Use TextureDecoderNEON funcs statically on arm64.

Not sure if they were being used before (HAVE_ARMV7...?)

GPU/Common/TextureDecoder.cpp

@@ -207,9 +207,11 @@ void DoUnswizzleTex16Basic(const u8 *texptr, u32 *ydestp, int bxc, int byc, u32
 }
 
 #ifndef _M_SSE
+#ifndef ARM64
 QuickTexHashFunc DoQuickTexHash = &QuickTexHashBasic;
 UnswizzleTex16Func DoUnswizzleTex16 = &DoUnswizzleTex16Basic;
 ReliableHash32Func DoReliableHash32 = &XXH32;
+#endif
 ReliableHash64Func DoReliableHash64 = &XXH64;
 #endif
 

GPU/Common/TextureDecoder.h

@@ -21,10 +21,12 @@
 #include "Core/MemMap.h"
 #include "GPU/ge_constants.h"
 #include "GPU/GPUState.h"
+#include "GPU/Common/TextureDecoderNEON.h"
 
 void SetupTextureDecoder();
 
-#ifdef _M_SSE
+// For SSE, we statically link the SSE2 algorithms.
+#if defined(_M_SSE)
 u32 QuickTexHashSSE2(const void *checkp, u32 size);
 #define DoQuickTexHash QuickTexHashSSE2
 
@@ -42,6 +44,20 @@ typedef u64 ReliableHashType;
 #define DoReliableHash XXH32
 typedef u32 ReliableHashType;
 #endif
+
+// For ARM64, NEON is mandatory, so we also statically link.
+#elif defined(ARM64)
+#define DoQuickTexHash QuickTexHashNEON
+#define DoUnswizzleTex16 DoUnswizzleTex16NEON
+#define DoReliableHash32 ReliableHash32NEON
+
+// TODO: NEON version of this too?  Since we're 64, might be faster.
+typedef u64 (*ReliableHash64Func)(const void *input, size_t len, u64 seed);
+extern ReliableHash64Func DoReliableHash64;
+
+#define DoReliableHash DoReliableHash32
+typedef u32 ReliableHashType;
+
 #else
 typedef u32 (*QuickTexHashFunc)(const void *checkp, u32 size);
 extern QuickTexHashFunc DoQuickTexHash;

android/jni/Android.mk

@@ -82,6 +82,7 @@ ARCH_FILES := \
   $(SRC)/Core/Util/AudioFormatNEON.cpp \
   $(SRC)/Common/Arm64Emitter.cpp \
   $(SRC)/Common/ArmCPUDetect.cpp \
+  $(SRC)/Common/ColorConvNEON.cpp \
   $(SRC)/Core/MIPS/ARM64/Arm64CompALU.cpp \
   $(SRC)/Core/MIPS/ARM64/Arm64CompBranch.cpp \
   $(SRC)/Core/MIPS/ARM64/Arm64CompFPU.cpp \