experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix a potential overflow bug.

Browse Source 
Also remove an unnecessary capture.

PiperOrigin-RevId: 308582689
Change-Id: Ifa9734cd77fe41e79bbf89e10e6a1bd0b0021ddf
This commit is contained in:
Adrian Kuegel 2020-04-27 02:32:38 -07:00 committed by TensorFlower Gardener
parent 70462ca5d6
commit 036b75a818
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/compiler/xla/service/gpu/ir_emitter_unnested.cc
View File

@ -1866,7 +1866,7 @@ namespace {
bool MayPreventVectorization(const HloInstruction& hlo) {
if (hlo.opcode() == HloOpcode::kFusion) {
return absl::c_any_of(hlo.fused_instructions_computation()->instructions(),
[&](const HloInstruction* instr) {
[](const HloInstruction* instr) {
switch (instr->opcode()) {
case HloOpcode::kReduceWindow:
case HloOpcode::kSort:
@ -1963,7 +1963,7 @@ static void UnrollInnerTileLoop(
IrArray::Index source_idx_x_base = source_idx.AddOffsetToDim(y_loc, kDimY, b);
for (int64 j = 0; j < x_num_steps / vector_size; j++) {
for (int64 i = 0; i < vector_size; i++) {
int linear_index = j * vector_size + i;
int64 linear_index = j * vector_size + i;
llvm::Value* x_loc = b->CreateAdd(constant(j * step_x * vector_size + i),
start_offset_x, "x_loc");
IrArray::Index source_idx_x = source_idx_x_base.AddOffsetToDim(