koffi 0.9.25 → 0.9.28

package/README.md

@@ -261,84 +261,84 @@ At this stage, two benchmarks are implemented:
 
 In order to run it, go to `koffi/benchmark` and run `../../cnoke/cnoke.js` (or `node ..\..\cnoke\cnoke.js` on Windows) before doing anything else.
 
-Once this is done, you can execute each implementation, e.g. `build/atoi_cc 20000000` or `./atoi_koffi.js 20000000`.
+Once this is done, you can execute each implementation, e.g. `build/atoi_cc` or `./atoi_koffi.js`. You can optionally define a custom number of iterations, e.g. `./atoi_koffi.js 10000000`.
 
 ## atoi results
 
-Here are some results from 2022-04-24 on my Linux machine (AMD® Ryzen™ 7 5800H 16G):
+Here are some results from 2022-04-24 on Linux on my machine (AMD® Ryzen™ 7 5800H 16G):
 
 ```sh
-$ build/atoi_cc 20000000
+$ build/atoi_cc
 Iterations: 20000000
 Time: 0.24s
 
-$ ./atoi_napi.js 20000000
+$ ./atoi_napi.js
 Iterations: 20000000
-Time: 1.56s
+Time: 1.10s
 
-$ ./atoi_koffi.js 20000000
+$ ./atoi_koffi.js
 Iterations: 20000000
-Time: 2.41s
+Time: 1.91s
 
 # Note: the Node-FFI version does a few setTimeout calls to force the GC to run (around 20
 # for the example below), without which Node will consume all memory because the GC never appears
 # to run, or not enough. It's not ideal but on the other hand it counts as another limitation
 # to Node-FFI performance.
-$ ./atoi_node_ffi.js 20000000
+$ ./atoi_node_ffi.js
 Iterations: 20000000
 Time: 640.49s
 ```
 
-And on my Windows machine (Intel® Core™ i5-4460 16G):
+And on Windows on the same machine (AMD® Ryzen™ 7 5800H 16G):
 
 ```sh
-$ build\atoi_cc.exe 20000000
+$ build\atoi_cc.exe
 Iterations: 20000000
-Time: 0.66s
+Time: 0.25s
 
-$ node atoi_napi.js 20000000
+$ node atoi_napi.js
 Iterations: 20000000
-Time: 3.52s
+Time: 1.94s
 
-$ node atoi_koffi.js 20000000
+$ node atoi_koffi.js
 Iterations: 20000000
-Time: 4.81s
+Time: 3.15s
 
-$ node atoi_node_ffi.js 20000000
+$ node atoi_node_ffi.js
 Iterations: 20000000
-Time: 491.99s
+Time: 267.20s
 ```
 
 ## Raylib results
 
-Here are some results from 2022-04-24 on my Linux machine (AMD® Ryzen™ 7 5800H 16G):
+Here are some results from 2022-04-24 on Linux on my machine (AMD® Ryzen™ 7 5800H 16G):
 
 ```sh
-$ build/raylib_cc 100
+$ build/raylib_cc
 Iterations: 100
 Time: 4.14s
 
-$ ./raylib_koffi.js 100
+$ ./raylib_koffi.js
 Iterations: 100
 Time: 6.25s
 
-$ ./raylib_node_ffi.js 100
+$ ./raylib_node_ffi.js
 Iterations: 100
 Time: 27.13s
 ```
 
-And on my Windows machine (Intel® Core™ i5-4460 16G):
+And on Windows on the same machine (AMD® Ryzen™ 7 5800H 16G):
 
 ```sh
-$ build\raylib_cc.exe 100
+$ build\raylib_cc.exe
 Iterations: 100
-Time: 10.53s
+Time: 8.39s
 
-$ node raylib_koffi.js 100
+$ node raylib_koffi.js
 Iterations: 100
-Time: 14.60s
+Time: 11.51s
 
-$ node raylib_node_ffi.js 100
+$ node raylib_node_ffi.js
 Iterations: 100
-Time: 44.97s
+Time: 32.47s
 ```

package/benchmark/CMakeLists.txt

@@ -16,15 +16,21 @@ project(koffi C CXX ASM)
 
 find_package(CNoke)
 
-set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD 20)
 
-add_subdirectory(.. koffi)
+set(THREADS_PREFER_PTHREAD_FLAG ON)
+find_package(Threads REQUIRED)
+
+if(NOT TARGET koffi)
+    add_subdirectory(.. koffi)
+endif()
 add_subdirectory(../test test)
 
 # ---- atoi ----
 
 add_executable(atoi_cc atoi_cc.cc ../vendor/libcc/libcc.cc)
 target_include_directories(atoi_cc PRIVATE ..)
+target_link_libraries(atoi_cc PRIVATE Threads::Threads)
 
 if(WIN32)
     target_compile_definitions(atoi_cc PRIVATE _CRT_SECURE_NO_WARNINGS _CRT_NONSTDC_NO_DEPRECATE)
@@ -33,19 +39,24 @@ endif()
 
 add_node_addon(NAME atoi_napi SOURCES atoi_napi.cc ../vendor/libcc/libcc.cc)
 target_include_directories(atoi_napi PRIVATE .. ../vendor/node-addon-api)
+target_link_libraries(atoi_napi PRIVATE Threads::Threads)
 
 if(WIN32)
     target_compile_definitions(atoi_napi PRIVATE _CRT_SECURE_NO_WARNINGS _CRT_NONSTDC_NO_DEPRECATE)
     target_link_libraries(atoi_napi PRIVATE ws2_32)
+else()
+    target_link_libraries(atoi_napi PRIVATE dl)
 endif()
 
 # ---- Raylib ----
 
-add_executable(raylib_cc atoi_cc.cc ../vendor/libcc/libcc.cc)
+add_executable(raylib_cc raylib_cc.cc ../vendor/libcc/libcc.cc)
 target_include_directories(raylib_cc PRIVATE ..)
-add_dependencies(raylib_cc raylib)
+target_link_libraries(raylib_cc PRIVATE Threads::Threads raylib)
 
 if(WIN32)
     target_compile_definitions(raylib_cc PRIVATE _CRT_SECURE_NO_WARNINGS _CRT_NONSTDC_NO_DEPRECATE)
     target_link_libraries(raylib_cc PRIVATE ws2_32)
+else()
+    target_link_libraries(raylib_cc PRIVATE dl)
 endif()

package/benchmark/atoi_cc.cc

@@ -25,21 +25,18 @@ volatile uint64_t sum = 0;
 
 int Main(int argc, char **argv)
 {
-    if (argc < 2) {
-        LogError("Missing number of iterations");
-        LogInfo("Usage: atoi_cc <iterations>");
-        return 1;
-    }
+    int iterations = 20000000;
 
-    int iterations = 0;
-    if (!ParseInt(argv[1], &iterations))
-        return 1;
+    if (argc >= 2) {
+        if (!ParseInt(argv[1], &iterations))
+            return 1;
+    }
     LogInfo("Iterations: %1", iterations);
 
     int64_t start = GetMonotonicTime();
 
     for (int i = 0; i < iterations; i++) {
-        sum += (uint64_t)atoi(strings[i % RG_LEN(strings)]);
+        sum = sum + (uint64_t)atoi(strings[i % RG_LEN(strings)]);
     }
 
     // Help prevent optimisation of loop

package/benchmark/atoi_koffi.js

@@ -13,7 +13,7 @@
 // You should have received a copy of the GNU Affero General Public License
 // along with this program. If not, see https://www.gnu.org/licenses/.
 
-const koffi = require('../build/koffi.node');
+const koffi = require('./build/koffi.node');
 
 const strings = [
     '424242',
@@ -26,14 +26,15 @@ let sum = 0;
 main();
 
 function main() {
-    if (process.argv.length < 3)
-        throw new Error('Missing number of iterations');
-
-    let iterations = parseInt(process.argv[2], 10);
-    if (Number.isNaN(iterations))
-        throw new Error('Not a valid number');
-    if (iterations < 1)
-        throw new Error('Value must be positive');
+    let iterations = 20000000;
+
+    if (process.argv.length >= 3) {
+        iterations = parseInt(process.argv[2], 10);
+        if (Number.isNaN(iterations))
+            throw new Error('Not a valid number');
+        if (iterations < 1)
+            throw new Error('Value must be positive');
+    }
     console.log('Iterations:', iterations);
 
     let lib = koffi.load(process.platform == 'win32' ? 'msvcrt.dll' : null);

package/benchmark/atoi_napi.cc

@@ -30,20 +30,26 @@ static Napi::Value RunAtoi(const Napi::CallbackInfo &info)
 {
     Napi::Env env = info.Env();
 
-    // We want maximum performance here, this is a benchmark
-#if 0
-    if (info.Length() < 1) {
+    if (RG_UNLIKELY(info.Length() < 1)) {
         ThrowError<Napi::TypeError>(env, "Expected 1 argument, got %1", info.Length());
         return env.Null();
     }
-    if (!info[0].IsString()) {
-        ThrowError<Napi::TypeError>(env, "Unexpected value for str, expected string");
-        return env.Null();
+
+    char str[64];
+    {
+        napi_status status = napi_get_value_string_utf8(env, info[0], str, RG_SIZE(str), nullptr);
+
+        if (RG_UNLIKELY(status != napi_ok)) {
+            if (status == napi_string_expected) {
+                ThrowError<Napi::TypeError>(env, "Unexpected value for str, expected string");
+            } else {
+                ThrowError<Napi::TypeError>(env, "Failed to read JS string");
+            }
+            return env.Null();
+        }
     }
-#endif
 
-    std::string name = info[0].As<Napi::String>();
-    int value = atoi(name.c_str());
+    int value = atoi(str);
 
     return Napi::Number::New(env, value);
 }

package/benchmark/atoi_napi.js

@@ -26,14 +26,15 @@ let sum = 0;
 main();
 
 function main() {
-    if (process.argv.length < 3)
-        throw new Error('Missing number of iterations');
-
-    let iterations = parseInt(process.argv[2], 10);
-    if (Number.isNaN(iterations))
-        throw new Error('Not a valid number');
-    if (iterations < 1)
-        throw new Error('Value must be positive');
+    let iterations = 20000000;
+
+    if (process.argv.length >= 3) {
+        iterations = parseInt(process.argv[2], 10);
+        if (Number.isNaN(iterations))
+            throw new Error('Not a valid number');
+        if (iterations < 1)
+            throw new Error('Value must be positive');
+    }
     console.log('Iterations:', iterations);
 
     let start = performance.now();

package/benchmark/atoi_node_ffi.js

@@ -16,7 +16,6 @@
 const ref = require('ref-napi');
 const ffi = require('ffi-napi');
 const struct = require('ref-struct-di')(ref);
-const path = require('path');
 
 const strings = [
     '424242',
@@ -29,14 +28,15 @@ let sum = 0;
 main();
 
 async function main() {
-    if (process.argv.length < 3)
-        throw new Error('Missing number of iterations');
-
-    let iterations = parseInt(process.argv[2], 10);
-    if (Number.isNaN(iterations))
-        throw new Error('Not a valid number');
-    if (iterations < 1)
-        throw new Error('Value must be positive');
+    let iterations = 20000000;
+
+    if (process.argv.length >= 3) {
+        iterations = parseInt(process.argv[2], 10);
+        if (Number.isNaN(iterations))
+            throw new Error('Not a valid number');
+        if (iterations < 1)
+            throw new Error('Value must be positive');
+    }
     console.log('Iterations:', iterations);
 
     const lib = ffi.Library(process.platform == 'win32' ? 'msvcrt.dll' : null, {

package/benchmark/raylib_cc.cc

@@ -18,15 +18,12 @@ namespace RG {
 
 int Main(int argc, char **argv)
 {
-    if (argc < 2) {
-        LogError("Missing number of iterations");
-        LogInfo("Usage: raylib_cc <iterations>");
-        return 1;
-    }
+    int iterations = 100;
 
-    int iterations = 0;
-    if (!ParseInt(argv[1], &iterations))
-        return 1;
+    if (argc >= 2) {
+        if (!ParseInt(argv[1], &iterations))
+            return 1;
+    }
     LogInfo("Iterations: %1", iterations);
 
     // We need to call InitWindow before using anything else (such as fonts)
@@ -34,13 +31,13 @@ int Main(int argc, char **argv)
     SetWindowState(FLAG_WINDOW_HIDDEN);
     InitWindow(640, 480, "Raylib Test");
 
-    Image img = GenImageColor(800, 600, (Color){ .r = 0, .g = 0, .b = 0, .a = 255 });
+    Image img = GenImageColor(800, 600, Color { .r = 0, .g = 0, .b = 0, .a = 255 });
     Font font = GetFontDefault();
 
     int64_t start = GetMonotonicTime();
 
     for (int i = 0; i < iterations; i++) {
-        ImageClearBackground(&img, (Color){ .r = 0, .g = 0, .b = 0, .a = 255 });
+        ImageClearBackground(&img, Color { .r = 0, .g = 0, .b = 0, .a = 255 });
 
         for (int j = 0; j < 3600; j++) {
             const char *text = "Hello World!";

package/benchmark/raylib_koffi.js

@@ -13,7 +13,7 @@
 // You should have received a copy of the GNU Affero General Public License
 // along with this program. If not, see https://www.gnu.org/licenses/.
 
-const koffi = require('../build/koffi.node');
+const koffi = require('./build/koffi.node');
 const path = require('path');
 
 const Color = koffi.struct('Color', {
@@ -71,17 +71,18 @@ const Font = koffi.struct('Font', {
 main();
 
 function main() {
-    if (process.argv.length < 3)
-        throw new Error('Missing number of iterations');
-
-    let iterations = parseInt(process.argv[2], 10);
-    if (Number.isNaN(iterations))
-        throw new Error('Not a valid number');
-    if (iterations < 1)
-        throw new Error('Value must be positive');
+    let iterations = 100;
+
+    if (process.argv.length >= 3) {
+        iterations = parseInt(process.argv[2], 10);
+        if (Number.isNaN(iterations))
+            throw new Error('Not a valid number');
+        if (iterations < 1)
+            throw new Error('Value must be positive');
+    }
     console.log('Iterations:', iterations);
 
-    let lib_filename = path.dirname(__filename) + '/../test/build/raylib' + koffi.extension;
+    let lib_filename = path.dirname(__filename) + '/build/raylib' + koffi.extension;
     let lib = koffi.load(lib_filename);
 
     const InitWindow = lib.cdecl('InitWindow', 'void', ['int', 'int', 'string']);

package/benchmark/raylib_node_ffi.js

@@ -16,7 +16,7 @@
 const ref = require('ref-napi');
 const ffi = require('ffi-napi');
 const struct = require('ref-struct-di')(ref);
-const koffi = require('../build/koffi.node');
+const koffi = require('./build/koffi.node');
 const path = require('path');
 
 const Color = struct({
@@ -87,17 +87,18 @@ const Font = struct({
 main();
 
 function main() {
-    if (process.argv.length < 3)
-        throw new Error('Missing number of iterations');
-
-    let iterations = parseInt(process.argv[2], 10);
-    if (Number.isNaN(iterations))
-        throw new Error('Not a valid number');
-    if (iterations < 1)
-        throw new Error('Value must be positive');
+    let iterations = 100;
+
+    if (process.argv.length >= 3) {
+        iterations = parseInt(process.argv[2], 10);
+        if (Number.isNaN(iterations))
+            throw new Error('Not a valid number');
+        if (iterations < 1)
+            throw new Error('Value must be positive');
+    }
     console.log('Iterations:', iterations);
 
-    let lib_filename = path.dirname(__filename) + '/../test/build/raylib' + koffi.extension;
+    let lib_filename = path.dirname(__filename) + '/build/raylib' + koffi.extension;
 
     const r = ffi.Library(lib_filename, {
         InitWindow: ['void', ['int', 'int', 'string']],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "koffi",
-  "version": "0.9.25",
+  "version": "0.9.28",
   "description": "Fast and simple FFI (foreign function interface) for Node.js",
   "keywords": [
     "foreign",
@@ -23,7 +23,7 @@
   },
   "license": "AGPL-3.0",
   "dependencies": {
-    "cnoke": "^0.9.26"
+    "cnoke": "^1.0.1"
   },
   "devDependencies": {
     "chalk": "^4.1.2",
@@ -35,11 +35,13 @@
   },
   "files": [
     "src",
-    "benchmark",
-    "test/files",
+    "benchmark/CMakeLists.txt",
+    "benchmark/atoi_*",
+    "benchmark/raylib_*",
+    "test/CMakeLists.txt",
+    "test/test.js",
     "test/registry",
     "test/tests",
-    "test/test.js",
     "vendor",
     "LICENSE.txt",
     "README.md",

package/src/call_arm32.cc

@@ -233,10 +233,10 @@ static Napi::Value TranslateCall(const Napi::CallbackInfo &info)
     // Return through registers unless it's too big
     if (RG_UNLIKELY(!call.AllocStack(func->args_size, 16, &args_ptr)))
         return env.Null();
-    if (RG_UNLIKELY(!call.AllocStack(4 * 4, 8, &gpr_ptr)))
-        return env.Null();
     if (RG_UNLIKELY(!call.AllocStack(8 * 8, 8, &vec_ptr)))
         return env.Null();
+    if (RG_UNLIKELY(!call.AllocStack(4 * 4, 8, &gpr_ptr)))
+        return env.Null();
     if (func->ret.use_memory) {
         if (RG_UNLIKELY(!call.AllocHeap(func->ret.type->size, 16, &return_ptr)))
             return env.Null();

package/src/call_arm32_fwd.S

@@ -56,22 +56,22 @@
 
 // Prepare general purpose argument registers from array passed by caller.
 .macro forward_int
-    ldr r3, [r1, 76]
-    ldr r2, [r1, 72]
-    ldr r0, [r1, 64]
-    ldr r1, [r1, 68]
+    ldr r3, [r1, 12]
+    ldr r2, [r1, 8]
+    ldr r0, [r1, 0]
+    ldr r1, [r1, 4]
 .endm
 
 // Prepare vector argument registers from array passed by caller.
 .macro forward_vec
-    vldr d7, [r1, 56]
-    vldr d6, [r1, 48]
-    vldr d5, [r1, 40]
-    vldr d4, [r1, 32]
-    vldr d3, [r1, 24]
-    vldr d2, [r1, 16]
-    vldr d1, [r1, 8]
-    vldr d0, [r1, 0]
+    vldr d7, [r1, 72]
+    vldr d6, [r1, 64]
+    vldr d5, [r1, 56]
+    vldr d4, [r1, 48]
+    vldr d3, [r1, 40]
+    vldr d2, [r1, 32]
+    vldr d1, [r1, 24]
+    vldr d0, [r1, 16]
 .endm
 
 ForwardCallGG:

package/src/ffi.cc

@@ -116,7 +116,7 @@ static Napi::Value CreateStructType(const Napi::CallbackInfo &info)
         if (!member.type)
             return env.Null();
 
-        type->size = AlignLen(type->size, member.type->align) + member.type->size;
+        type->size = (int16_t)(AlignLen(type->size, member.type->align) + member.type->size);
         type->align = std::max(type->align, member.type->align);
 
         type->members.Append(member);

package/src/util.cc

@@ -81,23 +81,38 @@ const char *CallData::PushString(const Napi::Value &value)
 {
     RG_ASSERT(value.IsString());
 
+    const Size SmallSize = 32;
+
     Napi::Env env = value.Env();
     napi_status status;
 
+    Span<char> buf;
     size_t len = 0;
-    status = napi_get_value_string_utf8(env, value, nullptr, 0, &len);
-    RG_ASSERT(status == napi_ok);
 
-    Span<char> buf;
-    buf.len = (Size)len + 1;
+    // Optimize for small strings
+    buf.len = SmallSize;
     if (RG_UNLIKELY(!AllocHeap(buf.len, 1, &buf.ptr)))
         return nullptr;
-
     if (RG_UNLIKELY(napi_get_value_string_utf8(env, value, buf.ptr, (size_t)buf.len, &len) != napi_ok)) {
         ThrowError<Napi::Error>(env, "Failed to convert string to UTF-8");
         return nullptr;
     }
 
+    // Slow path for bigger strings
+    if (len >= SmallSize - 1) {
+        status = napi_get_value_string_utf8(env, value, nullptr, 0, &len);
+        RG_ASSERT(status == napi_ok);
+        RG_ASSERT(len >= SmallSize);
+
+        buf.len = (Size)len + 1;
+        if (RG_UNLIKELY(!AllocHeap(buf.len - SmallSize, 1)))
+            return nullptr;
+        if (RG_UNLIKELY(napi_get_value_string_utf8(env, value, buf.ptr, (size_t)buf.len, &len) != napi_ok)) {
+            ThrowError<Napi::Error>(env, "Failed to convert string to UTF-8");
+            return nullptr;
+        }
+    }
+
     return buf.ptr;
 }
 

package/src/util.hh

@@ -101,8 +101,8 @@ public:
 
     template <typename T = void>
     bool AllocStack(Size size, Size align, T **out_ptr = nullptr);
-    template <typename T>
-    bool AllocHeap(Size size, Size align, T **out_ptr);
+    template <typename T = void>
+    bool AllocHeap(Size size, Size align, T **out_ptr = nullptr);
 
     const char *PushString(const Napi::Value &value);
     bool PushObject(const Napi::Object &obj, const TypeInfo *type, uint8_t *dest);
@@ -151,7 +151,9 @@ bool CallData::AllocHeap(Size size, Size align, T **out_ptr)
     heap_mem->ptr += delta;
     heap_mem->len -= delta;
 
-    *out_ptr = (T *)ptr;
+    if (out_ptr) {
+        *out_ptr = (T *)ptr;
+    }
     return true;
 }