From 9ed465f4ac6cc9723824cb5e6b2e32b99b9e50cb Mon Sep 17 00:00:00 2001
From: degasus <wickmarkus@web.de>
Date: Fri, 4 Mar 2016 21:58:17 +0100
Subject: [PATCH] JitArm64: Implement mulhwx

---
 Source/Core/Common/Arm64Emitter.cpp           |  4 ++
 Source/Core/Common/Arm64Emitter.h             |  1 +
 Source/Core/Core/PowerPC/JitArm64/Jit.h       |  2 +
 .../PowerPC/JitArm64/JitArm64_Integer.cpp     | 50 +++++++++++++++++++
 .../Core/PowerPC/JitArm64/JitArm64_Tables.cpp |  4 +-
 5 files changed, 59 insertions(+), 2 deletions(-)

diff --git a/Source/Core/Common/Arm64Emitter.cpp b/Source/Core/Common/Arm64Emitter.cpp
index 2c2ede99ea..d7c53ba289 100644
--- a/Source/Core/Common/Arm64Emitter.cpp
+++ b/Source/Core/Common/Arm64Emitter.cpp
@@ -1389,6 +1389,10 @@ void ARM64XEmitter::UMADDL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, ARM64Reg Ra)
 {
 	EncodeData3SrcInst(5, Rd, Rn, Rm, Ra);
 }
+void ARM64XEmitter::UMULL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm)
+{
+	UMADDL(Rd, Rn, Rm, SP);
+}
 void ARM64XEmitter::UMSUBL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, ARM64Reg Ra)
 {
 	EncodeData3SrcInst(6, Rd, Rn, Rm, Ra);
diff --git a/Source/Core/Common/Arm64Emitter.h b/Source/Core/Common/Arm64Emitter.h
index 9041cd32bb..098b48fd25 100644
--- a/Source/Core/Common/Arm64Emitter.h
+++ b/Source/Core/Common/Arm64Emitter.h
@@ -517,6 +517,7 @@ public:
 	void SMSUBL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, ARM64Reg Ra);
 	void SMULH(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
 	void UMADDL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, ARM64Reg Ra);
+	void UMULL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
 	void UMSUBL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, ARM64Reg Ra);
 	void UMULH(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
 	void MUL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
diff --git a/Source/Core/Core/PowerPC/JitArm64/Jit.h b/Source/Core/Core/PowerPC/JitArm64/Jit.h
index 02064515a5..cb6162ba87 100644
--- a/Source/Core/Core/PowerPC/JitArm64/Jit.h
+++ b/Source/Core/Core/PowerPC/JitArm64/Jit.h
@@ -87,6 +87,8 @@ public:
 	void rlwnmx(UGeckoInstruction inst);
 	void srawix(UGeckoInstruction inst);
 	void mullwx(UGeckoInstruction inst);
+	void mulhwx(UGeckoInstruction inst);
+	void mulhwux(UGeckoInstruction inst);
 	void addic(UGeckoInstruction inst);
 	void mulli(UGeckoInstruction inst);
 	void addzex(UGeckoInstruction inst);
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
index d9eb780d28..058341b1cf 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
@@ -747,6 +747,56 @@ void JitArm64::mullwx(UGeckoInstruction inst)
 	}
 }
 
+void JitArm64::mulhwx(UGeckoInstruction inst)
+{
+	INSTRUCTION_START
+	JITDISABLE(bJITIntegerOff);
+
+	int a = inst.RA, b = inst.RB, d = inst.RD;
+
+	if (gpr.IsImm(a) && gpr.IsImm(b))
+	{
+		s32 i = (s32)gpr.GetImm(a), j = (s32)gpr.GetImm(b);
+		gpr.SetImmediate(d, (u32)((u64)(((s64)i * (s64)j) ) >> 32));
+		if (inst.Rc)
+			ComputeRC(gpr.GetImm(d), 0);
+	}
+	else
+	{
+		gpr.BindToRegister(d, d == a || d == b);
+		SMULL(EncodeRegTo64(gpr.R(d)), gpr.R(a), gpr.R(b));
+		LSR(EncodeRegTo64(gpr.R(d)), EncodeRegTo64(gpr.R(d)), 32);
+
+		if (inst.Rc)
+			ComputeRC(gpr.R(d), 0);
+	}
+}
+
+void JitArm64::mulhwux(UGeckoInstruction inst)
+{
+	INSTRUCTION_START
+	JITDISABLE(bJITIntegerOff);
+
+	int a = inst.RA, b = inst.RB, d = inst.RD;
+
+	if (gpr.IsImm(a) && gpr.IsImm(b))
+	{
+		u32 i = gpr.GetImm(a), j = gpr.GetImm(b);
+		gpr.SetImmediate(d, (u32)( ( (u64)i * (u64)j) >> 32) );
+		if (inst.Rc)
+			ComputeRC(gpr.GetImm(d), 0);
+	}
+	else
+	{
+		gpr.BindToRegister(d, d == a || d == b);
+		UMULL(EncodeRegTo64(gpr.R(d)), gpr.R(a), gpr.R(b));
+		LSR(EncodeRegTo64(gpr.R(d)), EncodeRegTo64(gpr.R(d)), 32);
+
+		if (inst.Rc)
+			ComputeRC(gpr.R(d), 0);
+	}
+}
+
 void JitArm64::addzex(UGeckoInstruction inst)
 {
 	INSTRUCTION_START
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_Tables.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_Tables.cpp
index 96b3f9046f..09fdd271ce 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_Tables.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_Tables.cpp
@@ -184,8 +184,8 @@ static GekkoOPTemplate table31[] =
 	{1003, &JitArm64::FallBackToInterpreter},   // divwox
 	{459,  &JitArm64::divwux},                  // divwux
 	{971,  &JitArm64::divwux},                  // divwuox
-	{75,   &JitArm64::FallBackToInterpreter},   // mulhwx
-	{11,   &JitArm64::FallBackToInterpreter},   // mulhwux
+	{75,   &JitArm64::mulhwx},                  // mulhwx
+	{11,   &JitArm64::mulhwux},                 // mulhwux
 	{235,  &JitArm64::mullwx},                  // mullwx
 	{747,  &JitArm64::mullwx},                  // mullwox
 	{104,  &JitArm64::negx},                    // negx