Skip to content

Commit

Permalink
improvements gamma transfer, Rec 709, libwebp version changed
Browse files Browse the repository at this point in the history
  • Loading branch information
@awxkee
awxkee committed Oct 12, 2023
1 parent 3b55b4c commit c29f126
Show file tree
Hide file tree
Showing 7 changed files with 67 additions and 49 deletions.
4 changes: 2 additions & 2 deletions Package.swift
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,7 @@ let package = Package(
.package(url: "https://github.com/awxkee/libaom.swift.git", "1.0.0"..<"1.1.0"),
.package(url: "https://github.com/awxkee/libdav1d.swift.git", "1.0.0"..<"1.1.0"),
.package(url: "https://github.com/awxkee/libyuv.swift.git", "1.0.0"..<"1.1.0"),
.package(url: "https://github.com/awxkee/libwebp-ios.git", "1.0.0"..<"1.1.0"),
.package(url: "https://github.com/awxkee/libwebp-ios.git", "1.1.0"..<"1.2.0"),
.package(url: "https://github.com/kean/Nuke.git", "12.0.0"..<"13.0.0")
],
targets: [
Expand All @@ -39,7 +39,7 @@ let package = Package(
.product(name: "libaom", package: "libaom.swift"),
.product(name: "libdav1d", package: "libdav1d.swift"),
.product(name: "libyuv", package: "libyuv.swift"),
.product(name: "libwebp", package: "libwebp-ios")],
.product(name: "libwebp-ios", package: "libwebp-ios")],
publicHeadersPath: "include",
cSettings: [
.define("AVIF_CODEC_AOM_ENCODE", to: "1"),
Expand Down
22 changes: 21 additions & 1 deletion Sources/avifc/AVIFImageXForm.mm
View file
Original file line number Diff line number Diff line change
Expand Up @@ -222,6 +222,26 @@ - (_Nullable CGImageRef)formCGImage:(nonnull avifDecoder*)decoder scale:(CGFloat
U16:depth > 8 depth:depth half:depth > 8
primaries:lumaPrimaries components:components
gammaCorrection:gamma function:function matrix:nullptr];
} else if (colorPrimaries == AVIF_COLOR_PRIMARIES_BT709 /* Rec 709 */ &&
(transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_SMPTE2084
|| transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_HLG
|| transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_SMPTE428)) {
float lumaPrimaries[3] = { 0.2627f, 0.6780f, 0.0593f };
ColorGammaCorrection gamma = Rec709;
TransferFunction function;
if (transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_SMPTE2084) {
function = PQ;
} else if (transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_HLG) {
function = HLG;
} else {
function = SMPTE428;
}
colorSpace = CGColorSpaceCreateWithName(kCGColorSpaceITUR_709);
[HDRColorTransfer transfer:reinterpret_cast<uint8_t*>(pixelsData)
stride:stride width:newWidth height:newHeight
U16:depth > 8 depth:depth half:depth > 8
primaries:lumaPrimaries components:components
gammaCorrection:gamma function:function matrix:nullptr];
} else if (colorPrimaries == AVIF_COLOR_PRIMARIES_SMPTE432 /* Display P3 */ &&
transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_LINEAR) {
CGColorSpaceRef p3linear = NULL;
Expand All @@ -234,7 +254,7 @@ - (_Nullable CGImageRef)formCGImage:(nonnull avifDecoder*)decoder scale:(CGFloat
} else if (transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_SMPTE2084
|| transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_HLG
|| transferCharacteristics == AVIF_TRANSFER_CHARACTERISTICS_SMPTE428) {
// IF Transfer function but we don't know the color space the we will convert it always to display P3
// IF Transfer function but we don't know the color space the we will convert it always to Display P3
float lumaPrimaries[3] = { 0.2627f, 0.6780f, 0.0593f };
ColorGammaCorrection gamma = DisplayP3;
TransferFunction function;
Expand Down
8 changes: 8 additions & 0 deletions Sources/avifc/Color/Gamma.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,14 @@
#pragma clang fp contract(on) exceptions(ignore) reassociate(on)
#endif

float LinearITUR709ToITUR709(const float linear) {
if (linear <= 0.018f) {
return 4.5f * linear;
} else {
return 1.099f * powf_c(linear, 0.45f) - 0.099f;
}
}

float LinearSRGBToSRGB(const float linearValue) {
if (linearValue <= 0.0031308) {
return 12.92f * linearValue;
Expand Down
50 changes: 21 additions & 29 deletions Sources/avifc/Color/Gamma.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@ constexpr float alphaRec2020 = 1.09929682680944f;

float LinearSRGBToSRGB(const float linearValue);
float LinearRec2020ToRec2020(const float linear);
float LinearITUR709ToITUR709(const float linear);
float dciP3PQGammaCorrection(const float linear);

#if __arm64__
Expand All @@ -51,61 +52,52 @@ float dciP3PQGammaCorrection(const float linear);
static inline float32x4_t LinearITUR709ToITUR709(const float32x4_t linear) {
const float32x4_t level = vdupq_n_f32(0.018);

uint32x4_t mask = vcgtq_f32(linear, level);
uint32x4_t maskHigh = vcltq_f32(linear, level);
const uint32x4_t mask = vcgtq_f32(linear, level);
const uint32x4_t maskHigh = vcltq_f32(linear, level);

float32x4_t low = vbslq_f32(mask, vdupq_n_f32(0), linear);
float32x4_t high = vbslq_f32(maskHigh, vdupq_n_f32(0), linear);
float32x4_t low = linear;
float32x4_t high = linear;
low = vmulq_n_f32(low, 4.5f);

high = vsubq_f32(vmulq_n_f32(vpowq_f32(high, 0.45f), 1.099f), vdupq_n_f32(0.099f));

low = vbslq_f32(mask, vdupq_n_f32(0), low);
high = vbslq_f32(maskHigh, vdupq_n_f32(0), high);

float32x4_t result = vmaxq_f32(vaddq_f32(low, high), vdupq_n_f32(0));
return result;
}

static inline float32x4_t LinearSRGBToSRGB(const float32x4_t linear) {
const float32x4_t level = vdupq_n_f32(0.0031308);

uint32x4_t mask = vcgtq_f32(linear, level);
uint32x4_t maskHigh = vcltq_f32(linear, level);
const uint32x4_t mask = vcgtq_f32(linear, level);
const uint32x4_t maskHigh = vcltq_f32(linear, level);

float32x4_t low = vbslq_f32(mask, vdupq_n_f32(0), linear);
float32x4_t high = vbslq_f32(maskHigh, vdupq_n_f32(0), linear);
float32x4_t low = linear;
float32x4_t high = linear;
low = vmulq_n_f32(low, 12.92f);

high = vsubq_f32(vmulq_n_f32(vpowq_f32(high, 1.0f/2.4f), 1.055f), vdupq_n_f32(0.055f));
low = vbslq_f32(mask, vdupq_n_f32(0), low);
high = vbslq_f32(maskHigh, vdupq_n_f32(0), high);
float32x4_t result = vmaxq_f32(vaddq_f32(low, high), vdupq_n_f32(0));
return result;
}

static inline float32x4_t LinearRec2020ToRec2020(const float32x4_t linear) {
uint32x4_t mask = vcgtq_f32(linear, vdupq_n_f32(betaRec2020));
uint32x4_t maskHigh = vcltq_f32(linear, vdupq_n_f32(betaRec2020));

float32x4_t low = vbslq_f32(mask, vdupq_n_f32(0), linear);
float32x4_t high = vbslq_f32(maskHigh, vdupq_n_f32(0), linear);
const uint32x4_t mask = vcgtq_f32(linear, vdupq_n_f32(betaRec2020));
const uint32x4_t maskHigh = vcltq_f32(linear, vdupq_n_f32(betaRec2020));

float32x4_t low = linear;
float32x4_t high = linear;
low = vmulq_n_f32(low, 4.5f);
constexpr float fk = alphaRec2020 - 1;
high = vsubq_f32(vmulq_n_f32(vpowq_f32(high, 0.45f), alphaRec2020), vdupq_n_f32(fk));

low = vbslq_f32(mask, vdupq_n_f32(0), low);
high = vbslq_f32(maskHigh, vdupq_n_f32(0), high);
return vaddq_f32(low, high);
}

__attribute__((always_inline))
static inline float32x4_t applyMatrixNEON(vector<vector<float>> matrix, const float32x4_t v) {
const float32x4_t row1 = { matrix[0][0], matrix[0][1], matrix[0][2], 0.0f };
const float32x4_t row2 = { matrix[1][0], matrix[1][1], matrix[1][2], 0.0f };
const float32x4_t row3 = { matrix[2][0], matrix[2][1], matrix[2][2], 0.0f };
const float32x4_t v1 = vmulq_f32(v, row1);
const float32x4_t v2 = vmulq_f32(v, row2);
const float32x4_t v3 = vmulq_f32(v, row3);
const float r = vsumq_f32(v1);
const float g = vsumq_f32(v2);
const float b = vsumq_f32(v3);
const float32x4_t res = { r, g, b, 0.0f };
return res;
}
#endif

#endif /* Gamma_hpp */
2 changes: 1 addition & 1 deletion Sources/avifc/HDRColorTransfer.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@
#import "Colorspace.h"

enum ColorGammaCorrection {
Linear, Rec2020, DisplayP3
Linear, Rec2020, DisplayP3, Rec709
};

enum TransferFunction {
Expand Down
18 changes: 13 additions & 5 deletions Sources/avifc/HDRColorTransfer.mm
View file
Original file line number Diff line number Diff line change
Expand Up @@ -133,6 +133,10 @@ void TransferROW_U16HFloats(uint16_t *data, ColorGammaCorrection gammaCorrection
data[0] = half(clamp(LinearSRGBToSRGB(r), 0.0f, 1.0f)).data_;
data[1] = half(clamp(LinearSRGBToSRGB(g), 0.0f, 1.0f)).data_;
data[2] = half(clamp(LinearSRGBToSRGB(b), 0.0f, 1.0f)).data_;
} else if (gammaCorrection == Rec709) {
data[0] = half(clamp(LinearITUR709ToITUR709(r), 0.0f, 1.0f)).data_;
data[1] = half(clamp(LinearITUR709ToITUR709(g), 0.0f, 1.0f)).data_;
data[2] = half(clamp(LinearITUR709ToITUR709(b), 0.0f, 1.0f)).data_;
} else {
data[0] = half(clamp(r, 0.0f, 1.0f)).data_;
data[1] = half(clamp(g, 0.0f, 1.0f)).data_;
Expand All @@ -142,11 +146,6 @@ void TransferROW_U16HFloats(uint16_t *data, ColorGammaCorrection gammaCorrection

#if __arm64__

__attribute__((always_inline))
inline float32x4_t dcpi3GammaCorrection(float32x4_t linear) {
return vpowq_f32(linear, 1.0f/2.6f);
}

__attribute__((always_inline))
inline void SetPixelsRGB(float16x4_t rgb, uint16_t *vector, int components) {
uint16x4_t t = vreinterpret_u16_f16(rgb);
Expand Down Expand Up @@ -219,6 +218,11 @@ inline float32x4x4_t Transfer(float32x4_t rChan, float32x4_t gChan,
r.val[1] = vclampq_n_f32(LinearSRGBToSRGB(r.val[1]), 0.0f, 1.0f);
r.val[2] = vclampq_n_f32(LinearSRGBToSRGB(r.val[2]), 0.0f, 1.0f);
r.val[3] = vclampq_n_f32(LinearSRGBToSRGB(r.val[3]), 0.0f, 1.0f);
} else if (gammaCorrection == Rec709) {
r.val[0] = vclampq_n_f32(LinearITUR709ToITUR709(r.val[0]), 0.0f, 1.0f);
r.val[1] = vclampq_n_f32(LinearITUR709ToITUR709(r.val[1]), 0.0f, 1.0f);
r.val[2] = vclampq_n_f32(LinearITUR709ToITUR709(r.val[2]), 0.0f, 1.0f);
r.val[3] = vclampq_n_f32(LinearITUR709ToITUR709(r.val[3]), 0.0f, 1.0f);
} else {
r.val[0] = vclampq_n_f32(r.val[0], 0.0f, 1.0f);
r.val[1] = vclampq_n_f32(r.val[1], 0.0f, 1.0f);
Expand Down Expand Up @@ -280,6 +284,10 @@ void TransferROW_U8(uint8_t *data, float maxColors,
r = LinearSRGBToSRGB(r);
g = LinearSRGBToSRGB(g);
b = LinearSRGBToSRGB(b);
} else if (gammaCorrection == Rec709) {
r = LinearITUR709ToITUR709(r);
g = LinearITUR709ToITUR709(g);
b = LinearITUR709ToITUR709(b);
}

data[0] = (uint8_t) clamp((float) round(r * maxColors), 0.0f, maxColors);
Expand Down
12 changes: 1 addition & 11 deletions Sources/avifc/NEMath.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -200,9 +200,6 @@ static inline float32x4_t vlog10q_f32(float32x4_t x)
poly = vmlaq_f32(poly, vcvtq_f32_s32(m), CONST_LOG10);

return poly;
// static const float32x4_t CONST_LN10 = vdupq_n_f32(2.30258509299); // ln(2)
// const float32x4_t v = vlogq_f32(x);
// return vdivq_f32(v, CONST_LN10);
}

__attribute__((always_inline))
Expand Down Expand Up @@ -230,8 +227,7 @@ static inline float16x8_t vclampq_n_f16(const float16x8_t t, const float16_t min
const float vMax = max;
const float32x4_t low = vclampq_n_f32(vcvt_f32_f16(vget_low_f16(t)), vMin, vMax);
const float32x4_t high = vclampq_n_f32(vcvt_f32_f16(vget_high_f16(t)), vMin, vMax);
const float16x8_t r = vcombine_f16(vcvt_f16_f32(low), vcvt_f16_f32(high));
return r;
return vcombine_f16(vcvt_f16_f32(low), vcvt_f16_f32(high));
#endif
}

Expand All @@ -255,12 +251,6 @@ static inline float32x4x4_t MatTransponseQF32(const float32x4x4_t matrix)
return r;
}

__attribute__((always_inline))
static inline float32x4_t vreinhardq_f32(const float32x4_t t) {
float32x4_t v = vaddq_f32(t, vdupq_n_f32(1.0f));
return vdivq_f32(t, v);
}

__attribute__((always_inline))
static inline uint32x4_t vhtonlq_u32(const uint32x4_t hostlong) {
uint8x8_t low = vreinterpret_u8_u32(vget_low_u32(hostlong));
Expand Down

0 comments on commit c29f126

Please sign in to comment.