From 3354f068fcde2b19f2e15592ef1ae958ad68ec90 Mon Sep 17 00:00:00 2001
From: Nekotekina <nekotekina@gmail.com>
Date: Tue, 2 Apr 2019 13:12:05 +0300
Subject: [PATCH] PPU/SPU transactions: ease cache line interference (TSX path)

Touch memory on the same memory page, but different cache lines.
---
 rpcs3/Emu/Cell/PPUThread.cpp |  8 ++++++++
 rpcs3/Emu/Cell/SPUThread.cpp | 12 ++++++++++++
 2 files changed, 20 insertions(+)

diff --git a/rpcs3/Emu/Cell/PPUThread.cpp b/rpcs3/Emu/Cell/PPUThread.cpp
index fa8319d5b1..bc23813dbc 100644
--- a/rpcs3/Emu/Cell/PPUThread.cpp
+++ b/rpcs3/Emu/Cell/PPUThread.cpp
@@ -1054,8 +1054,12 @@ const auto ppu_stwcx_tx = build_function_asm<bool(*)(u32 raddr, u64 rtime, u64 r
 	c.jz(fail);
 	c.sar(x86::eax, 24);
 	c.js(fail);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.lock().add(x86::dword_ptr(x86::r11), 0);
 	c.lock().add(x86::qword_ptr(x86::r10), 0);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.jmp(begin);
 
 	c.bind(fail);
@@ -1148,8 +1152,12 @@ const auto ppu_stdcx_tx = build_function_asm<bool(*)(u32 raddr, u64 rtime, u64 r
 	c.jz(fail);
 	c.sar(x86::eax, 24);
 	c.js(fail);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.lock().add(x86::qword_ptr(x86::r11), 0);
 	c.lock().add(x86::qword_ptr(x86::r10), 0);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.jmp(begin);
 
 	c.bind(fail);
diff --git a/rpcs3/Emu/Cell/SPUThread.cpp b/rpcs3/Emu/Cell/SPUThread.cpp
index 23dc4753a8..7995a857bf 100644
--- a/rpcs3/Emu/Cell/SPUThread.cpp
+++ b/rpcs3/Emu/Cell/SPUThread.cpp
@@ -247,8 +247,12 @@ const auto spu_putllc_tx = build_function_asm<u32(*)(u32 raddr, u64 rtime, const
 	c.js(fail);
 	c.sub(args[0].r32(), 1);
 	c.jz(retry);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.lock().add(x86::qword_ptr(x86::r11), 0);
 	c.lock().add(x86::qword_ptr(x86::r10), 0);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 #ifdef _WIN32
 	c.vmovups(x86::ymm4, x86::yword_ptr(args[3], 0));
 	c.vmovups(x86::ymm5, x86::yword_ptr(args[3], 96));
@@ -298,8 +302,12 @@ const auto spu_getll_tx = build_function_asm<u64(*)(u32 raddr, void* rdata)>([](
 	// Touch memory after transaction failure
 	c.bind(fall);
 	c.pause();
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.mov(x86::rax, x86::qword_ptr(x86::r11));
 	c.mov(x86::rax, x86::qword_ptr(x86::r10));
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.sub(args[0], 1);
 	c.jnz(begin);
 	c.mov(x86::eax, 1);
@@ -343,8 +351,12 @@ const auto spu_putlluc_tx = build_function_asm<bool(*)(u32 raddr, const void* rd
 	// Touch memory after transaction failure
 	c.bind(fall);
 	c.pause();
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.lock().add(x86::qword_ptr(x86::r11), 0);
 	c.lock().add(x86::qword_ptr(x86::r10), 0);
+	c.xor_(x86::r11, 0xf80);
+	c.xor_(x86::r10, 0xf80);
 	c.sub(args[0], 1);
 	c.jnz(begin);
 	c.xor_(x86::eax, x86::eax);