Skip to content

feat(ascend): integrate InfiniOps backend #1537

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
voltjia merged 1 commit into from
Aug 26, 2026
Merged

feat(ascend): integrate InfiniOps backend #1537

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion src/infinicore/nn/rope.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,8 @@ void RoPE::initialize_cache() {
|| device_.getType() == Device::Type::METAX
|| device_.getType() == Device::Type::ILUVATAR
|| device_.getType() == Device::Type::CAMBRICON
|| device_.getType() == Device::Type::HYGON)
|| device_.getType() == Device::Type::HYGON
|| device_.getType() == Device::Type::ASCEND)
&& !mrope_section_) {
INFINICORE_NN_BUFFER_INIT(cos_sin_cache, ({max_seq_len_, rotary_dim_}, dtype_, device_));
}
Expand Down
4 changes: 4 additions & 0 deletions src/infinicore/ops/infiniops_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,6 +59,8 @@ inline infini::ops::Device toInfiniOpsDevice(const Device &device) {
return infini::ops::Device{infini::ops::Device::Type::kCambricon, static_cast<int>(device.getIndex())};
case Device::Type::HYGON:
return infini::ops::Device{infini::ops::Device::Type::kHygon, static_cast<int>(device.getIndex())};
case Device::Type::ASCEND:
return infini::ops::Device{infini::ops::Device::Type::kAscend, static_cast<int>(device.getIndex())};
default:
throw std::runtime_error("InfiniOps backend does not support this device type.");
}
Expand All @@ -72,6 +74,7 @@ inline bool isSupportedDevice(Device::Type device_type) {
case Device::Type::ILUVATAR:
case Device::Type::CAMBRICON:
case Device::Type::HYGON:
case Device::Type::ASCEND:
return true;
default:
return false;
Expand All @@ -86,6 +89,7 @@ void registerSupportedDevices(Dispatcher &dispatcher, Function function) {
dispatcher.registerDevice(Device::Type::ILUVATAR, function);
dispatcher.registerDevice(Device::Type::CAMBRICON, function);
dispatcher.registerDevice(Device::Type::HYGON, function);
dispatcher.registerDevice(Device::Type::ASCEND, function);
}

struct TensorMeta {
Expand Down
6 changes: 5 additions & 1 deletion src/infinicore/ops/rotary_embedding/rotary_embedding_infiniops.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,8 @@ void *plan(const Tensor &positions,
|| device_type == Device::Type::METAX
|| device_type == Device::Type::ILUVATAR
|| device_type == Device::Type::CAMBRICON
|| device_type == Device::Type::HYGON);
|| device_type == Device::Type::HYGON
|| device_type == Device::Type::ASCEND);
return new PlannedMeta{
TensorMeta(positions),
TensorMeta(query),
Expand Down Expand Up @@ -94,6 +95,9 @@ static bool registered = []() {
RotaryEmbedding::plan_dispatcher().registerDevice(Device::Type::HYGON, &plan);
RotaryEmbedding::run_dispatcher().registerDevice(Device::Type::HYGON, &run);
RotaryEmbedding::cleanup_dispatcher().registerDevice(Device::Type::HYGON, &cleanup);
RotaryEmbedding::plan_dispatcher().registerDevice(Device::Type::ASCEND, &plan);
RotaryEmbedding::run_dispatcher().registerDevice(Device::Type::ASCEND, &run);
RotaryEmbedding::cleanup_dispatcher().registerDevice(Device::Type::ASCEND, &cleanup);
return true;
}();

Expand Down
2 changes: 2 additions & 0 deletions xmake.lua
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -470,6 +470,8 @@ local function configure_infiniops_ops(infiniops_ops, xmake_os, json)
local implementations = "all"
if use_linked_implementation then
implementations = {16}
elseif has_config("ascend-npu") and op == "argmax" then
implementations = {0}
elseif op == "argmax" or op == "index_select" then
implementations = {8}
elseif op == "rms_norm" or op == "silu_and_mul" or op == "topk_softmax" then
Expand Down
Loading