From 9dd09cb60dfd861f54aab948652cc06ab92da351 Mon Sep 17 00:00:00 2001
From: 1 <1>
Date: Wed, 4 Oct 2017 14:25:02 +0100
Subject: [PATCH] AVX

---
 src/common/CMakeLists.txt                |  1 +
 src/common/avx_utils.h                   | 22 ++++++++++++++++++++++
 src/core/arm/dynarmic/arm_dynarmic.cpp   |  2 ++
 src/video_core/shader/shader_jit_x64.cpp |  2 ++
 4 files changed, 27 insertions(+)
 create mode 100644 src/common/avx_utils.h

diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 7e83e64b0..1126f4f29 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -45,6 +45,7 @@ set(SRCS
 set(HEADERS
             alignment.h
             assert.h
+            avx_utils.h
             bit_field.h
             bit_set.h
             break_points.h
diff --git a/src/common/avx_utils.h b/src/common/avx_utils.h
new file mode 100644
index 000000000..6dfcf3ca0
--- /dev/null
+++ b/src/common/avx_utils.h
@@ -0,0 +1,22 @@
+
+#pragma once
+
+#if defined(_MSC_VER)
+/* Microsoft C/C++-compatible compiler */
+#include <intrin.h>
+#elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
+/* GCC-compatible compiler, targeting x86/x86-64 */
+#include <x86intrin.h>
+#endif
+
+// It seems a lot of folks out there compile citra with avx. This lil function
+// fixes the penalty by mixing SSE instructions with AVX. For further info:
+// https://software.intel.com/en-us/articles/avoiding-avx-sse-transition-penalties
+// call this function before running dynamicly generated code.
+inline void ZeroUpperAVX() {
+#ifdef __AVX__
+    _mm256_zeroupper();
+#else
+// Do Nothing
+#endif
+}
diff --git a/src/core/arm/dynarmic/arm_dynarmic.cpp b/src/core/arm/dynarmic/arm_dynarmic.cpp
index 6e238009f..375eec39c 100644
--- a/src/core/arm/dynarmic/arm_dynarmic.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic.cpp
@@ -5,6 +5,7 @@
 #include <cstring>
 #include <dynarmic/dynarmic.h>
 #include "common/assert.h"
+#include "common/avx_utils.h"
 #include "common/microprofile.h"
 #include "core/arm/dynarmic/arm_dynarmic.h"
 #include "core/arm/dynarmic/arm_dynarmic_cp15.h"
@@ -130,6 +131,7 @@ void ARM_Dynarmic::ExecuteInstructions(int num_instructions) {
     ASSERT(Memory::GetCurrentPageTable() == current_page_table);
     MICROPROFILE_SCOPE(ARM_Jit);
 
+    ZeroUpperAVX();
     std::size_t ticks_executed = jit->Run(static_cast<unsigned>(num_instructions));
 
     CoreTiming::AddTicks(4000);
diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp
index 73c21871c..33b061d9f 100644
--- a/src/video_core/shader/shader_jit_x64.cpp
+++ b/src/video_core/shader/shader_jit_x64.cpp
@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/avx_utils.h"
 #include "common/hash.h"
 #include "common/microprofile.h"
 #include "video_core/shader/shader.h"
@@ -41,6 +42,7 @@ void JitX64Engine::Run(const ShaderSetup& setup, UnitState& state) const {
     MICROPROFILE_SCOPE(GPU_Shader);
 
     const JitShader* shader = static_cast<const JitShader*>(setup.engine_data.cached_shader);
+    ZeroUpperAVX();
     shader->Run(setup, state, setup.engine_data.entry_point);
 }