fff055cc5f
These are basically inlined, 4-at-a-time versions of our existing functions,
but cut down to avoid any work that's only necessary outside [0,1].
Both f16 and f32 denorms should work fine modulo the usual ARMv7 NEON denorm==zero caveat.
In exchange for a little speed, f32->f16 does not round properly.
Instead it truncates, so it's never off by more than 1 bit.
Support for finite values >1 or <0 is straightforward to add back.
>1 might already work as-is.
Getting close to _u16 performance:
micros bench
261.13 xferu64_bw_1_opaque_u16
1833.51 xferu64_bw_1_alpha_u16
2762.32 ? xferu64_aa_1_opaque_u16
3334.29 xferu64_aa_1_alpha_u16
249.78 xferu64_bw_1_opaque_f16
3383.18 xferu64_bw_1_alpha_f16
4214.72 xferu64_aa_1_opaque_f16
4701.19 xferu64_aa_1_alpha_f16
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1685133005
Committed: https://skia.googlesource.com/skia/+/9ea11a4235b3e3521cc8bf914a27c2d0dc062db9
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
Review URL: https://codereview.chromium.org/1685133005
102 lines
2.8 KiB
C++
102 lines
2.8 KiB
C++
/*
|
|
* Copyright 2016 Google Inc.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license that can be
|
|
* found in the LICENSE file.
|
|
*/
|
|
|
|
#include "Test.h"
|
|
#include "SkColor.h"
|
|
#include "SkHalf.h"
|
|
#include "SkOpts.h"
|
|
#include "SkPixmap.h"
|
|
#include "SkRandom.h"
|
|
|
|
static bool eq_within_half_float(float a, float b) {
|
|
const float kTolerance = 1.0f / (1 << (8 + 10));
|
|
|
|
SkHalf ha = SkFloatToHalf(a);
|
|
SkHalf hb = SkFloatToHalf(b);
|
|
float a2 = SkHalfToFloat(ha);
|
|
float b2 = SkHalfToFloat(hb);
|
|
return fabsf(a2 - b2) <= kTolerance;
|
|
}
|
|
|
|
static bool eq_within_half_float(const SkPM4f& a, const SkPM4f& b) {
|
|
for (int i = 0; i < 4; ++i) {
|
|
if (!eq_within_half_float(a.fVec[i], b.fVec[i])) {
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
DEF_TEST(color_half_float, reporter) {
|
|
const int w = 100;
|
|
const int h = 100;
|
|
|
|
SkImageInfo info = SkImageInfo::Make(w, h, kRGBA_F16_SkColorType, kPremul_SkAlphaType);
|
|
|
|
SkAutoPixmapStorage pm;
|
|
pm.alloc(info);
|
|
REPORTER_ASSERT(reporter, pm.getSafeSize() == SkToSizeT(w * h * sizeof(uint64_t)));
|
|
|
|
SkColor4f c4 { 0.5f, 1, 0.5f, 0.25f };
|
|
pm.erase(c4);
|
|
|
|
SkPM4f origpm4 = c4.premul();
|
|
for (int y = 0; y < pm.height(); ++y) {
|
|
for (int x = 0; x < pm.width(); ++x) {
|
|
SkPM4f pm4 = SkPM4f::FromF16(pm.addrF16(x, y));
|
|
REPORTER_ASSERT(reporter, eq_within_half_float(origpm4, pm4));
|
|
}
|
|
}
|
|
}
|
|
|
|
DEF_TEST(float_to_half, reporter) {
|
|
const float fs[] = { 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0 };
|
|
const uint16_t hs[] = { 0x3c00, 0x4000, 0x4200, 0x4400, 0x4500, 0x4600, 0x4700 };
|
|
|
|
uint16_t hscratch[7];
|
|
SkOpts::float_to_half(hscratch, fs, 7);
|
|
REPORTER_ASSERT(reporter, 0 == memcmp(hscratch, hs, sizeof(hs)));
|
|
|
|
float fscratch[7];
|
|
SkOpts::half_to_float(fscratch, hs, 7);
|
|
REPORTER_ASSERT(reporter, 0 == memcmp(fscratch, fs, sizeof(fs)));
|
|
}
|
|
|
|
DEF_TEST(HalfToFloat_01, r) {
|
|
for (uint16_t h = 0; h < 0x8000; h++) {
|
|
float f = SkHalfToFloat(h);
|
|
if (f >= 0 && f <= 1) {
|
|
REPORTER_ASSERT(r, SkHalfToFloat_01(h)[0] == f);
|
|
REPORTER_ASSERT(r, SkFloatToHalf_01(SkHalfToFloat_01(h)) == h);
|
|
}
|
|
}
|
|
}
|
|
|
|
DEF_TEST(FloatToHalf_01, r) {
|
|
#if 0
|
|
for (uint32_t bits = 0; bits < 0x80000000; bits++) {
|
|
#else
|
|
SkRandom rand;
|
|
for (int i = 0; i < 1000000; i++) {
|
|
uint32_t bits = rand.nextU();
|
|
#endif
|
|
float f;
|
|
memcpy(&f, &bits, 4);
|
|
if (f >= 0 && f <= 1) {
|
|
uint16_t h1 = (uint16_t)SkFloatToHalf_01(Sk4f(f,0,0,0)),
|
|
h2 = SkFloatToHalf(f);
|
|
bool ok = (h1 == h2 || h1 == h2-1);
|
|
REPORTER_ASSERT(r, ok);
|
|
if (!ok) {
|
|
SkDebugf("%08x (%d) -> %04x (%d), want %04x (%d)\n",
|
|
bits, bits>>23, h1, h1>>10, h2, h2>>10);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|