diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.h b/src/xenia/gpu/d3d12/d3d12_command_processor.h
index ba464d8e8..92e0f5c02 100644
--- a/src/xenia/gpu/d3d12/d3d12_command_processor.h
+++ b/src/xenia/gpu/d3d12/d3d12_command_processor.h
@@ -376,7 +376,7 @@ class D3D12CommandProcessor : public CommandProcessor {
   CommandAllocator* command_allocator_submitted_last_ = nullptr;
   ID3D12GraphicsCommandList* command_list_ = nullptr;
   ID3D12GraphicsCommandList1* command_list_1_ = nullptr;
-  std::unique_ptr<DeferredCommandList> deferred_command_list_ = nullptr;
+  std::unique_ptr<DeferredCommandList> deferred_command_list_;
 
   // Should bindless textures and samplers be used - many times faster
   // UpdateBindings than bindful (that becomes a significant bottleneck with
@@ -388,14 +388,12 @@ class D3D12CommandProcessor : public CommandProcessor {
   // targets.
   bool edram_rov_used_ = false;
 
-  std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> constant_buffer_pool_ =
-      nullptr;
+  std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> constant_buffer_pool_;
 
   static constexpr uint32_t kViewBindfulHeapSize = 32768;
   static_assert(kViewBindfulHeapSize <=
                 D3D12_MAX_SHADER_VISIBLE_DESCRIPTOR_HEAP_SIZE_TIER_1);
-  std::unique_ptr<ui::d3d12::D3D12DescriptorHeapPool> view_bindful_heap_pool_ =
-      nullptr;
+  std::unique_ptr<ui::d3d12::D3D12DescriptorHeapPool> view_bindful_heap_pool_;
   // Currently bound descriptor heap - updated by RequestViewBindfulDescriptors.
   ID3D12DescriptorHeap* view_bindful_heap_current_;
   // Rationale: textures have 4 KB alignment in guest memory, and there can be
@@ -426,7 +424,7 @@ class D3D12CommandProcessor : public CommandProcessor {
   static constexpr uint32_t kSamplerHeapSize = 2000;
   static_assert(kSamplerHeapSize <= D3D12_MAX_SHADER_VISIBLE_SAMPLER_HEAP_SIZE);
   std::unique_ptr<ui::d3d12::D3D12DescriptorHeapPool>
-      sampler_bindful_heap_pool_ = nullptr;
+      sampler_bindful_heap_pool_;
   ID3D12DescriptorHeap* sampler_bindful_heap_current_;
   ID3D12DescriptorHeap* sampler_bindless_heap_current_ = nullptr;
   D3D12_CPU_DESCRIPTOR_HANDLE sampler_bindless_heap_cpu_start_;
@@ -452,15 +450,15 @@ class D3D12CommandProcessor : public CommandProcessor {
   ID3D12RootSignature* root_signature_bindless_vs_ = nullptr;
   ID3D12RootSignature* root_signature_bindless_ds_ = nullptr;
 
-  std::unique_ptr<SharedMemory> shared_memory_ = nullptr;
+  std::unique_ptr<SharedMemory> shared_memory_;
 
-  std::unique_ptr<PipelineCache> pipeline_cache_ = nullptr;
+  std::unique_ptr<PipelineCache> pipeline_cache_;
 
-  std::unique_ptr<TextureCache> texture_cache_ = nullptr;
+  std::unique_ptr<TextureCache> texture_cache_;
 
-  std::unique_ptr<RenderTargetCache> render_target_cache_ = nullptr;
+  std::unique_ptr<RenderTargetCache> render_target_cache_;
 
-  std::unique_ptr<PrimitiveConverter> primitive_converter_ = nullptr;
+  std::unique_ptr<PrimitiveConverter> primitive_converter_;
 
   // Mip 0 contains the normal gamma ramp (256 entries), mip 1 contains the PWL
   // ramp (128 entries). DXGI_FORMAT_R10G10B10A2_UNORM 1D.
diff --git a/src/xenia/gpu/d3d12/d3d12_graphics_system.cc b/src/xenia/gpu/d3d12/d3d12_graphics_system.cc
index 835c94c07..e50bbbaac 100644
--- a/src/xenia/gpu/d3d12/d3d12_graphics_system.cc
+++ b/src/xenia/gpu/d3d12/d3d12_graphics_system.cc
@@ -142,6 +142,7 @@ X_STATUS D3D12GraphicsSystem::Setup(cpu::Processor* processor,
   stretch_pipeline_desc.VS.BytecodeLength = sizeof(fullscreen_vs);
   stretch_pipeline_desc.PS.pShaderBytecode = stretch_ps;
   stretch_pipeline_desc.PS.BytecodeLength = sizeof(stretch_ps);
+  // The shader will set alpha to 1, don't use output-merger to preserve it.
   stretch_pipeline_desc.BlendState.RenderTarget[0].RenderTargetWriteMask =
       D3D12_COLOR_WRITE_ENABLE_ALL;
   stretch_pipeline_desc.SampleMask = UINT_MAX;
diff --git a/src/xenia/gpu/d3d12/pipeline_cache.h b/src/xenia/gpu/d3d12/pipeline_cache.h
index 30fd68a4e..cdc6ed5f3 100644
--- a/src/xenia/gpu/d3d12/pipeline_cache.h
+++ b/src/xenia/gpu/d3d12/pipeline_cache.h
@@ -247,7 +247,7 @@ class PipelineCache {
   uint32_t resolution_scale_;
 
   // Reusable shader translator.
-  std::unique_ptr<DxbcShaderTranslator> shader_translator_ = nullptr;
+  std::unique_ptr<DxbcShaderTranslator> shader_translator_;
 
   // Command processor thread DXIL conversion/disassembly interfaces, if DXIL
   // disassembly is enabled.
@@ -344,7 +344,7 @@ class PipelineCache {
   // Manual-reset event set when the last queued pipeline state object is
   // created and there are no more pipeline state objects to create. This is
   // triggered by the thread creating the last pipeline state object.
-  std::unique_ptr<xe::threading::Event> creation_completion_event_ = nullptr;
+  std::unique_ptr<xe::threading::Event> creation_completion_event_;
   // Whether setting the event on completion is queued. Protected with
   // creation_request_lock_, notify_one creation_request_cond_ when set.
   bool creation_completion_set_event_ = false;
diff --git a/src/xenia/gpu/d3d12/primitive_converter.h b/src/xenia/gpu/d3d12/primitive_converter.h
index c00d29f7c..4d5c80f2d 100644
--- a/src/xenia/gpu/d3d12/primitive_converter.h
+++ b/src/xenia/gpu/d3d12/primitive_converter.h
@@ -107,7 +107,7 @@ class PrimitiveConverter {
   Memory& memory_;
   TraceWriter& trace_writer_;
 
-  std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> buffer_pool_ = nullptr;
+  std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> buffer_pool_;
 
   // Static index buffers for emulating unsupported primitive types when drawing
   // without an index buffer.
diff --git a/src/xenia/gpu/d3d12/render_target_cache.h b/src/xenia/gpu/d3d12/render_target_cache.h
index de3846635..cf575dcdf 100644
--- a/src/xenia/gpu/d3d12/render_target_cache.h
+++ b/src/xenia/gpu/d3d12/render_target_cache.h
@@ -603,7 +603,7 @@ class RenderTargetCache {
   // For traces only.
   ID3D12Resource* edram_snapshot_download_buffer_ = nullptr;
   std::unique_ptr<ui::d3d12::D3D12UploadBufferPool>
-      edram_snapshot_restore_pool_ = nullptr;
+      edram_snapshot_restore_pool_;
 };
 
 }  // namespace d3d12
diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.cso b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.cso
index 6a1fe6724..229d21858 100644
Binary files a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.cso and b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.cso differ
diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.h b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.h
index c41cf6287..e6fd53b4f 100644
--- a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.h
+++ b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.h
@@ -1,11 +1,11 @@
 // generated from `xb buildhlsl`
 // source: stretch_gamma.ps.hlsl
 const uint8_t stretch_gamma_ps[] = {
-    0x44, 0x58, 0x42, 0x43, 0x8D, 0x1B, 0xAE, 0x52, 0x96, 0x87, 0x7B, 0x8F,
-    0xCB, 0xD3, 0xBC, 0x4B, 0xCF, 0xAF, 0x66, 0xA9, 0x01, 0x00, 0x00, 0x00,
-    0x10, 0x05, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
+    0x44, 0x58, 0x42, 0x43, 0x46, 0xCD, 0x15, 0xA8, 0x4D, 0x5B, 0xFD, 0x07,
+    0x4C, 0x62, 0xA7, 0xD0, 0x61, 0x89, 0x85, 0xA6, 0x01, 0x00, 0x00, 0x00,
+    0x24, 0x05, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
     0x14, 0x02, 0x00, 0x00, 0x48, 0x02, 0x00, 0x00, 0x7C, 0x02, 0x00, 0x00,
-    0x74, 0x04, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x01, 0x00, 0x00,
+    0x88, 0x04, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x01, 0x00, 0x00,
     0x01, 0x00, 0x00, 0x00, 0x2C, 0x01, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
     0x3C, 0x00, 0x00, 0x00, 0x01, 0x05, 0xFF, 0xFF, 0x00, 0x05, 0x00, 0x00,
     0xB0, 0x01, 0x00, 0x00, 0x13, 0x13, 0x44, 0x25, 0x3C, 0x00, 0x00, 0x00,
@@ -16,7 +16,7 @@ const uint8_t stretch_gamma_ps[] = {
     0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0xF4, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
     0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
-    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00,
     0x02, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
     0xFF, 0xFF, 0xFF, 0xFF, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
@@ -54,8 +54,8 @@ const uint8_t stretch_gamma_ps[] = {
     0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0x00, 0x00, 0x00,
     0x53, 0x56, 0x5F, 0x54, 0x61, 0x72, 0x67, 0x65, 0x74, 0x00, 0xAB, 0xAB,
-    0x53, 0x48, 0x45, 0x58, 0xF0, 0x01, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00,
-    0x7C, 0x00, 0x00, 0x00, 0x6A, 0x08, 0x00, 0x01, 0x59, 0x00, 0x00, 0x07,
+    0x53, 0x48, 0x45, 0x58, 0x04, 0x02, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00,
+    0x81, 0x00, 0x00, 0x00, 0x6A, 0x08, 0x00, 0x01, 0x59, 0x00, 0x00, 0x07,
     0x46, 0x8E, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x5A, 0x00, 0x00, 0x06, 0x46, 0x6E, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00,
@@ -67,20 +67,20 @@ const uint8_t stretch_gamma_ps[] = {
     0x55, 0x55, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x62, 0x10, 0x00, 0x03,
     0x32, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x03,
     0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x02,
-    0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D, 0xF2, 0x00, 0x10, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D, 0x72, 0x00, 0x10, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x46, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x46, 0x7E, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A,
-    0x12, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0A, 0x80, 0x30, 0x80,
+    0x82, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A, 0x80, 0x30, 0x80,
     0x41, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3F,
-    0x38, 0x00, 0x00, 0x09, 0x22, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x38, 0x00, 0x00, 0x09, 0x12, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
     0x0A, 0x80, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F,
     0x32, 0x00, 0x00, 0x09, 0x72, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x46, 0x02, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x10, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x56, 0x05, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x46, 0x02, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0xF6, 0x0F, 0x10, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
     0x48, 0x00, 0x00, 0x0D, 0x12, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x0A, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x46, 0x7E, 0x20, 0x00,
     0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x60, 0x20, 0x00,
@@ -94,19 +94,21 @@ const uint8_t stretch_gamma_ps[] = {
     0x00, 0x00, 0x00, 0x00, 0x46, 0x7E, 0x20, 0x00, 0x01, 0x00, 0x00, 0x00,
     0x01, 0x00, 0x00, 0x00, 0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x36, 0x00, 0x00, 0x05, 0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x46, 0x0E, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x00, 0x01,
-    0x53, 0x54, 0x41, 0x54, 0x94, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
-    0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x36, 0x00, 0x00, 0x05, 0x72, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x46, 0x02, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x05,
+    0x82, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00,
+    0x00, 0x00, 0x80, 0x3F, 0x3E, 0x00, 0x00, 0x01, 0x53, 0x54, 0x41, 0x54,
+    0x94, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 };
diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.txt b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.txt
index c4addc470..2a06727e9 100644
--- a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.txt
+++ b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.txt
@@ -17,7 +17,7 @@
 // Name                                 Type  Format         Dim      ID      HLSL Bind  Count
 // ------------------------------ ---------- ------- ----------- ------- -------------- ------
 // xe_sampler_linear_clamp           sampler      NA          NA      S0             s0      1 
-// xe_texture                        texture  float4          2d      T0             t0      1 
+// xe_texture                        texture  float3          2d      T0             t0      1 
 // xe_gamma_ramp                     texture  float3          1d      T1             t1      1 
 // XeStretchGammaRootConstants       cbuffer      NA          NA     CB0            cb0      1 
 //
@@ -45,13 +45,14 @@ dcl_resource_texture1d (float,float,float,float) T1[1:1], space=0
 dcl_input_ps linear v0.xy
 dcl_output o0.xyzw
 dcl_temps 2
-sample_l r0.xyzw, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
-add r1.x, -CB0[0][0].x, l(1.000000)
-mul r1.y, CB0[0][0].x, l(0.500000)
-mad r0.xyz, r0.xyzx, r1.xxxx, r1.yyyy
+sample_l r0.xyz, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
+add r0.w, -CB0[0][0].x, l(1.000000)
+mul r1.x, CB0[0][0].x, l(0.500000)
+mad r0.xyz, r0.xyzx, r0.wwww, r1.xxxx
 sample_l r0.x, r0.x, T1[1].xyzw, S0[0], l(0.000000)
 sample_l r0.y, r0.y, T1[1].xyzw, S0[0], l(0.000000)
 sample_l r0.z, r0.z, T1[1].xyzw, S0[0], l(0.000000)
-mov o0.xyzw, r0.xyzw
+mov o0.xyz, r0.xyzx
+mov o0.w, l(1.000000)
 ret 
-// Approximately 9 instruction slots used
+// Approximately 10 instruction slots used
diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.cso b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.cso
index 1843dd639..c5a49abfd 100644
Binary files a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.cso and b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.cso differ
diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.h b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.h
index 5aa4fdb1e..7f850d127 100644
--- a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.h
+++ b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.h
@@ -1,11 +1,11 @@
 // generated from `xb buildhlsl`
 // source: stretch.ps.hlsl
 const uint8_t stretch_ps[] = {
-    0x44, 0x58, 0x42, 0x43, 0x84, 0x8B, 0x62, 0x99, 0x4D, 0x9D, 0x1A, 0x13,
-    0x49, 0x13, 0xF7, 0x6B, 0x9F, 0x69, 0x25, 0x8A, 0x01, 0x00, 0x00, 0x00,
-    0xB0, 0x02, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
+    0x44, 0x58, 0x42, 0x43, 0x0B, 0x39, 0x5E, 0x9C, 0x0F, 0x3E, 0xAC, 0xF6,
+    0x9E, 0xB3, 0x11, 0xC0, 0x9B, 0x25, 0x2E, 0x6E, 0x01, 0x00, 0x00, 0x00,
+    0xE0, 0x02, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
     0x14, 0x01, 0x00, 0x00, 0x48, 0x01, 0x00, 0x00, 0x7C, 0x01, 0x00, 0x00,
-    0x14, 0x02, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x00, 0x00, 0x00,
+    0x44, 0x02, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
     0x3C, 0x00, 0x00, 0x00, 0x01, 0x05, 0xFF, 0xFF, 0x00, 0x05, 0x00, 0x00,
     0xAF, 0x00, 0x00, 0x00, 0x13, 0x13, 0x44, 0x25, 0x3C, 0x00, 0x00, 0x00,
@@ -16,7 +16,7 @@ const uint8_t stretch_ps[] = {
     0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0xA4, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
     0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
-    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x73,
     0x61, 0x6D, 0x70, 0x6C, 0x65, 0x72, 0x5F, 0x6C, 0x69, 0x6E, 0x65, 0x61,
     0x72, 0x5F, 0x63, 0x6C, 0x61, 0x6D, 0x70, 0x00, 0x78, 0x65, 0x5F, 0x74,
@@ -33,26 +33,30 @@ const uint8_t stretch_ps[] = {
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x53, 0x56, 0x5F, 0x54,
     0x61, 0x72, 0x67, 0x65, 0x74, 0x00, 0xAB, 0xAB, 0x53, 0x48, 0x45, 0x58,
-    0x90, 0x00, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
+    0xC0, 0x00, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00,
     0x6A, 0x08, 0x00, 0x01, 0x5A, 0x00, 0x00, 0x06, 0x46, 0x6E, 0x30, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x58, 0x18, 0x00, 0x07, 0x46, 0x7E, 0x30, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x55, 0x55, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x62, 0x10, 0x00, 0x03,
     0x32, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x03,
-    0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D,
-    0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x46, 0x10, 0x10, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x46, 0x7E, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x02,
+    0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D, 0x72, 0x00, 0x10, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x46, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x46, 0x7E, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x05,
+    0x72, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x46, 0x02, 0x10, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x05, 0x82, 0x20, 0x10, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3F,
     0x3E, 0x00, 0x00, 0x01, 0x53, 0x54, 0x41, 0x54, 0x94, 0x00, 0x00, 0x00,
-    0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
diff --git a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.txt b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.txt
index ff7c711a5..8495d8f55 100644
--- a/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.txt
+++ b/src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.txt
@@ -7,7 +7,7 @@
 // Name                                 Type  Format         Dim      ID      HLSL Bind  Count
 // ------------------------------ ---------- ------- ----------- ------- -------------- ------
 // xe_sampler_linear_clamp           sampler      NA          NA      S0             s0      1 
-// xe_texture                        texture  float4          2d      T0             t0      1 
+// xe_texture                        texture  float3          2d      T0             t0      1 
 //
 //
 //
@@ -30,6 +30,9 @@ dcl_sampler S0[0:0], mode_default, space=0
 dcl_resource_texture2d (float,float,float,float) T0[0:0], space=0
 dcl_input_ps linear v0.xy
 dcl_output o0.xyzw
-sample_l o0.xyzw, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
+dcl_temps 1
+sample_l r0.xyz, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
+mov o0.xyz, r0.xyzx
+mov o0.w, l(1.000000)
 ret 
-// Approximately 2 instruction slots used
+// Approximately 4 instruction slots used
diff --git a/src/xenia/gpu/d3d12/shaders/stretch.ps.hlsl b/src/xenia/gpu/d3d12/shaders/stretch.ps.hlsl
index 681552587..c7a568434 100644
--- a/src/xenia/gpu/d3d12/shaders/stretch.ps.hlsl
+++ b/src/xenia/gpu/d3d12/shaders/stretch.ps.hlsl
@@ -1,6 +1,8 @@
-Texture2D<float4> xe_texture : register(t0);
+Texture2D<float3> xe_texture : register(t0);
 SamplerState xe_sampler_linear_clamp : register(s0);
 
 float4 main(float2 xe_texcoord : TEXCOORD) : SV_Target {
-  return xe_texture.SampleLevel(xe_sampler_linear_clamp, xe_texcoord, 0.0f);
+  // Force alpha to 1 to make sure the surface won't be translucent.
+  return float4(
+      xe_texture.SampleLevel(xe_sampler_linear_clamp, xe_texcoord, 0.0f), 1.0f);
 }
diff --git a/src/xenia/gpu/d3d12/shaders/stretch_gamma.ps.hlsl b/src/xenia/gpu/d3d12/shaders/stretch_gamma.ps.hlsl
index cfa545aa1..a13e333a8 100644
--- a/src/xenia/gpu/d3d12/shaders/stretch_gamma.ps.hlsl
+++ b/src/xenia/gpu/d3d12/shaders/stretch_gamma.ps.hlsl
@@ -1,4 +1,4 @@
-Texture2D<float4> xe_texture : register(t0);
+Texture2D<float3> xe_texture : register(t0);
 Texture1D<float3> xe_gamma_ramp : register(t1);
 SamplerState xe_sampler_linear_clamp : register(s0);
 cbuffer XeStretchGammaRootConstants : register(b0) {
@@ -6,14 +6,15 @@ cbuffer XeStretchGammaRootConstants : register(b0) {
 };
 
 float4 main(float2 xe_texcoord : TEXCOORD) : SV_Target {
-  float4 color =
+  float3 color =
       xe_texture.SampleLevel(xe_sampler_linear_clamp, xe_texcoord, 0.0f);
   // The center of the first texel of the LUT contains the value for 0, and the
   // center of the last texel contains the value for 1.
-  color.rgb = color.rgb * (1.0f - xe_gamma_ramp_inv_size) +
-              (0.5 * xe_gamma_ramp_inv_size);
+  color =
+      color * (1.0f - xe_gamma_ramp_inv_size) + (0.5 * xe_gamma_ramp_inv_size);
   color.r = xe_gamma_ramp.SampleLevel(xe_sampler_linear_clamp, color.r, 0.0f).r;
   color.g = xe_gamma_ramp.SampleLevel(xe_sampler_linear_clamp, color.g, 0.0f).g;
   color.b = xe_gamma_ramp.SampleLevel(xe_sampler_linear_clamp, color.b, 0.0f).b;
-  return color;
+  // Force alpha to 1 to make sure the surface won't be translucent.
+  return float4(color, 1.0f);
 }
diff --git a/src/xenia/gpu/d3d12/shared_memory.h b/src/xenia/gpu/d3d12/shared_memory.h
index 676803f15..86a55b2b7 100644
--- a/src/xenia/gpu/d3d12/shared_memory.h
+++ b/src/xenia/gpu/d3d12/shared_memory.h
@@ -212,8 +212,7 @@ class SharedMemory {
   std::vector<UploadRange> upload_ranges_;
   void GetRangesToUpload(uint32_t request_page_first,
                          uint32_t request_page_last);
-  std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> upload_buffer_pool_ =
-      nullptr;
+  std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> upload_buffer_pool_;
 
   // GPU-written memory downloading for traces.
   // Start page, length in pages.
diff --git a/src/xenia/ui/d3d12/d3d12_context.h b/src/xenia/ui/d3d12/d3d12_context.h
index c9f235b97..8ae3a0b2c 100644
--- a/src/xenia/ui/d3d12/d3d12_context.h
+++ b/src/xenia/ui/d3d12/d3d12_context.h
@@ -99,7 +99,7 @@ class D3D12Context : public GraphicsContext {
   //     kSwapCommandAllocatorCount.
   ID3D12GraphicsCommandList* swap_command_list_ = nullptr;
 
-  std::unique_ptr<D3D12ImmediateDrawer> immediate_drawer_ = nullptr;
+  std::unique_ptr<D3D12ImmediateDrawer> immediate_drawer_;
 };
 
 }  // namespace d3d12
diff --git a/src/xenia/ui/d3d12/d3d12_immediate_drawer.cc b/src/xenia/ui/d3d12/d3d12_immediate_drawer.cc
index 565bf59a7..c1bcac5e6 100644
--- a/src/xenia/ui/d3d12/d3d12_immediate_drawer.cc
+++ b/src/xenia/ui/d3d12/d3d12_immediate_drawer.cc
@@ -158,7 +158,7 @@ bool D3D12ImmediateDrawer::Initialize() {
   }
   {
     auto& root_parameter =
-        root_parameters[size_t(RootParameter::kViewportInvSize)];
+        root_parameters[size_t(RootParameter::kViewportSizeInv)];
     root_parameter.ParameterType = D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
     root_parameter.Constants.ShaderRegister = 0;
     root_parameter.Constants.RegisterSpace = 0;
@@ -179,7 +179,7 @@ bool D3D12ImmediateDrawer::Initialize() {
     return false;
   }
 
-  // Create the pipelines.
+  // Create the pipeline states.
   D3D12_GRAPHICS_PIPELINE_STATE_DESC pipeline_desc = {};
   pipeline_desc.pRootSignature = root_signature_;
   pipeline_desc.VS.pShaderBytecode = immediate_vs;
@@ -192,10 +192,13 @@ bool D3D12ImmediateDrawer::Initialize() {
   pipeline_blend_desc.SrcBlend = D3D12_BLEND_SRC_ALPHA;
   pipeline_blend_desc.DestBlend = D3D12_BLEND_INV_SRC_ALPHA;
   pipeline_blend_desc.BlendOp = D3D12_BLEND_OP_ADD;
-  pipeline_blend_desc.SrcBlendAlpha = D3D12_BLEND_SRC_ALPHA;
-  pipeline_blend_desc.DestBlendAlpha = D3D12_BLEND_INV_SRC_ALPHA;
+  // Don't change alpha (always 1).
+  pipeline_blend_desc.SrcBlendAlpha = D3D12_BLEND_ZERO;
+  pipeline_blend_desc.DestBlendAlpha = D3D12_BLEND_ONE;
   pipeline_blend_desc.BlendOpAlpha = D3D12_BLEND_OP_ADD;
-  pipeline_blend_desc.RenderTargetWriteMask = D3D12_COLOR_WRITE_ENABLE_ALL;
+  pipeline_blend_desc.RenderTargetWriteMask = D3D12_COLOR_WRITE_ENABLE_RED |
+                                              D3D12_COLOR_WRITE_ENABLE_GREEN |
+                                              D3D12_COLOR_WRITE_ENABLE_BLUE;
   pipeline_desc.SampleMask = UINT_MAX;
   pipeline_desc.RasterizerState.FillMode = D3D12_FILL_MODE_SOLID;
   pipeline_desc.RasterizerState.CullMode = D3D12_CULL_MODE_NONE;
@@ -295,6 +298,7 @@ bool D3D12ImmediateDrawer::Initialize() {
 
   // Reset the current state.
   current_command_list_ = nullptr;
+  batch_open_ = false;
 
   return true;
 }
@@ -415,6 +419,9 @@ void D3D12ImmediateDrawer::UpdateTexture(ImmediateTexture* texture,
 
 void D3D12ImmediateDrawer::Begin(int render_target_width,
                                  int render_target_height) {
+  assert_null(current_command_list_);
+  assert_false(batch_open_);
+
   auto device = context_.GetD3D12Provider().GetDevice();
 
   // Use the compositing command list.
@@ -479,13 +486,17 @@ void D3D12ImmediateDrawer::Begin(int render_target_width,
   viewport.MinDepth = 0.0f;
   viewport.MaxDepth = 1.0f;
   current_command_list_->RSSetViewports(1, &viewport);
+  current_scissor_.left = 0;
+  current_scissor_.top = 0;
+  current_scissor_.right = 0;
+  current_scissor_.bottom = 0;
 
   current_command_list_->SetGraphicsRootSignature(root_signature_);
   float viewport_inv_size[2];
   viewport_inv_size[0] = 1.0f / viewport.Width;
   viewport_inv_size[1] = 1.0f / viewport.Height;
   current_command_list_->SetGraphicsRoot32BitConstants(
-      UINT(RootParameter::kViewportInvSize), 2, viewport_inv_size, 0);
+      UINT(RootParameter::kViewportSizeInv), 2, viewport_inv_size, 0);
 
   current_primitive_topology_ = D3D_PRIMITIVE_TOPOLOGY_UNDEFINED;
   current_texture_ = nullptr;
@@ -493,21 +504,18 @@ void D3D12ImmediateDrawer::Begin(int render_target_width,
 }
 
 void D3D12ImmediateDrawer::BeginDrawBatch(const ImmediateDrawBatch& batch) {
+  assert_false(batch_open_);
   assert_not_null(current_command_list_);
-  if (current_command_list_ == nullptr) {
-    return;
-  }
-  uint64_t current_fence_value = context_.GetSwapCurrentFenceValue();
 
-  batch_open_ = false;
+  uint64_t current_fence_value = context_.GetSwapCurrentFenceValue();
 
   // Bind the vertices.
   D3D12_VERTEX_BUFFER_VIEW vertex_buffer_view;
   vertex_buffer_view.StrideInBytes = UINT(sizeof(ImmediateVertex));
   vertex_buffer_view.SizeInBytes =
-      batch.vertex_count * uint32_t(sizeof(ImmediateVertex));
+      UINT(sizeof(ImmediateVertex)) * batch.vertex_count;
   void* vertex_buffer_mapping = vertex_buffer_pool_->Request(
-      current_fence_value, vertex_buffer_view.SizeInBytes, sizeof(uint32_t),
+      current_fence_value, vertex_buffer_view.SizeInBytes, sizeof(float),
       nullptr, nullptr, &vertex_buffer_view.BufferLocation);
   if (vertex_buffer_mapping == nullptr) {
     XELOGE("Failed to get a buffer for {} vertices in the immediate drawer",
@@ -522,7 +530,7 @@ void D3D12ImmediateDrawer::BeginDrawBatch(const ImmediateDrawBatch& batch) {
   batch_has_index_buffer_ = batch.indices != nullptr;
   if (batch_has_index_buffer_) {
     D3D12_INDEX_BUFFER_VIEW index_buffer_view;
-    index_buffer_view.SizeInBytes = batch.index_count * sizeof(uint16_t);
+    index_buffer_view.SizeInBytes = UINT(sizeof(uint16_t)) * batch.index_count;
     index_buffer_view.Format = DXGI_FORMAT_R16_UINT;
     void* index_buffer_mapping = vertex_buffer_pool_->Request(
         current_fence_value, index_buffer_view.SizeInBytes, sizeof(uint16_t),
@@ -541,11 +549,6 @@ void D3D12ImmediateDrawer::BeginDrawBatch(const ImmediateDrawBatch& batch) {
 }
 
 void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
-  assert_not_null(current_command_list_);
-  if (current_command_list_ == nullptr) {
-    return;
-  }
-
   if (!batch_open_) {
     // Could be an error while obtaining the vertex and index buffers.
     return;
@@ -554,6 +557,32 @@ void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
   auto& provider = context_.GetD3D12Provider();
   auto device = provider.GetDevice();
 
+  // Set the scissor rectangle if enabled.
+  D3D12_RECT scissor;
+  if (draw.scissor) {
+    scissor.left = draw.scissor_rect[0];
+    scissor.top = current_render_target_height_ -
+                  (draw.scissor_rect[1] + draw.scissor_rect[3]);
+    scissor.right = scissor.left + draw.scissor_rect[2];
+    scissor.bottom = scissor.top + draw.scissor_rect[3];
+  } else {
+    scissor.left = 0;
+    scissor.top = 0;
+    scissor.right = current_render_target_width_;
+    scissor.bottom = current_render_target_height_;
+  }
+  if (scissor.right <= scissor.left || scissor.bottom <= scissor.top) {
+    // Nothing is visible (used as the default current_scissor_ value also).
+    return;
+  }
+  if (current_scissor_.left != scissor.left ||
+      current_scissor_.top != scissor.top ||
+      current_scissor_.right != scissor.right ||
+      current_scissor_.bottom != scissor.bottom) {
+    current_scissor_ = scissor;
+    current_command_list_->RSSetScissorRects(1, &scissor);
+  }
+
   // Bind the texture.
   auto texture = reinterpret_cast<D3D12ImmediateTexture*>(draw.texture_handle);
   ID3D12Resource* texture_resource;
@@ -580,6 +609,7 @@ void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
     current_command_list_->SetDescriptorHeaps(2, descriptor_heaps);
   }
   if (bind_texture) {
+    current_texture_ = texture;
     D3D12_SHADER_RESOURCE_VIEW_DESC texture_view_desc;
     texture_view_desc.Format = D3D12ImmediateTexture::kFormat;
     texture_view_desc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
@@ -599,7 +629,6 @@ void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
         provider.OffsetViewDescriptor(
             texture_descriptor_pool_->GetLastRequestHeapGPUStart(),
             texture_descriptor_index));
-    current_texture_ = texture;
   }
 
   // Bind the sampler.
@@ -616,11 +645,11 @@ void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
     sampler_index = SamplerIndex::kNearestClamp;
   }
   if (current_sampler_index_ != sampler_index) {
+    current_sampler_index_ = sampler_index;
     current_command_list_->SetGraphicsRootDescriptorTable(
         UINT(RootParameter::kSampler),
         provider.OffsetSamplerDescriptor(sampler_heap_gpu_start_,
                                          uint32_t(sampler_index)));
-    current_sampler_index_ = sampler_index;
   }
 
   // Set whether texture coordinates need to be restricted.
@@ -646,27 +675,11 @@ void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
       return;
   }
   if (current_primitive_topology_ != primitive_topology) {
+    current_primitive_topology_ = primitive_topology;
     current_command_list_->IASetPrimitiveTopology(primitive_topology);
     current_command_list_->SetPipelineState(pipeline);
-    current_primitive_topology_ = primitive_topology;
   }
 
-  // Set the scissor rectangle if enabled.
-  D3D12_RECT scissor;
-  if (draw.scissor) {
-    scissor.left = draw.scissor_rect[0];
-    scissor.top = current_render_target_height_ -
-                  (draw.scissor_rect[1] + draw.scissor_rect[3]);
-    scissor.right = scissor.left + draw.scissor_rect[2];
-    scissor.bottom = scissor.top + draw.scissor_rect[3];
-  } else {
-    scissor.left = 0;
-    scissor.top = 0;
-    scissor.right = current_render_target_width_;
-    scissor.bottom = current_render_target_height_;
-  }
-  current_command_list_->RSSetScissorRects(1, &scissor);
-
   // Draw.
   if (batch_has_index_buffer_) {
     current_command_list_->DrawIndexedInstanced(
@@ -678,7 +691,10 @@ void D3D12ImmediateDrawer::Draw(const ImmediateDraw& draw) {
 
 void D3D12ImmediateDrawer::EndDrawBatch() { batch_open_ = false; }
 
-void D3D12ImmediateDrawer::End() { current_command_list_ = nullptr; }
+void D3D12ImmediateDrawer::End() {
+  assert_false(batch_open_);
+  current_command_list_ = nullptr;
+}
 
 }  // namespace d3d12
 }  // namespace ui
diff --git a/src/xenia/ui/d3d12/d3d12_immediate_drawer.h b/src/xenia/ui/d3d12/d3d12_immediate_drawer.h
index ce3fbcd84..53ae81054 100644
--- a/src/xenia/ui/d3d12/d3d12_immediate_drawer.h
+++ b/src/xenia/ui/d3d12/d3d12_immediate_drawer.h
@@ -54,7 +54,7 @@ class D3D12ImmediateDrawer : public ImmediateDrawer {
     kRestrictTextureSamples,
     kTexture,
     kSampler,
-    kViewportInvSize,
+    kViewportSizeInv,
 
     kCount
   };
@@ -75,8 +75,8 @@ class D3D12ImmediateDrawer : public ImmediateDrawer {
   D3D12_CPU_DESCRIPTOR_HANDLE sampler_heap_cpu_start_;
   D3D12_GPU_DESCRIPTOR_HANDLE sampler_heap_gpu_start_;
 
-  std::unique_ptr<D3D12UploadBufferPool> vertex_buffer_pool_ = nullptr;
-  std::unique_ptr<D3D12DescriptorHeapPool> texture_descriptor_pool_ = nullptr;
+  std::unique_ptr<D3D12UploadBufferPool> vertex_buffer_pool_;
+  std::unique_ptr<D3D12DescriptorHeapPool> texture_descriptor_pool_;
   uint64_t texture_descriptor_pool_heap_index_;
 
   struct PendingTextureUpload {
@@ -95,6 +95,7 @@ class D3D12ImmediateDrawer : public ImmediateDrawer {
   int current_render_target_width_, current_render_target_height_;
   bool batch_open_ = false;
   bool batch_has_index_buffer_;
+  D3D12_RECT current_scissor_;
   D3D_PRIMITIVE_TOPOLOGY current_primitive_topology_;
   ImmediateTexture* current_texture_;
   SamplerIndex current_sampler_index_;
diff --git a/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.cso b/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.cso
index 81917561c..5a4a854bf 100644
Binary files a/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.cso and b/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.cso differ
diff --git a/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.h b/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.h
index 03a09ce46..074264ab7 100644
--- a/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.h
+++ b/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.h
@@ -1,8 +1,8 @@
 // generated from `xb buildhlsl`
 // source: immediate.vs.hlsl
 const uint8_t immediate_vs[] = {
-    0x44, 0x58, 0x42, 0x43, 0x6A, 0xC1, 0xC3, 0xE8, 0xA7, 0x09, 0x30, 0xF3,
-    0x48, 0x84, 0xA3, 0x3C, 0x7D, 0x90, 0xA1, 0x09, 0x01, 0x00, 0x00, 0x00,
+    0x44, 0x58, 0x42, 0x43, 0xC3, 0x3E, 0x79, 0xCB, 0x09, 0x65, 0x04, 0xF0,
+    0x71, 0x43, 0x47, 0x45, 0xAC, 0xE1, 0xA9, 0x03, 0x01, 0x00, 0x00, 0x00,
     0x00, 0x04, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
     0x54, 0x01, 0x00, 0x00, 0xC4, 0x01, 0x00, 0x00, 0x34, 0x02, 0x00, 0x00,
     0x64, 0x03, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0x18, 0x01, 0x00, 0x00,
@@ -21,8 +21,8 @@ const uint8_t immediate_vs[] = {
     0x08, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0xCC, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00,
     0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x76,
-    0x69, 0x65, 0x77, 0x70, 0x6F, 0x72, 0x74, 0x5F, 0x69, 0x6E, 0x76, 0x5F,
-    0x73, 0x69, 0x7A, 0x65, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x32, 0x00,
+    0x69, 0x65, 0x77, 0x70, 0x6F, 0x72, 0x74, 0x5F, 0x73, 0x69, 0x7A, 0x65,
+    0x5F, 0x69, 0x6E, 0x76, 0x00, 0x66, 0x6C, 0x6F, 0x61, 0x74, 0x32, 0x00,
     0x01, 0x00, 0x03, 0x00, 0x01, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
     0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xC5, 0x00, 0x00, 0x00,
diff --git a/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.txt b/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.txt
index fe5f8134f..fbda7ff07 100644
--- a/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.txt
+++ b/src/xenia/ui/d3d12/shaders/dxbc/immediate_vs.txt
@@ -7,7 +7,7 @@
 // cbuffer $Globals
 // {
 //
-//   float2 xe_viewport_inv_size;       // Offset:    0 Size:     8
+//   float2 xe_viewport_size_inv;       // Offset:    0 Size:     8
 //
 // }
 //
diff --git a/src/xenia/ui/d3d12/shaders/immediate.vs.hlsl b/src/xenia/ui/d3d12/shaders/immediate.vs.hlsl
index 7c3e22530..a79da7b1c 100644
--- a/src/xenia/ui/d3d12/shaders/immediate.vs.hlsl
+++ b/src/xenia/ui/d3d12/shaders/immediate.vs.hlsl
@@ -1,4 +1,4 @@
-float2 xe_viewport_inv_size : register(b0);
+float2 xe_viewport_size_inv : register(b0);
 
 struct XeVertexShaderInput {
   float2 position : POSITION;
@@ -15,7 +15,7 @@ struct XeVertexShaderOutput {
 XeVertexShaderOutput main(XeVertexShaderInput input) {
   XeVertexShaderOutput output;
   output.position = float4(
-      input.position * xe_viewport_inv_size * float2(2.0, -2.0) +
+      input.position * xe_viewport_size_inv * float2(2.0, -2.0) +
       float2(-1.0, 1.0), 0.0, 1.0);
   output.texcoord = input.texcoord;
   output.color = input.color;