NEON armv7a & aarch64

Auburn · Nov 2, 2022 · 31a065f · 31a065f
1 parent c6b370f
commit 31a065f
Show file tree

Hide file tree

Showing 6 changed files with 632 additions and 412 deletions.
diff --git a/README.md b/README.md
@@ -15,11 +15,13 @@ Uses FastSIMD to compile classes with multiple SIMD types and selects the fastes
 - SSE4.1
 - AVX2
 - AVX512
+- NEON
 
 Supports:
 - 32/64 bit
 - Windows
 - Linux
+- Android
 - MacOS
 - MSVC
 - Clang

diff --git a/include/FastNoise/FastNoise_Config.h b/include/FastNoise/FastNoise_Config.h
@@ -16,7 +16,8 @@ namespace FastNoise
         FastSIMD::Level_SSE2   |
         FastSIMD::Level_SSE41  |
         FastSIMD::Level_AVX2   |
-        FastSIMD::Level_AVX512 ;
+        FastSIMD::Level_AVX512 |
+        FastSIMD::Level_NEON   ;
 
     class Generator;
     struct Metadata;
@@ -41,4 +42,4 @@ namespace FastNoise
 
 #if !FASTNOISE_USE_SHARED_PTR
 #include "SmartNode.h"
-#endif
+#endif
diff --git a/include/FastNoise/Generators/Utils.inl b/include/FastNoise/Generators/Utils.inl
@@ -71,6 +71,53 @@ namespace FastNoise
             // Bit-8 = Flip sign of a + b
             return ( a + b ) ^ FS_Casti32_f32( (index >> 3) << 31 );
         }
+        template<typename SIMD = FS, std::enable_if_t<SIMD::SIMD_Level == FastSIMD::Level_NEON>* = nullptr>
+        FS_INLINE static float32v GetGradientDotFancy( int32v hash, float32v fX, float32v fY )
+        {
+            int32v index = FS_Convertf32_i32( FS_Converti32_f32( hash & int32v( 0x3FFFFF ) ) * float32v( 1.3333333333333333f ) );
+
+            // Bit-4 = Choose X Y ordering
+            mask32v xy;
+
+//             if constexpr( FS::SIMD_Level == FastSIMD::Level_Scalar )
+//             {
+//                 xy = int32_t( index & int32v( 1 << 2 ) ) != 0;
+//             }
+//             else
+//             {
+                xy = index << 29;
+
+//                 if constexpr( FS::SIMD_Level < FastSIMD::Level_SSE41 )
+//                 {
+                    xy >>= 31;
+//                 }
+//             }
+
+            float32v a = FS_Select_f32( xy, fY, fX );
+            float32v b = FS_Select_f32( xy, fX, fY );
+
+            // Bit-1 = b flip sign
+            b ^= FS_Casti32_f32( index << 31 );
+
+            // Bit-2 = Mul a by 2 or Root3
+            mask32v aMul2;
+
+//             if constexpr( FS::SIMD_Level == FastSIMD::Level_Scalar )
+//             {
+//                 aMul2 = int32_t( index & int32v( 1 << 1 ) ) != 0;
+//             }
+//             else
+//             {
+                aMul2 = (index << 30) >> 31;
+//             }
+
+            a *= FS_Select_f32( aMul2, float32v( 2 ), float32v( ROOT3 ) );
+            // b zero value if a mul 2
+            b = FS_NMask_f32( b, aMul2 );
+
+            // Bit-8 = Flip sign of a + b
+            return ( a + b ) ^ FS_Casti32_f32( (index >> 3) << 31 );
+        }
 
         template<typename SIMD = FS, std::enable_if_t<SIMD::SIMD_Level == FastSIMD::Level_AVX2>* = nullptr>
         FS_INLINE static float32v GetGradientDotFancy( int32v hash, float32v fX, float32v fY )
@@ -128,6 +175,38 @@ namespace FastNoise
 
             return FS_FMulAdd_f32( float32v( 1.0f + ROOT2 ), a, b );
         }
+        template<typename SIMD = FS, std::enable_if_t<SIMD::SIMD_Level == FastSIMD::Level_NEON> * = nullptr>
+         FS_INLINE static float32v GetGradientDot( int32v hash, float32v fX, float32v fY )
+        {
+            // ( 1+R2, 1 ) ( -1-R2, 1 ) ( 1+R2, -1 ) ( -1-R2, -1 )
+            // ( 1, 1+R2 ) ( 1, -1-R2 ) ( -1, 1+R2 ) ( -1, -1-R2 )
+
+            int32v  bit1 = (hash << 31);
+            int32v  bit2 = (hash >> 1) << 31;
+            mask32v bit4;
+
+//             if constexpr( FS::SIMD_Level == FastSIMD::Level_Scalar )
+//             {
+//                 bit4 = int32_t( hash & int32v( 1 << 2 ) ) != 0;
+//             }
+//             else
+//             {
+                bit4 = hash << 29;
+// 
+//                 if constexpr( FS::SIMD_Level < FastSIMD::Level_SSE41 )
+//                 {
+                    bit4 >>= 31;
+//                 }
+//             }
+
+            fX ^= FS_Casti32_f32( bit1 );
+            fY ^= FS_Casti32_f32( bit2 );
+
+            float32v a = FS_Select_f32( bit4, fY, fX );
+            float32v b = FS_Select_f32( bit4, fX, fY );
+
+            return FS_FMulAdd_f32( float32v( 1.0f + ROOT2 ), a, b );
+        }
 
         template<typename SIMD = FS, std::enable_if_t<SIMD::SIMD_Level == FastSIMD::Level_AVX2>* = nullptr>
         FS_INLINE static float32v GetGradientDot( int32v hash, float32v fX, float32v fY )
@@ -311,4 +390,4 @@ namespace FastNoise
 }
 
 using FnUtils = FastNoise::Utils<FS_SIMD_CLASS>;
-namespace FnPrimes = FastNoise::Primes;
+namespace FnPrimes = FastNoise::Primes;
diff --git a/src/FastSIMD/FastSIMD.cpp b/src/FastSIMD/FastSIMD.cpp
@@ -3,12 +3,17 @@
 #include <algorithm>
 #include <cstdint>
 
+#if FASTSIMD_x86
+
 #ifdef __GNUG__
 #include <x86intrin.h>
 #else
 #include <intrin.h>
 #endif
 
+#endif
+
+
 #include "FastSIMD/SIMDTypeList.h"
 
 static_assert(FastSIMD::SIMDTypeList::MinimumCompiled & FastSIMD::COMPILED_SIMD_LEVELS, "FASTSIMD_FALLBACK_SIMD_LEVEL is not a compiled SIMD level, check FastSIMD_Config.h");