dx12 testing

src/config.h

@@ -80,7 +80,7 @@
 
 
 
-#define DX12_TEST 0
+#define DX12_TEST 1
 
 
 

src/gpu.h

@@ -15,7 +15,15 @@ struct gpu_startup_receipt gpu_startup(struct sys_window *window);
  * ========================== */
 
 struct gpu_handle {
-    u64 v;
+    union {
+        /* dx11 style */
+        u64 v;
+        /* dx12 style */
+        struct {
+            u64 gen;
+            u64 idx;
+        };
+    };
 };
 
 void gpu_release(struct gpu_handle handle);

src/gpu_dx11.c

@@ -338,12 +338,12 @@ struct gpu_startup_receipt gpu_startup(struct sys_window *window)
 
     /* DXGI Debug break */
     {
-        IDXGIInfoQueue *dxgiInfo;
+        IDXGIInfoQueue *dxgi_info;
-        hr = DXGIGetDebugInterface1(0, &IID_IDXGIInfoQueue, (void **)&dxgiInfo);
+        hr = DXGIGetDebugInterface1(0, &IID_IDXGIInfoQueue, (void **)&dxgi_info);
         ASSERT(SUCCEEDED(hr));
-        IDXGIInfoQueue_SetBreakOnSeverity(dxgiInfo, DXGI_DEBUG_ALL, DXGI_INFO_QUEUE_MESSAGE_SEVERITY_CORRUPTION, TRUE);
+        IDXGIInfoQueue_SetBreakOnSeverity(dxgi_info, DXGI_DEBUG_ALL, DXGI_INFO_QUEUE_MESSAGE_SEVERITY_CORRUPTION, TRUE);
-        IDXGIInfoQueue_SetBreakOnSeverity(dxgiInfo, DXGI_DEBUG_ALL, DXGI_INFO_QUEUE_MESSAGE_SEVERITY_ERROR, TRUE);
+        IDXGIInfoQueue_SetBreakOnSeverity(dxgi_info, DXGI_DEBUG_ALL, DXGI_INFO_QUEUE_MESSAGE_SEVERITY_ERROR, TRUE);
-        IDXGIInfoQueue_Release(dxgiInfo);
+        IDXGIInfoQueue_Release(dxgi_info);
     }
 #endif
 
@@ -388,6 +388,7 @@ struct gpu_startup_receipt gpu_startup(struct sys_window *window)
         /* Disable Alt+Enter changing monitor resolution to match window size */
         IDXGIFactory_MakeWindowAssociation(factory, hwnd, DXGI_MWA_NO_ALT_ENTER);
 
+        IDXGISwapChain1_Release(swapchain1);
         IDXGIFactory2_Release(factory);
         IDXGIAdapter_Release(dxgiAdapter);
         IDXGIDevice_Release(dxgiDevice);
@@ -2066,7 +2067,7 @@ void gpu_swap_backbuffer(i32 vsync)
     gpu_capture_image_for_profiler();
     {
         __profscope(Present);
-        IDXGISwapChain1_Present(G.swapchain, vsync, flags);
+        IDXGISwapChain2_Present(G.swapchain, vsync, flags);
         __prof_dx11_collect(G.profiling_ctx);
         __profframe(0);
     }

src/gpu_dx12.c

@@ -2,13 +2,73 @@
 
 #include "gpu.h"
 #include "sys.h"
+#include "arena.h"
+#include "memory.h"
+#include "string.h"
+#include "scratch.h"
+
+#pragma warning(push, 0)
+# define UNICODE
+# define COBJMACROS
+# include <Windows.h>
+# include <d3d12.h>
+# include <dxgidebug.h>
+# include <dxgi1_6.h>
+# include <combaseapi.h>
+#pragma warning(pop)
+
+#pragma comment(lib, "d3d12")
+#pragma comment(lib, "dxgi")
+#pragma comment(lib, "dxguid")
+
+#define DX12_WAIT_FRAME_LATENCY 1
+#define DX12_ALLOW_TEARING 1
+
+#define DX12_SWAPCHAIN_BUFFER_COUNT     (3)
+#define DX12_SWAPCHAIN_FLAGS            ((DX12_ALLOW_TEARING * DXGI_SWAP_CHAIN_FLAG_ALLOW_TEARING) | (DX12_WAIT_FRAME_LATENCY * DXGI_SWAP_CHAIN_FLAG_FRAME_LATENCY_WAITABLE_OBJECT))
+#define DX12_SWAPCHAIN_FORMAT           (DXGI_FORMAT_R8G8B8A8_UNORM)
+//#define DX12_SWAPCHAIN_RTV_FORMAT  (DXGI_FORMAT_R8G8B8A8_UNORM_SRGB)
+
+#if RTC
+# define DX12_DEBUG 1
+# define DX12_SHADER_DEBUG 1
+#else
+# define DX12_DEBUG 0
+# define DX12_SHADER_DEBUG 0
+#endif
+
+enum dx12_handle_kind {
+    DX12_HANDLE_KIND_NONE,
+    DX12_HANDLE_KIND_TEXTURE,
+
+    NUM_DX12_HANDLE_KINDS
+};
+
+struct dx12_handle_entry {
+    enum dx12_handle_kind kind;
+    u64 gen;
+    u64 idx;
+    void *data;
+
+    struct dx12_handle_entry *next_free;
+};
+
+struct dx12_texture {
+    i32 _;
+};
 
 /* ========================== *
  * Global state
  * ========================== */
 
 GLOBAL struct {
-    i32 _;
+    /* Handles pool */
+    struct sys_mutex handle_entries_mutex;
+    struct arena handle_entries_arena;
+    struct dx12_handle_entry *first_free_handle_entry;
+    u64 num_handle_entries_reserved;
+
+    /* Swapchain */
 } G = ZI, DEBUG_ALIAS(G, G_gpu_dx12);
 
 /* ========================== *
@@ -17,7 +77,163 @@ GLOBAL struct {
 
 struct gpu_startup_receipt gpu_startup(struct sys_window *window)
 {
-    (UNUSED)window;
+    struct temp_arena scratch = scratch_begin_no_conflict();
+    HRESULT hr;
+
+    /* Initialize handles pool */
+    G.handle_entries_mutex = sys_mutex_alloc();
+    G.handle_entries_arena = arena_alloc(GIGABYTE(64));
+
+    /* Create debug controller */
+    u32 dxgi_factory_flags = 0;
+#if DX12_DEBUG
+    ID3D12Debug *debug_controller = NULL;
+    {
+        hr = D3D12GetDebugInterface(&IID_ID3D12Debug, (void **)&debug_controller);
+        if (!SUCCEEDED(hr)) {
+            sys_panic(LIT("Failed to create D3D12 debug controller"));
+        }
+        ID3D12Debug_EnableDebugLayer(debug_controller);
+        ID3D12Debug_Release(debug_controller);
+        dxgi_factory_flags |= DXGI_CREATE_FACTORY_DEBUG;
+    }
+#endif
+
+    /* Create factory */
+    IDXGIFactory6 *factory = NULL;
+    hr = CreateDXGIFactory2(dxgi_factory_flags, &IID_IDXGIFactory6, (void **)&factory);
+    if (!SUCCEEDED(hr)) {
+        sys_panic(LIT("Failed to initialize DXGI factory"));
+    }
+
+    /* Create device */
+    ID3D12Device *device = NULL;
+    {
+        struct string error = LIT("Could not initialize GPU device.");
+        struct string first_gpu_name = ZI;
+        u32 adapter_index = 0;
+        while (true) {
+            IDXGIAdapter1 *adapter = NULL;
+            hr = IDXGIFactory6_EnumAdapterByGpuPreference(factory, adapter_index, DXGI_GPU_PREFERENCE_HIGH_PERFORMANCE, &IID_IDXGIAdapter1, (void **)&adapter);
+            if (SUCCEEDED(hr)) {
+                DXGI_ADAPTER_DESC1 desc;
+                IDXGIAdapter1_GetDesc1(adapter, &desc);
+                if (!(desc.Flags & DXGI_ADAPTER_FLAG_SOFTWARE)) {
+                    if (first_gpu_name.len == 0) {
+                        first_gpu_name = string_from_wstr_no_limit(scratch.arena, desc.Description);
+                    }
+                    /* Using feature level 12_0 instead of 12_2 because mesh feature check happens separately
+                     * (because 1600 series cards support mesh shaders but not the rest of 12_2) */
+                    hr = D3D12CreateDevice((IUnknown *)adapter, D3D_FEATURE_LEVEL_12_0, &IID_ID3D12Device, (void **)&device);
+                    if (SUCCEEDED(hr)) {
+                        D3D12_FEATURE_DATA_D3D12_OPTIONS7 features = ZI;
+                        hr = ID3D12Device_CheckFeatureSupport(device, D3D12_FEATURE_D3D12_OPTIONS7, &features, sizeof(features));
+                        if (SUCCEEDED(hr) && features.MeshShaderTier >= D3D12_MESH_SHADER_TIER_1) {
+                            IDXGIAdapter1_Release(adapter);
+                            adapter = NULL;
+                            break;
+                        }
+                    }
+                }
+                ID3D12Device_Release(device);
+                IDXGIAdapter1_Release(adapter);
+                adapter = NULL;
+                device = NULL;
+                ++adapter_index;
+            } else {
+                break;
+            }
+        }
+        /* TODO: Fall back to compute shaders if mesh shaders aren't supported */
+        if (!device) {
+            if (first_gpu_name.len > 0) {
+                struct string fmt = LIT("Device '%F' does not support DirectX 12_2. Ensure that your drivers are up to date.\n\n"
+                                        "Note that older GPUs such as those up to and including the GTX 1000 series, the RX 5000 series, and the Intel Xe-LP series do not support DirectX 12_2 features.");
+                error = string_format(scratch.arena, fmt, FMT_STR(first_gpu_name));
+            }
+            sys_panic(error);
+        }
+    }
+
+#if DX12_DEBUG
+    /* D3D12 Debug break */
+    {
+        ID3D12InfoQueue *info = NULL;
+        hr = ID3D12Device_QueryInterface(device, &IID_ID3D12InfoQueue, (void **)&info);
+        if (!SUCCEEDED(hr)) {
+            sys_panic(LIT("Failed to query ID3D12Device interface"));
+        }
+        ID3D12InfoQueue_SetBreakOnSeverity(info, D3D12_MESSAGE_SEVERITY_CORRUPTION, TRUE);
+        ID3D12InfoQueue_SetBreakOnSeverity(info, D3D12_MESSAGE_SEVERITY_ERROR, TRUE);
+        ID3D12InfoQueue_Release(info);
+    }
+
+    /* DXGI Debug break */
+    {
+        IDXGIInfoQueue *dxgi_info = NULL;
+        hr = DXGIGetDebugInterface1(0, &IID_IDXGIInfoQueue, (void **)&dxgi_info);
+        if (!SUCCEEDED(hr)) {
+            sys_panic(LIT("Failed to get DXGI debug interface"));
+        }
+        IDXGIInfoQueue_SetBreakOnSeverity(dxgi_info, DXGI_DEBUG_ALL, DXGI_INFO_QUEUE_MESSAGE_SEVERITY_CORRUPTION, TRUE);
+        IDXGIInfoQueue_SetBreakOnSeverity(dxgi_info, DXGI_DEBUG_ALL, DXGI_INFO_QUEUE_MESSAGE_SEVERITY_ERROR, TRUE);
+        IDXGIInfoQueue_Release(dxgi_info);
+    }
+#endif
+
+    /* Create command queue */
+    ID3D12CommandQueue *cq = NULL;
+    {
+        D3D12_COMMAND_QUEUE_DESC desc = ZI;
+        desc.Flags = D3D12_COMMAND_QUEUE_FLAG_NONE;
+        desc.Type = D3D12_COMMAND_LIST_TYPE_DIRECT;
+
+        hr = ID3D12Device_CreateCommandQueue(device, &desc, &IID_ID3D12CommandQueue, (void **)&cq);
+        if (!SUCCEEDED(hr)) {
+            sys_panic(LIT("Failed to create command queue"));
+        }
+    }
+
+    /* Create swapchain */
+    IDXGISwapChain3 *swapchain = NULL;
+    u32 swapchain_frame_index = 0;
+    {
+        HWND hwnd = (HWND)sys_window_get_internal_handle(window);
+        DXGI_SWAP_CHAIN_DESC1 desc = {
+            .Format = DX12_SWAPCHAIN_FORMAT,
+            .SampleDesc = { 1, 0 },
+            .BufferUsage = DXGI_USAGE_SHADER_INPUT | DXGI_USAGE_RENDER_TARGET_OUTPUT,
+            .BufferCount = 3,
+            .Scaling = DXGI_SCALING_NONE,
+            .Flags = DX12_SWAPCHAIN_FLAGS,
+            .AlphaMode = DXGI_ALPHA_MODE_IGNORE,
+            .SwapEffect = DXGI_SWAP_EFFECT_FLIP_DISCARD
+        };
+
+        /* Create swapchain1 */
+        IDXGISwapChain1 *swapchain1 = NULL;
+        hr = IDXGIFactory2_CreateSwapChainForHwnd(factory, (IUnknown *)cq, hwnd, &desc, NULL, NULL, &swapchain1);
+        if (!SUCCEEDED(hr)) {
+            sys_panic(LIT("Failed to create IDXGISwapChain1"));
+        }
+
+        /* Upgrade to swapchain3 */
+        hr = IDXGISwapChain1_QueryInterface(swapchain1, &IID_IDXGISwapChain3, (void **)&swapchain);
+        if (!SUCCEEDED(hr)) {
+            sys_panic(LIT("Failed to create IDXGISwapChain3"));
+        }
+
+        /* Disable Alt+Enter changing monitor resolution to match window size */
+        IDXGIFactory_MakeWindowAssociation(factory, hwnd, DXGI_MWA_NO_ALT_ENTER);
+
+        swapchain_frame_index = IDXGISwapChain3_GetCurrentBackBufferIndex(swapchain);
+        (UNUSED)swapchain_frame_index;
+
+        IDXGISwapChain1_Release(swapchain1);
+    }
+
+    IDXGIFactory6_Release(factory);
+    scratch_end(scratch);
     struct gpu_startup_receipt res = ZI;
     return res;
 }
@@ -26,23 +242,104 @@ struct gpu_startup_receipt gpu_startup(struct sys_window *window)
  * Handle
  * ========================== */
 
+INTERNAL void dx12_texture_release(struct dx12_texture *t);
+
+INTERNAL struct gpu_handle handle_alloc(enum dx12_handle_kind kind, void *data)
+{
+    u64 old_gen = 0;
+    u64 idx = 0;
+    struct dx12_handle_entry *entry = NULL;
+    {
+        struct sys_lock lock = sys_mutex_lock_e(&G.handle_entries_mutex);
+        if (G.first_free_handle_entry) {
+            entry = G.first_free_handle_entry;
+            G.first_free_handle_entry = entry->next_free;
+            old_gen = entry->gen;
+            idx = entry->idx;
+        } else {
+            entry = arena_push_no_zero(&G.handle_entries_arena, struct dx12_handle_entry);
+            idx = G.num_handle_entries_reserved++;
+        }
+        sys_mutex_unlock(&lock);
+    }
+    MEMZERO_STRUCT(entry);
+    entry->kind = kind;
+    entry->gen = old_gen + 1;
+    entry->idx = idx;
+    entry->data = data;
+
+    struct gpu_handle res = ZI;
+    res.gen = entry->gen;
+    res.idx = entry->idx;
+    return res;
+}
+
+INTERNAL struct dx12_handle_entry *handle_get_entry(struct gpu_handle handle, struct sys_lock *lock)
+{
+    sys_assert_locked_e_or_s(lock, &G.handle_entries_mutex);
+    struct dx12_handle_entry *res = NULL;
+    if (handle.idx > 0 && handle.idx < G.num_handle_entries_reserved) {
+        struct dx12_handle_entry *tmp = &((struct dx12_handle_entry *)G.handle_entries_arena.base)[handle.idx];
+        if (tmp->gen == handle.gen) {
+            res = tmp;
+        }
+    }
+    return res;
+}
+
+/* TODO: The GPU api should ensure that resources freed by the caller will not cause issues on the GPU (via fencing),
+ * however the caller is responsible for managing resource lifetimes on the CPU side (e.g. using sprites w/ sprite scopes
+ * to ensure freed textures aren't being used in pending command lists. */
 void gpu_release(struct gpu_handle handle)
 {
-    (UNUSED)handle;
+    enum dx12_handle_kind kind = NULL;
+    void *data = NULL;
+
+    /* Release handle entry */
+    struct sys_lock lock = sys_mutex_lock_e(&G.handle_entries_mutex);
+    {
+        struct dx12_handle_entry *entry = handle_get_entry(handle, &lock);
+        if (entry) {
+            kind = entry->kind;
+            data = entry->data;
+        }
+        ++entry->gen;
+        entry->next_free = G.first_free_handle_entry;
+        G.first_free_handle_entry = entry;
+    }
+    sys_mutex_unlock(&lock);
+
+    /* Release data */
+    if (data) {
+        switch (kind) {
+            default: break;
+
+            case DX12_HANDLE_KIND_TEXTURE:
+            {
+                dx12_texture_release(data);
+            } break;
+        }
+    }
 }
 
 /* ========================== *
  * Texture
  * ========================== */
 
+INTERNAL void dx12_texture_release(struct dx12_texture *t)
+{
+    (UNUSED)t;
+}
+
 struct gpu_handle gpu_texture_alloc(enum gpu_texture_format format, u32 flags, struct v2i32 size, void *initial_data)
 {
     (UNUSED)format;
     (UNUSED)flags;
     (UNUSED)size;
     (UNUSED)initial_data;
-    struct gpu_handle res = ZI;
+    struct dx12_texture *t = NULL;
-    return res;
+
+    return handle_alloc(DX12_HANDLE_KIND_TEXTURE, t);
 }
 
 void gpu_texture_clear(struct gpu_handle target_texture, u32 clear_color)