Add a NEON version of the tex hash.

Should be used only for NEON devices.  Currently only compiled on Android.

GPU/Common/TextureDecoder.cpp

@@ -15,10 +15,62 @@
 // Official git repository and contact information can be found at
 // https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
 
+#include "Common/CPUDetect.h"
 #include "GPU/Common/TextureDecoder.h"
+#include "GPU/Common/TextureDecoderNEON.h"
 
 // TODO: Move some common things into here.
 
+#ifdef _M_SSE
+#include <xmmintrin.h>
+#endif
+
+static u32 QuickTexHashSSE2(const void *checkp, u32 size) {
+	u32 check = 0;
+
+#ifdef _M_SSE
+	if (((intptr_t)checkp & 0xf) == 0 && (size & 0x3f) == 0) {
+		__m128i cursor = _mm_set1_epi32(0);
+		__m128i cursor2 = _mm_set_epi16(0x0001U, 0x0083U, 0x4309U, 0x4d9bU, 0xb651U, 0x4b73U, 0x9bd9U, 0xc00bU);
+		__m128i update = _mm_set1_epi16(0x2455U);
+		const __m128i *p = (const __m128i *)checkp;
+		for (u32 i = 0; i < size / 16; i += 4) {
+			__m128i chunk = _mm_mullo_epi16(_mm_load_si128(&p[i]), cursor2);
+			cursor = _mm_add_epi32(cursor, chunk);
+			cursor = _mm_xor_si128(cursor, _mm_load_si128(&p[i + 1]));
+			cursor = _mm_add_epi32(cursor, _mm_load_si128(&p[i + 2]));
+			chunk = _mm_mullo_epi16(_mm_load_si128(&p[i + 3]), cursor2);
+			cursor = _mm_xor_si128(cursor, chunk);
+			cursor2 = _mm_add_epi16(cursor2, update);
+		}
+		// Add the four parts into the low i32.
+		cursor = _mm_add_epi32(cursor, _mm_add_epi32(_mm_srli_si128(cursor, 8), cursor2));
+		cursor = _mm_add_epi32(cursor, _mm_srli_si128(cursor, 4));
+		check = _mm_cvtsi128_si32(cursor);
+	} else {
+#else
+	{
+#endif
+		const u32 *p = (const u32 *)checkp;
+		for (u32 i = 0; i < size / 8; ++i) {
+			check += *p++;
+			check ^= *p++;
+		}
+	}
+
+	return check;
+}
+
+QuickTexHashFunc DoQuickTexHash = &QuickTexHashSSE2;
+
+void SetupQuickTexHash() {
+#ifdef ARM
+	if (cpu_info.bNEON) {
+		DoQuickTexHash = &QuickTexHashNEON;
+	}
+#endif
+}
+
 static inline u32 makecol(int r, int g, int b, int a) {
 	return (a << 24) | (r << 16) | (g << 8) | b;
 }

GPU/Common/TextureDecoder.h

@@ -22,6 +22,10 @@
 #include "GPU/ge_constants.h"
 #include "GPU/GPUState.h"
 
+void SetupQuickTexHash();
+typedef u32 (*QuickTexHashFunc)(const void *checkp, u32 size);
+extern QuickTexHashFunc DoQuickTexHash;
+
 // All these DXT structs are in the reverse order, as compared to PC.
 // On PC, alpha comes before color, and interpolants are before the tile data.
 

GPU/Common/TextureDecoderNEON.cpp

@@ -0,0 +1,57 @@
+// Copyright (c) 2012- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+#include <arm_neon.h>
+#include "GPU/Common/TextureDecoder.h"
+
+#ifndef ARM
+#error Should not be compiled on non-ARM.
+#endif
+
+static const u16 MEMORY_ALIGNED16(QuickTexHashInitial[8]) = {0x0001U, 0x0083U, 0x4309U, 0x4d9bU, 0xb651U, 0x4b73U, 0x9bd9U, 0xc00bU};
+
+u32 QuickTexHashNEON(const void *checkp, u32 size) {
+	u32 check = 0;
+
+	if (((intptr_t)checkp & 0xf) == 0 && (size & 0x3f) == 0) {
+		uint32x4_t cursor = vdupq_n_u32(0);
+		uint32x4_t cursor2 = vld1q_u32((const u32 *)QuickTexHashInitial);
+		uint32x4_t update = vdupq_n_u32(0x24552455U);
+
+		const u32 *p = (const u32 *)checkp;
+		for (u32 i = 0; i < size / 16; i += 4) {
+			cursor = vmlaq_u32(cursor, vld1q_u32(&p[4 * 0]), cursor2);
+			cursor = veorq_u32(cursor, vld1q_u32(&p[4 * 1]));
+			cursor = vaddq_u32(cursor, vld1q_u32(&p[4 * 2]));
+			cursor = veorq_u32(cursor, vmulq_u32(vld1q_u32(&p[4 * 3]), cursor2));
+			cursor2 = vaddq_u32(cursor2, update);
+
+			p += 4 * 4;
+		}
+
+		cursor = vaddq_u32(cursor, cursor2);
+		check = vgetq_lane_u32(cursor, 0) + vgetq_lane_u32(cursor, 1) + vgetq_lane_u32(cursor, 2) + vgetq_lane_u32(cursor, 3);
+	} else {
+		const u32 *p = (const u32 *)checkp;
+		for (u32 i = 0; i < size / 8; ++i) {
+			check += *p++;
+			check ^= *p++;
+		}
+	}
+
+	return check;
+}

GPU/Common/TextureDecoderNEON.h

@@ -0,0 +1,20 @@
+// Copyright (c) 2012- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+#include "GPU/Common/TextureDecoder.h"
+
+u32 QuickTexHashNEON(const void *checkp, u32 size);

GPU/GLES/TextureCache.cpp

@@ -56,6 +56,7 @@ TextureCache::TextureCache() : clearCacheNextFrame_(false), lowMemoryMode_(false
 	clutBufConverted_ = (u32 *)AllocateAlignedMemory(4096 * sizeof(u32), 16);  // 16KB
 	clutBufRaw_ = (u32 *)AllocateAlignedMemory(4096 * sizeof(u32), 16);  // 16KB
 	glGetFloatv(GL_MAX_TEXTURE_MAX_ANISOTROPY_EXT, &maxAnisotropyLevel);
+	SetupQuickTexHash();
 }
 
 TextureCache::~TextureCache() {
@@ -678,39 +679,8 @@ static inline u32 QuickClutHash(const u8 *clut, u32 bytes) {
 static inline u32 QuickTexHash(u32 addr, int bufw, int w, int h, GETextureFormat format) {
 	const u32 sizeInRAM = (textureBitsPerPixel[format] * bufw * h) / 8;
 	const u32 *checkp = (const u32 *) Memory::GetPointer(addr);
-	u32 check = 0;
 
-#ifdef _M_SSE
-	if (((intptr_t)checkp & 0xf) == 0 && (sizeInRAM & 0x3f) == 0) {
-		__m128i cursor = _mm_set1_epi32(0);
-		__m128i cursor2 = _mm_set_epi16(0x0001U, 0x0083U, 0x4309U, 0x4d9bU, 0xb651U, 0x4b73U, 0x9bd9U, 0xc00bU);
-		__m128i update = _mm_set1_epi16(0x2455U);
-		const __m128i *p = (const __m128i *)checkp;
-		for (u32 i = 0; i < sizeInRAM / 16; i += 4) {
-			__m128i chunk = _mm_mullo_epi16(_mm_load_si128(&p[i]), cursor2);
-			cursor = _mm_add_epi32(cursor, chunk);
-			cursor = _mm_xor_si128(cursor, _mm_load_si128(&p[i + 1]));
-			cursor = _mm_add_epi32(cursor, _mm_load_si128(&p[i + 2]));
-			chunk = _mm_mullo_epi16(_mm_load_si128(&p[i + 3]), cursor2);
-			cursor = _mm_xor_si128(cursor, chunk);
-			cursor2 = _mm_add_epi16(cursor2, update);
-		}
-		// Add the four parts into the low i32.
-		cursor = _mm_add_epi32(cursor, _mm_add_epi32(_mm_srli_si128(cursor, 8), cursor2));
-		cursor = _mm_add_epi32(cursor, _mm_srli_si128(cursor, 4));
-		check = _mm_cvtsi128_si32(cursor);
-	} else {
-#else
-	// TODO: ARM NEON implementation (using CPUDetect to be sure it has NEON.)
-	{
-#endif
-		for (u32 i = 0; i < sizeInRAM / 8; ++i) {
-			check += *checkp++;
-			check ^= *checkp++;
-		}
-	}
-
-	return check;
+	return DoQuickTexHash(checkp, sizeInRAM);
 }
 
 inline bool TextureCache::TexCacheEntry::Matches(u16 dim2, u8 format2, int maxLevel2) {

GPU/GPU.vcxproj

@@ -159,6 +159,12 @@
     <ClInclude Include="Common\IndexGenerator.h" />
     <ClInclude Include="Common\PostShader.h" />
     <ClInclude Include="Common\SplineCommon.h" />
+    <ClInclude Include="Common\TextureDecoderNEON.h">
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
+    </ClInclude>
     <ClInclude Include="Common\VertexDecoderCommon.h" />
     <ClInclude Include="Debugger\Breakpoints.h" />
     <ClInclude Include="Debugger\Stepping.h" />
@@ -204,6 +210,12 @@
     <ClCompile Include="..\ext\xbrz\xbrz.cpp" />
     <ClCompile Include="Common\IndexGenerator.cpp" />
     <ClCompile Include="Common\PostShader.cpp" />
+    <ClCompile Include="Common\TextureDecoderNEON.cpp">
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
+      <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
+    </ClCompile>
     <ClCompile Include="Common\VertexDecoderCommon.cpp" />
     <ClCompile Include="Debugger\Breakpoints.cpp" />
     <ClCompile Include="Debugger\Stepping.cpp" />

GPU/GPU.vcxproj.filters

@@ -159,6 +159,9 @@
     <ClInclude Include="Common\PostShader.h">
       <Filter>Common</Filter>
     </ClInclude>
+    <ClInclude Include="Common\TextureDecoderNEON.h">
+      <Filter>Common</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="Math3D.cpp">
@@ -287,8 +290,11 @@
     <ClCompile Include="Common\PostShader.cpp">
       <Filter>Common</Filter>
     </ClCompile>
+    <ClCompile Include="Common\TextureDecoderNEON.cpp">
+      <Filter>Common</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <None Include="CMakeLists.txt" />
   </ItemGroup>
-</Project>
+</Project>

android/jni/Android.mk

@@ -147,6 +147,7 @@ EXEC_AND_LIB_FILES := \
   $(SRC)/GPU/Common/IndexGenerator.cpp.arm \
   $(SRC)/GPU/Common/VertexDecoderCommon.cpp.arm \
   $(SRC)/GPU/Common/TextureDecoder.cpp \
+  $(SRC)/GPU/Common/TextureDecoderNEON.cpp.neon \
   $(SRC)/GPU/Common/PostShader.cpp \
   $(SRC)/GPU/Debugger/Breakpoints.cpp \
   $(SRC)/GPU/Debugger/Stepping.cpp \