Add benchmark tests/ffi/benchmark/external_call (#243)

crates/lune-std-ffi/README.md

@@ -23,6 +23,8 @@ See [tests/ffi](../../tests/ffi/README.md)
 
   > For windows API
 
+- Add varargs support
+
 ## Code structure
 
 ### /c

crates/lune-std-ffi/src/data/box_data/mod.rs

@@ -107,6 +107,7 @@ impl BoxData {
     }
 
     // Get size of box
+    #[inline]
     pub fn size(&self) -> usize {
         self.data.len()
     }
@@ -121,18 +122,22 @@ impl Drop for BoxData {
 }
 
 impl FfiData for BoxData {
+    #[inline]
     fn check_inner_boundary(&self, offset: isize, size: usize) -> bool {
         if offset < 0 {
             return false;
         }
         self.size() - (offset as usize) >= size
     }
+    #[inline]
     unsafe fn get_inner_pointer(&self) -> *mut () {
         self.data.as_ptr().cast_mut().cast::<()>()
     }
+    #[inline]
     fn is_readable(&self) -> bool {
         true
     }
+    #[inline]
     fn is_writable(&self) -> bool {
         true
     }

crates/lune-std-ffi/src/data/ref_data/bounds.rs

@@ -16,11 +16,13 @@ impl RefBounds {
         Self { above, below }
     }
 
+    #[inline]
     pub fn is_unsized(&self) -> bool {
         self.above == usize::MAX && self.below == usize::MAX
     }
 
     // Check boundary
+    #[inline]
     pub fn check_boundary(&self, offset: isize) -> bool {
         if self.is_unsized() {
             return true;
@@ -39,6 +41,7 @@ impl RefBounds {
 
     // Check boundary
     // Check required here
+    #[inline]
     pub fn check_sized(&self, offset: isize, size: usize) -> bool {
         if self.is_unsized() {
             return true;
@@ -61,6 +64,7 @@ impl RefBounds {
 
     // Calculate new bounds from bounds and offset
     // No boundary checking in here
+    #[inline]
     pub fn offset(&self, offset: isize) -> Self {
         let sign = offset.signum();
         let offset_abs = offset.unsigned_abs();

crates/lune-std-ffi/src/data/ref_data/mod.rs

@@ -127,15 +127,19 @@ impl Drop for RefData {
 }
 
 impl FfiData for RefData {
+    #[inline]
     fn check_inner_boundary(&self, offset: isize, size: usize) -> bool {
         self.boundary.check_sized(offset, size)
     }
+    #[inline]
     unsafe fn get_inner_pointer(&self) -> *mut () {
         **self.ptr
     }
+    #[inline]
     fn is_readable(&self) -> bool {
         u8_test(self.flags, RefFlag::Readable.value())
     }
+    #[inline]
     fn is_writable(&self) -> bool {
         u8_test(self.flags, RefFlag::Writable.value())
     }

tests/ffi/README.md

@@ -4,7 +4,7 @@
 
 gcc for library compiling (for external-\*)
 
-## Results
+## Test Results
 
 **External tests**
 
@@ -12,9 +12,7 @@ gcc for library compiling (for external-\*)
 - [x] [external_pointer](./external_pointer/init.luau)
 - [x] [external_print](./external_print/init.luau)
 - [x] [external_struct](./external_struct/init.luau)
-- [ ] [external_closure](./external_closure/init.luau)
+- [x] [external_closure](./external_closure/init.luau)
-
-  > failed (segfault)
 
 **Luau-side**
 
@@ -34,3 +32,47 @@ gcc for library compiling (for external-\*)
 - [ ] [cast](./cast)
 
   > need assertion
+
+## Benchmark Results
+
+> Note: LuaJit's os.clock function returns process CPU time (used) which much smaller then Luau's os.clock output. In this benchmark, luau uses 'time.h' instead of os.clock. See [utility/proc_clock](./utility/proc_clock/init.luau)
+
+### [benchmark/external_call](./benchmark/external_call/init.luau)
+
+**Target external c function**
+
+```c
+int add(int a, int b) {
+    return a + b;
+}
+```
+
+bench_scale = 1000000
+
+**Lune ffi call function**
+
+> cargo run run tests/ffi/benchmark/external_call
+> cargo run --profile=release run tests/ffi/benchmark/external_call
+
+Lune release target: 0.205127 (sec)
+Lune dev target: 1.556489 (sec)
+
+**LuaJit ffi call function**
+
+> luajit tests/ffi/benchmark/external_call/luajit.lua
+
+LuaJIT 2.1.1727870382: 0.001682 (sec)
+flags = JIT ON SSE3 SSE4.1 BMI2 fold cse dce fwd dse narrow loop abc sink fuse
+
+**Deno ffi call function**
+
+> deno run --unstable-ffi --allow-ffi ./tests/ffi/benchmark/external_call/deno.ts
+
+Deno 1.46.3: 0.006384 (sec)
+v8 = 12.9.202.5-rusty
+
+**Sysinformation**
+
+> CPU: AMD Ryzen 5 7600 (12) @ 5.1
+> MEM: 61898MiB 5600 MT/s
+> KERNEL: 6.8.12-2-pve (Proxmox VE 8.2.7 x86_64)

tests/ffi/benchmark/external_call/deno.ts

@@ -0,0 +1,23 @@
+import { libSuffix } from "../../utility/deno.ts";
+import { get_clock, get_offset } from "../../utility/proc_clock/deno.ts";
+
+const library_file = "./tests/ffi/benchmark/external_call/lib."+libSuffix;
+let library = Deno.dlopen(library_file, {
+    add: {
+        parameters: ["i32", "i32"],
+        result: "i32",
+    },
+});
+
+function bench_add(bench_size: number) {
+    let add = library.symbols.add;
+    let value = 0;
+    const before = get_clock();
+    for (let i=0; i<bench_size; i++) {
+        value = add(value,1);
+    }
+    const after = get_clock();
+    console.log(get_offset(before,after))
+}
+
+bench_add(1000000);

tests/ffi/benchmark/external_call/init.luau

@@ -0,0 +1,35 @@
+local ffi = require("@lune/ffi")
+local c = ffi.c
+
+local proc_clock = require("../../utility/proc_clock")
+local before, after = proc_clock.new_box()
+local get_clock = proc_clock.get_clock
+
+local testdir = "./tests/ffi/benchmark/external_call"
+local compile = require("../../utility/compile")
+compile(`{testdir}/lib.c`, `{testdir}/lib.so`)
+local lib = ffi.open(`{testdir}/lib.so`)
+
+local function bench_add(bench_scale: number)
+	local add_info = c.fn({ c.int, c.int }, c.int)
+
+	local add_callable = add_info:callable(lib:find("add"))
+
+	local a = c.int:box(0)
+	local delta = c.int:box(1)
+
+	local a_ref = a:ref()
+	local delta_ref = delta:ref()
+
+	get_clock(before)
+	for i = 1, bench_scale do
+		add_callable(a, a_ref, delta_ref)
+	end
+	get_clock(after)
+	print(proc_clock.get_offset(before, after))
+
+	local result = c.int:readData(a)
+	assert(result == bench_scale, `bench_add failed. result expected {bench_scale}, got {result}`)
+end
+
+bench_add(1000000)

tests/ffi/benchmark/external_call/lib.c

@@ -0,0 +1,4 @@
+
+int add(int a, int b) {
+    return a + b;
+}

tests/ffi/benchmark/external_call/luajit.lua

@@ -0,0 +1,27 @@
+--!nolint
+--!nocheck
+
+local ffi = require("ffi")
+
+local function bench_add(bench_scale)
+	ffi.cdef([[
+		int add(int a, int b);
+    ]])
+	local lib = ffi.load("./tests/ffi/benchmark/external_call/lib.so")
+	local add = lib.add
+	local a = 0
+
+	local before = os.clock()
+	for i = 1, bench_scale do
+		a = add(a, 1)
+	end
+	local after = os.clock()
+
+	print(after - before)
+	assert(
+		a == bench_scale,
+		string.format("bench_add failed. result expected %d, got %d", bench_scale, a)
+	)
+end
+
+bench_add(1000000)

tests/ffi/external_closure/init.luau

@@ -1,37 +1,36 @@
 local ffi = require("@lune/ffi")
+local c = ffi.c
 
 local testdir = "./tests/ffi/external_closure"
-
 local compile = require("../utility/compile")
 compile(`{testdir}/lib.c`, `{testdir}/lib.so`)
-
 local lib = ffi.open(`{testdir}/lib.so`)
 
 local function test_closure()
-	local callback_info = ffi.c.fn({ ffi.c.int, ffi.c.int }, ffi.c.int)
+	local callback_info = c.fn({ c.int, c.int }, c.int)
 	local callback_closure = callback_info:closure(function(ret, a, b)
-		ffi.c.int:writeData(ret, ffi.c.int:readData(a) + ffi.c.int:readData(b))
+		c.int:writeData(ret, c.int:readData(a) + c.int:readData(b))
 	end)
 
-	local closure_test_info = ffi.c.fn({ callback_info }, ffi.c.int)
+	local closure_test_info = c.fn({ callback_info }, c.int)
 
 	local closure_test_callable = closure_test_info:callable(lib:find("closure"))
 
-	local result_box = ffi.box(ffi.c.int.size)
+	local result_box = ffi.box(c.int.size)
 	closure_test_callable(result_box, callback_closure:ref())
-	local result = ffi.c.int:readData(result_box)
+	local result = c.int:readData(result_box)
 	assert(result == 72, `test_closure failed. result expected 20000, got {result}`)
 end
 
 test_closure()
 
 local function test_hello_world()
-	local callback_info = ffi.c.fn({}, ffi.c.void)
+	local callback_info = c.fn({}, c.void)
 	local callback_closure = callback_info:closure(function()
 		print("Hello world in lua closure!")
 	end)
 
-	local closure_test_info = ffi.c.fn({ callback_info }, ffi.c.void)
+	local closure_test_info = c.fn({ callback_info }, c.void)
 
 	local closure_test_callable = closure_test_info:callable(lib:find("hello_world"))
 

tests/ffi/utility/deno.ts

@@ -0,0 +1,12 @@
+export let libSuffix = "";
+switch (Deno.build.os) {
+  case "windows":
+    libSuffix = "dll";
+    break;
+  case "darwin":
+    libSuffix = "dylib";
+    break;
+  case "linux":
+    libSuffix = "so";
+    break;
+}

tests/ffi/utility/proc_clock/deno.ts

@@ -0,0 +1,25 @@
+import { libSuffix } from "../deno.ts";
+
+const library_file = "./tests/ffi/utility/proc_clock/lib."+libSuffix;
+let library = Deno.dlopen(library_file, {
+    sizeof_clock: {
+        parameters: [],
+        result: "i32",
+    },
+});
+const sizeof_clock = library.symbols.sizeof_clock();
+const type_clock_t = "u" + (sizeof_clock * 8);
+library.close();
+library = Deno.dlopen(library_file, {
+    get_clock: {
+        parameters: [],
+        result: type_clock_t,
+    },
+    get_offset: {
+        parameters: [type_clock_t, type_clock_t],
+        result: "f64",
+    },
+});
+
+export const get_clock = library.symbols.get_clock;
+export const get_offset = library.symbols.get_offset;

tests/ffi/utility/proc_clock/init.luau

@@ -0,0 +1,37 @@
+local ffi = require("@lune/ffi")
+local c = ffi.c
+
+local proc_clock = {}
+
+local libdir = "./tests/ffi/utility/proc_clock"
+local compile = require("../compile")
+compile(`{libdir}/lib.c`, `{libdir}/lib.so`)
+local lib = ffi.open(`{libdir}/lib.so`)
+
+-- sizeof_clock
+local sizeof_clock = c.fn({}, c.int):callable(lib:find("sizeof_clock"))
+function proc_clock.sizeof_clock()
+	local result = ffi.box(c.int.size)
+	sizeof_clock(result)
+	return c.int:readData(result)
+end
+
+-- get_clock
+local clock_t = ffi["u" .. (proc_clock.sizeof_clock() * 8)]
+assert(clock_t, "clock_t is unknown type")
+local get_clock = c.fn({}, clock_t):callable(lib:find("get_clock"))
+proc_clock.get_clock = get_clock
+
+-- get_offset
+local get_offset = c.fn({ clock_t, clock_t }, ffi.f64):callable(lib:find("get_offset"))
+function proc_clock.get_offset(before, after)
+	local result = ffi.box(ffi.f64.size)
+	get_offset(result, before:ref(), after:ref())
+	return ffi.f64:readData(result)
+end
+
+function proc_clock.new_box()
+	return ffi.box(clock_t.size), ffi.box(clock_t.size)
+end
+
+return proc_clock

tests/ffi/utility/proc_clock/lib.c

@@ -0,0 +1,12 @@
+#include <time.h>
+clock_t get_clock() {
+    return clock();
+}
+
+int sizeof_clock() {
+    return sizeof(clock_t);
+}
+
+double get_offset(clock_t before, clock_t after) {
+    return (double)(after - before) / CLOCKS_PER_SEC;
+}

types/ffi.luau

@@ -119,6 +119,15 @@ export type CallableData = (ret: (RefData|BoxData)?, ...RefData)->() & {
 	A lua function wrapper for function pointer
 ]=]
 export type ClosureData = {
+	--[=[
+		@within ClosureData
+		@tag Method
+		@method ref
+
+		Create a reference of the closure. usually can be used for passing function pointer as argument
+
+		@return A reference of the closure
+	]=]
 	ref: (self: ClosureData)->RefData,
 }