getchar on non-terminal exit

.github/workflows/CI.yml

@@ -25,6 +25,18 @@ jobs:
         cmake -A x64 ..
         cmake --build . --config Release -j 4
 
+  windows-arm:
+    runs-on: windows-11-arm
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        submodules: 'recursive'
+    - name: build
+      run: |
+        mkdir build; cd build
+        cmake -A arm64 ..
+        cmake --build . --config Release -j 4
+
   ubuntu:
     runs-on: ubuntu-latest
     steps:

.github/workflows/release.yml

@@ -120,8 +120,36 @@ jobs:
         name: ${{ env.PACKAGENAME }}
         path: ${{ env.PACKAGENAME }}.zip
 
+  windows-arm:
+    needs: [setup]
+    runs-on: windows-11-arm
+    env:
+      UseMultiToolTask: true
+      PACKAGENAME: ${{ needs.setup.outputs.APPNAME }}-${{ needs.setup.outputs.VERSION }}-windows-arm64
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        submodules: 'recursive'
+    - name: build
+      run: |
+        mkdir build; cd build
+        cmake -A arm64 ..
+        cmake --build . --config Release -j 4
+    - name: package
+      run: |
+        mkdir ${{ env.PACKAGENAME }}
+        Copy-Item -Verbose -Path "README.md" -Destination "${{ env.PACKAGENAME }}"
+        Copy-Item -Verbose -Path "LICENSE" -Destination "${{ env.PACKAGENAME }}"
+        Copy-Item -Verbose -Path "build\Release\${{ needs.setup.outputs.APPNAME }}.exe" -Destination "${{ env.PACKAGENAME }}"
+        7z a -r ${{ env.PACKAGENAME }}.zip ${{ env.PACKAGENAME }}
+    - name: upload
+      uses: actions/upload-artifact@v4
+      with:
+        name: ${{ env.PACKAGENAME }}
+        path: ${{ env.PACKAGENAME }}.zip
+
   release:
-    needs: [setup, ubuntu, macos, windows]
+    needs: [setup, ubuntu, macos, windows, windows-arm]
     runs-on: ubuntu-latest
     steps:
     - name: download

vkpeak.cpp

@@ -1,5 +1,16 @@
 // vkpeak implemented with ncnn library
 
+#include <stdio.h>
+#include <stdlib.h>
+
+#ifdef _WIN32
+#include <conio.h>
+#include <io.h>
+#include <windows.h>
+#include <tlhelp32.h>
+#include <psapi.h>
+#endif
+
 #include <benchmark.h>
 #include <command.h>
 #include <gpu.h>
@@ -2320,9 +2331,51 @@ static double vkpeak(int device_id, int storage_type, int arithmetic_type, int p
     return max_gflops;
 }
 
+#ifdef _WIN32
+static int is_running_in_console()
+{
+    if (!_isatty(_fileno(stdin)))
+        return 0;
+
+    HWND consoleWindow = GetConsoleWindow();
+    if (consoleWindow == NULL)
+        return 0;
+
+    DWORD currentProcessId = GetCurrentProcessId();
+    DWORD parentProcessId = 0;
+
+    HANDLE hSnapshot = CreateToolhelp32Snapshot(TH32CS_SNAPPROCESS, 0);
+    if (hSnapshot == INVALID_HANDLE_VALUE)
+        return 0;
+
+    PROCESSENTRY32 pe32;
+    pe32.dwSize = sizeof(PROCESSENTRY32);
+
+    // Find parent process ID
+    if (Process32First(hSnapshot, &pe32)) {
+        do {
+            if (pe32.th32ProcessID == currentProcessId) {
+                parentProcessId = pe32.th32ParentProcessID;
+                break;
+            }
+        } while (Process32Next(hSnapshot, &pe32));
+    }
+    CloseHandle(hSnapshot);
+
+    if (parentProcessId == 0)
+        return 0;
+
+    HANDLE hParent = OpenProcess(PROCESS_QUERY_INFORMATION | PROCESS_VM_READ, FALSE, parentProcessId);
+    if (hParent == NULL)
+        return 0;
+
+    return 1;
+}
+#endif // _WIN32
+
 int main(int argc, char** argv)
 {
-    if (argc != 2)
+    if (argc != 1 && argc != 2)
     {
         fprintf(stderr, "Usage: %s [device_id]\n", argv[0]);
         return -1;
@@ -2337,7 +2390,7 @@ int main(int argc, char** argv)
         return -1;
     }
 
-    const int device_id = atoi(argv[1]);
+    const int device_id = argc == 2 ? atoi(argv[1]) : ncnn::get_default_gpu_index();
     if (device_id < 0 || device_id >= gpu_count)
     {
         fprintf(stderr, "No vulkan device for %d\n", device_id);
@@ -2351,6 +2404,18 @@ int main(int argc, char** argv)
         return -1;
     }
 
+    if (argc == 1)
+    {
+        fprintf(stderr, "Available devices:\n");
+
+        for (int i = 0; i < gpu_count; i++)
+        {
+            fprintf(stderr, "%d = %s\n", i, ncnn::get_gpu_info(i).device_name());
+        }
+
+        fprintf(stderr, "Device %d will be used\n\n", device_id);
+    }
+
     fprintf(stderr, "device       = %s\n", ncnn::get_gpu_info(device_id).device_name());
 
     // device_id        = 0
@@ -2389,5 +2454,33 @@ int main(int argc, char** argv)
 
     ncnn::destroy_gpu_instance();
 
+#ifdef _WIN32
+    if (!is_running_in_console())
+    {
+        fprintf(stderr, "\nPress any key to continue...\n");
+
+        HANDLE hStdin = GetStdHandle(STD_INPUT_HANDLE);
+
+        // mltiple flush strategy
+        FlushConsoleInputBuffer(hStdin);
+        while (_kbhit()) _getch();
+        fflush(stdin);
+
+        // delay to ensure all input is processed
+        Sleep(100);
+
+        // flush again
+        FlushConsoleInputBuffer(hStdin);
+        while (_kbhit()) _getch();
+
+        // use low-level API to ensure waiting for new input
+        INPUT_RECORD record;
+        DWORD read;
+        do {
+            ReadConsoleInput(hStdin, &record, 1, &read);
+        } while (record.EventType != KEY_EVENT || !record.Event.KeyEvent.bKeyDown);
+    }
+#endif // _WIN32
+
     return 0;
 }