working host-memory gpu resources

src/config.h

@@ -70,7 +70,7 @@
 #define FLOOD_DEBUG 0
 
 #define GPU_DEBUG 1
-#define GPU_DEBUG_VALIDATION 0
+#define GPU_DEBUG_VALIDATION 1
 
 /* If enabled, bitbuffs will insert/verify magic numbers & length for each read & write */
 #define BITBUFF_DEBUG 0

src/gpu/gpu_core.h

@@ -298,6 +298,8 @@ Enum(G_ResourceFlag)
     G_ResourceFlag_AllowShaderReadWrite    = (1 << 0),
     G_ResourceFlag_AllowRenderTarget       = (1 << 1),
     G_ResourceFlag_AllowDepthStencil       = (1 << 2),
+    G_ResourceFlag_HostMemory              = (1 << 3),
+    G_ResourceFlag_WriteCombineHostMemory  = (1 << 4),
 };
 
 ////////////////////////////////////////////////////////////
@@ -568,6 +570,11 @@ i32     G_CountDepth(G_ResourceHandle texture);
 
 #define G_CountBuffer(buffer, type) G_CountBufferBytes(buffer) / sizeof(type)
 
+//- Map
+
+void *G_HostPointerFromResource(G_ResourceHandle resource);
+#define G_StructFromResource(resource, type) (type *)G_HostPointerFromResource(resource)
+
 ////////////////////////////////////////////////////////////
 //~ @hookdecl Shader resource reference
 

src/gpu/gpu_dx12/gpu_dx12_core.c

@@ -300,7 +300,12 @@ void G_Bootstrap(void)
                 if (kind != G_QueueKind_AsyncCopy)
                 {
                     G_ArenaHandle gpu_perm = G_PermArena();
-                    queue->debug_print_buffer = G_PushBuffer(gpu_perm, u8, print_buffer_size, .flags = G_ResourceFlag_AllowShaderReadWrite);
+                    queue->debug_print_buffer = G_PushBuffer(
+                        gpu_perm,
+                        u8,
+                        print_buffer_size,
+                        .flags = G_ResourceFlag_AllowShaderReadWrite | G_ResourceFlag_HostMemory
+                    );
                     queue->debug_print_buffer_ref = G_PushRWByteAddressBufferRef(gpu_perm, queue->debug_print_buffer);
                 }
             }
@@ -718,6 +723,111 @@ void G_D12_CommitRawCommandList(G_D12_RawCommandList *cl)
     }
 }
 
+////////////////////////////////////////////////////////////
+//~ Arena
+
+G_D12_ResourceHeap *G_D12_ResourceHeapFromArena(G_D12_Arena *gpu_arena, G_D12_ResourceHeapKind kind)
+{
+    G_D12_SharedState *g = &G_D12_shared_state;
+    G_D12_ResourceHeap *heap = &gpu_arena->resource_heaps[kind];
+
+    b32 is_mappable = 0;
+
+    /* Initialize heap */
+    /* FIXME: Dynamic size */
+    if (heap->d3d_heap == 0)
+    {
+        HRESULT hr = 0;
+
+        /* Create d3d heap */
+        {
+            D3D12_HEAP_DESC d3d_desc = ZI;
+            d3d_desc.SizeInBytes = Mebi(512);
+            if (kind == G_D12_ResourceHeapKind_CpuWriteBack)
+            {
+                d3d_desc.Properties.Type = D3D12_HEAP_TYPE_CUSTOM;
+                d3d_desc.Properties.CPUPageProperty = D3D12_CPU_PAGE_PROPERTY_WRITE_BACK;
+                d3d_desc.Properties.MemoryPoolPreference = D3D12_MEMORY_POOL_L0;
+                is_mappable = 1;
+            }
+            else if (kind == G_D12_ResourceHeapKind_CpuWriteCombine)
+            {
+                d3d_desc.Properties.Type = D3D12_HEAP_TYPE_CUSTOM;
+                d3d_desc.Properties.CPUPageProperty = D3D12_CPU_PAGE_PROPERTY_WRITE_COMBINE;
+                d3d_desc.Properties.MemoryPoolPreference = D3D12_MEMORY_POOL_L0;
+                is_mappable = 1;
+            }
+            else
+            {
+                d3d_desc.Properties.Type = D3D12_HEAP_TYPE_DEFAULT;
+            }
+            d3d_desc.Flags |= D3D12_HEAP_FLAG_CREATE_NOT_ZEROED;
+            d3d_desc.Flags |= D3D12_HEAP_FLAG_ALLOW_ALL_BUFFERS_AND_TEXTURES;  /* TODO: Remove this and support tier 1 resource heaps */
+            hr = ID3D12Device_CreateHeap(g->device, &d3d_desc, &IID_ID3D12Heap, (void **)&heap->d3d_heap);
+            heap->size = d3d_desc.SizeInBytes;
+        }
+
+        /* Map resource */
+        if (is_mappable)
+        {
+            /* Create heap-sized resource for mapping */
+            if (SUCCEEDED(hr))
+            {
+                D3D12_RESOURCE_DESC1 d3d_desc  = ZI;
+                d3d_desc.Dimension             = D3D12_RESOURCE_DIMENSION_BUFFER;
+                d3d_desc.Layout                = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
+                d3d_desc.Format                = DXGI_FORMAT_UNKNOWN;
+                d3d_desc.Width                 = heap->size;
+                d3d_desc.Height                = 1;
+                d3d_desc.DepthOrArraySize      = 1;
+                d3d_desc.MipLevels             = 1;
+                d3d_desc.SampleDesc.Count      = 1;
+                d3d_desc.SampleDesc.Quality    = 0;
+                d3d_desc.Flags                |= D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE;
+
+                u64 alloc_size = 0;
+                u64 alloc_align = 0;
+                {
+                    D3D12_RESOURCE_ALLOCATION_INFO alloc_info = ZI;
+                    ID3D12Device_GetResourceAllocationInfo(g->device, &alloc_info, 0, 1, (D3D12_RESOURCE_DESC *)&d3d_desc);
+                    alloc_size = alloc_info.SizeInBytes;
+                    alloc_align = alloc_info.Alignment;
+                }
+
+                if (alloc_size > heap->size)
+                {
+                    Panic(Lit("Gpu heap overflow"));
+                }
+
+                hr = ID3D12Device10_CreatePlacedResource2(g->device,
+                                                          heap->d3d_heap,
+                                                          0,
+                                                          &d3d_desc,
+                                                          D3D12_BARRIER_LAYOUT_UNDEFINED,
+                                                          0,
+                                                          0,
+                                                          0,
+                                                          &IID_ID3D12Resource,
+                                                          (void **)&heap->d3d_mapped_resource);
+            }
+            if (SUCCEEDED(hr))
+            {
+                D3D12_RANGE read_range = ZI;
+                hr = ID3D12Resource_Map(heap->d3d_mapped_resource, 0, &read_range, &heap->mapped);
+            }
+        }
+
+        if (!SUCCEEDED(hr))
+        {
+            /* TODO: Don't panic */
+            Panic(Lit("Failed to create D3D12 resource heap"));
+        }
+
+    }
+
+    return heap;
+}
+
 ////////////////////////////////////////////////////////////
 //~ @hookimpl Arena
 
@@ -732,6 +842,11 @@ G_ArenaHandle G_AcquireArena(void)
     }
     gpu_arena->arena = AcquireArena(Gibi(1));
 
+    for (u64 heap_idx = 0; heap_idx < countof(gpu_arena->resource_heaps); ++heap_idx)
+    {
+        gpu_arena->resource_heaps[heap_idx].kind = (G_D12_ResourceHeapKind)heap_idx;
+    }
+
     return G_D12_MakeHandle(G_ArenaHandle, gpu_arena);
 }
 
@@ -741,7 +856,7 @@ void G_ReleaseArena(G_ArenaHandle arena)
 }
 
 ////////////////////////////////////////////////////////////
-//~ Resource helpers
+//~ Resource
 
 G_D12_Descriptor *G_D12_DescriptorFromIndex(G_D12_DescriptorHeapKind heap_kind, u32 index)
 {
@@ -826,30 +941,21 @@ G_ResourceHandle G_PushBufferEx(G_ArenaHandle arena_handle, G_BufferResourceDesc
     G_D12_SharedState *g = &G_D12_shared_state;
     G_D12_Arena *gpu_arena = G_D12_ArenaFromHandle(arena_handle);
 
-    /* Create resource heap */
+    /* Fetch heap */
-    if (!gpu_arena->d3d_resource_heap)
+    G_D12_ResourceHeapKind heap_kind = G_D12_ResourceHeapKind_Default;
+    if (desc.flags & G_ResourceFlag_HostMemory)
     {
-        /* FIXME: Dynamic size */
+        heap_kind = G_D12_ResourceHeapKind_CpuWriteBack;
-        D3D12_HEAP_DESC d3d_desc = ZI;
+        if (desc.flags & G_ResourceFlag_WriteCombineHostMemory)
-        d3d_desc.SizeInBytes = Mebi(512);
-        d3d_desc.Flags |= D3D12_HEAP_FLAG_CREATE_NOT_ZEROED;
-        d3d_desc.Flags |= D3D12_HEAP_FLAG_ALLOW_ALL_BUFFERS_AND_TEXTURES;  /* TODO: Remove this and support tier 1 resource heaps */
-        d3d_desc.Properties.Type = D3D12_HEAP_TYPE_DEFAULT;
-
-        ID3D12Heap *heap = 0;
-        HRESULT hr = ID3D12Device_CreateHeap(g->device, &d3d_desc, &IID_ID3D12Heap, (void **)&heap);
-        if (!SUCCEEDED(hr))
         {
-            /* TODO: Don't panic */
+            heap_kind = G_D12_ResourceHeapKind_CpuWriteCombine;
-            Panic(Lit("Failed to create D3D12 resource heap"));
         }
-
-        gpu_arena->d3d_resource_heap = heap;
-        gpu_arena->heap_size = d3d_desc.SizeInBytes;
     }
+    G_D12_ResourceHeap *heap = G_D12_ResourceHeapFromArena(gpu_arena, heap_kind);
 
     /* Create d3d resource */
     ID3D12Resource *d3d_resource = 0;
+    u64 pos_in_heap = 0;
     u64 aligned_size = AlignU64(MaxU64(desc.size, 1), 4);
     {
         D3D12_RESOURCE_DESC1 d3d_desc  = ZI;
@@ -873,18 +979,18 @@ G_ResourceHandle G_PushBufferEx(G_ArenaHandle arena_handle, G_BufferResourceDesc
             alloc_align = alloc_info.Alignment;
         }
 
-        u64 alloc_pos = gpu_arena->heap_pos;
+        pos_in_heap = heap->pos;
-        alloc_pos = AlignU64(alloc_pos, alloc_align);
+        pos_in_heap = AlignU64(pos_in_heap, alloc_align);
-        gpu_arena->heap_pos = alloc_pos + alloc_size;
+        heap->pos = pos_in_heap + alloc_size;
 
-        if (alloc_pos + alloc_size > gpu_arena->heap_size)
+        if (pos_in_heap + alloc_size > heap->size)
         {
             Panic(Lit("Gpu arena overflow"));
         }
 
         HRESULT hr = ID3D12Device10_CreatePlacedResource2(g->device,
-                                                          gpu_arena->d3d_resource_heap,
+                                                          heap->d3d_heap,
-                                                          alloc_pos,
+                                                          pos_in_heap,
                                                           &d3d_desc,
                                                           D3D12_BARRIER_LAYOUT_UNDEFINED,
                                                           0,
@@ -895,6 +1001,8 @@ G_ResourceHandle G_PushBufferEx(G_ArenaHandle arena_handle, G_BufferResourceDesc
     }
 
     G_D12_Resource *resource = PushStruct(gpu_arena->arena, G_D12_Resource);
+    resource->heap = heap;
+    resource->pos_in_heap = pos_in_heap;
     resource->d3d_resource = d3d_resource;
     resource->uid = Atomic64FetchAdd(&g->resource_creation_gen.v, 1) + 1;
     resource->flags = desc.flags;
@@ -912,30 +1020,21 @@ G_ResourceHandle G_PushTextureEx(G_ArenaHandle arena_handle, G_TextureResourceDe
     G_D12_Arena *gpu_arena = G_D12_ArenaFromHandle(arena_handle);
     D3D12_BARRIER_LAYOUT initial_layout = G_D12_BarrierLayoutFromLayout(desc.initial_layout);
 
-    /* Create resource heap */
+    /* Fetch heap */
-    if (!gpu_arena->d3d_resource_heap)
+    G_D12_ResourceHeapKind heap_kind = G_D12_ResourceHeapKind_Default;
+    if (desc.flags & G_ResourceFlag_HostMemory)
     {
-        /* FIXME: Dynamic size */
+        heap_kind = G_D12_ResourceHeapKind_CpuWriteBack;
-        D3D12_HEAP_DESC d3d_desc = ZI;
+        if (desc.flags & G_ResourceFlag_WriteCombineHostMemory)
-        d3d_desc.SizeInBytes = Mebi(512);
-        d3d_desc.Flags |= D3D12_HEAP_FLAG_CREATE_NOT_ZEROED;
-        d3d_desc.Flags |= D3D12_HEAP_FLAG_ALLOW_ALL_BUFFERS_AND_TEXTURES;  /* TODO: Remove this and support tier 1 resource heaps */
-        d3d_desc.Properties.Type = D3D12_HEAP_TYPE_DEFAULT;
-
-        ID3D12Heap *heap = 0;
-        HRESULT hr = ID3D12Device_CreateHeap(g->device, &d3d_desc, &IID_ID3D12Heap, (void **)&heap);
-        if (!SUCCEEDED(hr))
         {
-            /* TODO: Don't panic */
+            heap_kind = G_D12_ResourceHeapKind_CpuWriteCombine;
-            Panic(Lit("Failed to create D3D12 resource heap"));
         }
-
-        gpu_arena->d3d_resource_heap = heap;
-        gpu_arena->heap_size = d3d_desc.SizeInBytes;
     }
+    G_D12_ResourceHeap *heap = G_D12_ResourceHeapFromArena(gpu_arena, heap_kind);
 
     /* Create d3d resource */
     ID3D12Resource *d3d_resource = 0;
+    u64 pos_in_heap;
     {
         D3D12_RESOURCE_DESC1 d3d_desc = ZI;
         d3d_desc.Dimension = desc.kind == G_TextureKind_1D ? D3D12_RESOURCE_DIMENSION_TEXTURE1D :
@@ -961,11 +1060,11 @@ G_ResourceHandle G_PushTextureEx(G_ArenaHandle arena_handle, G_TextureResourceDe
             alloc_align = alloc_info.Alignment;
         }
 
-        u64 alloc_pos = gpu_arena->heap_pos;
+        pos_in_heap = heap->pos;
-        alloc_pos = AlignU64(alloc_pos, alloc_align);
+        pos_in_heap = AlignU64(pos_in_heap, alloc_align);
-        gpu_arena->heap_pos = alloc_pos + alloc_size;
+        heap->pos = pos_in_heap + alloc_size;
 
-        if (alloc_pos + alloc_size > gpu_arena->heap_size)
+        if (pos_in_heap + alloc_size > heap->size)
         {
             Panic(Lit("Gpu arena overflow"));
         }
@@ -979,8 +1078,8 @@ G_ResourceHandle G_PushTextureEx(G_ArenaHandle arena_handle, G_TextureResourceDe
         };
 
         HRESULT hr = ID3D12Device10_CreatePlacedResource2(g->device,
-                                                          gpu_arena->d3d_resource_heap,
+                                                          heap->d3d_heap,
-                                                          alloc_pos,
+                                                          pos_in_heap,
                                                           &d3d_desc,
                                                           initial_layout,
                                                           (d3d_desc.Flags & D3D12_RESOURCE_FLAG_ALLOW_RENDER_TARGET) ? &clear_value : 0,
@@ -991,6 +1090,8 @@ G_ResourceHandle G_PushTextureEx(G_ArenaHandle arena_handle, G_TextureResourceDe
     }
 
     G_D12_Resource *resource = PushStruct(gpu_arena->arena, G_D12_Resource);
+    resource->heap = heap;
+    resource->pos_in_heap = pos_in_heap;
     resource->d3d_resource = d3d_resource;
     resource->uid          = Atomic64FetchAdd(&g->resource_creation_gen.v, 1) + 1;
     resource->flags        = desc.flags;
@@ -1055,20 +1156,17 @@ u32 G_PushRef(G_ArenaHandle arena_handle, G_ResourceHandle resource_handle, G_Re
     G_D12_Descriptor *descriptor = 0;
     if (is_buffer)
     {
-        descriptor = G_D12_PushDescriptor(gpu_arena, G_D12_DescriptorHeapKind_CbvSrvUav);
-        u64 num_elements_in_buffer = 0;
-        u64 num_elements_after_offset = 0;
         if (is_raw)
         {
-            num_elements_in_buffer = 0;
+            ref_desc.element_size = 4;
-            num_elements_after_offset = 0;
+            ref_desc.element_offset /= 4;
         }
-        else
+
-        {
+        descriptor = G_D12_PushDescriptor(gpu_arena, G_D12_DescriptorHeapKind_CbvSrvUav);
         u64 buffer_size_aligned = resource->buffer_size_aligned;
-            num_elements_in_buffer = buffer_size_aligned / ref_desc.element_size;
+        u64 num_elements_in_buffer = buffer_size_aligned / ref_desc.element_size;
-            num_elements_after_offset = num_elements_in_buffer > ref_desc.element_offset ? num_elements_in_buffer - ref_desc.element_offset : 0;
+        u64 num_elements_after_offset = num_elements_in_buffer > ref_desc.element_offset ? num_elements_in_buffer - ref_desc.element_offset : 0;
-        }
+
         if (num_elements_after_offset > 0)
         {
             if (is_uav)
@@ -1076,7 +1174,7 @@ u32 G_PushRef(G_ArenaHandle arena_handle, G_ResourceHandle resource_handle, G_Re
                 D3D12_UNORDERED_ACCESS_VIEW_DESC desc = ZI;
                 {
                     desc.Format = DXGI_FORMAT_UNKNOWN;
-                    desc.ViewDimension = D3D12_UAV_DIMENSION_BUFFER ;
+                    desc.ViewDimension = D3D12_UAV_DIMENSION_BUFFER;
                     desc.Buffer.FirstElement = ref_desc.element_offset;
                     desc.Buffer.NumElements = num_elements_after_offset;
                     desc.Buffer.StructureByteStride = ref_desc.element_size;
@@ -1086,6 +1184,7 @@ u32 G_PushRef(G_ArenaHandle arena_handle, G_ResourceHandle resource_handle, G_Re
                 {
                     desc.Format = DXGI_FORMAT_R32_TYPELESS;
                     desc.Buffer.Flags = D3D12_BUFFER_UAV_FLAG_RAW;
+                    desc.Buffer.StructureByteStride = 0;
                 }
                 ID3D12Device_CreateUnorderedAccessView(g->device, resource->d3d_resource, 0, &desc, descriptor->handle);
             }
@@ -1105,6 +1204,7 @@ u32 G_PushRef(G_ArenaHandle arena_handle, G_ResourceHandle resource_handle, G_Re
                 {
                     desc.Format = DXGI_FORMAT_R32_TYPELESS;
                     desc.Buffer.Flags = D3D12_BUFFER_SRV_FLAG_RAW;
+                    desc.Buffer.StructureByteStride = 0;
                 }
                 ID3D12Device_CreateShaderResourceView(g->device, resource->d3d_resource, &desc, descriptor->handle);
             }
@@ -1199,6 +1299,15 @@ i32 G_CountDepth(G_ResourceHandle texture)
     return resource->texture_dims.z;
 }
 
+//- Map
+
+void *G_HostPointerFromResource(G_ResourceHandle resource_handle)
+{
+    G_D12_Resource *resource = G_D12_ResourceFromHandle(resource_handle);
+    G_D12_ResourceHeap *heap = resource->heap;
+    return ((u8 *)heap->mapped) + resource->pos_in_heap;
+}
+
 ////////////////////////////////////////////////////////////
 //~ Command helpers
 
@@ -2112,7 +2221,11 @@ void G_ResetArena(G_CommandListHandle cl_handle, G_ArenaHandle arena_handle)
     /* FIXME: Move descriptors into committed lists */
 
     /* FIXME: Release id3d12 resource com object references */
-    gpu_arena->heap_pos = 0;
+    for (u64 heap_idx = 0; heap_idx < countof(gpu_arena->resource_heaps); ++heap_idx)
+    {
+        G_D12_ResourceHeap *heap = &gpu_arena->resource_heaps[heap_idx];
+        heap->pos = 0;
+    }
 }
 
 //- Cpu -> Gpu copy
@@ -2422,41 +2535,6 @@ void G_SyncOtherQueues(G_QueueKind completion_queue_kind)
 //     ID3D12Resource_Unmap(r->d3d_resource, 0, 0);
 // }
 
-// void G_CopyBytesToFootprint(void *dst, void *src, G_Resource *footprint_reference)
-// {
-//     G_D12_SharedState *g = &G_D12_shared_state;
-
-//     D3D12_RESOURCE_DESC desc = ZI;
-//     ID3D12Resource_GetDesc(((G_D12_Resource *)footprint_reference)->d3d_resource, &desc);
-
-//     u64 upload_size = 0;
-//     u64 upload_row_size = 0;
-//     u32 upload_num_rows = 0;
-//     D3D12_PLACED_SUBRESOURCE_FOOTPRINT placed_footprint = ZI;
-//     ID3D12Device_GetCopyableFootprints(g->device, &desc, 0, 1, 0, &placed_footprint, &upload_num_rows, &upload_row_size, &upload_size);
-//     D3D12_SUBRESOURCE_FOOTPRINT footprint = placed_footprint.Footprint;
-
-//     {
-//         D3D12_RANGE read_range = ZI;
-//         u8 *dst_base = (u8 *)dst + placed_footprint.Offset;
-//         u8 *src_base = src;
-
-//         u32 z_size = upload_row_size * upload_num_rows;
-
-//         b32 src_overflow = 0;
-//         for (u32 z = 0; !src_overflow && z < desc.DepthOrArraySize; ++z)
-//         {
-//             u32 z_offset = z * z_size;
-//             for (u32 y = 0; !src_overflow && y < upload_num_rows; ++y)
-//             {
-//                 u8 *dst_row = dst_base + y * footprint.RowPitch + z_offset;
-//                 u8 *src_row = src_base + y * upload_row_size + z_offset;
-//                 CopyBytes(dst_row, src_row, upload_row_size);
-//             }
-//         }
-//     }
-// }
-
 ////////////////////////////////////////////////////////////
 //~ @hookimpl Statistics
 
@@ -2709,5 +2787,24 @@ void G_D12_WorkerEntry(WaveLaneCtx *lane)
 
     for (;;)
     {
+        /* FIXME: Remove this */
+
+        Sleep(500);
+
+        G_ResourceHandle debug_print_buff = queue->debug_print_buffer;
+        G_D12_Resource *resource = G_D12_ResourceFromHandle(debug_print_buff);
+
+        if (!G_IsResourceNil(debug_print_buff))
+        {
+            u8 *base = G_StructFromResource(debug_print_buff, u8);
+            u32 size = *((u32 *)base);
+            String text = STRING(size, base + 4);
+
+            if (queue_kind == G_QueueKind_Direct)
+            {
+                DEBUGBREAKABLE;
+            }
+        }
+
     }
 }

src/gpu/gpu_dx12/gpu_dx12_core.h

@@ -98,6 +98,36 @@ Struct(G_D12_DescriptorList)
     G_D12_Descriptor *last;
 };
 
+////////////////////////////////////////////////////////////
+//~ Resource heap types
+
+/* TODO:
+ * To support D3D12_RESOURCE_HEAP_TIER_1 devices, create separate heaps for:
+ * - Buffers
+ * - Non-render target & non-depth stencil textures
+ * - Render target or depth stencil textures
+ */
+Enum(G_D12_ResourceHeapKind)
+{
+    G_D12_ResourceHeapKind_Default,
+    G_D12_ResourceHeapKind_CpuWriteBack,
+    G_D12_ResourceHeapKind_CpuWriteCombine,
+
+    G_D12_ResourceHeapKind_Count,
+};
+
+Struct(G_D12_ResourceHeap)
+{
+    G_D12_ResourceHeapKind kind;
+
+    ID3D12Heap *d3d_heap;
+    ID3D12Resource *d3d_mapped_resource;
+    void *mapped;
+
+    u64 pos;
+    u64 size;
+};
+
 ////////////////////////////////////////////////////////////
 //~ Arena types
 
@@ -106,15 +136,7 @@ Struct(G_D12_Arena)
     Arena *arena;
     G_D12_DescriptorList committed_descriptors_by_heap_and_queue[G_D12_DescriptorHeapKind_Count][G_NumQueues];
 
-    /* TODO:
+    G_D12_ResourceHeap resource_heaps[G_D12_ResourceHeapKind_Count];
-     * To support D3D12_RESOURCE_HEAP_TIER_1 devices, create separate heaps for:
-     * - Buffers
-     * - Non-render target & non-depth stencil textures
-     * - Render target or depth stencil textures
-     */
-    ID3D12Heap *d3d_resource_heap;
-    u64 heap_pos;
-    u64 heap_size;
 };
 
 ////////////////////////////////////////////////////////////
@@ -124,6 +146,9 @@ Struct(G_D12_Resource)
 {
     G_D12_Resource *next_free;
 
+    G_D12_ResourceHeap *heap;
+    u64 pos_in_heap;
+
     ID3D12Resource *d3d_resource;
     u64 uid;
     G_ResourceFlag flags;
@@ -414,7 +439,12 @@ G_D12_Pipeline *G_D12_PipelineFromDesc(G_D12_PipelineDesc desc);
 G_D12_Queue *G_D12_QueueFromKind(G_QueueKind kind);
 
 ////////////////////////////////////////////////////////////
-//~ Resource helpers
+//~ Arena
+
+G_D12_ResourceHeap *G_D12_ResourceHeapFromArena(G_D12_Arena *gpu_arena, G_D12_ResourceHeapKind kind);
+
+////////////////////////////////////////////////////////////
+//~ Resource
 
 G_D12_Descriptor *G_D12_DescriptorFromIndex(G_D12_DescriptorHeapKind heap_kind, u32 index);
 G_D12_Descriptor *G_D12_PushDescriptor(G_D12_Arena *gpu_arena, G_D12_DescriptorHeapKind heap_kind);

src/gpu/gpu_shader_core.cgh

@@ -110,12 +110,12 @@ G_ForceDeclConstant(G_RWByteAddressBufferRef, G_DebugPrintBuffer, 8);
 #if IsLanguageG
 
     /* This technique comes from MJP's article: https://therealmjp.github.io/posts/hlsl-printf/ */
-    #if G__DEBUG
+    #if GPU_DEBUG
         #define G_DebugPrint(fmt_cstr) do {                                                 \
             u32 __strlen = 0;                                                               \
             for (;;)  { if (U32FromChar(fmt_cstr[__strlen]) == 0) { break; } ++__strlen; }  \
             RWByteAddressBuffer __print_buff;                                               \
-            __print_buff = RWByteAddressBufferFromRef(G_DebugPrintBuffer);                  \
+            __print_buff = G_RWByteAddressBufferFromRef(G_DebugPrintBuffer);                \
             u32 __pos;                                                                      \
             __print_buff.InterlockedAdd(0, __strlen, __pos);                                \
             if (__pos < countof(__print_buff))                                              \