diff options
Diffstat (limited to 'files/source/scale_common.cc')
-rw-r--r-- | files/source/scale_common.cc | 191 |
1 files changed, 178 insertions, 13 deletions
diff --git a/files/source/scale_common.cc b/files/source/scale_common.cc index b02bdafd..77455903 100644 --- a/files/source/scale_common.cc +++ b/files/source/scale_common.cc @@ -23,6 +23,25 @@ namespace libyuv { extern "C" { #endif +#ifdef __cplusplus +#define STATIC_CAST(type, expr) static_cast<type>(expr) +#else +#define STATIC_CAST(type, expr) (type)(expr) +#endif + +// TODO(fbarchard): make clamp255 preserve negative values. +static __inline int32_t clamp255(int32_t v) { + return (-(v >= 255) | v) & 255; +} + +// Use scale to convert lsb formats to msb, depending how many bits there are: +// 32768 = 9 bits +// 16384 = 10 bits +// 4096 = 12 bits +// 256 = 16 bits +// TODO(fbarchard): change scale to bits +#define C16TO8(v, scale) clamp255(((v) * (scale)) >> 16) + static __inline int Abs(int v) { return v >= 0 ? v : -v; } @@ -62,6 +81,50 @@ void ScaleRowDown2_16_C(const uint16_t* src_ptr, } } +void ScaleRowDown2_16To8_C(const uint16_t* src_ptr, + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width, + int scale) { + int x; + (void)src_stride; + assert(scale >= 256); + assert(scale <= 32768); + for (x = 0; x < dst_width - 1; x += 2) { + dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[1], scale)); + dst[1] = STATIC_CAST(uint8_t, C16TO8(src_ptr[3], scale)); + dst += 2; + src_ptr += 4; + } + if (dst_width & 1) { + dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[1], scale)); + } +} + +void ScaleRowDown2_16To8_Odd_C(const uint16_t* src_ptr, + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width, + int scale) { + int x; + (void)src_stride; + assert(scale >= 256); + assert(scale <= 32768); + dst_width -= 1; + for (x = 0; x < dst_width - 1; x += 2) { + dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[1], scale)); + dst[1] = STATIC_CAST(uint8_t, C16TO8(src_ptr[3], scale)); + dst += 2; + src_ptr += 4; + } + if (dst_width & 1) { + dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[1], scale)); + dst += 1; + src_ptr += 2; + } + dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[0], scale)); +} + void ScaleRowDown2Linear_C(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst, @@ -98,6 +161,52 @@ void ScaleRowDown2Linear_16_C(const uint16_t* src_ptr, } } +void ScaleRowDown2Linear_16To8_C(const uint16_t* src_ptr, + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width, + int scale) { + const uint16_t* s = src_ptr; + int x; + (void)src_stride; + assert(scale >= 256); + assert(scale <= 32768); + for (x = 0; x < dst_width - 1; x += 2) { + dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + s[1] + 1) >> 1, scale)); + dst[1] = STATIC_CAST(uint8_t, C16TO8((s[2] + s[3] + 1) >> 1, scale)); + dst += 2; + s += 4; + } + if (dst_width & 1) { + dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + s[1] + 1) >> 1, scale)); + } +} + +void ScaleRowDown2Linear_16To8_Odd_C(const uint16_t* src_ptr, + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width, + int scale) { + const uint16_t* s = src_ptr; + int x; + (void)src_stride; + assert(scale >= 256); + assert(scale <= 32768); + dst_width -= 1; + for (x = 0; x < dst_width - 1; x += 2) { + dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + s[1] + 1) >> 1, scale)); + dst[1] = STATIC_CAST(uint8_t, C16TO8((s[2] + s[3] + 1) >> 1, scale)); + dst += 2; + s += 4; + } + if (dst_width & 1) { + dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + s[1] + 1) >> 1, scale)); + dst += 1; + s += 2; + } + dst[0] = STATIC_CAST(uint8_t, C16TO8(s[0], scale)); +} + void ScaleRowDown2Box_C(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst, @@ -160,6 +269,61 @@ void ScaleRowDown2Box_16_C(const uint16_t* src_ptr, } } +void ScaleRowDown2Box_16To8_C(const uint16_t* src_ptr, + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width, + int scale) { + const uint16_t* s = src_ptr; + const uint16_t* t = src_ptr + src_stride; + int x; + assert(scale >= 256); + assert(scale <= 32768); + for (x = 0; x < dst_width - 1; x += 2) { + dst[0] = STATIC_CAST(uint8_t, + C16TO8((s[0] + s[1] + t[0] + t[1] + 2) >> 2, scale)); + dst[1] = STATIC_CAST(uint8_t, + C16TO8((s[2] + s[3] + t[2] + t[3] + 2) >> 2, scale)); + dst += 2; + s += 4; + t += 4; + } + if (dst_width & 1) { + dst[0] = STATIC_CAST(uint8_t, + C16TO8((s[0] + s[1] + t[0] + t[1] + 2) >> 2, scale)); + } +} + +void ScaleRowDown2Box_16To8_Odd_C(const uint16_t* src_ptr, + ptrdiff_t src_stride, + uint8_t* dst, + int dst_width, + int scale) { + const uint16_t* s = src_ptr; + const uint16_t* t = src_ptr + src_stride; + int x; + assert(scale >= 256); + assert(scale <= 32768); + dst_width -= 1; + for (x = 0; x < dst_width - 1; x += 2) { + dst[0] = STATIC_CAST(uint8_t, + C16TO8((s[0] + s[1] + t[0] + t[1] + 2) >> 2, scale)); + dst[1] = STATIC_CAST(uint8_t, + C16TO8((s[2] + s[3] + t[2] + t[3] + 2) >> 2, scale)); + dst += 2; + s += 4; + t += 4; + } + if (dst_width & 1) { + dst[0] = STATIC_CAST(uint8_t, + C16TO8((s[0] + s[1] + t[0] + t[1] + 2) >> 2, scale)); + dst += 1; + s += 2; + t += 2; + } + dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + t[0] + 1) >> 1, scale)); +} + void ScaleRowDown4_C(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst, @@ -1116,18 +1280,13 @@ void ScaleUVRowDown2_C(const uint8_t* src_uv, ptrdiff_t src_stride, uint8_t* dst_uv, int dst_width) { - const uint16_t* src = (const uint16_t*)(src_uv); - uint16_t* dst = (uint16_t*)(dst_uv); int x; (void)src_stride; - for (x = 0; x < dst_width - 1; x += 2) { - dst[0] = src[1]; - dst[1] = src[3]; - src += 2; - dst += 2; - } - if (dst_width & 1) { - dst[0] = src[1]; + for (x = 0; x < dst_width; ++x) { + dst_uv[0] = src_uv[2]; // Store the 2nd UV + dst_uv[1] = src_uv[3]; + src_uv += 4; + dst_uv += 2; } } @@ -1469,7 +1628,7 @@ void ScalePlaneVertical(int src_height, enum FilterMode filtering) { // TODO(fbarchard): Allow higher bpp. int dst_width_bytes = dst_width * bpp; - void (*InterpolateRow)(uint8_t * dst_argb, const uint8_t* src_argb, + void (*InterpolateRow)(uint8_t* dst_argb, const uint8_t* src_argb, ptrdiff_t src_stride, int dst_width, int source_y_fraction) = InterpolateRow_C; const int max_y = (src_height > 1) ? ((src_height - 1) << 16) - 1 : 0; @@ -1519,6 +1678,12 @@ void ScalePlaneVertical(int src_height, } } #endif +#if defined(HAS_INTERPOLATEROW_RVV) + if (TestCpuFlag(kCpuHasRVV)) { + InterpolateRow = InterpolateRow_RVV; + } +#endif + for (j = 0; j < dst_height; ++j) { int yi; int yf; @@ -1548,7 +1713,7 @@ void ScalePlaneVertical_16(int src_height, enum FilterMode filtering) { // TODO(fbarchard): Allow higher wpp. int dst_width_words = dst_width * wpp; - void (*InterpolateRow)(uint16_t * dst_argb, const uint16_t* src_argb, + void (*InterpolateRow)(uint16_t* dst_argb, const uint16_t* src_argb, ptrdiff_t src_stride, int dst_width, int source_y_fraction) = InterpolateRow_16_C; const int max_y = (src_height > 1) ? ((src_height - 1) << 16) - 1 : 0; @@ -1627,7 +1792,7 @@ void ScalePlaneVertical_16To8(int src_height, // TODO(fbarchard): Allow higher wpp. int dst_width_words = dst_width * wpp; // TODO(https://crbug.com/libyuv/931): Add NEON 32 bit and AVX2 versions. - void (*InterpolateRow_16To8)(uint8_t * dst_argb, const uint16_t* src_argb, + void (*InterpolateRow_16To8)(uint8_t* dst_argb, const uint16_t* src_argb, ptrdiff_t src_stride, int scale, int dst_width, int source_y_fraction) = InterpolateRow_16To8_C; const int max_y = (src_height > 1) ? ((src_height - 1) << 16) - 1 : 0; |