From 26d78c7cd1baa781b2efde4740f76e62d664e1e1 Mon Sep 17 00:00:00 2001
From: "Unknown W. Brackets" <checkins@unknownbrackets.org>
Date: Sat, 16 Mar 2019 16:17:24 -0700
Subject: [PATCH] interp: Cleanup prefix/size in vcrsp/vqmul.

---
 Core/MIPS/IR/IRCompVFPU.cpp | 10 +++----
 Core/MIPS/MIPSIntVFPU.cpp   | 57 ++++++++++++++++++++++++++++---------
 2 files changed, 49 insertions(+), 18 deletions(-)

diff --git a/Core/MIPS/IR/IRCompVFPU.cpp b/Core/MIPS/IR/IRCompVFPU.cpp
index 434c4fb57e..aaeb70be85 100644
--- a/Core/MIPS/IR/IRCompVFPU.cpp
+++ b/Core/MIPS/IR/IRCompVFPU.cpp
@@ -1628,13 +1628,12 @@ namespace MIPSComp {
 
 	void IRFrontend::Comp_VCrossQuat(MIPSOpcode op) {
 		CONDITIONAL_DISABLE(VFPU_VEC);
-		// TODO: Does this instruction even look at prefixes at all?
-		if (js.HasUnknownPrefix())
+		if (!js.HasNoPrefix())
 			DISABLE;
 
-		// Vector cross product (n = 3)
+		// Vector cross product (n = 3, weird prefixes)
 		// d[0 .. 2] = s[0 .. 2] X t[0 .. 2]
-		// Vector quaternion product (n = 4)
+		// Vector quaternion product (n = 4, weird prefixes)
 		// d[0 .. 2] = t[0 .. 2] X s[0 .. 2] + s[3] * t[0 .. 2] + t[3] * s[0 .. 2]
 		// d[3] = s[3]*t[3] - s[0 .. 2] dot t[0 .. 3]
 
@@ -1674,13 +1673,14 @@ namespace MIPSComp {
 			ir.Write(IROp::FSub, tempregs[2], temp0, temp1);
 		} else if (sz == V_Quad) {
 			DISABLE;
+		} else {
+			DISABLE;
 		}
 
 		for (int i = 0; i < n; i++) {
 			if (tempregs[i] != dregs[i])
 				ir.Write(IROp::FMov, dregs[i], tempregs[i]);
 		}
-		// No D prefix supported
 	}
 
 	void IRFrontend::Comp_Vcmp(MIPSOpcode op) {
diff --git a/Core/MIPS/MIPSIntVFPU.cpp b/Core/MIPS/MIPSIntVFPU.cpp
index 84917eb83c..e80ce283bb 100644
--- a/Core/MIPS/MIPSIntVFPU.cpp
+++ b/Core/MIPS/MIPSIntVFPU.cpp
@@ -1956,38 +1956,69 @@ namespace MIPSInt
 		EatPrefixes();
 	}
 	
-	void Int_CrossQuat(MIPSOpcode op)
-	{
+	void Int_CrossQuat(MIPSOpcode op) {
+		float s[4]{}, t[4]{}, d[4];
 		int vd = _VD;
 		int vs = _VS;
 		int vt = _VT;
 		VectorSize sz = GetVecSize(op);
-		float s[4];
-		float t[4];
-		float d[4];
+		int n = GetNumVectorElements(sz);
 		ReadVector(s, sz, vs);
 		ReadVector(t, sz, vt);
-		switch (sz)
-		{
+
+		u32 tprefixRemove = VFPU_ANY_SWIZZLE() | VFPU_NEGATE(1, 1, 1, 1);
+		u32 tprefixAdd;
+
+		switch (sz) {
 		case V_Triple:  // vcrsp.t
 			d[0] = s[1]*t[2] - s[2]*t[1];
 			d[1] = s[2]*t[0] - s[0]*t[2];
-			d[2] = s[0]*t[1] - s[1]*t[0];
+
+			// T prefix forces siwzzle and negate, can be used to have weird constants.
+			tprefixAdd = VFPU_SWIZZLE(1, 0, 3, 2) | VFPU_NEGATE(0, 1, 0, 0);
+			ApplyPrefixST(t, VFPURewritePrefix(VFPU_CTRL_TPREFIX, tprefixRemove, tprefixAdd), V_Quad);
+			ApplySwizzleS(s, V_Quad);
+			d[2] = s[0] * t[0] + s[1] * t[1] + s[2] * t[2] + s[3] * t[3];
 			break;
 
 		case V_Quad:   // vqmul.q
 			d[0] = s[0]*t[3] + s[1]*t[2] - s[2]*t[1] + s[3]*t[0];
 			d[1] = -s[0]*t[2] + s[1]*t[3] + s[2]*t[0] + s[3]*t[1];
 			d[2] = s[0]*t[1] - s[1]*t[0] + s[2]*t[3] + s[3]*t[2];
-			d[3] = -s[0]*t[0] - s[1]*t[1] - s[2]*t[2] + s[3]*t[3];
+
+			// T prefix forces siwzzle and negate, can be used to have weird constants.
+			tprefixAdd = VFPU_SWIZZLE(0, 1, 2, 3) | VFPU_NEGATE(1, 1, 1, 0);
+			ApplyPrefixST(t, VFPURewritePrefix(VFPU_CTRL_TPREFIX, tprefixRemove, tprefixAdd), V_Quad);
+			ApplySwizzleS(s, sz);
+			d[3] = s[0] * t[0] + s[1] * t[1] + s[2] * t[2] + s[3] * t[3];
 			break;
 
-		default:
-			Reporting::ReportMessage("CrossQuat instruction with wrong size");
-			_dbg_assert_msg_(CPU,0,"Trying to interpret instruction that can't be interpreted");
+		case V_Pair:
+			// t swizzles invalid so the multiply is always zero.
 			d[0] = 0;
-			d[1] = 0;
+
+			tprefixAdd = VFPU_SWIZZLE(0, 0, 0, 0) | VFPU_NEGATE(0, 0, 0, 0);
+			ApplyPrefixST(t, VFPURewritePrefix(VFPU_CTRL_TPREFIX, tprefixRemove, tprefixAdd), V_Quad);
+			ApplySwizzleS(s, V_Quad);
+			// It's possible to populate a value by swizzling s[2].
+			d[1] = s[2] * t[2];
 			break;
+
+		case V_Single:
+			// t swizzles invalid so the multiply is always zero.
+			d[0] = 0;
+			break;
+		}
+
+		// D prefix applies to the last element only (mask and sat) for pair and larger.
+		if (sz != V_Single) {
+			u32 lastmask = (currentMIPS->vfpuCtrl[VFPU_CTRL_DPREFIX] & (1 << 8)) << (n - 1);
+			u32 lastsat = (currentMIPS->vfpuCtrl[VFPU_CTRL_DPREFIX] & 3) << (n + n - 2);
+			currentMIPS->vfpuCtrl[VFPU_CTRL_DPREFIX] = lastmask | lastsat;
+			ApplyPrefixD(d, sz);
+		} else {
+			// Single always seems to write out zero.
+			currentMIPS->vfpuCtrl[VFPU_CTRL_DPREFIX] = 0;
 		}
 		WriteVector(d, sz, vd);
 		PC += 4;