SkNx Load/store: take any pointer.

src/opts/SkNx_sse.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_sse_DEFINED
 #define SkNx_sse_DEFINED
 
 // This file may assume <= SSE2, but must check SK_CPU_SSE_LEVEL for anything more recent.
 
 #define SKNX_IS_FAST
 
 namespace {  // See SkNx.h
 
 
 template <>
 class SkNx<2, float> {
 public:
     SkNx(const __m128& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(float val) : fVec(_mm_set1_ps(val)) {}
-    static SkNx Load(const float vals[2]) {
-        return _mm_castsi128_ps(_mm_loadl_epi64((const __m128i*)vals));
+    static SkNx Load(const void* ptr) {
+        return _mm_castsi128_ps(_mm_loadl_epi64((const __m128i*)ptr));
     }
     SkNx(float a, float b) : fVec(_mm_setr_ps(a,b,0,0)) {}
 
-    void store(float vals[2]) const { _mm_storel_pi((__m64*)vals, fVec); }
+    void store(void* ptr) const { _mm_storel_pi((__m64*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_ps(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_ps(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return _mm_mul_ps(fVec, o.fVec); }
     SkNx operator / (const SkNx& o) const { return _mm_div_ps(fVec, o.fVec); }
 
     SkNx operator == (const SkNx& o) const { return _mm_cmpeq_ps (fVec, o.fVec); }
     SkNx operator != (const SkNx& o) const { return _mm_cmpneq_ps(fVec, o.fVec); }
     SkNx operator  < (const SkNx& o) const { return _mm_cmplt_ps (fVec, o.fVec); }
     SkNx operator  > (const SkNx& o) const { return _mm_cmpgt_ps (fVec, o.fVec); }
@@ skipping 23 matching lines @@
     __m128 fVec;
 };
 
 template <>
 class SkNx<2, double> {
 public:
     SkNx(const __m128d& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(double val) : fVec(_mm_set1_pd(val)) {}
-    static SkNx Load(const double vals[2]) { return _mm_loadu_pd(vals); }
+    static SkNx Load(const void* ptr) { return _mm_loadu_pd((const double*)ptr); }
     SkNx(double a, double b) : fVec(_mm_setr_pd(a,b)) {}
 
-    void store(double vals[2]) const { _mm_storeu_pd(vals, fVec); }
+    void store(void* ptr) const { _mm_storeu_pd((double*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_pd(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_pd(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return _mm_mul_pd(fVec, o.fVec); }
     SkNx operator / (const SkNx& o) const { return _mm_div_pd(fVec, o.fVec); }
 
     SkNx operator == (const SkNx& o) const { return _mm_cmpeq_pd (fVec, o.fVec); }
     SkNx operator != (const SkNx& o) const { return _mm_cmpneq_pd(fVec, o.fVec); }
     SkNx operator  < (const SkNx& o) const { return _mm_cmplt_pd (fVec, o.fVec); }
     SkNx operator  > (const SkNx& o) const { return _mm_cmpgt_pd (fVec, o.fVec); }
@@ skipping 22 matching lines @@
     __m128d fVec;
 };
 
 template <>
 class SkNx<4, int> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(int val) : fVec(_mm_set1_epi32(val)) {}
-    static SkNx Load(const int vals[4]) { return _mm_loadu_si128((const __m128i*)vals); }
+    static SkNx Load(const void* ptr) { return _mm_loadu_si128((const __m128i*)ptr); }
     SkNx(int a, int b, int c, int d) : fVec(_mm_setr_epi32(a,b,c,d)) {}
 
-    void store(int vals[4]) const { _mm_storeu_si128((__m128i*)vals, fVec); }
+    void store(void* ptr) const { _mm_storeu_si128((__m128i*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_epi32(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_epi32(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const {
         __m128i mul20 = _mm_mul_epu32(fVec, o.fVec),
                 mul31 = _mm_mul_epu32(_mm_srli_si128(fVec, 4), _mm_srli_si128(o.fVec, 4));
         return _mm_unpacklo_epi32(_mm_shuffle_epi32(mul20, _MM_SHUFFLE(0,0,2,0)),
                                   _mm_shuffle_epi32(mul31, _MM_SHUFFLE(0,0,2,0)));
     }
 
@@ skipping 14 matching lines @@
     __m128i fVec;
 };
 
 template <>
 class SkNx<4, float> {
 public:
     SkNx(const __m128& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(float val)           : fVec( _mm_set1_ps(val) ) {}
-    static SkNx Load(const float vals[4]) { return _mm_loadu_ps(vals); }
+    static SkNx Load(const void* ptr) { return _mm_loadu_ps((const float*)ptr); }
 
     SkNx(float a, float b, float c, float d) : fVec(_mm_setr_ps(a,b,c,d)) {}
 
-    void store(float vals[4]) const { _mm_storeu_ps(vals, fVec); }
+    void store(void* ptr) const { _mm_storeu_ps((float*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_ps(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_ps(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return _mm_mul_ps(fVec, o.fVec); }
     SkNx operator / (const SkNx& o) const { return _mm_div_ps(fVec, o.fVec); }
 
     SkNx operator == (const SkNx& o) const { return _mm_cmpeq_ps (fVec, o.fVec); }
     SkNx operator != (const SkNx& o) const { return _mm_cmpneq_ps(fVec, o.fVec); }
     SkNx operator  < (const SkNx& o) const { return _mm_cmplt_ps (fVec, o.fVec); }
     SkNx operator  > (const SkNx& o) const { return _mm_cmpgt_ps (fVec, o.fVec); }
@@ skipping 30 matching lines @@
     __m128 fVec;
 };
 
 template <>
 class SkNx<4, uint16_t> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(uint16_t val) : fVec(_mm_set1_epi16(val)) {}
-    static SkNx Load(const uint16_t vals[4]) { return _mm_loadl_epi64((const __m128i*)vals); }
+    static SkNx Load(const void* ptr) { return _mm_loadl_epi64((const __m128i*)ptr); }
     SkNx(uint16_t a, uint16_t b, uint16_t c, uint16_t d) : fVec(_mm_setr_epi16(a,b,c,d,0,0,0,0)) {}
 
-    void store(uint16_t vals[4]) const { _mm_storel_epi64((__m128i*)vals, fVec); }
+    void store(void* ptr) const { _mm_storel_epi64((__m128i*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_epi16(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_epi16(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return _mm_mullo_epi16(fVec, o.fVec); }
 
     SkNx operator << (int bits) const { return _mm_slli_epi16(fVec, bits); }
     SkNx operator >> (int bits) const { return _mm_srli_epi16(fVec, bits); }
 
     template <int k> uint16_t kth() const {
         SkASSERT(0 <= k && k < 4);
         return _mm_extract_epi16(fVec, k);
     }
 
     __m128i fVec;
 };
 
 template <>
 class SkNx<8, uint16_t> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(uint16_t val) : fVec(_mm_set1_epi16(val)) {}
-    static SkNx Load(const uint16_t vals[8]) { return _mm_loadu_si128((const __m128i*)vals); }
+    static SkNx Load(const void* ptr) { return _mm_loadu_si128((const __m128i*)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(_mm_setr_epi16(a,b,c,d,e,f,g,h)) {}
 
-    void store(uint16_t vals[8]) const { _mm_storeu_si128((__m128i*)vals, fVec); }
+    void store(void* ptr) const { _mm_storeu_si128((__m128i*)ptr, fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_epi16(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_epi16(fVec, o.fVec); }
     SkNx operator * (const SkNx& o) const { return _mm_mullo_epi16(fVec, o.fVec); }
 
     SkNx operator << (int bits) const { return _mm_slli_epi16(fVec, bits); }
     SkNx operator >> (int bits) const { return _mm_srli_epi16(fVec, bits); }
 
     static SkNx Min(const SkNx& a, const SkNx& b) {
         // No unsigned _mm_min_epu16, so we'll shift into a space where we can use the
@@ skipping 16 matching lines @@
 
     __m128i fVec;
 };
 
 template <>
 class SkNx<4, uint8_t> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
-    static SkNx Load(const uint8_t vals[4]) { return _mm_cvtsi32_si128(*(const int*)vals); }
-    void store(uint8_t vals[4]) const { *(int*)vals = _mm_cvtsi128_si32(fVec); }
+    static SkNx Load(const void* ptr) { return _mm_cvtsi32_si128(*(const int*)ptr); }
+    void store(void* ptr) const { *(int*)ptr = _mm_cvtsi128_si32(fVec); }
 
     // TODO as needed
 
     __m128i fVec;
 };
 
 template <>
 class SkNx<8, uint8_t> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
-    static SkNx Load(const uint8_t vals[8]) { return _mm_loadl_epi64((const __m128i*)vals); }
-    void store(uint8_t vals[8]) const { _mm_storel_epi64((__m128i*)vals, fVec); }
+    static SkNx Load(const void* ptr) { return _mm_loadl_epi64((const __m128i*)ptr); }
+    void store(void* ptr) const { _mm_storel_epi64((__m128i*)ptr, fVec); }
 
     // TODO as needed
 
     __m128i fVec;
 };
 
 template <>
 class SkNx<16, uint8_t> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
     SkNx() {}
     SkNx(uint8_t val) : fVec(_mm_set1_epi8(val)) {}
-    static SkNx Load(const uint8_t vals[16]) { return _mm_loadu_si128((const __m128i*)vals); }
+    static SkNx Load(const void* ptr) { return _mm_loadu_si128((const __m128i*)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(_mm_setr_epi8(a,b,c,d, e,f,g,h, i,j,k,l, m,n,o,p)) {}
 
-    void store(uint8_t vals[16]) const { _mm_storeu_si128((__m128i*)vals, fVec); }
+    void store(void* ptr) const { _mm_storeu_si128((__m128i*)ptr, fVec); }
 
     SkNx saturatedAdd(const SkNx& o) const { return _mm_adds_epu8(fVec, o.fVec); }
 
     SkNx operator + (const SkNx& o) const { return _mm_add_epi8(fVec, o.fVec); }
     SkNx operator - (const SkNx& o) const { return _mm_sub_epi8(fVec, o.fVec); }
 
     static SkNx Min(const SkNx& a, const SkNx& b) { return _mm_min_epu8(a.fVec, b.fVec); }
     SkNx operator < (const SkNx& o) const {
         // There's no unsigned _mm_cmplt_epu8, so we flip the sign bits then use a signed compare.
         auto flip = _mm_set1_epi8(char(0x80));
@@ skipping 56 matching lines @@
 }
 
 template<> inline Sk4b SkNx_cast<uint8_t, uint16_t, 4>(const Sk4h& src) {
     return _mm_packus_epi16(src.fVec, src.fVec);
 }
 
 
 }  // namespace
 
 #endif//SkNx_sse_DEFINED