Sk4px_NEON.h 2.4 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556
  1. /*
  2. * Copyright 2015 Google Inc.
  3. *
  4. * Use of this source code is governed by a BSD-style license that can be
  5. * found in the LICENSE file.
  6. */
  7. namespace { // NOLINT(google-build-namespaces)
  8. inline Sk4px::Wide Sk4px::widen() const {
  9. return Sk16h(vmovl_u8(vget_low_u8 (this->fVec)),
  10. vmovl_u8(vget_high_u8(this->fVec)));
  11. }
  12. inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
  13. return Sk16h(vmull_u8(vget_low_u8 (this->fVec), vget_low_u8 (other.fVec)),
  14. vmull_u8(vget_high_u8(this->fVec), vget_high_u8(other.fVec)));
  15. }
  16. inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
  17. const Sk4px::Wide o(other); // Should be no code, but allows us to access fLo, fHi.
  18. return Sk16b(vcombine_u8(vaddhn_u16(this->fLo.fVec, o.fLo.fVec),
  19. vaddhn_u16(this->fHi.fVec, o.fHi.fVec)));
  20. }
  21. inline Sk4px Sk4px::Wide::div255() const {
  22. // Calculated as (x + (x+128)>>8 +128) >> 8. The 'r' in each instruction provides each +128.
  23. return Sk16b(vcombine_u8(vraddhn_u16(this->fLo.fVec, vrshrq_n_u16(this->fLo.fVec, 8)),
  24. vraddhn_u16(this->fHi.fVec, vrshrq_n_u16(this->fHi.fVec, 8))));
  25. }
  26. inline Sk4px Sk4px::alphas() const {
  27. auto as = vshrq_n_u32((uint32x4_t)fVec, SK_A32_SHIFT); // ___3 ___2 ___1 ___0
  28. return Sk16b((uint8x16_t)vmulq_n_u32(as, 0x01010101)); // 3333 2222 1111 0000
  29. }
  30. inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
  31. uint8x16_t a8 = vdupq_n_u8(0); // ____ ____ ____ ____
  32. a8 = vld1q_lane_u8(a+0, a8, 0); // ____ ____ ____ ___0
  33. a8 = vld1q_lane_u8(a+1, a8, 4); // ____ ____ ___1 ___0
  34. a8 = vld1q_lane_u8(a+2, a8, 8); // ____ ___2 ___1 ___0
  35. a8 = vld1q_lane_u8(a+3, a8, 12); // ___3 ___2 ___1 ___0
  36. auto a32 = (uint32x4_t)a8; //
  37. return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101)); // 3333 2222 1111 0000
  38. }
  39. inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
  40. uint8x16_t a8 = vdupq_n_u8(0); // ____ ____ ____ ____
  41. a8 = vld1q_lane_u8(a+0, a8, 0); // ____ ____ ____ ___0
  42. a8 = vld1q_lane_u8(a+1, a8, 4); // ____ ____ ___1 ___0
  43. auto a32 = (uint32x4_t)a8; //
  44. return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101)); // ____ ____ 1111 0000
  45. }
  46. } // namespace