Skip to content

issue/1287 - direct gemm to InfiniOps #1288

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

Open
wooway777 wants to merge 1 commit into
base: main
Choose a base branch
from
+263 −0
Open

issue/1287 - direct gemm to InfiniOps #1288

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
59 changes: 59 additions & 0 deletions src/infinicore/ops/add/add_infiniops.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,59 @@
#include "infinicore/ops/add.hpp"

#ifdef ENABLE_INFINIOPS_API
#include "../infiniops_impl.hpp"

namespace infinicore::op::add_impl::infiniops {
namespace {

using TensorMeta = ::infinicore::op::infiniops::TensorMeta;

struct PlannedMeta {
TensorMeta c, a, b;
graph::GraphTensor c_tensor, a_tensor, b_tensor;
};

} // namespace

void *plan(Tensor c, const Tensor &a, const Tensor &b) {
INFINICORE_ASSERT(c->device().getType() == Device::Type::NVIDIA);
INFINICORE_ASSERT_TENSORS_SAME_DEVICE(c, a, b);

return new PlannedMeta{
TensorMeta(c),
TensorMeta(a),
TensorMeta(b),
graph::GraphTensor(c),
graph::GraphTensor(a),
graph::GraphTensor(b)};
}

void run(void *planned_meta) {
auto planned = reinterpret_cast<PlannedMeta *>(planned_meta);

infini::ops::Handle handle;
handle.set_stream(context::getStream());
infini::ops::Config config;

infini::ops::Operator<infini::ops::Add>::Call(
handle,
config,
planned->a.tensor(planned->a_tensor),
planned->b.tensor(planned->b_tensor),
planned->c.tensor(planned->c_tensor));
}

void cleanup(void **planned_meta_ptr) {
delete *reinterpret_cast<PlannedMeta **>(planned_meta_ptr);
*planned_meta_ptr = nullptr;
}

static bool registered = []() {
Add::plan_dispatcher().registerDevice(Device::Type::NVIDIA, &plan);
Add::run_dispatcher().registerDevice(Device::Type::NVIDIA, &run);
Add::cleanup_dispatcher().registerDevice(Device::Type::NVIDIA, &cleanup);
return true;
}();

} // namespace infinicore::op::add_impl::infiniops
#endif
68 changes: 68 additions & 0 deletions src/infinicore/ops/gemm/gemm_infiniops.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,68 @@
#include "infinicore/ops/gemm.hpp"

#ifdef ENABLE_INFINIOPS_API
#include "../infiniops_impl.hpp"

#include <optional>

namespace infinicore::op::gemm_impl::infiniops {
namespace {

using TensorMeta = ::infinicore::op::infiniops::TensorMeta;

struct PlannedMeta {
TensorMeta c, a, b;
graph::GraphTensor c_tensor, a_tensor, b_tensor;
float alpha, beta;
};

} // namespace

void *plan(Tensor c, const Tensor &a, const Tensor &b, float alpha, float beta) {
INFINICORE_ASSERT(c->device().getType() == Device::Type::NVIDIA);
INFINICORE_ASSERT_TENSORS_SAME_DEVICE(c, a, b);

return new PlannedMeta{
TensorMeta(c),
TensorMeta(a),
TensorMeta(b),
graph::GraphTensor(c),
graph::GraphTensor(a),
graph::GraphTensor(b),
alpha,
beta};
}

void run(void *planned_meta) {
auto planned = reinterpret_cast<PlannedMeta *>(planned_meta);

infini::ops::Handle handle;
handle.set_stream(context::getStream());
infini::ops::Config config;

infini::ops::Operator<infini::ops::Gemm>::Call(
handle,
config,
planned->a.tensor(planned->a_tensor),
planned->b.tensor(planned->b_tensor),
std::optional<float>{planned->alpha},
std::optional<float>{planned->beta},
std::optional<int>{},
std::optional<int>{},
planned->c.tensor(planned->c_tensor));
}

void cleanup(void **planned_meta_ptr) {
delete *reinterpret_cast<PlannedMeta **>(planned_meta_ptr);
*planned_meta_ptr = nullptr;
}

static bool registered = []() {
Gemm::plan_dispatcher().registerDevice(Device::Type::NVIDIA, &plan);
Gemm::run_dispatcher().registerDevice(Device::Type::NVIDIA, &run);
Gemm::cleanup_dispatcher().registerDevice(Device::Type::NVIDIA, &cleanup);
return true;
}();

} // namespace infinicore::op::gemm_impl::infiniops
#endif
72 changes: 72 additions & 0 deletions src/infinicore/ops/infiniops_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,72 @@
#pragma once

#include "../utils.hpp"
#include "infinicore/tensor.hpp"

#include <stdexcept>

#include "infini/operator_call_instantiations.h"
#include "tensor.h"

namespace infinicore::op::infiniops {

inline infini::ops::DataType toInfiniOpsDtype(DataType dtype) {
switch (dtype) {
case DataType::I8:
return infini::ops::DataType::kInt8;
case DataType::I16:
return infini::ops::DataType::kInt16;
case DataType::I32:
return infini::ops::DataType::kInt32;
case DataType::I64:
return infini::ops::DataType::kInt64;
case DataType::U8:
case DataType::BYTE:
return infini::ops::DataType::kUInt8;
case DataType::U16:
return infini::ops::DataType::kUInt16;
case DataType::U32:
return infini::ops::DataType::kUInt32;
case DataType::U64:
return infini::ops::DataType::kUInt64;
case DataType::F16:
return infini::ops::DataType::kFloat16;
case DataType::BF16:
return infini::ops::DataType::kBFloat16;
case DataType::F32:
return infini::ops::DataType::kFloat32;
case DataType::F64:
return infini::ops::DataType::kFloat64;
default:
throw std::runtime_error("InfiniOps backend does not support this tensor dtype.");
}
}

inline infini::ops::Device toInfiniOpsDevice(const Device &device) {
INFINICORE_ASSERT(device.getType() == Device::Type::NVIDIA);
return infini::ops::Device{infini::ops::Device::Type::kNvidia, static_cast<int>(device.getIndex())};
}

struct TensorMeta {
Shape shape;
Strides strides;
infini::ops::DataType dtype;
infini::ops::Device device;

explicit TensorMeta(const Tensor &tensor)
: shape(tensor->shape()),
strides(tensor->strides()),
dtype(toInfiniOpsDtype(tensor->dtype())),
device(toInfiniOpsDevice(tensor->device())) {}

infini::ops::Tensor tensor(const void *data) const {
return infini::ops::Tensor(
const_cast<void *>(data), shape, dtype, device, strides);
}

infini::ops::Tensor tensor(const Tensor &tensor) const {
return this->tensor(tensor->data());
}
};

} // namespace infinicore::op::infiniops
64 changes: 64 additions & 0 deletions xmake.lua
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -266,6 +266,19 @@ if has_config("ccl") then
add_defines("ENABLE_CCL")
end

-- InfiniOps
option("infiniops")
set_default(false)
set_showmenu(true)
set_description("Whether to use InfiniOps kernels where adapters are available")
option_end()

option("infiniops-root")
set_default("submodules/InfiniOps")
set_showmenu(true)
set_description("Path to the InfiniOps repository used by --infiniops")
option_end()

-- Mutual Awareness Analyzer
option("mutual-awareness")
set_default(false)
Expand Down Expand Up @@ -353,6 +366,11 @@ target("infiniop")
set_kind("shared")
add_deps("infinirt")

if has_config("nv-gpu") then
local cuda_root = os.getenv("CUDA_HOME") or os.getenv("CUDA_PATH") or get_config("cuda") or "/usr/local/cuda"
add_includedirs(cuda_root .. "/include")
end

if has_config("cpu") then
add_deps("infiniop-cpu")
end
Expand Down Expand Up @@ -467,6 +485,40 @@ target("infinicore_cpp_api")

add_includedirs("include")
add_includedirs(INFINI_ROOT.."/include", { public = true })
if has_config("nv-gpu") then
local cuda_root = os.getenv("CUDA_HOME") or os.getenv("CUDA_PATH") or get_config("cuda") or "/usr/local/cuda"
add_includedirs(cuda_root .. "/include")
end
if has_config("infiniops") then
local infiniops_root = path.absolute(get_config("infiniops-root") or "submodules/InfiniOps", os.projectdir())
local infiniops_builddir = path.join(infiniops_root, "build")
if not os.isdir(infiniops_root) then
raise("InfiniOps root not found: " .. infiniops_root)
end
if not has_config("nv-gpu") then
raise("InfiniOps integration currently has adapters only for NVIDIA")
end
add_defines("ENABLE_INFINIOPS_API")
add_includedirs(infiniops_root .. "/src", infiniops_root .. "/include", infiniops_root .. "/generated/include")
add_linkdirs(infiniops_builddir .. "/src")
add_links("infiniops")
add_rpathdirs(infiniops_builddir .. "/src")
add_installfiles(infiniops_builddir .. "/src/libinfiniops.so", {prefixdir = "lib"})
before_build(function (target)
import("core.base.option")
local infiniops_root = path.absolute(get_config("infiniops-root") or "submodules/InfiniOps", os.projectdir())
local infiniops_builddir = path.join(infiniops_root, "build")
os.execv("cmake", {
"-S", infiniops_root,
"-B", infiniops_builddir,
"-DWITH_NVIDIA=ON",
"-DGENERATE_OPERATOR_CALL_INSTANTIATIONS=ON",
"-DGENERATE_PYTHON_BINDINGS=OFF",
"-DCMAKE_BUILD_TYPE=Release"
})
os.execv("cmake", {"--build", infiniops_builddir, "--target", "infiniops"})
end)
end

add_linkdirs(INFINI_ROOT.."/lib")
add_links("infiniop", "infinirt", "infiniccl")
Expand Down Expand Up @@ -675,6 +727,18 @@ target("_infinicore")

add_files("src/infinicore/pybind11/**.cc")

if has_config("infiniops") then
after_install(function (target)
local INFINI_ROOT = os.getenv("INFINI_ROOT") or (os.getenv(is_host("windows") and "HOMEPATH" or "HOME") .. "/.infini")
local infiniops_root = path.absolute(get_config("infiniops-root") or "submodules/InfiniOps", os.projectdir())
local infiniops_lib = path.join(infiniops_root, "build", "src", "libinfiniops.so")
os.mkdir(path.join(INFINI_ROOT, "lib"))
os.cp(infiniops_lib, path.join(INFINI_ROOT, "lib"))
os.mkdir(path.join(os.projectdir(), "python", "infinicore", "lib"))
os.cp(infiniops_lib, path.join(os.projectdir(), "python", "infinicore", "lib"))
end)
end

set_installdir("python/infinicore")
target_end()

Expand Down
Loading