OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2015 Google Inc. | 2 * Copyright 2015 Google Inc. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 #ifndef SkNx_neon_DEFINED | 8 #ifndef SkNx_neon_DEFINED |
9 #define SkNx_neon_DEFINED | 9 #define SkNx_neon_DEFINED |
10 | 10 |
(...skipping 370 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
381 union { int32x4_t v; int is[4]; } pun = {fVec}; | 381 union { int32x4_t v; int is[4]; } pun = {fVec}; |
382 return pun.is[k&3]; | 382 return pun.is[k&3]; |
383 } | 383 } |
384 | 384 |
385 SkNx operator + (const SkNx& o) const { return vaddq_s32(fVec, o.fVec); } | 385 SkNx operator + (const SkNx& o) const { return vaddq_s32(fVec, o.fVec); } |
386 SkNx operator - (const SkNx& o) const { return vsubq_s32(fVec, o.fVec); } | 386 SkNx operator - (const SkNx& o) const { return vsubq_s32(fVec, o.fVec); } |
387 SkNx operator * (const SkNx& o) const { return vmulq_s32(fVec, o.fVec); } | 387 SkNx operator * (const SkNx& o) const { return vmulq_s32(fVec, o.fVec); } |
388 | 388 |
389 SkNx operator & (const SkNx& o) const { return vandq_s32(fVec, o.fVec); } | 389 SkNx operator & (const SkNx& o) const { return vandq_s32(fVec, o.fVec); } |
390 SkNx operator | (const SkNx& o) const { return vorrq_s32(fVec, o.fVec); } | 390 SkNx operator | (const SkNx& o) const { return vorrq_s32(fVec, o.fVec); } |
391 SkNx operator ^ (const SkNx& o) const { return veorq_s32(fVec, o.fVec); } | |
392 | 391 |
393 SkNx operator << (int bits) const { SHIFT32(vshlq_n_s32, fVec, bits); } | 392 SkNx operator << (int bits) const { SHIFT32(vshlq_n_s32, fVec, bits); } |
394 SkNx operator >> (int bits) const { SHIFT32(vshrq_n_s32, fVec, bits); } | 393 SkNx operator >> (int bits) const { SHIFT32(vshrq_n_s32, fVec, bits); } |
395 | 394 |
396 SkNx operator == (const SkNx& o) const { | |
397 return vreinterpretq_s32_u32(vceqq_s32(fVec, o.fVec)); | |
398 } | |
399 SkNx operator < (const SkNx& o) const { | |
400 return vreinterpretq_s32_u32(vcltq_s32(fVec, o.fVec)); | |
401 } | |
402 SkNx operator > (const SkNx& o) const { | |
403 return vreinterpretq_s32_u32(vcgtq_s32(fVec, o.fVec)); | |
404 } | |
405 | |
406 static SkNx Min(const SkNx& a, const SkNx& b) { return vminq_s32(a.fVec, b.f
Vec); } | 395 static SkNx Min(const SkNx& a, const SkNx& b) { return vminq_s32(a.fVec, b.f
Vec); } |
407 // TODO as needed | 396 // TODO as needed |
408 | 397 |
409 SkNx thenElse(const SkNx& t, const SkNx& e) const { | |
410 return vbslq_s32(vreinterpretq_u32_s32(fVec), t.fVec, e.fVec); | |
411 } | |
412 | |
413 int32x4_t fVec; | 398 int32x4_t fVec; |
414 }; | 399 }; |
415 | 400 |
416 #undef SHIFT32 | 401 #undef SHIFT32 |
417 #undef SHIFT16 | 402 #undef SHIFT16 |
418 #undef SHIFT8 | 403 #undef SHIFT8 |
419 | 404 |
420 template<> inline Sk4i SkNx_cast<int, float>(const Sk4f& src) { | 405 template<> inline Sk4i SkNx_cast<int, float>(const Sk4f& src) { |
421 return vcvtq_s32_f32(src.fVec); | 406 return vcvtq_s32_f32(src.fVec); |
422 | 407 |
(...skipping 41 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
464 | 449 |
465 template<> inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) { | 450 template<> inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) { |
466 return vmovn_u16(vcombine_u16(src.fVec, src.fVec)); | 451 return vmovn_u16(vcombine_u16(src.fVec, src.fVec)); |
467 } | 452 } |
468 | 453 |
469 template<> inline Sk4b SkNx_cast<uint8_t, int>(const Sk4i& src) { | 454 template<> inline Sk4b SkNx_cast<uint8_t, int>(const Sk4i& src) { |
470 uint16x4_t _16 = vqmovun_s32(src.fVec); | 455 uint16x4_t _16 = vqmovun_s32(src.fVec); |
471 return vqmovn_u16(vcombine_u16(_16, _16)); | 456 return vqmovn_u16(vcombine_u16(_16, _16)); |
472 } | 457 } |
473 | 458 |
474 template<> inline Sk4i SkNx_cast<int, uint16_t>(const Sk4h& src) { | |
475 return vreinterpretq_s32_u32(vmovl_u16(src.fVec)); | |
476 } | |
477 | |
478 template<> inline Sk4h SkNx_cast<uint16_t, int>(const Sk4i& src) { | |
479 return vmovn_u32(vreinterpretq_u32_s32(src.fVec)); | |
480 } | |
481 | |
482 static inline Sk4i Sk4f_round(const Sk4f& x) { | 459 static inline Sk4i Sk4f_round(const Sk4f& x) { |
483 return vcvtq_s32_f32((x + 0.5f).fVec); | 460 return vcvtq_s32_f32((x + 0.5f).fVec); |
484 } | 461 } |
485 | 462 |
486 #endif//SkNx_neon_DEFINED | 463 #endif//SkNx_neon_DEFINED |
OLD | NEW |