replace granular gpu sync, access, & texture-layouts with queue-exclusive/shared layouts & tracked sync

.gitattributes

@@ -23,8 +23,3 @@
 *.ttf     filter=lfs diff=lfs merge=lfs -text
 *.mp3     filter=lfs diff=lfs merge=lfs -text
 *.dat     filter=lfs diff=lfs merge=lfs -text
-
-##############################
-#- Libraries
-
-# vulkan/*  filter=lfs diff=lfs merge=lfs -text

src/base/base.cgh

@@ -798,6 +798,7 @@ Inline u64 MixU64s(u64 seed_a, u64 seed_b)
   String GetAppDirectory(void);
   void Echo(String msg);
   b32 Panic(String msg);
+  b32 DebugBreakPrompt(String title, String msg);
   Callstack CaptureCallstack(u64 skip_frames);
   b32 IsRunningInDebugger(void);
   b32 IsRunningInWine(void);

src/base/base_win32/base_win32.c

@@ -96,6 +96,34 @@ b32 Panic(String msg)
   return 0;
 }
 
+b32 DebugBreakPrompt(String title, String msg)
+{
+  LogInfoF("[DEBUG BREAK PROMPT]: %F", FmtString(msg));
+  TempArena scratch = BeginScratchNoConflict();
+  i32 result = 0;
+  {
+    wchar_t *title_wstr = WstrFromString(scratch.arena, title);
+    wchar_t *msg_wstr = WstrFromString(scratch.arena, msg);
+    b32 is_debug = IsRunningInDebugger();
+    i32 mb_result = 0;
+    {
+      u32 mb_flags = MB_SETFOREGROUND | MB_ICONWARNING;
+      if (is_debug)
+      {
+        mb_flags |= MB_CANCELTRYCONTINUE;
+      }
+      mb_result = MessageBoxExW(0, msg_wstr, title_wstr, mb_flags, 0);
+    }
+    if (mb_result == IDCANCEL)
+    {
+      ExitProcess(1);
+    }
+    result = is_debug && mb_result != IDCONTINUE;
+  }
+  EndScratch(scratch);
+  return result;
+}
+
 Callstack CaptureCallstack(u64 skip_frames)
 {
   Callstack result;

src/gpu/gpu.lay

@@ -27,5 +27,4 @@
 
 @IncludeC gpu_common.c
 
-// @DefaultDownstream Win32 gpu_dx12
+@DefaultDownstream Win32 gpu_dx12
-@DefaultDownstream Win32 gpu_vk

src/gpu/gpu_core.h

@@ -189,112 +189,22 @@ Enum(G_Format)
 };
 
 ////////////////////////////////////////////////////////////
-//~ Memory sync types
+//~ Texture layout types
-
-Enum(G_Stage)
-{
-  G_Stage_None           = 0,
-
-  // Compute stages
-  G_Stage_ComputeShading = (1 << 1),
-
-  // Draw stages
-  G_Stage_IndexAssembly  = (1 << 2),
-  G_Stage_VertexShading  = (1 << 3),
-  G_Stage_PixelShading   = (1 << 4),
-  G_Stage_DepthStencil   = (1 << 5),
-  G_Stage_RenderTarget   = (1 << 6),
-
-  // Copy stages
-  G_Stage_Copy           = (1 << 7),
-
-  // Indirect stages
-  G_Stage_Indirect       = (1 << 8),
-
-  // Aggregate stages
-  G_Stage_Drawing = G_Stage_IndexAssembly |
-                    G_Stage_VertexShading |
-                    G_Stage_PixelShading  |
-                    G_Stage_DepthStencil  |
-                    G_Stage_RenderTarget,
-
-  G_Stage_Shading = G_Stage_ComputeShading |
-                    G_Stage_VertexShading  |
-                    G_Stage_PixelShading,
-
-  G_Stage_All            = 0xFFFFFFFF
-};
-
-Enum(G_Access)
-{
-  G_Access_None                 = 0,
-
-  G_Access_ShaderReadWrite      = (1 << 1),
-  G_Access_ShaderRead           = (1 << 2),
-
-  G_Access_CopyWrite            = (1 << 3),
-  G_Access_CopyRead             = (1 << 4),
-
-  G_Access_DepthStencilRead     = (1 << 5),
-  G_Access_DepthStencilWrite    = (1 << 6),
-  G_Access_RenderTargetWrite    = (1 << 7),
-
-  G_Access_IndexBuffer          = (1 << 8),
-  G_Access_IndirectArgument     = (1 << 9),
-
-  G_Access_All                  = 0xFFFFFFFF  // Represents all accesses relevant to the stage specified in the barrier
-};
 
 Enum(G_Layout)
 {
-  G_Layout_NoChange,
+  // Supports any read access with up to 1 write access to non overlapping regions from any queue.
-  G_Layout_Undefined,
+  // Cannot be transitioned to/from.
+  // Depth-stencil textures cannot use this layout.
+  G_Layout_Simultaneous,
 
-  //////////////////////////////
+  // Supports present, shader-read, and copy-read/write in any queue kind.
-  //- Queue-agnostic
+  // Transitionable from `G_Layout_Exclusive` in non-copy queue.
+  G_Layout_Common,
 
-  // Simultaneous layout allows a resource to be used on any queue with any
+  // Supports any access in the current queue kind.
-  // access type (except depth-stencil). Resources cannot transition to/from
+  // Transitionable from `G_Layout_Common` in non-copy queue.
-  // this layout, they must be created with it. Allows concurrent reads
+  G_Layout_Exclusive,
-  // with up to 1 write to non-overlapping regions.
-  G_Layout_Simultaneous,                       // Any access except depth-stencil                 <-- D3D12_BARRIER_LAYOUT_COMMON + D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS
-  G_Layout_Common,                             // ShaderRead/CopyRead/CopyWrite/Present           <-- D3D12_BARRIER_LAYOUT_COMMON
-
-  //////////////////////////////
-  //- Direct queue
-
-  G_Layout_DirectQueue_General,                // ShaderRead/ShaderReadWrite/CopyRead/CopyWrite   <-- D3D12_BARRIER_LAYOUT_DIRECT_QUEUE_COMMON
-  G_Layout_DirectQueue_Read,                   // ShaderRead/CopyRead/DepthStencilRead            <-- D3D12_BARRIER_LAYOUT_DIRECT_QUEUE_GENERIC_READ
-  G_Layout_DirectQueue_DepthStencil,           // DepthStencilRead/DepthStencilWrite              <-- D3D12_BARRIER_LAYOUT_DEPTH_STENCIL_WRITE
-  G_Layout_DirectQueue_RenderTarget,           // RenderTargetWrite                               <-- D3D12_BARRIER_LAYOUT_RENDER_TARGET
-
-  //////////////////////////////
-  //- Compute queue
-
-  G_Layout_ComputeQueue_General,               // ShaderRead/ShaderReadWrite/CopyRead/CopyWrite   <-- D3D12_BARRIER_LAYOUT_COMPUTE_QUEUE_COMMON
-
-  //////////////////////////////
-  //- Direct & Compute queue
-
-  G_Layout_DirectComputeQueue_Read,            // ShaderRead/CopyRead                             <-- D3D12_BARRIER_LAYOUT_GENERIC_READ
-  G_Layout_DirectComputeQueue_ShaderReadWrite, // ShaderReadWrite                                 <-- D3D12_BARRIER_LAYOUT_UNORDERED_ACCESS
-  G_Layout_DirectComputeQueue_CopyWrite,       // CopyWrite                                       <-- D3D12_BARRIER_LAYOUT_COPY_DEST
-};
-
-// Barrier will execute after stages specified by `stage_prev`, and before stages specified by `stage_next`.
-// When barrier executes:
-//   - Necessary resource flushes will occur based on `access_prev` & `access_next`
-//   - Texture layout will transition based on `layout` (if specified)
-Struct(G_MemoryBarrierDesc)
-{
-  G_ResourceHandle resource;
-  b32 is_global;
-  G_Stage stage_prev;
-  G_Stage stage_next;
-  G_Access access_prev;
-  G_Access access_next;
-  G_Layout layout;
-  RngI32 mips;  // Inclusive range of texture mip levels to sync
 };
 
 ////////////////////////////////////////////////////////////
@@ -390,14 +300,14 @@ Enum(G_ResourceKind)
 
 Enum(G_ResourceFlag)
 {
-  G_ResourceFlag_None                    = 0,
+  G_ResourceFlag_None                     = 0,
-  G_ResourceFlag_AllowShaderReadWrite    = (1 << 0),
+  G_ResourceFlag_AllowShaderReadWrite     = (1 << 1),
-  G_ResourceFlag_AllowRenderTarget       = (1 << 1),
+  G_ResourceFlag_AllowRenderTarget        = (1 << 2),
-  G_ResourceFlag_AllowDepthStencil       = (1 << 2),
+  G_ResourceFlag_AllowDepthStencil        = (1 << 3),
-  G_ResourceFlag_ZeroMemory              = (1 << 3),
+  G_ResourceFlag_ZeroMemory               = (1 << 4),
-  G_ResourceFlag_HostMemory              = (1 << 4),  // Resource will be mapped into the cpu's address space
+  G_ResourceFlag_HostMemory               = (1 << 5),  // Resource will be mapped into the cpu's address space
-  G_ResourceFlag_Uncached                = (1 << 5),  // Cpu writes will be combined & reads will be uncached
+  G_ResourceFlag_Uncached                 = (1 << 6),  // Cpu writes will be combined & reads will be uncached
-  G_ResourceFlag_ForceNoReuse            = (1 << 6),
+  G_ResourceFlag_ForceNoReuse             = (1 << 7),
 };
 
 Struct(G_BufferDesc)
@@ -454,18 +364,18 @@ Struct(G_RefDesc)
 };
 
 ////////////////////////////////////////////////////////////
-//~ Rasterization types
+//~ Draw types
 
-Enum(G_RasterMode)
+Enum(G_DrawMode)
 {
-  G_RasterMode_None,
+  G_DrawMode_None,
-  G_RasterMode_PointList,
+  G_DrawMode_PointList,
-  G_RasterMode_LineList,
+  G_DrawMode_LineList,
-  G_RasterMode_LineStrip,
+  G_DrawMode_LineStrip,
-  G_RasterMode_TriangleList,
+  G_DrawMode_TriangleList,
-  G_RasterMode_TriangleStrip,
+  G_DrawMode_TriangleStrip,
-  G_RasterMode_WireTriangleList,
+  G_DrawMode_WireTriangleList,
-  G_RasterMode_WireTriangleStrip,
+  G_DrawMode_WireTriangleStrip,
 };
 
 Enum(G_BlendMode)
@@ -684,52 +594,10 @@ void G_SetConstantEx(G_CommandListHandle cl, i32 slot, void *src_32bit, u32 size
     G_SetConstantEx((cl), (name), &__src, sizeof(__src));      \
   } while (0)
 
-//- Memory sync
+//- Sync
 
-void G_MemorySyncEx(G_CommandListHandle cl, G_MemoryBarrierDesc desc);
+void G_Sync(G_CommandListHandle cl);
-
+void G_SyncLayout(G_CommandListHandle cl, G_ResourceHandle resource, G_Layout layout);
-#define G_MemorySync(_cl, _resource, _stage_prev, _access_prev, _stage_next, _access_next, ...)                 \
-  G_MemorySyncEx((_cl), (G_MemoryBarrierDesc) {                                                                 \
-    .resource = (_resource),                                                                                    \
-    .stage_prev = _stage_prev,                                                                                  \
-    .access_prev = _access_prev,                                                                                \
-    .stage_next = _stage_next,                                                                                  \
-    .access_next = _access_next,                                                                                \
-    .mips.max = G_MaxMips,                                                                                      \
-    __VA_ARGS__                                                                                                 \
-  })
-
-#define G_MemoryLayoutSync(_cl, _resource, _stage_prev, _access_prev, _stage_next, _access_next, _layout, ...)  \
-  G_MemorySyncEx((_cl), (G_MemoryBarrierDesc) {                                                                 \
-    .resource = (_resource),                                                                                    \
-    .stage_prev = _stage_prev,                                                                                  \
-    .access_prev = _access_prev,                                                                                \
-    .stage_next = _stage_next,                                                                                  \
-    .access_next = _access_next,                                                                                \
-    .layout = _layout,                                                                                          \
-    .mips.max = G_MaxMips,                                                                                      \
-    __VA_ARGS__                                                                                                 \
-  })
-
-#define G_GlobalMemorySync(_cl, _stage_prev, _access_prev, _stage_next, _access_next, ...)                      \
-  G_MemorySyncEx((_cl), (G_MemoryBarrierDesc) {                                                                 \
-    .is_global = 1,                                                                                             \
-    .stage_prev = _stage_prev,                                                                                  \
-    .access_prev = _access_prev,                                                                                \
-    .stage_next = _stage_next,                                                                                  \
-    .access_next = _access_next,                                                                                \
-    .mips.max = G_MaxMips,                                                                                      \
-    __VA_ARGS__                                                                                                 \
-  })
-
-#define G_DumbMemorySync(cl, resource, ...) \
-  G_MemorySync((cl), (resource), G_Stage_All, G_Access_All, G_Stage_All, G_Access_All, __VA_ARGS__)
-
-#define G_DumbMemoryLayoutSync(cl, resource, layout, ...) \
-  G_MemoryLayoutSync((cl), (resource), G_Stage_All, G_Access_All, G_Stage_All, G_Access_All, (layout), __VA_ARGS__)
-
-#define G_DumbGlobalMemorySync(cl, ...) \
-  G_GlobalMemorySync((cl), G_Stage_All, G_Access_All, G_Stage_All, G_Access_All, __VA_ARGS__)
 
 //- Compute
 
@@ -739,25 +607,21 @@ void G_ComputeEx(G_CommandListHandle cl, ComputeShaderDesc cs, Vec3I32 threads);
 #define G_Compute2D(cl, cs, threads)  G_ComputeEx((cl), (cs), VEC3I32((threads).x, (threads).y, 1))
 #define G_Compute3D(cl, cs, threads)  G_ComputeEx((cl), (cs), VEC3I32((threads).x, (threads).y, (threads).z))
 
-//- Rasterize
+//- Draw
 
-void G_Rasterize(
+void G_Draw(
   G_CommandListHandle cl,
   VertexShaderDesc vs, PixelShaderDesc ps,
   u32 instances_count, G_IndexBufferDesc index_buffer,
   u32 render_targets_count, G_RenderTargetDesc *render_targets,
   Rng3 viewport, Rng2 scissor,
-  G_RasterMode raster_mode
+  G_DrawMode draw_mode
 );
 
 //- Clear
 
 void G_ClearRenderTarget(G_CommandListHandle cl, G_ResourceHandle render_target, Vec4 color, i32 mip);
 
-//- Log
-
-void G_LogResource(G_CommandListHandle cl, G_ResourceHandle resource);
-
 ////////////////////////////////////////////////////////////
 //~ @hookdecl Queue synchronization
 

src/gpu/gpu_dx12/gpu_dx12_core.h

@@ -87,7 +87,6 @@ Struct(G_D12_Resource)
   G_Format texture_format;
   Vec3I32 texture_dims;
   i32 texture_mips;
-  D3D12_BARRIER_LAYOUT cmdlist_texture_layouts[G_D12_MaxMips];
 
   // Sampler info
   G_SamplerDesc sampler_desc;
@@ -306,16 +305,14 @@ Enum(G_D12_CmdKind)
   G_D12_CmdKind_CopyBytes,
   G_D12_CmdKind_CopyTexels,
   G_D12_CmdKind_Compute,
-  G_D12_CmdKind_Rasterize,
+  G_D12_CmdKind_Draw,
   G_D12_CmdKind_ClearRtv,
-  G_D12_CmdKind_Log,
   G_D12_CmdKind_Discard,
 };
 
 Struct(G_D12_Cmd)
 {
   G_D12_CmdKind kind;
-  b32 skip;
   union
   {
     struct
@@ -326,11 +323,8 @@ Struct(G_D12_Cmd)
 
     struct
     {
-      G_MemoryBarrierDesc desc;
+      G_D12_Resource *resource;
-
+      b32 to_exclusive;
-      // Post-batch data
-      b32 is_end_of_batch;
-      u64 batch_gen;
     } barrier;
 
     struct
@@ -366,20 +360,15 @@ Struct(G_D12_Cmd)
       G_RenderTargetDesc render_target_descs[G_MaxRenderTargets];
       Rng3 viewport;
       Rng2 scissor;
-      G_RasterMode raster_mode;
+      G_DrawMode draw_mode;
-    } rasterize;
+    } draw;
-
-    struct
-    {
-      G_D12_Resource *render_target;
-      Vec4 color;
-      i32 mip;
-    } clear_rtv;
 
     struct
     {
       G_D12_Resource *resource;
-    } log;
+      Vec4 color;
+      i32 mip;
+    } clear_rtv;
 
     struct
     {
@@ -413,6 +402,86 @@ Struct(G_D12_CmdList)
   u64 cmds_count;
 };
 
+////////////////////////////////////////////////////////////
+//~ Command batching types
+
+// TODO: Use a dynamic bin count, since the maximum number of tracked resources in the list is known at command list creation time
+#define G_D12_TrackedResourceBinsCount 64
+
+Enum(G_D12_TrackedUsageKind)
+{
+  G_D12_TrackedUsageKind_Untracked,
+  G_D12_TrackedUsageKind_MakeExclusive,
+  G_D12_TrackedUsageKind_DepthStencilRead,
+  G_D12_TrackedUsageKind_DepthStencilReadWrite,
+  G_D12_TrackedUsageKind_RenderTarget,
+  G_D12_TrackedUsageKind_MakeCommon,
+};
+
+Struct(G_D12_TransitionNode)
+{
+  G_D12_TransitionNode *next;
+  G_D12_Resource *resource;
+
+  G_D12_TrackedUsageKind old;
+  G_D12_TrackedUsageKind new;
+  RngI32 mips;
+};
+
+Struct(G_D12_TrackedMip)
+{
+  G_D12_TrackedUsageKind prev_usage;
+  G_D12_TrackedUsageKind usage;
+};
+
+Struct(G_D12_TrackedResourceNode)
+{
+  G_D12_TrackedResourceNode *next;
+  G_D12_TrackedResourceNode *next_in_bin;
+
+  u64 hash;
+  G_D12_Resource *resource;
+  G_D12_TrackedMip mips[G_MaxMips];
+};
+
+Struct(G_D12_TrackedResourceBin)
+{
+  G_D12_TrackedResourceNode *first;
+};
+
+Struct(G_D12_BatchedCmdNode)
+{
+  G_D12_BatchedCmdNode *next;
+  G_D12_Cmd *cmd;
+};
+
+Struct(G_D12_CmdBatch)
+{
+  G_D12_CmdBatch *next;
+
+  u64 tracked_resources_count;
+  G_D12_TrackedResourceNode *first_tracked_resource;
+  G_D12_TrackedResourceNode *last_tracked_resource;
+  G_D12_TrackedResourceBin *tracked_resource_bins;
+
+  u64 transitions_count;
+  G_D12_TransitionNode *first_transition;
+  G_D12_TransitionNode *last_transition;
+
+  u64 cmds_count;
+  G_D12_BatchedCmdNode *first_cmd;
+  G_D12_BatchedCmdNode *last_cmd;
+
+  b32 contains_hazard;
+
+  b32 contains_compute_shader;
+  b32 contains_draw_shader;
+  b32 contains_rtv;
+  b32 contains_dsv;
+  b32 contains_indirect;
+  b32 contains_copy;
+};
+
 ////////////////////////////////////////////////////////////
 //~ Swapchain types
 
@@ -521,10 +590,7 @@ G_D12_Resource *G_D12_ResourceFromHandle(G_ResourceHandle handle);
 G_D12_Swapchain *G_D12_SwapchainFromHandle(G_SwapchainHandle handle);
 
 DXGI_FORMAT G_D12_DxgiFormatFromGpuFormat(G_Format format);
-D3D12_BARRIER_SYNC G_D12_BarrierSyncFromStages(G_Stage stages);
+D3D12_BARRIER_LAYOUT G_D12_BarrierLayoutFromUsageKind(G_QueueKind queue_kind, G_D12_TrackedUsageKind usage_kind);
-D3D12_BARRIER_ACCESS G_D12_BarrierAccessFromAccesses(G_Access accesses);
-D3D12_BARRIER_LAYOUT G_D12_BarrierLayoutFromLayout(G_Layout layout);
-String G_D12_NameFromBarrierLayout(D3D12_BARRIER_LAYOUT layout);
 
 void G_D12_InitRtv(G_D12_Resource *resource, D3D12_CPU_DESCRIPTOR_HANDLE rtv_handle, i32 mip);
 
@@ -565,6 +631,18 @@ G_D12_Descriptor *G_D12_PushDescriptor(G_D12_Arena *gpu_arena, G_D12_DescriptorH
 G_D12_Cmd *G_D12_PushCmd(G_D12_CmdList *cl);
 G_D12_Cmd *G_D12_PushConstCmd(G_D12_CmdList *cl, i32 slot, void *v);
 G_D12_StagingRegionNode *G_D12_PushStagingRegion(G_D12_CmdList *cl, u64 size);
+void G_D12_UpdateTrackedUsage(Arena *arena, G_D12_CmdBatch *batch, G_D12_Resource *resource, RngI32 mips, G_D12_TrackedUsageKind usage_kind);
+
+////////////////////////////////////////////////////////////
+//~ Debug
+
+void G_D12_DebugCallback(
+  D3D12_MESSAGE_CATEGORY category,
+  D3D12_MESSAGE_SEVERITY severity,
+  D3D12_MESSAGE_ID id,
+  LPCSTR description_cstr,
+  void *context
+);
 
 ////////////////////////////////////////////////////////////
 //~ Collection worker

src/pp/pp_vis/pp_vis_core.c

@@ -426,7 +426,7 @@ void V_TickForever(WaveLaneCtx *lane)
           gpu_perm, cl,
           G_Format_R8_Uint,
           tiles_dims,
-          G_Layout_DirectQueue_General,
+          G_Layout_Exclusive,
           .flags = G_ResourceFlag_ZeroMemory,
           .name = Lit("Tiles")
         );
@@ -451,7 +451,7 @@ void V_TickForever(WaveLaneCtx *lane)
             gpu_perm, cl,
             G_Format_R32_Uint,
             cells_dims,
-            G_Layout_DirectQueue_General,
+            G_Layout_Exclusive,
             .flags = G_ResourceFlag_ZeroMemory | G_ResourceFlag_AllowShaderReadWrite,
             .name = StringF(perm, "Particle cells - layer %F", FmtSint(layer))
           );
@@ -464,7 +464,7 @@ void V_TickForever(WaveLaneCtx *lane)
             gpu_perm, cl,
             G_Format_R32_Uint,
             cells_dims,
-            G_Layout_DirectQueue_General,
+            G_Layout_Exclusive,
             .flags = G_ResourceFlag_ZeroMemory | G_ResourceFlag_AllowShaderReadWrite,
             .name = StringF(perm, "Particle densities - layer %F", FmtSint(layer))
           );
@@ -479,7 +479,7 @@ void V_TickForever(WaveLaneCtx *lane)
           gpu_perm, cl,
           G_Format_R16G16B16A16_Float,
           cells_dims,
-          G_Layout_DirectQueue_General,
+          G_Layout_Exclusive,
           .flags = G_ResourceFlag_ZeroMemory | G_ResourceFlag_AllowShaderReadWrite,
           .name = Lit("Stains")
         );
@@ -491,7 +491,7 @@ void V_TickForever(WaveLaneCtx *lane)
           gpu_perm, cl,
           G_Format_R16G16B16A16_Float,
           cells_dims,
-          G_Layout_DirectQueue_General,
+          G_Layout_Exclusive,
           .flags = G_ResourceFlag_ZeroMemory | G_ResourceFlag_AllowShaderReadWrite,
           .name = Lit("Dry stains")
         );
@@ -503,7 +503,7 @@ void V_TickForever(WaveLaneCtx *lane)
           gpu_perm, cl,
           G_Format_R32_Float,
           cells_dims,
-          G_Layout_DirectQueue_General,
+          G_Layout_Exclusive,
           .flags = G_ResourceFlag_ZeroMemory | G_ResourceFlag_AllowShaderReadWrite,
           .name = Lit("Drynesses")
         );
@@ -515,7 +515,7 @@ void V_TickForever(WaveLaneCtx *lane)
           gpu_perm, cl,
           G_Format_R32_Uint,
           cells_dims,
-          G_Layout_DirectQueue_General,
+          G_Layout_Exclusive,
           .flags = G_ResourceFlag_ZeroMemory | G_ResourceFlag_AllowShaderReadWrite,
           .name = Lit("Occluders cells")
         );
@@ -5153,7 +5153,7 @@ void V_TickForever(WaveLaneCtx *lane)
         gpu_frame_arena, cl,
         G_Format_R16G16B16A16_Float,
         frame->screen_dims,
-        G_Layout_DirectQueue_General,
+        G_Layout_Exclusive,
         .flags = G_ResourceFlag_AllowShaderReadWrite | G_ResourceFlag_AllowRenderTarget,
         .name = StringF(frame->arena, "Screen target [%F]", FmtSint(frame->tick))
       );
@@ -5166,7 +5166,7 @@ void V_TickForever(WaveLaneCtx *lane)
         gpu_frame_arena, cl,
         G_Format_R16G16B16A16_Float,
         frame->screen_dims,
-        G_Layout_DirectQueue_RenderTarget,
+        G_Layout_Exclusive,
         .flags = G_ResourceFlag_AllowRenderTarget,
         .name = StringF(frame->arena, "Albedo target [%F]", FmtSint(frame->tick))
       );
@@ -5177,7 +5177,7 @@ void V_TickForever(WaveLaneCtx *lane)
         gpu_frame_arena, cl,
         G_Format_R16G16B16A16_Float,
         G_DimsFromMip2D(G_Count2D(screen_target), 1),
-        G_Layout_DirectQueue_General,
+        G_Layout_Exclusive,
         .flags = G_ResourceFlag_AllowShaderReadWrite,
         .name = StringF(frame->arena, "Backdrop target [%F]", FmtSint(frame->tick)),
         .max_mips = 4
@@ -5193,7 +5193,7 @@ void V_TickForever(WaveLaneCtx *lane)
         gpu_frame_arena, cl,
         G_Format_R16G16B16A16_Float,
         G_DimsFromMip2D(G_Count2D(screen_target), 1),
-        G_Layout_DirectQueue_General,
+        G_Layout_Exclusive,
         .flags = G_ResourceFlag_AllowShaderReadWrite,
         .name = StringF(frame->arena, "Bloom target [%F]", FmtSint(frame->tick)),
         .max_mips = G_MaxMips
@@ -5208,7 +5208,7 @@ void V_TickForever(WaveLaneCtx *lane)
         gpu_frame_arena, cl,
         G_Format_R16G16B16A16_Float,
         frame->shade_dims,
-        G_Layout_DirectQueue_General,
+        G_Layout_Exclusive,
         .flags = G_ResourceFlag_AllowShaderReadWrite,
         .name = StringF(frame->arena, "Shade target [%F]", FmtSint(frame->tick))
       );
@@ -5274,7 +5274,7 @@ void V_TickForever(WaveLaneCtx *lane)
       G_SetConstant(cl, V_GpuConst_NoiseTex, G_BasicNoiseTexture());
 
       // Sync
-      G_DumbGlobalMemorySync(cl);
+      G_Sync(cl);
 
       //////////////////////////////
       //- Initialization pass
@@ -5306,7 +5306,7 @@ void V_TickForever(WaveLaneCtx *lane)
             G_SetConstant(cl, V_GpuConst_MipIdx, mip_idx);
             G_Compute2D(cl, V_BackdropDownCS, down_dims);
 
-            G_DumbGlobalMemorySync(cl);
+            G_Sync(cl);
           }
 
           //- Upsample passes
@@ -5317,11 +5317,11 @@ void V_TickForever(WaveLaneCtx *lane)
             G_SetConstant(cl, V_GpuConst_MipIdx, mip_idx);
             G_Compute2D(cl, V_BackdropUpCS, up_dims);
 
-            G_DumbGlobalMemorySync(cl);
+            G_Sync(cl);
          }
         }
 
-        G_DumbGlobalMemorySync(cl);
+        G_Sync(cl);
       }
 
       //////////////////////////////
@@ -5331,21 +5331,20 @@ void V_TickForever(WaveLaneCtx *lane)
         G_ClearRenderTarget(cl, albedo_target, VEC4(0, 0, 0, 0), 0);
 
         // Draw quads
-        G_Rasterize(
+        G_Draw(
           cl,
           V_QuadVS, V_QuadPS,
           G_CountBuffer(quads_buff, V_Quad), G_QuadIndices(),
           1, &G_Rt(albedo_target, G_BlendMode_CompositeStraightAlpha),
           screen_viewport, screen_scissor,
-          G_RasterMode_TriangleList
+          G_DrawMode_TriangleList
         );
 
         // Emit particles
         G_Compute(cl, V_EmitParticlesCS, frame->emitters_count);
 
         // Sync particles, occluders, & albedo
-        G_DumbGlobalMemorySync(cl);
+        G_Sync(cl);
-        G_DumbMemoryLayoutSync(cl, albedo_target, G_Layout_DirectQueue_General);
       }
 
       //////////////////////////////
@@ -5356,7 +5355,7 @@ void V_TickForever(WaveLaneCtx *lane)
         G_Compute(cl, V_SimParticlesCS, V_ParticlesCap);
 
         // Sync cells
-        G_DumbGlobalMemorySync(cl);
+        G_Sync(cl);
       }
 
       //////////////////////////////
@@ -5368,7 +5367,7 @@ void V_TickForever(WaveLaneCtx *lane)
       {
         G_Compute2D(cl, V_ShadeCS, frame->shade_dims);
 
-        G_DumbGlobalMemorySync(cl);
+        G_Sync(cl);
       }
 
       //////////////////////////////
@@ -5378,7 +5377,7 @@ void V_TickForever(WaveLaneCtx *lane)
         G_Compute2D(cl, V_CompositeCS, frame->screen_dims);
 
         // Sync screen tex
-        G_DumbGlobalMemorySync(cl);
+        G_Sync(cl);
       }
 
       //////////////////////////////
@@ -5400,7 +5399,7 @@ void V_TickForever(WaveLaneCtx *lane)
           G_SetConstant(cl, V_GpuConst_MipIdx, mip_idx);
           G_Compute2D(cl, V_BloomDownCS, down_dims);
 
-          G_DumbGlobalMemorySync(cl);
+          G_Sync(cl);
         }
 
         //- Upsample passes
@@ -5411,7 +5410,7 @@ void V_TickForever(WaveLaneCtx *lane)
           G_SetConstant(cl, V_GpuConst_MipIdx, mip_idx);
           G_Compute2D(cl, V_BloomUpCS, up_dims);
 
-          G_DumbGlobalMemorySync(cl);
+          G_Sync(cl);
        }
       }
 
@@ -5421,7 +5420,7 @@ void V_TickForever(WaveLaneCtx *lane)
       {
         G_Compute2D(cl, V_FinalizeCS, frame->screen_dims);
 
-        G_DumbGlobalMemorySync(cl);
+        G_Sync(cl);
       }
 
       //////////////////////////////
@@ -5429,18 +5428,14 @@ void V_TickForever(WaveLaneCtx *lane)
 
       if (dvert_idxs_ib.count > 0)
       {
-        G_DumbMemoryLayoutSync(cl, screen_target, G_Layout_DirectQueue_RenderTarget);
+        G_Draw(
-
-        G_Rasterize(
           cl,
           V_DVertVS, V_DVertPS,
           1, dvert_idxs_ib,
           1, &G_Rt(screen_target, G_BlendMode_CompositeStraightAlpha),
           screen_viewport, screen_scissor,
-          G_RasterMode_TriangleList
+          G_DrawMode_TriangleList
         );
-
-        G_DumbMemoryLayoutSync(cl, screen_target, G_Layout_DirectQueue_General);
       }
 
       //////////////////////////////

src/proto/proto.c

@@ -5,6 +5,8 @@ void PT_RunForever(WaveLaneCtx *lane)
   for (;;)
   {
     WND_Frame window_frame = WND_BeginFrame(G_Format_R16G16B16A16_Float, WND_BackbufferSizeMode_MatchWindow);
+    G_ResourceHandle backbuffer = window_frame.backbuffer;
+
     for (u64 cev_idx = 0; cev_idx < window_frame.controller_events.count; ++cev_idx)
     {
       ControllerEvent *cev = &window_frame.controller_events.events[cev_idx];
@@ -14,65 +16,57 @@ void PT_RunForever(WaveLaneCtx *lane)
       }
     }
 
+    G_CommandListHandle cl = G_PrepareCommandList(G_QueueKind_Direct);
     {
+      G_SyncLayout(cl, backbuffer, G_Layout_Exclusive);
+
+      Vec2I32 final_target_size = window_frame.draw_size;
+      G_ResourceHandle final_target_res = G_PushTexture2D(
+        gpu_frame_arena, cl,
+        G_Format_R16G16B16A16_Float,
+        final_target_size,
+        G_Layout_Exclusive,
+        .flags = G_ResourceFlag_AllowShaderReadWrite
+      );
+      G_Texture2DRef final_target = G_PushTexture2DRef(gpu_frame_arena, final_target_res);
+
+      // Prep test pass
       {
-        G_CommandListHandle cl = G_PrepareCommandList(G_QueueKind_Direct);
+        G_SetConstant(cl, PT_ShaderConst_TestTarget,          final_target);
-        {
+        G_SetConstant(cl, PT_ShaderConst_TestConst,           3.123);
-          Vec2I32 final_target_size = window_frame.draw_size;
+        G_SetConstant(cl, PT_ShaderConst_BlitSampler,         G_BasicSamplerFromKind(G_BasicSamplerKind_PointClamp));
-          G_ResourceHandle final_target_res = G_PushTexture2D(
+        G_SetConstant(cl, PT_ShaderConst_BlitSrc,             final_target);
-            gpu_frame_arena, cl,
+        G_SetConstant(cl, PT_ShaderConst_NoiseTex,            G_BasicNoiseTexture());
-            G_Format_R16G16B16A16_Float,
-            final_target_size,
-            G_Layout_DirectQueue_General,
-            .flags = G_ResourceFlag_AllowShaderReadWrite
-          );
-          G_Texture2DRef final_target = G_PushTexture2DRef(gpu_frame_arena, final_target_res);
-
-          // Prep test pass
-          {
-            G_SetConstant(cl, PT_ShaderConst_TestTarget,          final_target);
-            G_SetConstant(cl, PT_ShaderConst_TestConst,           3.123);
-            G_SetConstant(cl, PT_ShaderConst_BlitSampler,         G_BasicSamplerFromKind(G_BasicSamplerKind_PointClamp));
-            G_SetConstant(cl, PT_ShaderConst_BlitSrc,             final_target);
-            G_SetConstant(cl, PT_ShaderConst_NoiseTex,            G_BasicNoiseTexture());
-          }
-
-          // Test pass
-          {
-            G_Compute2D(cl, PT_TestCS, final_target_size);
-          }
-          G_DumbMemorySync(cl, final_target_res);
-
-          // Prep blit pass
-          {
-            G_DumbMemoryLayoutSync(cl, window_frame.backbuffer, G_Layout_DirectQueue_RenderTarget);
-          }
-
-          // Blit pass
-          {
-            G_Rasterize(
-              cl,
-              PT_BlitVS, PT_BlitPS,
-              1, G_QuadIndices(),
-              1, &G_Rt(window_frame.backbuffer, G_BlendMode_CompositeStraightAlpha),
-              G_ViewportFromTexture(window_frame.backbuffer), G_ScissorFromTexture(window_frame.backbuffer),
-              G_RasterMode_TriangleList
-            );
-          }
-
-          // Finalize backbuffer layout
-          {
-            G_DumbMemoryLayoutSync(cl, window_frame.backbuffer, G_Layout_Common);
-          }
-
-          // Reset
-          {
-            G_ResetArena(cl, gpu_frame_arena);
-          }
-        }
-        G_CommitCommandList(cl);
       }
+
+      // Test pass
+      {
+        G_Compute2D(cl, PT_TestCS, final_target_size);
+
+        G_Sync(cl);
+      }
+
+      // Blit pass
+      {
+        G_Draw(
+          cl,
+          PT_BlitVS, PT_BlitPS,
+          1, G_QuadIndices(),
+          1, &G_Rt(backbuffer, G_BlendMode_CompositeStraightAlpha),
+          G_ViewportFromTexture(backbuffer), G_ScissorFromTexture(backbuffer),
+          G_DrawMode_TriangleList
+        );
+      }
+
+      // Reset
+      {
+        G_ResetArena(cl, gpu_frame_arena);
+      }
+
+      G_SyncLayout(cl, backbuffer, G_Layout_Common);
     }
+    G_CommitCommandList(cl);
+
     WND_EndFrame(window_frame, 1);
   }
 }

src/ui/ui_core.c

@@ -1708,7 +1708,7 @@ void UI_EndFrame(UI_Frame *frame, i32 vsync)
       UI.gpu_frame_arena, UI.cl,
       G_Format_R16G16B16A16_Float,
       monitor_size,
-      G_Layout_DirectQueue_RenderTarget,
+      G_Layout_Exclusive,
       .flags = G_ResourceFlag_AllowRenderTarget,
       .name = Lit("UI draw target")
     );
@@ -1740,11 +1740,11 @@ void UI_EndFrame(UI_Frame *frame, i32 vsync)
     );
     G_StructuredBufferRef params_ro = G_PushStructuredBufferRef(UI.gpu_frame_arena, params_buff, UI_GpuParams);
 
-    // Initial constants
+    // Init constants
     G_SetConstant(UI.cl, UI_GpuConst_Params, params_ro);
 
     // Sync
-    G_DumbGlobalMemorySync(UI.cl);
+    G_Sync(UI.cl);
 
     //////////////////////////////
     //- Dispatch shaders
@@ -1760,47 +1760,44 @@ void UI_EndFrame(UI_Frame *frame, i32 vsync)
     if (rects_count > 0)
     {
       // Render rects
-      G_Rasterize(
+      G_Draw(
         UI.cl,
         UI_DRectVS, UI_DRectPS,
         rects_count, G_QuadIndices(),
         1, &G_Rt(draw_target, G_BlendMode_CompositePremultipliedAlpha),
         draw_viewport, draw_scissor,
-        G_RasterMode_TriangleList
+        G_DrawMode_TriangleList
       );
 
       // Render rect wireframes
       if (AnyBit(frame->frame_flags, UI_FrameFlag_Debug))
       {
         G_SetConstant(UI.cl, UI_GpuConst_DebugDraw, 1);
-        G_Rasterize(
+        G_Draw(
           UI.cl,
           UI_DRectVS, UI_DRectPS,
           rects_count, G_QuadIndices(),
           1, &G_Rt(draw_target, G_BlendMode_CompositePremultipliedAlpha),
           draw_viewport, draw_scissor,
-          G_RasterMode_WireTriangleList
+          G_DrawMode_WireTriangleList
         );
       }
     }
 
     //- Backbuffer blit pass
 
-    G_DumbMemoryLayoutSync(UI.cl, draw_target, G_Layout_DirectQueue_Read);
+    G_SyncLayout(UI.cl, backbuffer, G_Layout_Exclusive);
-    G_DumbMemoryLayoutSync(UI.cl, backbuffer, G_Layout_DirectQueue_RenderTarget);
-
     {
-      G_Rasterize(
+      G_Draw(
         UI.cl,
         UI_BlitVS, UI_BlitPS,
         1, G_QuadIndices(),
         1, &G_Rt(backbuffer, G_BlendMode_Opaque),
         monitor_viewport, monitor_scissor,
-        G_RasterMode_TriangleList
+        G_DrawMode_TriangleList
       );
     }
-
+    G_SyncLayout(UI.cl, backbuffer, G_Layout_Common);
-    G_DumbMemoryLayoutSync(UI.cl, backbuffer, G_Layout_Common);
   }
 
   //////////////////////////////