Add host lib (#1134)

* Format

* Format

* Format

* Remove const

* Use the right template

* Format

* Format

* add row/col instances

* Add missing file

* fixed

* Format

* Updates

* Format

* fixed rrr layout

* Format

* Update test and embed modules

* Restore older version

* Update year

* Set -fPIC

* Format

* Use double for isnan

* rename host folder to codegen + minor fix

* add codegen CI test

* add option to build components without building CK

* fix the groovy syntax

* fix typo

* use the correct function for the codegen stage

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>

[ROCm/composable_kernel commit: 8eff4d62b6]

codegen/test/rtc/include/rtc/compile_kernel.hpp

@@ -0,0 +1,27 @@
+#ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_COMPILE_KERNEL
+#define GUARD_HOST_TEST_RTC_INCLUDE_RTC_COMPILE_KERNEL
+
+#include <rtc/kernel.hpp>
+#include <filesystem>
+#include <string>
+
+namespace rtc {
+
+struct src_file
+{
+    std::filesystem::path path;
+    std::string_view content;
+};
+
+struct compile_options
+{
+    std::string flags       = "";
+    std::string kernel_name = "main";
+};
+
+kernel compile_kernel(const std::vector<src_file>& src,
+                      compile_options options = compile_options{});
+
+} // namespace rtc
+
+#endif

codegen/test/rtc/include/rtc/hip.hpp

@@ -0,0 +1,78 @@
+#ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_HIP
+#define GUARD_HOST_TEST_RTC_INCLUDE_RTC_HIP
+
+#include <hip/hip_runtime_api.h>
+#include <memory>
+#include <string>
+
+namespace rtc {
+
+template <class T>
+struct buffer
+{
+    buffer() : ptr(), n(0) {}
+    buffer(std::shared_ptr<T> p, std::size_t sz) : ptr(p), n(sz) {}
+    buffer(std::shared_ptr<void> p, std::size_t sz)
+        : ptr(std::reinterpret_pointer_cast<T>(p)), n(sz)
+    {
+    }
+    explicit buffer(std::size_t sz) : ptr(new T[sz]), n(sz) {}
+    T* begin() { return data(); }
+    T* end() { return data() + size(); }
+    const T* begin() const { return data(); }
+    const T* end() const { return data() + size(); }
+
+    T& front() { return data()[0]; }
+    T& back() { return data()[size() - 1]; }
+    T& operator[](std::size_t i) { return data()[i]; }
+    T& at(std::size_t i)
+    {
+        if(i >= size())
+            throw std::runtime_error("Out of bounds");
+        return data()[i];
+    }
+
+    const T& front() const { return data()[0]; }
+    const T& back() const { return data()[size() - 1]; }
+    const T& operator[](std::size_t i) const { return data()[i]; }
+    const T& at(std::size_t i) const
+    {
+        if(i >= size())
+            throw std::runtime_error("Out of bounds");
+        return data()[i];
+    }
+    const T* data() const { return ptr.get(); }
+    T* data() { return ptr.get(); }
+
+    std::size_t size() const { return n; }
+    std::size_t bytes() const { return size() * sizeof(T); }
+
+    bool empty() const { return size() == 0; }
+
+    private:
+    std::shared_ptr<T> ptr;
+    std::size_t n;
+};
+
+std::string get_device_name();
+std::string hip_error(int error);
+
+std::shared_ptr<void> allocate_gpu(std::size_t sz, bool host = false);
+std::shared_ptr<void> write_to_gpu(const void* x, std::size_t sz, bool host = false);
+std::shared_ptr<void> read_from_gpu(const void* x, std::size_t sz);
+
+template <class T>
+buffer<T> to_gpu(const buffer<T>& input)
+{
+    return {write_to_gpu(input.data(), input.bytes()), input.size()};
+}
+
+template <class T>
+buffer<T> from_gpu(const buffer<T>& input)
+{
+    return {read_from_gpu(input.data(), input.bytes()), input.size()};
+}
+
+} // namespace rtc
+
+#endif

codegen/test/rtc/include/rtc/kernel.hpp

@@ -0,0 +1,62 @@
+#ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_KERNEL
+#define GUARD_HOST_TEST_RTC_INCLUDE_RTC_KERNEL
+
+#include <hip/hip_runtime_api.h>
+#include <memory>
+#include <string>
+#include <vector>
+
+namespace rtc {
+
+struct kernel_argument
+{
+    template <class T,
+              class U = std::remove_reference_t<T>,
+              class   = std::enable_if_t<not std::is_base_of<kernel_argument, T>{}>>
+    kernel_argument(T&& x) : size(sizeof(U)), align(alignof(U)), data(&x) // NOLINT
+    {
+    }
+    std::size_t size;
+    std::size_t align;
+    void* data;
+};
+
+std::vector<char> pack_args(const std::vector<kernel_argument>& args);
+
+struct kernel_impl;
+
+struct kernel
+{
+    kernel() = default;
+    kernel(const char* image, const std::string& name);
+    template <class T>
+    kernel(const std::vector<T>& image, const std::string& name)
+        : kernel(reinterpret_cast<const char*>(image.data()), name)
+    {
+        static_assert(sizeof(T) == 1, "Only byte types");
+    }
+
+    void launch(hipStream_t stream,
+                std::size_t global,
+                std::size_t local,
+                const std::vector<kernel_argument>& args) const;
+
+    void launch(hipStream_t stream,
+                std::size_t global,
+                std::size_t local,
+                std::vector<void*> args) const;
+
+    template <class... Ts>
+    auto launch(hipStream_t stream, std::size_t global, std::size_t local, Ts... zs) const
+    {
+        return [=](auto&&... xs) {
+            launch(stream, global, local, std::vector<kernel_argument>{xs...}, zs...);
+        };
+    }
+
+    private:
+    std::shared_ptr<kernel_impl> impl;
+};
+} // namespace rtc
+
+#endif

codegen/test/rtc/include/rtc/manage_ptr.hpp

@@ -0,0 +1,55 @@
+#ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_MANAGE_POINTER
+#define GUARD_HOST_TEST_RTC_INCLUDE_RTC_MANAGE_POINTER
+
+#include <type_traits>
+#include <memory>
+
+namespace rtc {
+template <class F, F f>
+struct manage_deleter
+{
+    template <class T>
+    void operator()(T* x) const
+    {
+        if(x != nullptr)
+        {
+            (void)f(x);
+        }
+    }
+};
+
+struct null_deleter
+{
+    template <class T>
+    void operator()(T*) const
+    {
+    }
+};
+
+template <class T, class F, F f>
+using manage_ptr = std::unique_ptr<T, manage_deleter<F, f>>;
+
+template <class T>
+struct element_type
+{
+    using type = typename T::element_type;
+};
+
+template <class T>
+using remove_ptr = typename std::
+    conditional_t<std::is_pointer<T>{}, std::remove_pointer<T>, element_type<T>>::type;
+
+template <class T>
+using shared = std::shared_ptr<remove_ptr<T>>;
+
+template <class T>
+shared<T> share(T p)
+{
+    return shared<T>{std::move(p)};
+}
+
+#define RTC_MANAGE_PTR(T, F) rtc::manage_ptr<std::remove_pointer_t<T>, decltype(&F), &F>
+
+} // namespace rtc
+
+#endif

codegen/test/rtc/include/rtc/tmp_dir.hpp

@@ -0,0 +1,24 @@
+#ifndef GUARD_HOST_TEST_RTC_INCLUDE_RTC_TMP_DIR
+#define GUARD_HOST_TEST_RTC_INCLUDE_RTC_TMP_DIR
+
+#include <string>
+#include <filesystem>
+
+namespace rtc {
+
+struct tmp_dir
+{
+    std::filesystem::path path;
+    tmp_dir(const std::string& prefix = "");
+
+    void execute(const std::string& cmd) const;
+
+    tmp_dir(tmp_dir const&) = delete;
+    tmp_dir& operator=(tmp_dir const&) = delete;
+
+    ~tmp_dir();
+};
+
+} // namespace rtc
+
+#endif