Directly check PEB for IsDebuggerAttached

Add constexpr getters to magicdiv class so it can be used from jitted x64/dxbc
Track the guest return address as well for guest/host sync, if multiple entries have the same guest stack find the first one with a matching guest retaddr. this fixes epic mickey 2 (which the previous guest-stack change had allowed to go ingame for a bit) and potentially also a crash in fable3.
Break if under debugger when stackpoints are overflowed

Add much more useful output for host exceptions, print out xenia_canary.exe relative offsets if exception is in module, formatmessage for ntstatus/win32err, strerror

Minor d3d12 microoptimization, instead of doing SetEventOnCompletion + WaitForSingleObject do SetEventOnCompletion w/ nullptr so that the wait happens in kernel mode, avoiding two extra context switches

add unimplemented kernel functions:
ExAllocatePoolWithTag
ObReferenceObject

ObDereferenceObject has no return value.
Log a message when ObDereferenceObject/Reference receive unregistered guest kernel objects
gave ObLookupThreadByThreadId its correct error status
hoist object_types initialization out of ObReferenceObjectByHandle
Fix out parameter values on error for a few kernel funcs
add note about msr to KeSetCurrentStackPointers
add X_STATUS_OBJECT_TYPE_MISMATCH check for xeNtSetEvent
add msr_mask field to X_KPCR

src/xenia/gpu/d3d12/d3d12_command_processor.cc

@@ -3124,7 +3124,6 @@ bool D3D12CommandProcessor::IssueCopy_ReadbackResolvePath() {
             auto physaddr = memory_->TranslatePhysical(written_address);
             memory::vastcpy(physaddr, (uint8_t*)readback_mapping,
                             written_length);
-            // XEDmaCpy(physaddr, readback_mapping, written_length);
             D3D12_RANGE readback_write_range = {};
             readback_buffer->Unmap(0, &readback_write_range);
           }
@@ -3152,9 +3151,7 @@ void D3D12CommandProcessor::CheckSubmissionFence(uint64_t await_submission) {
               direct_queue->Signal(queue_operations_since_submission_fence_,
                                    fence_value) &&
               SUCCEEDED(queue_operations_since_submission_fence_
-                            ->SetEventOnCompletion(fence_value,
-                                                   fence_completion_event_)))) {
-        WaitForSingleObject(fence_completion_event_, INFINITE);
+                            ->SetEventOnCompletion(fence_value, nullptr)))) {
         queue_operations_done_since_submission_signal_ = false;
       } else {
         XELOGE(
@@ -3170,9 +3167,8 @@ void D3D12CommandProcessor::CheckSubmissionFence(uint64_t await_submission) {
   uint64_t submission_completed_before = submission_completed_;
   submission_completed_ = submission_fence_->GetCompletedValue();
   if (submission_completed_ < await_submission) {
-    if (SUCCEEDED(submission_fence_->SetEventOnCompletion(
-            await_submission, fence_completion_event_))) {
-      WaitForSingleObject(fence_completion_event_, INFINITE);
+    if (SUCCEEDED(submission_fence_->SetEventOnCompletion(await_submission,
+                                                          nullptr))) {
       submission_completed_ = submission_fence_->GetCompletedValue();
     }
   }

src/xenia/gpu/d3d12/d3d12_render_target_cache.cc

@@ -43,6 +43,7 @@ DEFINE_bool(
     "Allow stencil reference output usage on Direct3D 12 on Intel GPUs - not "
     "working on UHD Graphics 630 as of March 2021 (driver 27.20.0100.8336).",
     "GPU");
+DEFINE_bool(no_discard_stencil_in_transfer_pipelines, false, "bleh", "GPU");
 // TODO(Triang3l): Make ROV the default when it's optimized better (for
 // instance, using static shader modifications to pass render target
 // parameters).
@@ -2940,7 +2941,7 @@ D3D12RenderTargetCache::GetOrCreateTransferPipelines(TransferShaderKey key) {
   // r0.xy = destination pixel XY index within the 32bpp tile
   // r0.zw = 32bpp tile XY index
   a.OpUDiv(dxbc::Dest::R(0, 0b1100), dxbc::Dest::R(0, 0b0011),
-           dxbc::Src::R(0, 0b01000100),
+           dxbc::Src::R(0, dxbc::Src::kXYXY),
            dxbc::Src::LU(dest_tile_width_pixels, dest_tile_height_pixels,
                          dest_tile_width_pixels, dest_tile_height_pixels));
 
@@ -4189,12 +4190,14 @@ D3D12RenderTargetCache::GetOrCreateTransferPipelines(TransferShaderKey key) {
         break;
       case TransferOutput::kStencilBit:
         // Discard the sample if the needed stencil bit is not set.
-        assert_true(cbuffer_index_stencil_mask != UINT32_MAX);
-        a.OpAnd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
-                dxbc::Src::CB(cbuffer_index_stencil_mask,
-                              kTransferCBVRegisterStencilMask, 0,
-                              dxbc::Src::kXXXX));
-        a.OpDiscard(false, dxbc::Src::R(0, dxbc::Src::kXXXX));
+        if (!cvars::no_discard_stencil_in_transfer_pipelines) {
+          assert_true(cbuffer_index_stencil_mask != UINT32_MAX);
+          a.OpAnd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
+                  dxbc::Src::CB(cbuffer_index_stencil_mask,
+                                kTransferCBVRegisterStencilMask, 0,
+                                dxbc::Src::kXXXX));
+          a.OpDiscard(false, dxbc::Src::R(0, dxbc::Src::kXXXX));
+        }
         break;
     }
   }

src/xenia/gpu/dxbc.h

@@ -1046,6 +1046,7 @@ struct Src : OperandAddress {
     kYYYY = 0b01010101,
     kZZZZ = 0b10101010,
     kWWWW = 0b11111111,
+    kXYXY = 0b01000100
   };
 
   // Ignored for 0-component and 1-component operand types.