Make SkPMFloats store floats in [0,255] instead of [0,1].
This pushes the cost of the *255 and *1/255 conversions onto only those code
paths that need it. We're not doing it any more efficiently than can be done
with Sk4f.
In microbenchmark isolation, this is about a 15% speedup.
BUG=skia:
NOPRESUBMIT=true
Review URL: https://codereview.chromium.org/973603002
diff --git a/src/core/SkPMFloat.h b/src/core/SkPMFloat.h
index 936487e..b0cb314 100644
--- a/src/core/SkPMFloat.h
+++ b/src/core/SkPMFloat.h
@@ -5,7 +5,7 @@
#include "SkColor.h"
#include "Sk4x.h"
-// A pre-multiplied color storing each component as a float.
+// A pre-multiplied color storing each component as a float in the range [0, 255].
class SK_STRUCT_ALIGN(16) SkPMFloat {
public:
// Normal POD copies and do-nothing initialization.
@@ -30,10 +30,10 @@
void set(SkPMColor);
SkPMColor get() const; // May SkASSERT(this->isValid()). Some implementations may clamp.
- SkPMColor clamped() const; // Will clamp all values to [0,1], then SkASSERT(this->isValid()).
+ SkPMColor clamped() const; // Will clamp all values to [0, 255]. Then may assert isValid().
bool isValid() const {
- return this->a() >= 0 && this->a() <= 1
+ return this->a() >= 0 && this->a() <= 255
&& this->r() >= 0 && this->r() <= this->a()
&& this->g() >= 0 && this->g() <= this->a()
&& this->b() >= 0 && this->b() <= this->a();
diff --git a/src/opts/SkPMFloat_SSE2.h b/src/opts/SkPMFloat_SSE2.h
index c3dbbf2..f286080 100644
--- a/src/opts/SkPMFloat_SSE2.h
+++ b/src/opts/SkPMFloat_SSE2.h
@@ -2,11 +2,10 @@
#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.
+// then widen those to 8-bit-in-32-bits (fix8_32), and finally convert those to floats.
-// 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.
+// get() and clamped() do the opposite, working from floats 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) {
@@ -14,8 +13,7 @@
__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)));
+ _mm_store_ps(fColor, _mm_cvtepi32_ps(fix8_32));
SkASSERT(this->isValid());
}
@@ -25,8 +23,7 @@
}
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),
+ __m128i fix8_32 = _mm_cvtps_epi32(_mm_load_ps(fColor)),
fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
fix8 = _mm_packus_epi16(fix8_16, fix8_16);
SkPMColor c = _mm_cvtsi128_si32(fix8);
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h
index 004b658..15ba3a5 100644
--- a/src/opts/SkPMFloat_neon.h
+++ b/src/opts/SkPMFloat_neon.h
@@ -2,11 +2,10 @@
#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.
+// then widen those to 8-bit-in-32-bits (fix8_32), and finally convert those to floats.
-// 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.
+// get() and clamped() do the opposite, working from floats 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) {
@@ -14,15 +13,13 @@
uint8x8_t fix8 = (uint8x8_t)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)));
+ vst1q_f32(fColor, vcvtq_f32_u32(fix8_32));
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);
+ uint32x4_t fix8_32 = vcvtq_u32_f32(vld1q_f32(fColor));
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((uint32x2_t)fix8, 0);
@@ -31,8 +28,7 @@
}
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);
+ uint32x4_t fix8_32 = vcvtq_u32_f32(vld1q_f32(fColor));
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((uint32x2_t)fix8, 0);
diff --git a/src/opts/SkPMFloat_none.h b/src/opts/SkPMFloat_none.h
index a11fe24..448b509 100644
--- a/src/opts/SkPMFloat_none.h
+++ b/src/opts/SkPMFloat_none.h
@@ -1,17 +1,16 @@
#include "SkColorPriv.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);
+ this->setA(SkGetPackedA32(c));
+ this->setR(SkGetPackedR32(c));
+ this->setG(SkGetPackedG32(c));
+ this->setB(SkGetPackedB32(c));
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);
+ return SkPackARGB32(this->a(), this->r(), this->g(), this->b());
}
inline SkPMColor SkPMFloat::clamped() const {
@@ -19,9 +18,9 @@
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);
+ a = a < 0 ? 0 : (a > 255 ? 255 : a);
+ r = r < 0 ? 0 : (r > 255 ? 255 : r);
+ g = g < 0 ? 0 : (g > 255 ? 255 : g);
+ b = b < 0 ? 0 : (b > 255 ? 255 : b);
+ return SkPackARGB32(a, r, g, b);
}
diff --git a/tests/PMFloatTest.cpp b/tests/PMFloatTest.cpp
index f722e80..fceeb47 100644
--- a/tests/PMFloatTest.cpp
+++ b/tests/PMFloatTest.cpp
@@ -7,34 +7,34 @@
SkPMFloat pmf;
pmf.set(c);
- REPORTER_ASSERT(r, SkScalarNearlyEqual(1.0f, pmf.a()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.8f, pmf.r()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.6f, pmf.g()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.2f, pmf.b()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(255.0f, pmf.a()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(204.0f, pmf.r()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(153.0f, pmf.g()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual( 51.0f, pmf.b()));
REPORTER_ASSERT(r, c == pmf.get());
// Test clamping.
SkPMFloat unclamped;
- unclamped.setA(+2.0f);
- unclamped.setR(+0.2f);
- unclamped.setG(-0.2f);
- unclamped.setB(-5.0f);
+ unclamped.setA(+510.0f);
+ unclamped.setR(+153.0f);
+ unclamped.setG( +1.0f);
+ unclamped.setB( -0.2f);
SkPMFloat clamped;
clamped.set(unclamped.clamped());
- REPORTER_ASSERT(r, SkScalarNearlyEqual(1.0f, clamped.a()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.2f, clamped.r()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.0f, clamped.g()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.0f, clamped.b()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(255.0f, clamped.a()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(153.0f, clamped.r()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual( 1.0f, clamped.g()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual( 0.0f, clamped.b()));
// Test SkPMFloat <-> Sk4f conversion.
Sk4f fs = clamped;
- SkPMFloat scaled = fs.multiply(Sk4f(4,4,4,4));
+ SkPMFloat scaled = fs.multiply(0.25f);
- REPORTER_ASSERT(r, SkScalarNearlyEqual(4.0f, scaled.a()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.8f, scaled.r()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.0f, scaled.g()));
- REPORTER_ASSERT(r, SkScalarNearlyEqual(0.0f, scaled.b()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(63.75f, scaled.a()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual(38.25f, scaled.r()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual( 0.25f, scaled.g()));
+ REPORTER_ASSERT(r, SkScalarNearlyEqual( 0.00f, scaled.b()));
}
