Files
triton-tvm-ffi/examples/add/add.cc
2026-02-08 22:24:12 +08:00

38 lines
1.3 KiB
C++

#include <ATen/DLConvertor.h>
#include <ATen/dlpack.h>
#include <tvm/ffi/extra/cuda/cubin_launcher.h>
#include <tvm/ffi/function.h>
#include <tvm/ffi/tvm_ffi.h>
#ifndef ADD_KERNEL_STUB
#define ADD_KERNEL_STUB(grid, device, stream, args, kwargs)
#endif
#ifndef ADD_NAME
#define ADD_NAME ""
#endif
tvm::ffi::Tensor Add(tvm::ffi::Tensor x, tvm::ffi::Tensor y) {
at::Tensor xtorch = at::fromDLPack(x.ToDLPack());
at::Tensor otorch = at::empty_like(xtorch);
int32_t numel = otorch.numel();
tvm::ffi::Tensor output = tvm::ffi::Tensor::FromDLPack(at::toDLPack(otorch));
tvm::ffi::Function grid = tvm::ffi::Function::FromTyped(
[numel](const tvm::ffi::Map<tvm::ffi::String, tvm::ffi::Any> &meta)
-> tvm::ffi::Tuple<int32_t, int32_t, int32_t> {
const int32_t BLOCK_SIZE = meta["BLOCK_SIZE"].cast<int32_t>();
return tvm::ffi::Tuple((numel + BLOCK_SIZE - 1) / BLOCK_SIZE, 1, 1);
});
DLDevice device = x.device();
void *stream = TVMFFIEnvGetStream(device.device_type, device.device_id);
tvm::ffi::Array<tvm::ffi::Any> args = {x, y, output, numel, 1024};
tvm::ffi::Map<tvm::ffi::String, tvm::ffi::Any> kwargs = {};
ADD_KERNEL_STUB(grid, device.device_id, stream, args, kwargs);
return output;
}
TVM_FFI_STATIC_INIT_BLOCK() {
namespace refl = tvm::ffi::reflection;
refl::GlobalDef().def(ADD_NAME, Add);
}