diff options
author | mtklein <mtklein@chromium.org> | 2015-02-23 09:39:27 -0800 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2015-02-23 09:39:27 -0800 |
commit | 50d2b3114b3e59dc84811881591bf25b2c1ecb9f (patch) | |
tree | 29833d04dfc2c575bfabb7a0b561b7e88260fdc4 /src | |
parent | 259656779334689ab1624ec4e2e234b35fe4024b (diff) |
Sketch SkPMFloat
BUG=skia:
Review URL: https://codereview.chromium.org/936633002
Diffstat (limited to 'src')
-rw-r--r-- | src/core/SkPMFloat.h | 50 | ||||
-rw-r--r-- | src/opts/SkPMFloat_SSE2.h | 36 | ||||
-rw-r--r-- | src/opts/SkPMFloat_neon.h | 42 | ||||
-rw-r--r-- | src/opts/SkPMFloat_none.h | 28 |
4 files changed, 156 insertions, 0 deletions
diff --git a/src/core/SkPMFloat.h b/src/core/SkPMFloat.h new file mode 100644 index 0000000000..26bc9cce07 --- /dev/null +++ b/src/core/SkPMFloat.h @@ -0,0 +1,50 @@ +#ifndef SkPM_DEFINED +#define SkPM_DEFINED + +#include "SkTypes.h" +#include "SkColor.h" + +// As usual, there are two ways to increase alignment... the MSVC way and the everyone-else way. +#ifdef _MSC_VER + #define ALIGN(N) __declspec(align(N)) +#else + #define ALIGN(N) __attribute__((aligned(N))) +#endif + +// A pre-multiplied color in the same order as SkPMColor storing each component as a float. +struct ALIGN(16) SkPMFloat { + float fColor[4]; + + float a() const { return fColor[SK_A32_SHIFT / 8]; } + float r() const { return fColor[SK_R32_SHIFT / 8]; } + float g() const { return fColor[SK_G32_SHIFT / 8]; } + float b() const { return fColor[SK_B32_SHIFT / 8]; } + + void setA(float val) { fColor[SK_A32_SHIFT / 8] = val; } + void setR(float val) { fColor[SK_R32_SHIFT / 8] = val; } + void setG(float val) { fColor[SK_G32_SHIFT / 8] = val; } + void setB(float val) { fColor[SK_B32_SHIFT / 8] = val; } + + void set(SkPMColor); + + SkPMColor get() const; // May SkASSERT(this->isValid()). + SkPMColor clamped() const; // Will clamp all values to [0,1], then SkASSERT(this->isValid()). + + bool isValid() const { + return this->a() >= 0 && this->a() <= 1 + && this->r() >= 0 && this->r() <= this->a() + && this->g() >= 0 && this->g() <= this->a() + && this->b() >= 0 && this->b() <= this->a(); + } +}; +#undef ALIGN + +#if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2 + #include "../opts/SkPMFloat_SSE2.h" +#elif defined(__ARM_NEON__) + #include "../opts/SkPMFloat_neon.h" +#else + #include "../opts/SkPMFloat_none.h" +#endif + +#endif//SkPM_DEFINED diff --git a/src/opts/SkPMFloat_SSE2.h b/src/opts/SkPMFloat_SSE2.h new file mode 100644 index 0000000000..7bacf56af7 --- /dev/null +++ b/src/opts/SkPMFloat_SSE2.h @@ -0,0 +1,36 @@ +#include "SkColorPriv.h" +#include "SkPMFloat.h" +#include <emmintrin.h> + +// For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bits (fix8_16), +// then widen those to 8-bit-in-32-bits (fix8_32), convert those to floats (scaled), +// then finally scale those down from [0.0f, 255.0f] to [0.0f, 1.0f] into fColor. + +// get() and clamped() do the opposite, working from [0.0f, 1.0f] floats to [0.0f, 255.0f], +// to 8-bit-in-32-bit, to 8-bit-in-16-bit, back down to 8-bit components. +// _mm_packus_epi16() gives us clamping for free while narrowing. + +inline void SkPMFloat::set(SkPMColor c) { + SkPMColorAssert(c); + __m128i fix8 = _mm_set_epi32(0,0,0,c), + fix8_16 = _mm_unpacklo_epi8 (fix8, _mm_setzero_si128()), + fix8_32 = _mm_unpacklo_epi16(fix8_16, _mm_setzero_si128()); + __m128 scaled = _mm_cvtepi32_ps(fix8_32); + _mm_store_ps(fColor, _mm_mul_ps(scaled, _mm_set1_ps(1.0f/255.0f))); + SkASSERT(this->isValid()); +} + +inline SkPMColor SkPMFloat::get() const { + SkASSERT(this->isValid()); + return this->clamped(); // At the moment, we don't know anything faster. +} + +inline SkPMColor SkPMFloat::clamped() const { + __m128 scaled = _mm_mul_ps(_mm_load_ps(fColor), _mm_set1_ps(255.0f)); + __m128i fix8_32 = _mm_cvtps_epi32(scaled), + fix8_16 = _mm_packus_epi16(fix8_32, fix8_32), + fix8 = _mm_packus_epi16(fix8_16, fix8_16); + SkPMColor c = _mm_cvtsi128_si32(fix8); + SkPMColorAssert(c); + return c; +} diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h new file mode 100644 index 0000000000..12e857a50f --- /dev/null +++ b/src/opts/SkPMFloat_neon.h @@ -0,0 +1,42 @@ +#include "SkColorPriv.h" +#include "SkPMFloat.h" +#include <arm_neon.h> + +// For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bits (fix8_16), +// then widen those to 8-bit-in-32-bits (fix8_32), convert those to floats (scaled), +// then finally scale those down from [0.0f, 255.0f] to [0.0f, 1.0f] into fColor. + +// get() and clamped() do the opposite, working from [0.0f, 1.0f] floats to [0.0f, 255.0f], +// to 8-bit-in-32-bit, to 8-bit-in-16-bit, back down to 8-bit components. +// clamped() uses vqmovn to clamp while narrowing instead of just narrowing with vmovn. + +inline void SkPMFloat::set(SkPMColor c) { + SkPMColorAssert(c); + uint8x8_t fix8 = vdup_n_u32(c); + uint16x8_t fix8_16 = vmovl_u8(fix8); + uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16)); + float32x4_t scaled = vcvtq_f32_u32(fix8_32); + vst1q_f32(fColor, vmulq_f32(scaled, vdupq_n_f32(1.0f/255.0f))); + SkASSERT(this->isValid()); +} + +inline SkPMColor SkPMFloat::get() const { + SkASSERT(this->isValid()); + float32x4_t scaled = vmulq_f32(vld1q_f32(fColor), vdupq_n_f32(255.0f)); + uint32x4_t fix8_32 = vcvtq_u32_f32(scaled); + uint16x4_t fix8_16 = vmovn_u32(fix8_32); + uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); + SkPMColor c = vget_lane_u32(fix8, 0); + SkPMColorAssert(c); + return c; +} + +inline SkPMColor SkPMFloat::clamped() const { + float32x4_t scaled = vmulq_f32(vld1q_f32(fColor), vdupq_n_f32(255.0f)); + uint32x4_t fix8_32 = vcvtq_u32_f32(scaled); + uint16x4_t fix8_16 = vqmovn_u32(fix8_32); + uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0))); + SkPMColor c = vget_lane_u32(fix8, 0); + SkPMColorAssert(c); + return c; +} diff --git a/src/opts/SkPMFloat_none.h b/src/opts/SkPMFloat_none.h new file mode 100644 index 0000000000..dee7ce6862 --- /dev/null +++ b/src/opts/SkPMFloat_none.h @@ -0,0 +1,28 @@ +#include "SkColorPriv.h" +#include "SkPMFloat.h" + +inline void SkPMFloat::set(SkPMColor c) { + float scale = 1.0f / 255.0f; + this->setA(SkGetPackedA32(c) * scale); + this->setR(SkGetPackedR32(c) * scale); + this->setG(SkGetPackedG32(c) * scale); + this->setB(SkGetPackedB32(c) * scale); + SkASSERT(this->isValid()); +} + +inline SkPMColor SkPMFloat::get() const { + SkASSERT(this->isValid()); + return SkPackARGB32(this->a() * 255, this->r() * 255, this->g() * 255, this->b() * 255); +} + +inline SkPMColor SkPMFloat::clamped() const { + float a = this->a(), + r = this->r(), + g = this->g(), + b = this->b(); + a = a < 0 ? 0 : (a > 1 ? 1 : a); + r = r < 0 ? 0 : (r > 1 ? 1 : r); + g = g < 0 ? 0 : (g > 1 ? 1 : g); + b = b < 0 ? 0 : (b > 1 ? 1 : b); + return SkPackARGB32(a * 255, r * 255, g * 255, b * 255); +} |