Files
pytorch/torch/csrc/jit/tensorexpr/cpp_codegen.cpp
Cheng Chang 109bc1047e [NNC] Generate C++ code for Allocate and Free (#51070)
Summary:
This is the initial skeleton for C++ codegen, it includes generations for Allocate and Free.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/51070

Test Plan: New unit tests are added to `test_cpp_codegen.cpp`.

Reviewed By: ZolotukhinM

Differential Revision: D26061818

Pulled By: cheng-chang

fbshipit-source-id: b5256b2dcee6b2583ba73b6c9684994dbe7cdc1f
2021-02-01 13:06:51 -08:00

45 lines
1.2 KiB
C++

#include <torch/csrc/jit/tensorexpr/cpp_codegen.h>
namespace torch {
namespace jit {
namespace tensorexpr {
void CppPrinter::visit(const Allocate* alloc) {
constexpr size_t kAllocOnStackThresholdSize = 512;
size_t size = 1;
for (auto dim : alloc->dims()) {
const IntImm* v = dynamic_cast<const IntImm*>(dim);
if (v) {
size *= v->value();
} else {
throw std::runtime_error("Only IntImm dimensions are supported for now");
}
}
emitIndent();
if (size <= kAllocOnStackThresholdSize) {
os() << alloc->dtype().ToCppString() << " " << (*alloc->buffer_var()) << "["
<< size << "];" << std::endl;
} else {
size *= alloc->dtype().byte_size();
os() << alloc->dtype().ToCppString() << "* " << (*alloc->buffer_var())
<< " = static_cast<" << alloc->dtype().ToCppString() << "*>(malloc("
<< size << "));" << std::endl;
allocated_on_heap_.insert(alloc->buffer_var());
}
}
void CppPrinter::visit(const Free* free) {
const Var* var = free->buffer_var();
if (allocated_on_heap_.count(var)) {
emitIndent();
os() << "free(" << name_manager()->get_unique_name(var) << ");"
<< std::endl;
}
}
} // namespace tensorexpr
} // namespace jit
} // namespace torch