From 57e8f05234f2cf8ee29cffd3ab8fe57de38479b8 Mon Sep 17 00:00:00 2001
From: Triang3l <triang3l@yandex.ru>
Date: Tue, 29 Jan 2019 14:39:04 +0300
Subject: [PATCH] [D3D12] Make RG16/RGBA16 RTVs range switchable

---
 src/xenia/gpu/d3d12/d3d12_command_processor.cc | 5 +++--
 src/xenia/gpu/d3d12/render_target_cache.cc     | 8 +++++++-
 src/xenia/gpu/d3d12/render_target_cache.h      | 4 ++++
 3 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/src/xenia/gpu/d3d12/d3d12_command_processor.cc b/src/xenia/gpu/d3d12/d3d12_command_processor.cc
index 429680790..956e54b7b 100644
--- a/src/xenia/gpu/d3d12/d3d12_command_processor.cc
+++ b/src/xenia/gpu/d3d12/d3d12_command_processor.cc
@@ -2319,9 +2319,10 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
       // -32...32 range and expect shaders to give -32...32 values, but they're
       // emulated using normalized RG16/RGBA16 when not using the ROV, so the
       // value returned from the shader needs to be divided by 32 (blending will
-      // be incorrect in this case, but there's no other way without using ROV).
+      // be incorrect in this case, but there's no other way without using ROV,
+      // though there's an option to limit the range to -1...1).
       // http://www.students.science.uu.nl/~3220516/advancedgraphics/papers/inferred_lighting.pdf
-      if (!IsROVUsedForEDRAM()) {
+      if (!IsROVUsedForEDRAM() && FLAGS_d3d12_16bit_rtv_full_range) {
         color_exp_bias -= 5;
       }
     }
diff --git a/src/xenia/gpu/d3d12/render_target_cache.cc b/src/xenia/gpu/d3d12/render_target_cache.cc
index 4399116a1..c53cf4271 100644
--- a/src/xenia/gpu/d3d12/render_target_cache.cc
+++ b/src/xenia/gpu/d3d12/render_target_cache.cc
@@ -25,6 +25,9 @@
 #include "xenia/gpu/texture_util.h"
 #include "xenia/ui/d3d12/d3d12_util.h"
 
+DEFINE_bool(d3d12_16bit_rtv_full_range, true,
+            "Use full -32...32 range for RG16 and RGBA16 render targets "
+            "(at the expense of blending correctness) without ROV.");
 DEFINE_bool(d3d12_resolution_scale_resolve_edge_clamp, true,
             "When using resolution scale, apply the hack that duplicates the "
             "right/lower subpixel in the left and top sides of render target "
@@ -1171,7 +1174,10 @@ bool RenderTargetCache::ResolveCopy(SharedMemory* shared_memory,
       // sampling the host render target gives 1/32 of what is actually stored
       // there on the guest side.
       // http://www.students.science.uu.nl/~3220516/advancedgraphics/papers/inferred_lighting.pdf
-      dest_exp_bias += 5;
+      if (command_processor_->IsROVUsedForEDRAM() ||
+          FLAGS_d3d12_16bit_rtv_full_range) {
+        dest_exp_bias += 5;
+      }
     }
   }
   bool dest_swap = !is_depth && ((dest_info >> 24) & 0x1);
diff --git a/src/xenia/gpu/d3d12/render_target_cache.h b/src/xenia/gpu/d3d12/render_target_cache.h
index 38bddc32b..f3baf49b3 100644
--- a/src/xenia/gpu/d3d12/render_target_cache.h
+++ b/src/xenia/gpu/d3d12/render_target_cache.h
@@ -10,6 +10,8 @@
 #ifndef XENIA_GPU_D3D12_RENDER_TARGET_CACHE_H_
 #define XENIA_GPU_D3D12_RENDER_TARGET_CACHE_H_
 
+#include <gflags/gflags.h>
+
 #include <unordered_map>
 
 #include "xenia/gpu/d3d12/d3d12_shader.h"
@@ -20,6 +22,8 @@
 #include "xenia/memory.h"
 #include "xenia/ui/d3d12/d3d12_api.h"
 
+DECLARE_bool(d3d12_16bit_rtv_full_range);
+
 namespace xe {
 namespace gpu {
 namespace d3d12 {