56fd68e7dd
CCCL (CUDA C++ Core Libraries) provides: - CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk) - Thrust: high-level parallel algorithms (transform_reduce, sort, scan) - libcudacxx: CUDA C++ standard library (atomics, barriers, memory) - cudax: experimental features (memory resources, allocators) - Tuning policies: per-SM hardware-specific algorithm parameters Competition optimization vectors mapped to CCCL: - Output TPS (83% weight): warp_reduce, block_reduce, device_topk - Input TPS (14% weight): device_scan, block_load, prefetch - Cache TPS (3% weight): prefix caching strategy patterns - Memory (0.9 util): pooled/cached/buddy allocators Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only) License: Apache-2.0
118 lines
3.5 KiB
Plaintext
118 lines
3.5 KiB
Plaintext
//===----------------------------------------------------------------------===//
|
|
//
|
|
// Part of CUDA Experimental in CUDA C++ Core Libraries,
|
|
// under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include <cstdio>
|
|
#include <cstring>
|
|
|
|
#include <cccl/c/for.h>
|
|
#include <hostjit/codegen/cub_call.hpp>
|
|
#include <util/build_utils.h>
|
|
|
|
using namespace hostjit::codegen;
|
|
|
|
// d_in_0, num_items, op_0_state, stream
|
|
using for_fn_t = int (*)(void*, unsigned long long, void*, void*);
|
|
|
|
CUresult cccl_device_for_build_ex(
|
|
cccl_device_for_build_result_t* build_ptr,
|
|
cccl_iterator_t d_data,
|
|
cccl_op_t op,
|
|
int cc_major,
|
|
int cc_minor,
|
|
const char* cub_path,
|
|
const char* thrust_path,
|
|
const char* libcudacxx_path,
|
|
const char* ctk_path,
|
|
cccl_build_config* config)
|
|
try
|
|
{
|
|
if (build_ptr == nullptr)
|
|
{
|
|
return CUDA_ERROR_INVALID_VALUE;
|
|
}
|
|
std::string cccl_include_str = cccl::detail::parse_cccl_include_path(libcudacxx_path);
|
|
std::string ctk_root_str = cccl::detail::parse_ctk_root(ctk_path);
|
|
const char* cccl_include_path = cccl_include_str.empty() ? nullptr : cccl_include_str.c_str();
|
|
const char* ctk_root = ctk_root_str.empty() ? nullptr : ctk_root_str.c_str();
|
|
cccl::detail::MergedBuildConfig merged(config, cub_path, thrust_path);
|
|
|
|
auto result =
|
|
CubCall::from("cub/device/device_for.cuh")
|
|
.run("cub::DeviceFor::ForEachN")
|
|
.name("cccl_jit_for")
|
|
.with(in(d_data), num_items, for_each_op(op), stream)
|
|
.compile(cc_major, cc_minor, merged.get(), ctk_root, cccl_include_path);
|
|
|
|
build_ptr->cc = cc_major * 10 + cc_minor;
|
|
cccl::detail::copy_cubin(result.cubin, build_ptr->payload, build_ptr->payload_size);
|
|
build_ptr->jit_compiler = result.compiler;
|
|
build_ptr->for_fn = result.fn_ptr;
|
|
|
|
return CUDA_SUCCESS;
|
|
}
|
|
catch (const std::exception& exc)
|
|
{
|
|
fprintf(stderr, "\nEXCEPTION in cccl_device_for_build_ex(): %s\n", exc.what());
|
|
return CUDA_ERROR_UNKNOWN;
|
|
}
|
|
|
|
CUresult cccl_device_for(
|
|
cccl_device_for_build_result_t build, cccl_iterator_t d_data, uint64_t num_items, cccl_op_t op, CUstream stream)
|
|
try
|
|
{
|
|
if (!build.for_fn)
|
|
{
|
|
return CUDA_ERROR_INVALID_VALUE;
|
|
}
|
|
auto fn = reinterpret_cast<for_fn_t>(build.for_fn);
|
|
|
|
const int status = fn(d_data.state, num_items, op.state, reinterpret_cast<void*>(stream));
|
|
return (status == 0) ? CUDA_SUCCESS : CUDA_ERROR_UNKNOWN;
|
|
}
|
|
catch (const std::exception& exc)
|
|
{
|
|
fprintf(stderr, "\nEXCEPTION in cccl_device_for(): %s\n", exc.what());
|
|
return CUDA_ERROR_UNKNOWN;
|
|
}
|
|
|
|
CUresult cccl_device_for_build(
|
|
cccl_device_for_build_result_t* build,
|
|
cccl_iterator_t d_data,
|
|
cccl_op_t op,
|
|
int cc_major,
|
|
int cc_minor,
|
|
const char* cub_path,
|
|
const char* thrust_path,
|
|
const char* libcudacxx_path,
|
|
const char* ctk_path)
|
|
{
|
|
return cccl_device_for_build_ex(
|
|
build, d_data, op, cc_major, cc_minor, cub_path, thrust_path, libcudacxx_path, ctk_path, nullptr);
|
|
}
|
|
|
|
CUresult cccl_device_for_cleanup(cccl_device_for_build_result_t* build_ptr)
|
|
try
|
|
{
|
|
if (build_ptr == nullptr)
|
|
{
|
|
return CUDA_ERROR_INVALID_VALUE;
|
|
}
|
|
|
|
cccl::detail::release_jit_artifacts(build_ptr);
|
|
build_ptr->for_fn = nullptr;
|
|
|
|
return CUDA_SUCCESS;
|
|
}
|
|
catch (const std::exception& exc)
|
|
{
|
|
fprintf(stderr, "\nEXCEPTION in cccl_device_for_cleanup(): %s\n", exc.what());
|
|
return CUDA_ERROR_UNKNOWN;
|
|
}
|