/*
* Copyright 2013 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "SkMipMap.h"
#include "SkBitmap.h"
#include "SkColorData.h"
#include "SkHalf.h"
#include "SkImageInfoPriv.h"
#include "SkMathPriv.h"
#include "SkNx.h"
#include "SkTo.h"
#include "SkTypes.h"
#include <new>
//
// ColorTypeFilter is the "Type" we pass to some downsample template functions.
// It controls how we expand a pixel into a large type, with space between each component,
// so we can then perform our simple filter (either box or triangle) and store the intermediates
// in the expanded type.
//
struct ColorTypeFilter_8888 {
typedef uint32_t Type;
static Sk4h Expand(uint32_t x) {
return SkNx_cast<uint16_t>(Sk4b::Load(&x));
}
static uint32_t Compact(const Sk4h& x) {
uint32_t r;
SkNx_cast<uint8_t>(x).store(&r);
return r;
}
};
struct ColorTypeFilter_565 {
typedef uint16_t Type;
static uint32_t Expand(uint16_t x) {
return (x & ~SK_G16_MASK_IN_PLACE) | ((x & SK_G16_MASK_IN_PLACE) << 16);
}
static uint16_t Compact(uint32_t x) {
return (x & ~SK_G16_MASK_IN_PLACE) | ((x >> 16) & SK_G16_MASK_IN_PLACE);
}
};
struct ColorTypeFilter_4444 {
typedef uint16_t Type;
static uint32_t Expand(uint16_t x) {
return (x & 0xF0F) | ((x & ~0xF0F) << 12);
}
static uint16_t Compact(uint32_t x) {
return (x & 0xF0F) | ((x >> 12) & ~0xF0F);
}
};
struct ColorTypeFilter_8 {
typedef uint8_t Type;
static unsigned Expand(unsigned x) {
return x;
}
static uint8_t Compact(unsigned x) {
return (uint8_t)x;
}
};
struct ColorTypeFilter_F16 {
typedef uint64_t Type; // SkHalf x4
static Sk4f Expand(uint64_t x) {
return SkHalfToFloat_finite_ftz(x);
}
static uint64_t Compact(const Sk4f& x) {
uint64_t r;
SkFloatToHalf_finite_ftz(x).store(&r);
return r;
}
};
template <typename T> T add_121(const T& a, const T& b, const T& c) {
return a + b + b + c;
}
template <typename T> T shift_right(const T& x, int bits) {
return x >> bits;
}
Sk4f shift_right(const Sk4f& x, int bits) {
return x * (1.0f / (1 << bits));
}
template <typename T> T shift_left(const T& x, int bits) {
return x << bits;
}
Sk4f shift_left(const Sk4f& x, int bits) {
return x * (1 << bits);
}
//
// To produce each mip level, we need to filter down by 1/2 (e.g. 100x100 -> 50,50)
// If the starting dimension is odd, we floor the size of the lower level (e.g. 101 -> 50)
// In those (odd) cases, we use a triangle filter, with 1-pixel overlap between samplings,
// else for even cases, we just use a 2x box filter.
//
// This produces 4 possible isotropic filters: 2x2 2x3 3x2 3x3 where WxH indicates the number of
// src pixels we need to sample in each dimension to produce 1 dst pixel.
//
// OpenGL expects a full mipmap stack to contain anisotropic space as well.
// This means a 100x1 image would continue down to a 50x1 image, 25x1 image...
// Because of this, we need 4 more anisotropic filters: 1x2, 1x3, 2x1, 3x1.
template <typename F> void downsample_1_2(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto p1 = (const typename F::Type*)((const char*)p0 + srcRB);
auto d = static_cast<typename F::Type*>(dst);
for (int i = 0; i < count; ++i) {
auto c00 = F::Expand(p0[0]);
auto c10 = F::Expand(p1[0]);
auto c = c00 + c10;
d[i] = F::Compact(shift_right(c, 1));
p0 += 2;
p1 += 2;
}
}
template <typename F> void downsample_1_3(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto p1 = (const typename F::Type*)((const char*)p0 + srcRB);
auto p2 = (const typename F::Type*)((const char*)p1 + srcRB);
auto d = static_cast<typename F::Type*>(dst);
for (int i = 0; i < count; ++i) {
auto c00 = F::Expand(p0[0]);
auto c10 = F::Expand(p1[0]);
auto c20 = F::Expand(p2[0]);
auto c = add_121(c00, c10, c20);
d[i] = F::Compact(shift_right(c, 2));
p0 += 2;
p1 += 2;
p2 += 2;
}
}
template <typename F> void downsample_2_1(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto d = static_cast<typename F::Type*>(dst);
for (int i = 0; i < count; ++i) {
auto c00 = F::Expand(p0[0]);
auto c01 = F::Expand(p0[1]);
auto c = c00 + c01;
d[i] = F::Compact(shift_right(c, 1));
p0 += 2;
}
}
template <typename F> void downsample_2_2(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto p1 = (const typename F::Type*)((const char*)p0 + srcRB);
auto d = static_cast<typename F::Type*>(dst);
for (int i = 0; i < count; ++i) {
auto c00 = F::Expand(p0[0]);
auto c01 = F::Expand(p0[1]);
auto c10 = F::Expand(p1[0]);
auto c11 = F::Expand(p1[1]);
auto c = c00 + c10 + c01 + c11;
d[i] = F::Compact(shift_right(c, 2));
p0 += 2;
p1 += 2;
}
}
template <typename F> void downsample_2_3(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto p1 = (const typename F::Type*)((const char*)p0 + srcRB);
auto p2 = (const typename F::Type*)((const char*)p1 + srcRB);
auto d = static_cast<typename F::Type*>(dst);
for (int i = 0; i < count; ++i) {
auto c00 = F::Expand(p0[0]);
auto c01 = F::Expand(p0[1]);
auto c10 = F::Expand(p1[0]);
auto c11 = F::Expand(p1[1]);
auto c20 = F::Expand(p2[0]);
auto c21 = F::Expand(p2[1]);
auto c = add_121(c00, c10, c20) + add_121(c01, c11, c21);
d[i] = F::Compact(shift_right(c, 3));
p0 += 2;
p1 += 2;
p2 += 2;
}
}
template <typename F> void downsample_3_1(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto d = static_cast<typename F::Type*>(dst);
auto c02 = F::Expand(p0[0]);
for (int i = 0; i < count; ++i) {
auto c00 = c02;
auto c01 = F::Expand(p0[1]);
c02 = F::Expand(p0[2]);
auto c = add_121(c00, c01, c02);
d[i] = F::Compact(shift_right(c, 2));
p0 += 2;
}
}
template <typename F> void downsample_3_2(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto p1 = (const typename F::Type*)((const char*)p0 + srcRB);
auto d = static_cast<typename F::Type*>(dst);
// Given pixels:
// a0 b0 c0 d0 e0 ...
// a1 b1 c1 d1 e1 ...
// We want:
// (a0 + 2*b0 + c0 + a1 + 2*b1 + c1) / 8
// (c0 + 2*d0 + e0 + c1 + 2*d1 + e1) / 8
// ...
auto c0 = F::Expand(p0[0]);
auto c1 = F::Expand(p1[0]);
auto c = c0 + c1;
for (int i = 0; i < count; ++i) {
auto a = c;
auto b0 = F::Expand(p0[1]);
auto b1 = F::Expand(p1[1]);
auto b = b0 + b0 + b1 + b1;
c0 = F::Expand(p0[2]);
c1 = F::Expand(p1[2]);
c = c0 + c1;
auto sum = a + b + c;
d[i] = F::Compact(shift_right(sum, 3));
p0 += 2;
p1 += 2;
}
}
template <typename F> void downsample_3_3(void* dst, const void* src, size_t srcRB, int count) {
SkASSERT(count > 0);
auto p0 = static_cast<const typename F::Type*>(src);
auto p1 = (const typename F::Type*)((const char*)p0 + srcRB);
auto p2 = (const typename F::Type*)((const char*)p1 + srcRB);
auto d = static_cast<typename F::Type*>(dst);
// Given pixels:
// a0 b0 c0 d0 e0 ...
// a1 b1 c1 d1 e1 ...
// a2 b2 c2 d2 e2 ...
// We want:
// (a0 + 2*b0 + c0 + 2*a1 + 4*b1 + 2*c1 + a2 + 2*b2 + c2) / 16
// (c0 + 2*d0 + e0 + 2*c1 + 4*d1 + 2*e1 + c2 + 2*d2 + e2) / 16
// ...
auto c0 = F::Expand(p0[0]);
auto c1 = F::Expand(p1[0]);
auto c2 = F::Expand(p2[0]);
auto c = add_121(c0, c1, c2);
for (int i = 0; i < count; ++i) {
auto a = c;
auto b0 = F::Expand(p0[1]);
auto b1 = F::Expand(p1[1]);
auto b2 = F::Expand(p2[1]);
auto b = shift_left(add_121(b0, b1, b2), 1);
c0 = F::Expand(p0[2]);
c1 = F::Expand(p1[2]);
c2 = F::Expand(p2[2]);
c = add_121(c0, c1, c2);
auto sum = a + b + c;
d[i] = F::Compact(shift_right(sum, 4));
p0 += 2;
p1 += 2;
p2 += 2;
}
}
///////////////////////////////////////////////////////////////////////////////////////////////////
size_t SkMipMap::AllocLevelsSize(int levelCount, size_t pixelSize) {
if (levelCount < 0) {
return 0;
}
int64_t size = sk_64_mul(levelCount + 1, sizeof(Level)) + pixelSize;
if (!SkTFitsIn<int32_t>(size)) {
return 0;
}
return SkTo<int32_t>(size);
}
SkMipMap* SkMipMap::Build(const SkPixmap& src, SkDiscardableFactoryProc fact) {
typedef void FilterProc(void*, const void* srcPtr, size_t srcRB, int count);
FilterProc* proc_1_2 = nullptr;
FilterProc* proc_1_3 = nullptr;
FilterProc* proc_2_1 = nullptr;
FilterProc* proc_2_2 = nullptr;
FilterProc* proc_2_3 = nullptr;
FilterProc* proc_3_1 = nullptr;
FilterProc* proc_3_2 = nullptr;
FilterProc* proc_3_3 = nullptr;
const SkColorType ct = src.colorType();
const SkAlphaType at = src.alphaType();
switch (ct) {
case kRGBA_8888_SkColorType:
case kBGRA_8888_SkColorType:
proc_1_2 = downsample_1_2<ColorTypeFilter_8888>;
proc_1_3 = downsample_1_3<ColorTypeFilter_8888>;
proc_2_1 = downsample_2_1<ColorTypeFilter_8888>;
proc_2_2 = downsample_2_2<ColorTypeFilter_8888>;
proc_2_3 = downsample_2_3<ColorTypeFilter_8888>;
proc_3_1 = downsample_3_1<ColorTypeFilter_8888>;
proc_3_2 = downsample_3_2<ColorTypeFilter_8888>;
proc_3_3 = downsample_3_3<ColorTypeFilter_8888>;
break;
case kRGB_565_SkColorType:
proc_1_2 = downsample_1_2<ColorTypeFilter_565>;
proc_1_3 = downsample_1_3<ColorTypeFilter_565>;
proc_2_1 = downsample_2_1<ColorTypeFilter_565>;
proc_2_2 = downsample_2_2<ColorTypeFilter_565>;
proc_2_3 = downsample_2_3<ColorTypeFilter_565>;
proc_3_1 = downsample_3_1<ColorTypeFilter_565>;
proc_3_2 = downsample_3_2<ColorTypeFilter_565>;
proc_3_3 = downsample_3_3<ColorTypeFilter_565>;
break;
case kARGB_4444_SkColorType:
proc_1_2 = downsample_1_2<ColorTypeFilter_4444>;
proc_1_3 = downsample_1_3<ColorTypeFilter_4444>;
proc_2_1 = downsample_2_1<ColorTypeFilter_4444>;
proc_2_2 = downsample_2_2<ColorTypeFilter_4444>;
proc_2_3 = downsample_2_3<ColorTypeFilter_4444>;
proc_3_1 = downsample_3_1<ColorTypeFilter_4444>;
proc_3_2 = downsample_3_2<ColorTypeFilter_4444>;
proc_3_3 = downsample_3_3<ColorTypeFilter_4444>;
break;
case kAlpha_8_SkColorType:
case kGray_8_SkColorType:
proc_1_2 = downsample_1_2<ColorTypeFilter_8>;
proc_1_3 = downsample_1_3<ColorTypeFilter_8>;
proc_2_1 = downsample_2_1<ColorTypeFilter_8>;
proc_2_2 = downsample_2_2<ColorTypeFilter_8>;
proc_2_3 = downsample_2_3<ColorTypeFilter_8>;
proc_3_1 = downsample_3_1<ColorTypeFilter_8>;
proc_3_2 = downsample_3_2<ColorTypeFilter_8>;
proc_3_3 = downsample_3_3<ColorTypeFilter_8>;
break;
case kRGBA_F16_SkColorType:
proc_1_2 = downsample_1_2<ColorTypeFilter_F16>;
proc_1_3 = downsample_1_3<ColorTypeFilter_F16>;
proc_2_1 = downsample_2_1<ColorTypeFilter_F16>;
proc_2_2 = downsample_2_2<ColorTypeFilter_F16>;
proc_2_3 = downsample_2_3<ColorTypeFilter_F16>;
proc_3_1 = downsample_3_1<ColorTypeFilter_F16>;
proc_3_2 = downsample_3_2<ColorTypeFilter_F16>;
proc_3_3 = downsample_3_3<ColorTypeFilter_F16>;
break;
default:
return nullptr;
}
if (src.width() <= 1 && src.height() <= 1) {
return nullptr;
}
// whip through our loop to compute the exact size needed
size_t size = 0;
int countLevels = ComputeLevelCount(src.width(), src.height());
for (int currentMipLevel = countLevels; currentMipLevel >= 0; currentMipLevel--) {
SkISize mipSize = ComputeLevelSize(src.width(), src.height(), currentMipLevel);
size += SkColorTypeMinRowBytes(ct, mipSize.fWidth) * mipSize.fHeight;
}
size_t storageSize = SkMipMap::AllocLevelsSize(countLevels, size);
if (0 == storageSize) {
return nullptr;
}
SkMipMap* mipmap;
if (fact) {
SkDiscardableMemory* dm = fact(storageSize);
if (nullptr == dm) {
return nullptr;
}
mipmap = new SkMipMap(storageSize, dm);
} else {
mipmap = new SkMipMap(sk_malloc_throw(storageSize), storageSize);
}
// init
mipmap->fCS = sk_ref_sp(src.info().colorSpace());
mipmap->fCount = countLevels;
mipmap->fLevels = (Level*)mipmap->writable_data();
SkASSERT(mipmap->fLevels);
Level* levels = mipmap->fLevels;
uint8_t* baseAddr = (uint8_t*)&levels[countLevels];
uint8_t* addr = baseAddr;
int width = src.width();
int height = src.height();
uint32_t rowBytes;
SkPixmap srcPM(src);
// Depending on architecture and other factors, the pixel data alignment may need to be as
// large as 8 (for F16 pixels). See the comment on SkMipMap::Level.
SkASSERT(SkIsAlign8((uintptr_t)addr));
for (int i = 0; i < countLevels; ++i) {
FilterProc* proc;
if (height & 1) {
if (height == 1) { // src-height is 1
if (width & 1) { // src-width is 3
proc = proc_3_1;
} else { // src-width is 2
proc = proc_2_1;
}
} else { // src-height is 3
if (width & 1) {
if (width == 1) { // src-width is 1
proc = proc_1_3;
} else { // src-width is 3
proc = proc_3_3;
}
} else { // src-width is 2
proc = proc_2_3;
}
}
} else { // src-height is 2
if (width & 1) {
if (width == 1) { // src-width is 1
proc = proc_1_2;
} else { // src-width is 3
proc = proc_3_2;
}
} else { // src-width is 2
proc = proc_2_2;
}
}
width = SkTMax(1, width >> 1);
height = SkTMax(1, height >> 1);
rowBytes = SkToU32(SkColorTypeMinRowBytes(ct, width));
// We make the Info w/o any colorspace, since that storage is not under our control, and
// will not be deleted in a controlled fashion. When the caller is given the pixmap for
// a given level, we augment this pixmap with fCS (which we do manage).
new (&levels[i].fPixmap) SkPixmap(SkImageInfo::Make(width, height, ct, at), addr, rowBytes);
levels[i].fScale = SkSize::Make(SkIntToScalar(width) / src.width(),
SkIntToScalar(height) / src.height());
const SkPixmap& dstPM = levels[i].fPixmap;
const void* srcBasePtr = srcPM.addr();
void* dstBasePtr = dstPM.writable_addr();
const size_t srcRB = srcPM.rowBytes();
for (int y = 0; y < height; y++) {
proc(dstBasePtr, srcBasePtr, srcRB, width);
srcBasePtr = (char*)srcBasePtr + srcRB * 2; // jump two rows
dstBasePtr = (char*)dstBasePtr + dstPM.rowBytes();
}
srcPM = dstPM;
addr += height * rowBytes;
}
SkASSERT(addr == baseAddr + size);
SkASSERT(mipmap->fLevels);
return mipmap;
}
int SkMipMap::ComputeLevelCount(int baseWidth, int baseHeight) {
if (baseWidth < 1 || baseHeight < 1) {
return 0;
}
// OpenGL's spec requires that each mipmap level have height/width equal to
// max(1, floor(original_height / 2^i)
// (or original_width) where i is the mipmap level.
// Continue scaling down until both axes are size 1.
const int largestAxis = SkTMax(baseWidth, baseHeight);
if (largestAxis < 2) {
// SkMipMap::Build requires a minimum size of 2.
return 0;
}
const int leadingZeros = SkCLZ(static_cast<uint32_t>(largestAxis));
// If the value 00011010 has 3 leading 0s then it has 5 significant bits
// (the bits which are not leading zeros)
const int significantBits = (sizeof(uint32_t) * 8) - leadingZeros;
// This is making the assumption that the size of a byte is 8 bits
// and that sizeof(uint32_t)'s implementation-defined behavior is 4.
int mipLevelCount = significantBits;
// SkMipMap does not include the base mip level.
// For example, it contains levels 1-x instead of 0-x.
// This is because the image used to create SkMipMap is the base level.
// So subtract 1 from the mip level count.
if (mipLevelCount > 0) {
--mipLevelCount;
}
return mipLevelCount;
}
SkISize SkMipMap::ComputeLevelSize(int baseWidth, int baseHeight, int level) {
if (baseWidth < 1 || baseHeight < 1) {
return SkISize::Make(0, 0);
}
int maxLevelCount = ComputeLevelCount(baseWidth, baseHeight);
if (level >= maxLevelCount || level < 0) {
return SkISize::Make(0, 0);
}
// OpenGL's spec requires that each mipmap level have height/width equal to
// max(1, floor(original_height / 2^i)
// (or original_width) where i is the mipmap level.
// SkMipMap does not include the base mip level.
// For example, it contains levels 1-x instead of 0-x.
// This is because the image used to create SkMipMap is the base level.
// So subtract 1 from the mip level to get the index stored by SkMipMap.
int width = SkTMax(1, baseWidth >> (level + 1));
int height = SkTMax(1, baseHeight >> (level + 1));
return SkISize::Make(width, height);
}
///////////////////////////////////////////////////////////////////////////////
bool SkMipMap::extractLevel(const SkSize& scaleSize, Level* levelPtr) const {
if (nullptr == fLevels) {
return false;
}
SkASSERT(scaleSize.width() >= 0 && scaleSize.height() >= 0);
#ifndef SK_SUPPORT_LEGACY_ANISOTROPIC_MIPMAP_SCALE
// Use the smallest scale to match the GPU impl.
const SkScalar scale = SkTMin(scaleSize.width(), scaleSize.height());
#else
// Ideally we'd pick the smaller scale, to match Ganesh. But ignoring one of the
// scales can produce some atrocious results, so for now we use the geometric mean.
// (https://bugs.chromium.org/p/skia/issues/detail?id=4863)
const SkScalar scale = SkScalarSqrt(scaleSize.width() * scaleSize.height());
#endif
if (scale >= SK_Scalar1 || scale <= 0 || !SkScalarIsFinite(scale)) {
return false;
}
SkScalar L = -SkScalarLog2(scale);
if (!SkScalarIsFinite(L)) {
return false;
}
SkASSERT(L >= 0);
int level = SkScalarFloorToInt(L);
SkASSERT(level >= 0);
if (level <= 0) {
return false;
}
if (level > fCount) {
level = fCount;
}
if (levelPtr) {
*levelPtr = fLevels[level - 1];
// need to augment with our colorspace
levelPtr->fPixmap.setColorSpace(fCS);
}
return true;
}
// Helper which extracts a pixmap from the src bitmap
//
SkMipMap* SkMipMap::Build(const SkBitmap& src, SkDiscardableFactoryProc fact) {
SkPixmap srcPixmap;
if (!src.peekPixels(&srcPixmap)) {
return nullptr;
}
return Build(srcPixmap, fact);
}
int SkMipMap::countLevels() const {
return fCount;
}
bool SkMipMap::getLevel(int index, Level* levelPtr) const {
if (nullptr == fLevels) {
return false;
}
if (index < 0) {
return false;
}
if (index > fCount - 1) {
return false;
}
if (levelPtr) {
*levelPtr = fLevels[index];
}
return true;
}