From 1a1c161a0d9bf5eea8a0e76e0c49e051f3a75d60 Mon Sep 17 00:00:00 2001
From: Henrik Rydgard <hrydgard@gmail.com>
Date: Sat, 27 Apr 2013 20:52:42 +0200
Subject: [PATCH] Implement vmin/vmax in x86 jit, slots right into VecDo3

---
 Core/MIPS/ARM/ArmCompVFPU.cpp |  4 ----
 Core/MIPS/ARM/ArmJit.h        |  1 -
 Core/MIPS/MIPSIntVFPU.cpp     |  2 +-
 Core/MIPS/MIPSTables.cpp      |  4 ++--
 Core/MIPS/x86/CompVFPU.cpp    | 21 +++++++++++++++++----
 Core/MIPS/x86/Jit.h           |  1 -
 6 files changed, 20 insertions(+), 13 deletions(-)

diff --git a/Core/MIPS/ARM/ArmCompVFPU.cpp b/Core/MIPS/ARM/ArmCompVFPU.cpp
index 315542d7bc..b177f697d3 100644
--- a/Core/MIPS/ARM/ArmCompVFPU.cpp
+++ b/Core/MIPS/ARM/ArmCompVFPU.cpp
@@ -730,10 +730,6 @@ namespace MIPSComp
 		DISABLE;
 	}
 
-	void Jit::Comp_VMinMax(u32 op) {
-		DISABLE;
-	}
-
 	void Jit::Comp_VHdp(u32 op) {
 		DISABLE;
 	}
diff --git a/Core/MIPS/ARM/ArmJit.h b/Core/MIPS/ARM/ArmJit.h
index af0a286a05..efd2b76bcc 100644
--- a/Core/MIPS/ARM/ArmJit.h
+++ b/Core/MIPS/ARM/ArmJit.h
@@ -200,7 +200,6 @@ public:
 	void Comp_VHdp(u32 op);
 	void Comp_VCrs(u32 op);
 	void Comp_VDet(u32 op);
-	void Comp_VMinMax(u32 op);
 	void Comp_Vi2x(u32 op);
 	void Comp_Vx2i(u32 op);
 	void Comp_Vf2i(u32 op);
diff --git a/Core/MIPS/MIPSIntVFPU.cpp b/Core/MIPS/MIPSIntVFPU.cpp
index 485c5475af..8fcde3df42 100644
--- a/Core/MIPS/MIPSIntVFPU.cpp
+++ b/Core/MIPS/MIPSIntVFPU.cpp
@@ -1052,7 +1052,7 @@ namespace MIPSInt
 		int n = GetNumVectorElements(sz);
 		if (op & 0x10000)
 		{
-			//vbfy2
+			// vbfy2
 			d[0] = s[0] + s[2];
 			d[1] = s[1] + s[3];
 			d[2] = s[0] - s[2];
diff --git a/Core/MIPS/MIPSTables.cpp b/Core/MIPS/MIPSTables.cpp
index 7e7640cc2b..209351f4df 100644
--- a/Core/MIPS/MIPSTables.cpp
+++ b/Core/MIPS/MIPSTables.cpp
@@ -502,8 +502,8 @@ const MIPSInstruction tableVFPU3[8] = //011011 xxx
 {
 	INSTR("vcmp",&Jit::Comp_Generic, Dis_Vcmp, Int_Vcmp, IS_VFPU|OUT_EAT_PREFIX),
 	{-2},
-	INSTR("vmin",&Jit::Comp_VMinMax, Dis_VectorSet3, Int_Vminmax, IS_VFPU|OUT_EAT_PREFIX),
-	INSTR("vmax",&Jit::Comp_VMinMax, Dis_VectorSet3, Int_Vminmax, IS_VFPU|OUT_EAT_PREFIX),
+	INSTR("vmin",&Jit::Comp_VecDo3, Dis_VectorSet3, Int_Vminmax, IS_VFPU|OUT_EAT_PREFIX),
+	INSTR("vmax",&Jit::Comp_VecDo3, Dis_VectorSet3, Int_Vminmax, IS_VFPU|OUT_EAT_PREFIX),
 	{-2}, 
 	INSTR("vscmp",&Jit::Comp_Generic, Dis_VectorSet3, Int_Vscmp, IS_VFPU|OUT_EAT_PREFIX),
 	INSTR("vsge",&Jit::Comp_Generic, Dis_VectorSet3, Int_Vsge, IS_VFPU|OUT_EAT_PREFIX), 
diff --git a/Core/MIPS/x86/CompVFPU.cpp b/Core/MIPS/x86/CompVFPU.cpp
index a63b03cba0..5efe7ec787 100644
--- a/Core/MIPS/x86/CompVFPU.cpp
+++ b/Core/MIPS/x86/CompVFPU.cpp
@@ -435,13 +435,27 @@ void Jit::Comp_VecDo3(u32 op) {
 		}
 		break;
 	case 25: //VFPU1
-		switch ((op >> 23)&7)
+		switch ((op >> 23) & 7)
 		{
 		case 0: // d[i] = s[i] * t[i]; break; //vmul
 			xmmop = &XEmitter::MULSS;
 			break;
 		}
 		break;
+	case 27: //VFPU3
+		switch ((op >> 23) & 3)
+		{
+		case 2:  // vmin
+			xmmop = &XEmitter::MINSS;
+			break;
+		case 3:  // vmax
+			xmmop = &XEmitter::MAXSS;
+			break;
+		}
+		break;
+	default:
+		_dbg_assert_msg_(CPU,0,"invalid VecDo3");
+		break;
 	}
 
 	if (xmmop == NULL)
@@ -484,8 +498,10 @@ void Jit::Comp_VecDo3(u32 op) {
 		if (!fpr.V(sregs[i]).IsSimpleReg(tempxregs[i]))
 			MOVSS(tempxregs[i], fpr.V(sregs[i]));
 	}
+
 	for (int i = 0; i < n; ++i)
 		(this->*xmmop)(tempxregs[i], fpr.V(tregs[i]));
+
 	for (int i = 0; i < n; ++i)
 	{
 		if (!fpr.V(dregs[i]).IsSimpleReg(tempxregs[i]))
@@ -934,9 +950,6 @@ void Jit::Comp_Vtfm(u32 op) {
 	fpr.ReleaseSpillLocks();
 }
 
-void Jit::Comp_VMinMax(u32 op) {
-	DISABLE;
-}
 
 void Jit::Comp_VHdp(u32 op) {
 	DISABLE;
diff --git a/Core/MIPS/x86/Jit.h b/Core/MIPS/x86/Jit.h
index 8a38fc9ed1..705faedddb 100644
--- a/Core/MIPS/x86/Jit.h
+++ b/Core/MIPS/x86/Jit.h
@@ -210,7 +210,6 @@ public:
 	void Comp_VHdp(u32 op);
 	void Comp_VCrs(u32 op);
 	void Comp_VDet(u32 op);
-	void Comp_VMinMax(u32 op);
 	void Comp_Vi2x(u32 op);
 	void Comp_Vx2i(u32 op);
 	void Comp_Vf2i(u32 op);