summaryrefslogtreecommitdiffstats
path: root/chromium/third_party/libwebp/dsp/yuv.h
diff options
context:
space:
mode:
Diffstat (limited to 'chromium/third_party/libwebp/dsp/yuv.h')
-rw-r--r--chromium/third_party/libwebp/dsp/yuv.h261
1 files changed, 147 insertions, 114 deletions
diff --git a/chromium/third_party/libwebp/dsp/yuv.h b/chromium/third_party/libwebp/dsp/yuv.h
index 3844d8cab32..dd778f9cbe8 100644
--- a/chromium/third_party/libwebp/dsp/yuv.h
+++ b/chromium/third_party/libwebp/dsp/yuv.h
@@ -14,7 +14,7 @@
// Y = 0.2569 * R + 0.5044 * G + 0.0979 * B + 16
// U = -0.1483 * R - 0.2911 * G + 0.4394 * B + 128
// V = 0.4394 * R - 0.3679 * G - 0.0715 * B + 128
-// We use 16bit fixed point operations for RGB->YUV conversion.
+// We use 16bit fixed point operations for RGB->YUV conversion (YUV_FIX).
//
// For the Y'CbCr to RGB conversion, the BT.601 specification reads:
// R = 1.164 * (Y-16) + 1.596 * (V-128)
@@ -23,21 +23,24 @@
// where Y is in the [16,235] range, and U/V in the [16,240] range.
// In the table-lookup version (WEBP_YUV_USE_TABLE), the common factor
// "1.164 * (Y-16)" can be handled as an offset in the VP8kClip[] table.
-// So in this case the formulae should be read as:
+// So in this case the formulae should read:
// R = 1.164 * [Y + 1.371 * (V-128) ] - 18.624
// G = 1.164 * [Y - 0.698 * (V-128) - 0.336 * (U-128)] - 18.624
// B = 1.164 * [Y + 1.733 * (U-128)] - 18.624
-// once factorized. Here too, 16bit fixed precision is used.
+// once factorized.
+// For YUV->RGB conversion, only 14bit fixed precision is used (YUV_FIX2).
+// That's the maximum possible for a convenient ARM implementation.
//
// Author: Skal (pascal.massimino@gmail.com)
#ifndef WEBP_DSP_YUV_H_
#define WEBP_DSP_YUV_H_
+#include "./dsp.h"
#include "../dec/decode_vp8.h"
// Define the following to use the LUT-based code:
-#define WEBP_YUV_USE_TABLE
+// #define WEBP_YUV_USE_TABLE
#if defined(WEBP_EXPERIMENTAL_FEATURES)
// Do NOT activate this feature for real compression. This is only experimental!
@@ -52,53 +55,75 @@
//------------------------------------------------------------------------------
// YUV -> RGB conversion
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
extern "C" {
#endif
-enum { YUV_FIX = 16, // fixed-point precision
- YUV_HALF = 1 << (YUV_FIX - 1),
- YUV_MASK = (256 << YUV_FIX) - 1,
- YUV_RANGE_MIN = -227, // min value of r/g/b output
- YUV_RANGE_MAX = 256 + 226 // max value of r/g/b output
+enum {
+ YUV_FIX = 16, // fixed-point precision for RGB->YUV
+ YUV_HALF = 1 << (YUV_FIX - 1),
+ YUV_MASK = (256 << YUV_FIX) - 1,
+ YUV_RANGE_MIN = -227, // min value of r/g/b output
+ YUV_RANGE_MAX = 256 + 226, // max value of r/g/b output
+
+ YUV_FIX2 = 14, // fixed-point precision for YUV->RGB
+ YUV_HALF2 = 1 << (YUV_FIX2 - 1),
+ YUV_MASK2 = (256 << YUV_FIX2) - 1
};
-#ifdef WEBP_YUV_USE_TABLE
+// These constants are 14b fixed-point version of ITU-R BT.601 constants.
+#define kYScale 19077 // 1.164 = 255 / 219
+#define kVToR 26149 // 1.596 = 255 / 112 * 0.701
+#define kUToG 6419 // 0.391 = 255 / 112 * 0.886 * 0.114 / 0.587
+#define kVToG 13320 // 0.813 = 255 / 112 * 0.701 * 0.299 / 0.587
+#define kUToB 33050 // 2.018 = 255 / 112 * 0.886
+#define kRCst (-kYScale * 16 - kVToR * 128 + YUV_HALF2)
+#define kGCst (-kYScale * 16 + kUToG * 128 + kVToG * 128 + YUV_HALF2)
+#define kBCst (-kYScale * 16 - kUToB * 128 + YUV_HALF2)
-extern int16_t VP8kVToR[256], VP8kUToB[256];
-extern int32_t VP8kVToG[256], VP8kUToG[256];
-extern uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
-extern uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
+//------------------------------------------------------------------------------
+
+#if !defined(WEBP_YUV_USE_TABLE)
+
+// slower on x86 by ~7-8%, but bit-exact with the SSE2 version
+
+static WEBP_INLINE int VP8Clip8(int v) {
+ return ((v & ~YUV_MASK2) == 0) ? (v >> YUV_FIX2) : (v < 0) ? 0 : 255;
+}
+
+static WEBP_INLINE int VP8YUVToR(int y, int v) {
+ return VP8Clip8(kYScale * y + kVToR * v + kRCst);
+}
+
+static WEBP_INLINE int VP8YUVToG(int y, int u, int v) {
+ return VP8Clip8(kYScale * y - kUToG * u - kVToG * v + kGCst);
+}
+
+static WEBP_INLINE int VP8YUVToB(int y, int u) {
+ return VP8Clip8(kYScale * y + kUToB * u + kBCst);
+}
-static WEBP_INLINE void VP8YuvToRgb(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
uint8_t* const rgb) {
- const int r_off = VP8kVToR[v];
- const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
- const int b_off = VP8kUToB[u];
- rgb[0] = VP8kClip[y + r_off - YUV_RANGE_MIN];
- rgb[1] = VP8kClip[y + g_off - YUV_RANGE_MIN];
- rgb[2] = VP8kClip[y + b_off - YUV_RANGE_MIN];
+ rgb[0] = VP8YUVToR(y, v);
+ rgb[1] = VP8YUVToG(y, u, v);
+ rgb[2] = VP8YUVToB(y, u);
}
-static WEBP_INLINE void VP8YuvToBgr(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
uint8_t* const bgr) {
- const int r_off = VP8kVToR[v];
- const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
- const int b_off = VP8kUToB[u];
- bgr[0] = VP8kClip[y + b_off - YUV_RANGE_MIN];
- bgr[1] = VP8kClip[y + g_off - YUV_RANGE_MIN];
- bgr[2] = VP8kClip[y + r_off - YUV_RANGE_MIN];
+ bgr[0] = VP8YUVToB(y, u);
+ bgr[1] = VP8YUVToG(y, u, v);
+ bgr[2] = VP8YUVToR(y, v);
}
-static WEBP_INLINE void VP8YuvToRgb565(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
uint8_t* const rgb) {
- const int r_off = VP8kVToR[v];
- const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
- const int b_off = VP8kUToB[u];
- const uint8_t rg = ((VP8kClip[y + r_off - YUV_RANGE_MIN] & 0xf8) |
- (VP8kClip[y + g_off - YUV_RANGE_MIN] >> 5));
- const uint8_t gb = (((VP8kClip[y + g_off - YUV_RANGE_MIN] << 3) & 0xe0) |
- (VP8kClip[y + b_off - YUV_RANGE_MIN] >> 3));
+ const int r = VP8YUVToR(y, v); // 5 usable bits
+ const int g = VP8YUVToG(y, u, v); // 6 usable bits
+ const int b = VP8YUVToB(y, u); // 5 usable bits
+ const int rg = (r & 0xf8) | (g >> 5);
+ const int gb = ((g << 3) & 0xe0) | (b >> 3);
#ifdef WEBP_SWAP_16BIT_CSP
rgb[0] = gb;
rgb[1] = rg;
@@ -108,14 +133,13 @@ static WEBP_INLINE void VP8YuvToRgb565(uint8_t y, uint8_t u, uint8_t v,
#endif
}
-static WEBP_INLINE void VP8YuvToRgba4444(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
uint8_t* const argb) {
- const int r_off = VP8kVToR[v];
- const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
- const int b_off = VP8kUToB[u];
- const uint8_t rg = ((VP8kClip4Bits[y + r_off - YUV_RANGE_MIN] << 4) |
- VP8kClip4Bits[y + g_off - YUV_RANGE_MIN]);
- const uint8_t ba = (VP8kClip4Bits[y + b_off - YUV_RANGE_MIN] << 4) | 0x0f;
+ const int r = VP8YUVToR(y, v); // 4 usable bits
+ const int g = VP8YUVToG(y, u, v); // 4 usable bits
+ const int b = VP8YUVToB(y, u); // 4 usable bits
+ const int rg = (r & 0xf0) | (g >> 4);
+ const int ba = (b & 0xf0) | 0x0f; // overwrite the lower 4 bits
#ifdef WEBP_SWAP_16BIT_CSP
argb[0] = ba;
argb[1] = rg;
@@ -125,61 +149,45 @@ static WEBP_INLINE void VP8YuvToRgba4444(uint8_t y, uint8_t u, uint8_t v,
#endif
}
-#else // Table-free version (slower on x86)
-
-// These constants are 16b fixed-point version of ITU-R BT.601 constants
-#define kYScale 76309 // 1.164 = 255 / 219
-#define kVToR 104597 // 1.596 = 255 / 112 * 0.701
-#define kUToG 25674 // 0.391 = 255 / 112 * 0.886 * 0.114 / 0.587
-#define kVToG 53278 // 0.813 = 255 / 112 * 0.701 * 0.299 / 0.587
-#define kUToB 132201 // 2.018 = 255 / 112 * 0.886
-#define kRCst (-kYScale * 16 - kVToR * 128 + YUV_HALF)
-#define kGCst (-kYScale * 16 + kUToG * 128 + kVToG * 128 + YUV_HALF)
-#define kBCst (-kYScale * 16 - kUToB * 128 + YUV_HALF)
-
-static WEBP_INLINE uint8_t VP8Clip8(int v) {
- return ((v & ~YUV_MASK) == 0) ? (uint8_t)(v >> YUV_FIX)
- : (v < 0) ? 0u : 255u;
-}
-
-static WEBP_INLINE uint8_t VP8ClipN(int v, int N) { // clip to N bits
- return ((v & ~YUV_MASK) == 0) ? (uint8_t)(v >> (YUV_FIX + (8 - N)))
- : (v < 0) ? 0u : (255u >> (8 - N));
-}
-
-static WEBP_INLINE int VP8YUVToR(int y, int v) {
- return kYScale * y + kVToR * v + kRCst;
-}
+#else
-static WEBP_INLINE int VP8YUVToG(int y, int u, int v) {
- return kYScale * y - kUToG * u - kVToG * v + kGCst;
-}
+// Table-based version, not totally equivalent to the SSE2 version.
+// Rounding diff is only +/-1 though.
-static WEBP_INLINE int VP8YUVToB(int y, int u) {
- return kYScale * y + kUToB * u + kBCst;
-}
+extern int16_t VP8kVToR[256], VP8kUToB[256];
+extern int32_t VP8kVToG[256], VP8kUToG[256];
+extern uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
+extern uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
-static WEBP_INLINE void VP8YuvToRgb(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
uint8_t* const rgb) {
- rgb[0] = VP8Clip8(VP8YUVToR(y, v));
- rgb[1] = VP8Clip8(VP8YUVToG(y, u, v));
- rgb[2] = VP8Clip8(VP8YUVToB(y, u));
+ const int r_off = VP8kVToR[v];
+ const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
+ const int b_off = VP8kUToB[u];
+ rgb[0] = VP8kClip[y + r_off - YUV_RANGE_MIN];
+ rgb[1] = VP8kClip[y + g_off - YUV_RANGE_MIN];
+ rgb[2] = VP8kClip[y + b_off - YUV_RANGE_MIN];
}
-static WEBP_INLINE void VP8YuvToBgr(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
uint8_t* const bgr) {
- bgr[0] = VP8Clip8(VP8YUVToB(y, u));
- bgr[1] = VP8Clip8(VP8YUVToG(y, u, v));
- bgr[2] = VP8Clip8(VP8YUVToR(y, v));
+ const int r_off = VP8kVToR[v];
+ const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
+ const int b_off = VP8kUToB[u];
+ bgr[0] = VP8kClip[y + b_off - YUV_RANGE_MIN];
+ bgr[1] = VP8kClip[y + g_off - YUV_RANGE_MIN];
+ bgr[2] = VP8kClip[y + r_off - YUV_RANGE_MIN];
}
-static WEBP_INLINE void VP8YuvToRgb565(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
uint8_t* const rgb) {
- const int r = VP8Clip8(VP8YUVToR(y, u));
- const int g = VP8ClipN(VP8YUVToG(y, u, v), 6);
- const int b = VP8ClipN(VP8YUVToB(y, v), 5);
- const uint8_t rg = (r & 0xf8) | (g >> 3);
- const uint8_t gb = (g << 5) | b;
+ const int r_off = VP8kVToR[v];
+ const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
+ const int b_off = VP8kUToB[u];
+ const int rg = ((VP8kClip[y + r_off - YUV_RANGE_MIN] & 0xf8) |
+ (VP8kClip[y + g_off - YUV_RANGE_MIN] >> 5));
+ const int gb = (((VP8kClip[y + g_off - YUV_RANGE_MIN] << 3) & 0xe0) |
+ (VP8kClip[y + b_off - YUV_RANGE_MIN] >> 3));
#ifdef WEBP_SWAP_16BIT_CSP
rgb[0] = gb;
rgb[1] = rg;
@@ -189,13 +197,14 @@ static WEBP_INLINE void VP8YuvToRgb565(uint8_t y, uint8_t u, uint8_t v,
#endif
}
-static WEBP_INLINE void VP8YuvToRgba4444(uint8_t y, uint8_t u, uint8_t v,
+static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
uint8_t* const argb) {
- const int r = VP8Clip8(VP8YUVToR(y, u));
- const int g = VP8ClipN(VP8YUVToG(y, u, v), 4);
- const int b = VP8Clip8(VP8YUVToB(y, v));
- const uint8_t rg = (r & 0xf0) | g;
- const uint8_t ba = b | 0x0f; // overwrite the lower 4 bits
+ const int r_off = VP8kVToR[v];
+ const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
+ const int b_off = VP8kUToB[u];
+ const int rg = ((VP8kClip4Bits[y + r_off - YUV_RANGE_MIN] << 4) |
+ VP8kClip4Bits[y + g_off - YUV_RANGE_MIN]);
+ const int ba = (VP8kClip4Bits[y + b_off - YUV_RANGE_MIN] << 4) | 0x0f;
#ifdef WEBP_SWAP_16BIT_CSP
argb[0] = ba;
argb[1] = rg;
@@ -207,6 +216,9 @@ static WEBP_INLINE void VP8YuvToRgba4444(uint8_t y, uint8_t u, uint8_t v,
#endif // WEBP_YUV_USE_TABLE
+//-----------------------------------------------------------------------------
+// Alpha handling variants
+
static WEBP_INLINE void VP8YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
uint8_t* const argb) {
argb[0] = 0xff;
@@ -228,56 +240,77 @@ static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v,
// Must be called before everything, to initialize the tables.
void VP8YUVInit(void);
+//-----------------------------------------------------------------------------
+// SSE2 extra functions (mostly for upsampling_sse2.c)
+
+#if defined(WEBP_USE_SSE2)
+
+#if defined(FANCY_UPSAMPLING)
+// Process 32 pixels and store the result (24b or 32b per pixel) in *dst.
+void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+ uint8_t* dst);
+void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+ uint8_t* dst);
+void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+ uint8_t* dst);
+void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
+ uint8_t* dst);
+#endif // FANCY_UPSAMPLING
+
+// Must be called to initialize tables before using the functions.
+void VP8YUVInitSSE2(void);
+
+#endif // WEBP_USE_SSE2
+
//------------------------------------------------------------------------------
// RGB -> YUV conversion
-static WEBP_INLINE int VP8ClipUV(int v) {
- v = (v + (257 << (YUV_FIX + 2 - 1))) >> (YUV_FIX + 2);
- return ((v & ~0xff) == 0) ? v : (v < 0) ? 0 : 255;
+// Stub functions that can be called with various rounding values:
+static WEBP_INLINE int VP8ClipUV(int uv, int rounding) {
+ uv = (uv + rounding + (128 << (YUV_FIX + 2))) >> (YUV_FIX + 2);
+ return ((uv & ~0xff) == 0) ? uv : (uv < 0) ? 0 : 255;
}
#ifndef USE_YUVj
-static WEBP_INLINE int VP8RGBToY(int r, int g, int b) {
- const int kRound = (1 << (YUV_FIX - 1)) + (16 << YUV_FIX);
+static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
const int luma = 16839 * r + 33059 * g + 6420 * b;
- return (luma + kRound) >> YUV_FIX; // no need to clip
+ return (luma + rounding + (16 << YUV_FIX)) >> YUV_FIX; // no need to clip
}
-static WEBP_INLINE int VP8RGBToU(int r, int g, int b) {
+static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
const int u = -9719 * r - 19081 * g + 28800 * b;
- return VP8ClipUV(u);
+ return VP8ClipUV(u, rounding);
}
-static WEBP_INLINE int VP8RGBToV(int r, int g, int b) {
+static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
const int v = +28800 * r - 24116 * g - 4684 * b;
- return VP8ClipUV(v);
+ return VP8ClipUV(v, rounding);
}
#else
// This JPEG-YUV colorspace, only for comparison!
-// These are also 16-bit precision coefficients from Rec.601, but with full
+// These are also 16bit precision coefficients from Rec.601, but with full
// [0..255] output range.
-static WEBP_INLINE int VP8RGBToY(int r, int g, int b) {
- const int kRound = (1 << (YUV_FIX - 1));
+static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
const int luma = 19595 * r + 38470 * g + 7471 * b;
- return (luma + kRound) >> YUV_FIX; // no need to clip
+ return (luma + rounding) >> YUV_FIX; // no need to clip
}
-static WEBP_INLINE int VP8RGBToU(int r, int g, int b) {
+static WEBP_INLINE int VP8_RGB_TO_U(int r, int g, int b, int rounding) {
const int u = -11058 * r - 21710 * g + 32768 * b;
- return VP8ClipUV(u);
+ return VP8ClipUV(u, rounding);
}
-static WEBP_INLINE int VP8RGBToV(int r, int g, int b) {
+static WEBP_INLINE int VP8_RGB_TO_V(int r, int g, int b, int rounding) {
const int v = 32768 * r - 27439 * g - 5329 * b;
- return VP8ClipUV(v);
+ return VP8ClipUV(v, rounding);
}
#endif // USE_YUVj
-#if defined(__cplusplus) || defined(c_plusplus)
+#ifdef __cplusplus
} // extern "C"
#endif
generated by cgit v1.2.3 (git 2.25.1) at 2024-05-07 09:40:15 +0000