diff options
| -rw-r--r-- | src/core/SkNx.h | 10 | ||||
| -rw-r--r-- | src/opts/SkNx_neon.h | 8 | ||||
| -rw-r--r-- | src/opts/SkNx_sse.h | 5 | ||||
| -rw-r--r-- | tests/SkNxTest.cpp | 11 |
4 files changed, 34 insertions, 0 deletions
diff --git a/src/core/SkNx.h b/src/core/SkNx.h index 14af1186d4..0b41365e82 100644 --- a/src/core/SkNx.h +++ b/src/core/SkNx.h @@ -97,6 +97,11 @@ struct SkNx { Half::Store3(ptr, a.fLo, b.fLo, c.fLo); Half::Store3(ptr + 3*N/2*sizeof(T), a.fHi, b.fHi, c.fHi); } + AI static void Store2(void* vptr, const SkNx& a, const SkNx& b) { + auto ptr = (char*)vptr; + Half::Store2(ptr, a.fLo, b.fLo); + Half::Store2(ptr + 2*N/2*sizeof(T), a.fHi, b.fHi); + } AI bool anyTrue() const { return fLo.anyTrue() || fHi.anyTrue(); } AI bool allTrue() const { return fLo.allTrue() && fHi.allTrue(); } @@ -200,6 +205,11 @@ struct SkNx<1,T> { b.store(ptr + 1*sizeof(T)); c.store(ptr + 2*sizeof(T)); } + AI static void Store2(void* vptr, const SkNx& a, const SkNx& b) { + auto ptr = (char*)vptr; + a.store(ptr + 0*sizeof(T)); + b.store(ptr + 1*sizeof(T)); + } AI bool anyTrue() const { return fVal != 0; } AI bool allTrue() const { return fVal != 0; } diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h index 52247c8f66..232903ab67 100644 --- a/src/opts/SkNx_neon.h +++ b/src/opts/SkNx_neon.h @@ -34,6 +34,14 @@ public: AI static SkNx Load(const void* ptr) { return vld1_f32((const float*)ptr); } AI void store(void* ptr) const { vst1_f32((float*)ptr, fVec); } + AI static void Store2(void* dst, const SkNx& a, const SkNx& b) { + float32x2x2_t ab = {{ + a.fVec, + b.fVec, + }}; + vst2_f32((float*) dst, ab); + } + AI static void Store3(void* dst, const SkNx& a, const SkNx& b, const SkNx& c) { float32x2x3_t abc = {{ a.fVec, diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h index b4ae0cbe2b..bd5c58e261 100644 --- a/src/opts/SkNx_sse.h +++ b/src/opts/SkNx_sse.h @@ -29,6 +29,11 @@ public: AI void store(void* ptr) const { _mm_storel_pi((__m64*)ptr, fVec); } + AI static void Store2(void* dst, const SkNx& a, const SkNx& b) { + auto vals = _mm_unpacklo_ps(a.fVec, b.fVec); + _mm_storeu_ps((float*)dst, vals); + } + AI static void Store3(void* dst, const SkNx& a, const SkNx& b, const SkNx& c) { auto lo = _mm_setr_ps(a[0], b[0], c[0], a[1]), hi = _mm_setr_ps(b[1], c[1], 0, 0); diff --git a/tests/SkNxTest.cpp b/tests/SkNxTest.cpp index 9901f7a41d..c27b88c12b 100644 --- a/tests/SkNxTest.cpp +++ b/tests/SkNxTest.cpp @@ -424,6 +424,17 @@ DEF_TEST(Sk4f_Load2, r) { REPORTER_ASSERT(r, y[3] == 7); } +DEF_TEST(Sk2f_Store2, r) { + Sk2f p0{0, 2}; + Sk2f p1{1, 3}; + float dst[4]; + Sk2f::Store2(dst, p0, p1); + REPORTER_ASSERT(r, dst[0] == 0); + REPORTER_ASSERT(r, dst[1] == 1); + REPORTER_ASSERT(r, dst[2] == 2); + REPORTER_ASSERT(r, dst[3] == 3); +} + DEF_TEST(Sk2f_Store3, r) { Sk2f p0{0, 3}; Sk2f p1{1, 4}; |