From b49067d165326b82dd64f9633553a1cf68e132b3 Mon Sep 17 00:00:00 2001
From: Connor McLaughlin <stenzek@gmail.com>
Date: Sun, 22 Nov 2020 02:00:16 +1000
Subject: [PATCH] Build: Fix Android armv8 performance regression

---
 android/app/build.gradle | 2 +-
 src/core/gpu_sw.cpp      | 6 ++----
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/android/app/build.gradle b/android/app/build.gradle
index 7bac3f3f7..050425ca8 100644
--- a/android/app/build.gradle
+++ b/android/app/build.gradle
@@ -30,7 +30,7 @@ android {
     defaultConfig {
         externalNativeBuild {
             cmake {
-                arguments "-DCMAKE_BUILD_TYPE=Release -DANDROID_ARM_NEON=ON"
+                arguments "-DCMAKE_BUILD_TYPE=Release"
                 abiFilters "arm64-v8a", "armeabi-v7a"
             }
         }
diff --git a/src/core/gpu_sw.cpp b/src/core/gpu_sw.cpp
index a38bdf40b..5a69c1cb8 100644
--- a/src/core/gpu_sw.cpp
+++ b/src/core/gpu_sw.cpp
@@ -11,8 +11,6 @@ Log_SetChannel(GPU_SW);
 
 #if defined(CPU_X64)
 #include <emmintrin.h>
-#elif defined(CPU_AARCH32)
-#include <arm_neon.h>
 #elif defined(CPU_AARCH64)
 #ifdef _MSC_VER
 #include <arm64_neon.h>
@@ -157,7 +155,7 @@ ALWAYS_INLINE void CopyOutRow16<HostDisplayPixelFormat::RGBA5551, u16>(const u16
     _mm_storeu_si128(reinterpret_cast<__m128i*>(dst_ptr), value);
     dst_ptr += 8;
   }
-#elif defined(CPU_AARCH32) || defined(CPU_AARCH64)
+#elif defined(CPU_AARCH64)
   const u32 aligned_width = Common::AlignDownPow2(width, 8);
   for (; col < aligned_width; col += 8)
   {
@@ -197,7 +195,7 @@ ALWAYS_INLINE void CopyOutRow16<HostDisplayPixelFormat::RGB565, u16>(const u16*
     _mm_storeu_si128(reinterpret_cast<__m128i*>(dst_ptr), value);
     dst_ptr += 8;
   }
-#elif defined(CPU_AARCH32) || defined(CPU_AARCH64)
+#elif defined(CPU_AARCH64)
   const u32 aligned_width = Common::AlignDownPow2(width, 8);
   const uint16x8_t single_mask = vdupq_n_u16(0x1F);
   for (; col < aligned_width; col += 8)