CRC.cpp: add ARMv8 crc intrinsics

Speedup:
CRC_Calculate() up to 11x
CRC_CalculatePalette() up to 4x

Will be only used if the right compiler flags are set.
rpi3: -march=armv8-a+crc -mtune=cortex-a53

http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0801f/pge
1427897662265.html

If CRC32 can be replaced with CRC32C SSE4.2 CRC32C intrinsics could be
added as well.
https://github.com/gonetz/GLideN64/pull/1056#issuecomment-244530166

projects/cmake/Readme.txt

@@ -5,6 +5,7 @@ cmake [-DCMAKE_BUILD_TYPE=Debug] [-DVEC4_OPT=On] [-DCRC_OPT=On] [-DNEON_OPT=On]
 -DCMAKE_BUILD_TYPE=Debug - optional parameter, if you want debug build. Default buid type is Release
 -DVEC4_OPT=On  - optional parameter. set it if you want to enable additional VEC4 optimization (can cause additional bugs).
 -DCRC_OPT=On  - optional parameter. set it if you want to enable additional CRC optimization (can cause additional bugs).
+-DCRC_ARMV8=On  - optional parameter. set it if you want to enable armv8 hardware CRC. It will be ignored if -DCRC_OPT=On.
 -DNEON_OPT=On - optional parameter. set it if you want to enable additional ARM NEON optimization (can cause additional bugs).
 -DNOHQ=On - build without realtime texture enhancer library (GLideNHQ).
 -DUSE_UNIFORMBLOCK=On - Use uniform blocks in shaders. May help to improve performance. Not supported by GLES2 hardware.

src/CMakeLists.txt

@@ -185,14 +185,20 @@ if(VEC4_OPT)
   )
 endif(VEC4_OPT)
 
-if(CRC_OPT)
-  list(APPEND GLideN64_SOURCES
-    CRC_OPT.cpp
-  )
+if(CRC_OPT OR CRC_ARMV8)
   list(REMOVE_ITEM GLideN64_SOURCES
     CRC.cpp
   )
-endif(CRC_OPT)
+  if(CRC_OPT)
+    list(APPEND GLideN64_SOURCES
+      CRC_OPT.cpp
+    )
+  elseif(CRC_ARMV8)
+    list(APPEND GLideN64_SOURCES
+      CRC_ARMV8.cpp
+    )
+  endif(CRC_OPT)
+endif(CRC_OPT OR CRC_ARMV8)
 
 if(NEON_OPT)
   add_definitions(

src/CRC_ARMV8.cpp

@@ -0,0 +1,51 @@
+#include "CRC.h"
+#include <arm_acle.h>
+
+void CRC_BuildTable()
+{
+}
+
+u32 CRC_Calculate( u32 crc, const void * buffer, u32 count )
+{
+	u8 *p;
+	u32 orig = crc;
+
+	p = (u8*) buffer;
+
+	// use eight byte crc intrinsic __crc32d if count is high enough.
+	// __crc32d, __crc32w, __crc32h and __crc32b use polynomial 0x04C11DB7
+	while (count >= 8) {
+		crc = __crc32d(crc, *((u64*)p));
+		p += 8;
+		count -= 8;
+	}
+	if (count >= 4) {
+		crc = __crc32w(crc, *((u32*)p));
+		p += 4;
+		count -= 4;
+	}
+	if (count >= 2) {
+		crc = __crc32h(crc, *((u16*)p));
+		p += 2;
+		count -= 2;
+	}
+	if (count == 1)
+		crc = __crc32b(crc, *p);
+
+	return crc ^ orig;
+}
+
+u32 CRC_CalculatePalette(u32 crc, const void * buffer, u32 count )
+{
+	u8 *p;
+	u32 orig = crc;
+
+	p = (u8*) buffer;
+	while (count--) {
+		// use two byte intrinsic __crc32h
+		crc = __crc32h(crc, *((u16*)p));
+		p += 8;
+	}
+
+	return crc ^ orig;
+}