Skip to content

Commit

Permalink
#0: Restrict forced single core untilize on BH to non-sharded cases (#…
Browse files Browse the repository at this point in the history
…15353)

### Problem description
Forcing single core on BH for sharded tensors was causing failures in
recently-enabled tests in BH post-commit.

### What's changed
This commit adds another check that only forces single core on BH when
the input and output tensors are not sharded.

### Checklist
- [ ] Post commit CI passes
- [x] Blackhole Post commit
(https://github.com/tenstorrent/tt-metal/actions/runs/11963996332)
- [ ] Model regression CI testing passes (if applicable)
- [ ] Device performance regression CI testing passes (if applicable)
- [ ] New/Existing tests provide coverage for changes
  • Loading branch information
jaykru-tt authored Nov 22, 2024
1 parent 9df39e7 commit e7cd350
Showing 1 changed file with 2 additions and 1 deletion.
Original file line number Diff line number Diff line change
Expand Up @@ -110,8 +110,9 @@ operation::ProgramWithCallbacks Untilize::create_program(
const auto& input_tensor_a = input_tensors.at(0);
auto& output_tensor = output_tensors.at(0);
auto device_is_blackhole = input_tensor_a.device()->arch() == tt::ARCH::BLACKHOLE;
auto in_or_out_sharded = input_tensor_a.memory_config().is_sharded() || output_tensor.memory_config().is_sharded();
// FIXME: Remove this restriction once multicore untilize is supported on blackhole
if (this->use_multicore && !device_is_blackhole) {
if (this->use_multicore && (in_or_out_sharded || !device_is_blackhole)) {
return detail::untilize_multi_core(input_tensor_a, output_tensor, this->use_pack_untilize, this->fp32_dest_acc_en);
} else {
return detail::untilize_single_core(input_tensor_a, output_tensor, this->use_pack_untilize, this->fp32_dest_acc_en);
Expand Down

0 comments on commit e7cd350

Please sign in to comment.