We never generate kernels for unfused GTEs. Instead, if an // unfused GTE is an input to a kernel (including a fusion kernel), we @@ -41,7 +41,7 @@ bool IsFusile(const HloInstruction& hlo) { hlo.opcode() == HloOpcode::kDynamicUpdateSlice || hlo.opcode() == HloOpcode::kFusion || hlo.opcode() == HloOpcode::kGather || - hlo.opcode() == HloOpcode::kPad || + hlo.opcode() == HloOpcode::kIota || hlo.opcode() == HloOpcode::kPad || hlo.opcode() == HloOpcode::kReduce || hlo.opcode() == HloOpcode::kReduceWindow || hlo.opcode() == HloOpcode::kReshape || @@ -245,7 +245,7 @@ bool GpuInstructionFusion::ShouldFuse(HloInstruction* consumer, return true; } - if (!IsFusile(*producer) || !IsFusile(*consumer) || + if (!IsFusible(*producer) || !IsFusible(*consumer) || !InstructionFusion::ShouldFuse(consumer, operand_index)) { return false; } |