From fad46729b097ecd808508e9f57cc5609ed335054 Mon Sep 17 00:00:00 2001
From: Ryan Houdek <Sonicadvance1@gmail.com>
Date: Mon, 2 Mar 2015 03:41:45 -0600
Subject: [PATCH] [AArch64] Implemented paired pushing/popping for the VFP.

A bit more efficient if we are only pushing two VFP registers.
We can probably be a bit more efficient in the future by mixing paired loadstores in to the other paths as well.
---
 Source/Core/Common/Arm64Emitter.cpp | 27 ++++++++++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)

diff --git a/Source/Core/Common/Arm64Emitter.cpp b/Source/Core/Common/Arm64Emitter.cpp
index 4fc7c38ae0..4cb4086d7c 100644
--- a/Source/Core/Common/Arm64Emitter.cpp
+++ b/Source/Core/Common/Arm64Emitter.cpp
@@ -2877,6 +2877,19 @@ void ARM64FloatEmitter::FMUL(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm, u8
 void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers)
 {
 	bool bundled_loadstore = false;
+	int num_regs = registers.Count();
+
+	if (num_regs == 2)
+	{
+		int i = 0;
+		ARM64Reg regs[2];
+		for (auto it : registers)
+			regs[i++] = (ARM64Reg)(Q0 + it);
+
+		STP(128, INDEX_PRE, regs[0], regs[1], SP, -32);
+		return;
+	}
+
 	for (int i = 0; i < 32; ++i)
 	{
 		if (!registers[i])
@@ -2898,7 +2911,6 @@ void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers)
 	}
 	else
 	{
-		int num_regs = registers.Count();
 		// Violating the AAPCS64 never felt so right.
 		m_emit->SUB(SP, SP, num_regs * 16);
 		for (int i = 0; i < 32; ++i)
@@ -2925,6 +2937,19 @@ void ARM64FloatEmitter::ABI_PushRegisters(BitSet32 registers)
 void ARM64FloatEmitter::ABI_PopRegisters(BitSet32 registers)
 {
 	bool bundled_loadstore = false;
+	int num_regs = registers.Count();
+
+	if (num_regs == 2)
+	{
+		int i = 0;
+		ARM64Reg regs[2];
+		for (auto it : registers)
+			regs[i++] = (ARM64Reg)(Q0 + it);
+
+		LDP(128, INDEX_POST, regs[0], regs[1], SP, 32);
+		return;
+	}
+
 	for (int i = 0; i < 32; ++i)
 	{
 		if (!registers[i])