Add more GPU architectures support (#112)

* Add more GPU architectures support

* Update layout.py

* Optimize performance, Add SM90 support, Add 1D2D SM100 support

* Add fmtlib submodule at commit 553ec11

---------

Co-authored-by: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com>

csrc/utils/system.hpp

@@ -0,0 +1,70 @@
+#pragma once
+
+#include <random>
+#include <string>
+#include <memory>
+
+#include "exception.hpp"
+
+namespace deep_gemm {
+
+// ReSharper disable once CppNotAllPathsReturnValue
+template <typename dtype_t>
+static dtype_t get_env(const std::string& name, const dtype_t& default_value = dtype_t()) {
+    const auto& c_str = std::getenv(name.c_str());
+    if (c_str == nullptr)
+        return default_value;
+
+    // Read the env and convert to the desired type
+    if constexpr (std::is_same_v<dtype_t, std::string>) {
+        return std::string(c_str);
+    } else if constexpr (std::is_same_v<dtype_t, int>) {
+        int value;
+        std::sscanf(c_str, "%d", &value);
+        return value;
+    } else {
+        DG_HOST_ASSERT(false and "Unexpected type");
+    }
+}
+
+static std::tuple<int, std::string> call_external_command(std::string command) {
+    command = command + " 2>&1";
+    const auto& deleter = [](FILE* f) { if (f) pclose(f); };
+    std::unique_ptr<FILE, decltype(deleter)> pipe(popen(command.c_str(), "r"), deleter);
+    DG_HOST_ASSERT(pipe != nullptr);
+
+    std::array<char, 512> buffer;
+    std::string output;
+    while (fgets(buffer.data(), buffer.size(), pipe.get()))
+        output += buffer.data();
+    const auto& exit_code = WEXITSTATUS(pclose(pipe.release()));
+    return {exit_code, output};
+}
+
+static std::filesystem::path make_dirs(const std::filesystem::path& path) {
+    // OK if existed
+    std::error_code capture;
+    const bool& created = std::filesystem::create_directories(path, capture);
+    DG_HOST_ASSERT(created or capture.value() == 0);
+    if (created and get_env<int>("DG_JIT_DEBUG"))
+        printf("Create directory: %s\n", path.c_str());
+    return path;
+}
+
+static std::string get_uuid() {
+    static std::random_device rd;
+    static std::mt19937 gen([]() {
+        return rd() ^ std::chrono::steady_clock::now().time_since_epoch().count();
+    }());
+    static std::uniform_int_distribution<uint32_t> dist;
+
+    std::stringstream ss;
+    ss << getpid() << "-"
+       << std::hex << std::setfill('0')
+       << std::setw(8) << dist(gen) << "-"
+       << std::setw(8) << dist(gen) << "-"
+       << std::setw(8) << dist(gen);
+    return ss.str();
+}
+
+} // deep_gemm