SkNx Load/store: take any pointer.

src/opts/SkNx_neon.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkNx_neon_DEFINED
 #define SkNx_neon_DEFINED
 
@@ skipping 23 matching lines @@
     case 28: return op(v, 28);  case 29: return op(v, 29);  case 30: return op(v, 30); \
     case 31: return op(v, 31); } return fVec
 
 template <>
 class SkNx<2, float> {
 public:
     SkNx(float32x2_t vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(float val)           : fVec(vdup_n_f32(val)) {}
-    static SkNx Load(const float vals[2]) { return vld1_f32(vals); }
+    static SkNx Load(const void* ptr) { return vld1_f32((const float*)ptr); }
     SkNx(float a, float b) { fVec = (float32x2_t) { a, b }; }
 
-    void store(float vals[2]) const { vst1_f32(vals, fVec); }
+    void store(void* ptr) const { vst1_f32((float*)ptr, fVec); }
 
     SkNx approxInvert() const {
         float32x2_t est0 = vrecpe_f32(fVec),
                     est1 = vmul_f32(vrecps_f32(est0, fVec), est0);
         return est1;
     }
     SkNx invert() const {
         float32x2_t est1 = this->approxInvert().fVec,
                     est2 = vmul_f32(vrecps_f32(est1, fVec), est1);
         return est2;
@@ skipping 57 matching lines @@
     float32x2_t fVec;
 };
 
 template <>
 class SkNx<4, int> {
 public:
     SkNx(const int32x4_t& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(int val) : fVec(vdupq_n_s32(val)) {}
-    static SkNx Load(const int vals[4]) { return vld1q_s32(vals); }
+    static SkNx Load(const void* ptr) { return vld1q_s32((const int*)ptr); }
     SkNx(int a, int b, int c, int d) { fVec = (int32x4_t) { a, b, c, d }; }
 
-    void store(int vals[4]) const { vst1q_s32(vals, fVec); }
+    void store(void* ptr) const { vst1q_s32((int*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return vaddq_s32(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return vsubq_s32(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return vmulq_s32(fVec, o.fVec); }
 
     SkNx operator << (int bits) const { SHIFT32(vshlq_n_s32, fVec, bits); }
     SkNx operator >> (int bits) const { SHIFT32(vshrq_n_s32, fVec, bits); }
 
     template <int k> int kth() const {
         SkASSERT(0 <= k && k < 4);
         return vgetq_lane_s32(fVec, k&3);
     }
 
     int32x4_t fVec;
 };
 
 template <>
 class SkNx<4, float> {
 public:
     SkNx(float32x4_t vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(float val)           : fVec(vdupq_n_f32(val)) {}
-    static SkNx Load(const float vals[4]) { return vld1q_f32(vals); }
+    static SkNx Load(const void* ptr) { return vld1q_f32((const float*)ptr); }
     SkNx(float a, float b, float c, float d) { fVec = (float32x4_t) { a, b, c, d }; }
 
-    void store(float vals[4]) const { vst1q_f32(vals, fVec); }
+    void store(void* ptr) const { vst1q_f32((float*)ptr, fVec); }
     SkNx approxInvert() const {
         float32x4_t est0 = vrecpeq_f32(fVec),
                     est1 = vmulq_f32(vrecpsq_f32(est0, fVec), est0);
         return est1;
     }
     SkNx invert() const {
         float32x4_t est1 = this->approxInvert().fVec,
                     est2 = vmulq_f32(vrecpsq_f32(est1, fVec), est1);
         return est2;
     }
@@ skipping 67 matching lines @@
 // It's possible that for our current use cases, representing this as
 // half a uint16x8_t might be better than representing it as a uint16x4_t.
 // It'd make conversion to Sk4b one step simpler.
 template <>
 class SkNx<4, uint16_t> {
 public:
     SkNx(const uint16x4_t& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(uint16_t val) : fVec(vdup_n_u16(val)) {}
-    static SkNx Load(const uint16_t vals[4]) { return vld1_u16(vals); }
+    static SkNx Load(const void* ptr) { return vld1_u16((const uint16_t*)ptr); }
 
     SkNx(uint16_t a, uint16_t b, uint16_t c, uint16_t d) {
         fVec = (uint16x4_t) { a,b,c,d };
     }
 
-    void store(uint16_t vals[4]) const { vst1_u16(vals, fVec); }
+    void store(void* ptr) const { vst1_u16((uint16_t*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return vadd_u16(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return vsub_u16(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return vmul_u16(fVec, o.fVec); }
 
     SkNx operator << (int bits) const { SHIFT16(vshl_n_u16, fVec, bits); }
     SkNx operator >> (int bits) const { SHIFT16(vshr_n_u16, fVec, bits); }
 
     static SkNx Min(const SkNx& a, const SkNx& b) { return vmin_u16(a.fVec, b.fVec); }
 
     template <int k> uint16_t kth() const {
         SkASSERT(0 <= k && k < 4);
         return vget_lane_u16(fVec, k&3);
     }
 
     SkNx thenElse(const SkNx& t, const SkNx& e) const {
         return vbsl_u16(fVec, t.fVec, e.fVec);
     }
 
     uint16x4_t fVec;
 };
 
 template <>
 class SkNx<8, uint16_t> {
 public:
     SkNx(const uint16x8_t& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(uint16_t val) : fVec(vdupq_n_u16(val)) {}
-    static SkNx Load(const uint16_t vals[8]) { return vld1q_u16(vals); }
+    static SkNx Load(const void* ptr) { return vld1q_u16((const uint16_t*)ptr); }
 
     SkNx(uint16_t a, uint16_t b, uint16_t c, uint16_t d,
          uint16_t e, uint16_t f, uint16_t g, uint16_t h) {
         fVec = (uint16x8_t) { a,b,c,d, e,f,g,h };
     }
 
-    void store(uint16_t vals[8]) const { vst1q_u16(vals, fVec); }
+    void store(void* ptr) const { vst1q_u16((uint16_t*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return vaddq_u16(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return vsubq_u16(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return vmulq_u16(fVec, o.fVec); }
 
     SkNx operator << (int bits) const { SHIFT16(vshlq_n_u16, fVec, bits); }
     SkNx operator >> (int bits) const { SHIFT16(vshrq_n_u16, fVec, bits); }
 
     static SkNx Min(const SkNx& a, const SkNx& b) { return vminq_u16(a.fVec, b.fVec); }
 
     template <int k> uint16_t kth() const {
         SkASSERT(0 <= k && k < 8);
         return vgetq_lane_u16(fVec, k&7);
     }
 
     SkNx thenElse(const SkNx& t, const SkNx& e) const {
         return vbslq_u16(fVec, t.fVec, e.fVec);
     }
 
     uint16x8_t fVec;
 };
 
 template <>
 class SkNx<4, uint8_t> {
 public:
     SkNx(const uint8x8_t& vec) : fVec(vec) {}
 
     SkNx() {}
-    static SkNx Load(const uint8_t vals[4]) {
-        return (uint8x8_t)vld1_dup_u32((const uint32_t*)vals);
+    static SkNx Load(const void* ptr) {
+        return (uint8x8_t)vld1_dup_u32((const uint32_t*)ptr);
     }
-    void store(uint8_t vals[4]) const {
-        return vst1_lane_u32((uint32_t*)vals, (uint32x2_t)fVec, 0);
+    void store(void* ptr) const {
+        return vst1_lane_u32((uint32_t*)ptr, (uint32x2_t)fVec, 0);
     }
 
     // TODO as needed
 
     uint8x8_t fVec;
 };
 
 template <>
 class SkNx<16, uint8_t> {
 public:
     SkNx(const uint8x16_t& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(uint8_t val) : fVec(vdupq_n_u8(val)) {}
-    static SkNx Load(const uint8_t vals[16]) { return vld1q_u8(vals); }
+    static SkNx Load(const void* ptr) { return vld1q_u8((const uint8_t*)ptr); }
 
     SkNx(uint8_t a, uint8_t b, uint8_t c, uint8_t d,
          uint8_t e, uint8_t f, uint8_t g, uint8_t h,
          uint8_t i, uint8_t j, uint8_t k, uint8_t l,
          uint8_t m, uint8_t n, uint8_t o, uint8_t p) {
         fVec = (uint8x16_t) { a,b,c,d, e,f,g,h, i,j,k,l, m,n,o,p };
     }
 
-    void store(uint8_t vals[16]) const { vst1q_u8(vals, fVec); }
+    void store(void* ptr) const { vst1q_u8((uint8_t*)ptr, fVec); }
 
     SkNx saturatedAdd(const SkNx& o) const { return vqaddq_u8(fVec, o.fVec); }
 
     SkNx operator + (const SkNx& o) const { return vaddq_u8(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return vsubq_u8(fVec, o.fVec); }
 
     static SkNx Min(const SkNx& a, const SkNx& b) { return vminq_u8(a.fVec, b.fVec); }
     SkNx operator < (const SkNx& o) const { return vcltq_u8(fVec, o.fVec); }
 
     template <int k> uint8_t kth() const {
@@ skipping 40 matching lines @@
     return vget_low_u16(vmovl_u8(src.fVec));
 }
 
 template<> inline Sk4b SkNx_cast<uint8_t, uint16_t, 4>(const Sk4h& src) {
     return vmovn_u16(vcombine_u16(src.fVec, src.fVec));
 }
 
 }  // namespace
 
 #endif//SkNx_neon_DEFINED