From 06fcdef96825a9dd4f94dcdda605d696227a8758 Mon Sep 17 00:00:00 2001
From: Triang3l <hwguy.siplus@gmail.com>
Date: Thu, 31 May 2018 16:17:27 +0300
Subject: [PATCH] [CPU] Bring back saturation in D3DCOLOR vpkd3d128

---
 src/xenia/cpu/backend/x64/x64_emitter.cc   |  1 +
 src/xenia/cpu/backend/x64/x64_emitter.h    |  1 +
 src/xenia/cpu/backend/x64/x64_sequences.cc | 24 +++++++++-------------
 3 files changed, 12 insertions(+), 14 deletions(-)

diff --git a/src/xenia/cpu/backend/x64/x64_emitter.cc b/src/xenia/cpu/backend/x64/x64_emitter.cc
index 6479da0e3..bff56de0a 100644
--- a/src/xenia/cpu/backend/x64/x64_emitter.cc
+++ b/src/xenia/cpu/backend/x64/x64_emitter.cc
@@ -626,6 +626,7 @@ static const vec128_t xmm_consts[] = {
     vec128i(0x01000302u, 0x05040706u, 0x09080B0Au, 0x0D0C0F0Eu),
     /* XMMPermuteControl15    */ vec128b(15),
     /* XMMPermuteByteMask     */ vec128b(0x1F),
+    /* XMMPackD3DCOLORSat     */ vec128i(0x404000FFu),
     /* XMMPackD3DCOLOR        */
     vec128i(0xFFFFFFFFu, 0xFFFFFFFFu, 0xFFFFFFFFu, 0x0C000408u),
     /* XMMUnpackD3DCOLOR      */
diff --git a/src/xenia/cpu/backend/x64/x64_emitter.h b/src/xenia/cpu/backend/x64/x64_emitter.h
index 039d67509..5d9b91241 100644
--- a/src/xenia/cpu/backend/x64/x64_emitter.h
+++ b/src/xenia/cpu/backend/x64/x64_emitter.h
@@ -64,6 +64,7 @@ enum XmmConst {
   XMMByteOrderMask,
   XMMPermuteControl15,
   XMMPermuteByteMask,
+  XMMPackD3DCOLORSat,
   XMMPackD3DCOLOR,
   XMMUnpackD3DCOLOR,
   XMMPackFLOAT16_2,
diff --git a/src/xenia/cpu/backend/x64/x64_sequences.cc b/src/xenia/cpu/backend/x64/x64_sequences.cc
index 84b91ea10..b8b337230 100644
--- a/src/xenia/cpu/backend/x64/x64_sequences.cc
+++ b/src/xenia/cpu/backend/x64/x64_sequences.cc
@@ -6964,24 +6964,20 @@ struct PACK : Sequence<PACK, I<OPCODE_PACK, V128Op, V128Op, V128Op>> {
   }
   static void EmitD3DCOLOR(X64Emitter& e, const EmitArgType& i) {
     assert_true(i.src2.value->IsConstantZero());
-    // No saturation done here.
-    // Unpacking D3DCOLOR gives (1.0f | bits), or from 3F800000 to 3F8000FF.
-    // However, you can pack 3.0f + (value / (float) (1 << 22)), which creates
-    // a number between 40400000 and 404000FF:
-    // https://github.com/ValveSoftware/source-sdk-2013/blob/master/sp/src/public/pixelwriter.h#L648
-    // With saturation, you will get 0 when re-packing after unpacking.
-    // The above code also has to perform clamping explicitly.
-
+    // Saturate to [3,3....] so that only values between 3...[00] and 3...[FF]
+    // are valid.
+    if (i.src1.is_constant) {
+      e.LoadConstantXmm(i.dest, i.src1.constant());
+      e.vminps(i.dest, i.dest, e.GetXmmConstPtr(XMMPackD3DCOLORSat));
+    } else {
+      e.vminps(i.dest, i.src1, e.GetXmmConstPtr(XMMPackD3DCOLORSat));
+    }
+    e.vmaxps(i.dest, i.dest, e.GetXmmConstPtr(XMM3333));
     // Extract bytes.
     // RGBA (XYZW) -> ARGB (WXYZ)
     // w = ((src1.uw & 0xFF) << 24) | ((src1.ux & 0xFF) << 16) |
     //     ((src1.uy & 0xFF) << 8) | (src1.uz & 0xFF)
-    if (i.src1.is_constant) {
-      e.LoadConstantXmm(i.dest, i.src1.constant());
-      e.vpshufb(i.dest, i.dest, e.GetXmmConstPtr(XMMPackD3DCOLOR));
-    } else {
-      e.vpshufb(i.dest, i.src1, e.GetXmmConstPtr(XMMPackD3DCOLOR));
-    }
+    e.vpshufb(i.dest, i.dest, e.GetXmmConstPtr(XMMPackD3DCOLOR));
   }
   static __m128i EmulateFLOAT16_2(void*, __m128 src1) {
     alignas(16) float a[4];