[D3D12] Add red/blue swap to EDRAM loads and describe resolve modes

src/xenia/gpu/d3d12/shaders/edram_load_color_32bpp.cs.hlsl

@@ -8,6 +8,15 @@ void main(uint3 xe_group_id : SV_GroupID,
   tile_dword_index.x *= 4u;
   uint4 pixels = xe_edram_load_store_source.Load4(
       XeEDRAMOffset(xe_group_id.xy, tile_dword_index));
+  if (xe_edram_swap_red_blue != 0u) {
+    // Not a very long shift, just 16 or 20.
+    uint blue_shift = xe_edram_swap_red_blue >> 16u;
+    uint red_mask = xe_edram_swap_red_blue & 0xFFFFu;
+    uint blue_mask = red_mask << blue_shift;
+    pixels = (pixels & ~(red_mask | blue_mask)) |
+             ((pixels & red_mask) << blue_shift) |
+             ((pixels >> blue_shift) & red_mask);
+  }
   uint rt_offset = xe_thread_id.y * xe_edram_rt_color_depth_pitch +
                    xe_thread_id.x * 16u;
   xe_edram_load_store_dest.Store4(rt_offset, pixels);

src/xenia/gpu/d3d12/shaders/edram_load_color_64bpp.cs.hlsl

@@ -13,6 +13,11 @@ void main(uint3 xe_group_id : SV_GroupID,
   }
   uint4 pixels = xe_edram_load_store_source.Load4(
       XeEDRAMOffset(xe_group_id.xy, tile_dword_index));
+  if (xe_edram_swap_red_blue != 0u) {
+    // The only 64-bit formats with a blue component are 16_16_16_16 and
+    // 16_16_16_16_FLOAT.
+    pixels = (pixels.yxwz & 0xFFFFu) | (pixels & 0xFFFF0000u);
+  }
   uint rt_offset = xe_thread_id.y * xe_edram_rt_color_depth_pitch +
                    xe_thread_id.x * 16u;
   xe_edram_load_store_dest.Store4(rt_offset, pixels);

src/xenia/gpu/d3d12/shaders/edram_load_color_7e3.cs.hlsl

@@ -14,6 +14,10 @@ void main(uint3 xe_group_id : SV_GroupID,
   uint4 pixels_f16u32_packed =
       uint4(pixel_0_f16u32.xz, pixel_1_f16u32.xz) |
       (uint4(pixel_0_f16u32.yw, pixel_1_f16u32.yw) << 16u);
+  if (xe_edram_swap_red_blue != 0u) {
+    pixels_f16u32_packed = (pixels_f16u32_packed.yxwz & 0xFFFFu) |
+                           (pixels_f16u32_packed & 0xFFFF0000u);
+  }
   uint rt_offset = xe_thread_id.y * xe_edram_rt_color_depth_pitch +
                    xe_thread_id.x * 16u;
   xe_edram_load_store_dest.Store4(rt_offset, pixels_f16u32_packed);

src/xenia/gpu/d3d12/shaders/edram_load_store.hlsli

@@ -5,9 +5,13 @@ cbuffer XeEDRAMLoadStoreConstants : register(b0) {
   uint xe_edram_base_tiles;
   uint xe_edram_pitch_tiles;
   uint xe_edram_rt_color_depth_pitch;
-  uint xe_edram_rt_stencil_offset;
+  uint xe_edram_rt_stencil_offset_or_swap_red_blue;
   uint xe_edram_rt_stencil_pitch;
 };
+#define xe_edram_rt_stencil_offset xe_edram_rt_stencil_offset_or_swap_red_blue
+// For loads only. How exactly it's handled depends on the specific load shader,
+// but 0 always means red and blue shouldn't be swapped.
+#define xe_edram_swap_red_blue xe_edram_rt_stencil_offset_or_swap_red_blue
 
 ByteAddressBuffer xe_edram_load_store_source : register(t0);
 RWByteAddressBuffer xe_edram_load_store_dest : register(u0);