diff options
Diffstat (limited to 'src/gui/painting/qdrawhelper_neon.cpp')
| -rw-r--r-- | src/gui/painting/qdrawhelper_neon.cpp | 409 |
1 files changed, 367 insertions, 42 deletions
diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp index 77c5202..ee5f24a 100644 --- a/src/gui/painting/qdrawhelper_neon.cpp +++ b/src/gui/painting/qdrawhelper_neon.cpp @@ -40,10 +40,13 @@ ****************************************************************************/ #include <private/qdrawhelper_p.h> +#include <private/qblendfunctions_p.h> +#include <private/qmath_p.h> #ifdef QT_HAVE_NEON #include <private/qdrawhelper_neon_p.h> +#include <private/qpaintengine_raster_p.h> #include <arm_neon.h> QT_BEGIN_NAMESPACE @@ -87,60 +90,142 @@ static inline uint16x8_t qvsource_over_u16(uint16x8_t src16, uint16x8_t dst16, u return vaddq_u16(src16, qvbyte_mul_u16(dst16, alpha16, half)); } -void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl, - const uchar *srcPixels, int sbpl, - int w, int h, - int const_alpha) +extern "C" void +pixman_composite_over_8888_0565_asm_neon (int32_t w, + int32_t h, + uint16_t *dst, + int32_t dst_stride, + uint32_t *src, + int32_t src_stride); + +extern "C" void +pixman_composite_over_8888_8888_asm_neon (int32_t w, + int32_t h, + uint32_t *dst, + int32_t dst_stride, + uint32_t *src, + int32_t src_stride); + +extern "C" void +pixman_composite_src_0565_8888_asm_neon (int32_t w, + int32_t h, + uint32_t *dst, + int32_t dst_stride, + uint16_t *src, + int32_t src_stride); + +extern "C" void +pixman_composite_over_n_8_0565_asm_neon (int32_t w, + int32_t h, + uint16_t *dst, + int32_t dst_stride, + uint32_t src, + int32_t unused, + uint8_t *mask, + int32_t mask_stride); + +extern "C" void +pixman_composite_scanline_over_asm_neon (int32_t w, + const uint32_t *dst, + const uint32_t *src); + +// qblendfunctions.cpp +void qt_blend_argb32_on_rgb16_const_alpha(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + int w, int h, + int const_alpha); + +void qt_blend_rgb16_on_argb32_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + int w, int h, + int const_alpha) { - const uint *src = (const uint *) srcPixels; - uint *dst = (uint *) destPixels; - uint16x8_t half = vdupq_n_u16(0x80); - uint16x8_t full = vdupq_n_u16(0xff); - if (const_alpha == 256) { - for (int y = 0; y < h; ++y) { - int x = 0; - for (; x < w-3; x += 4) { - uint32x4_t src32 = vld1q_u32((uint32_t *)&src[x]); - if ((src[x] & src[x+1] & src[x+2] & src[x+3]) >= 0xff000000) { - // all opaque - vst1q_u32((uint32_t *)&dst[x], src32); - } else if (src[x] | src[x+1] | src[x+2] | src[x+3]) { - uint32x4_t dst32 = vld1q_u32((uint32_t *)&dst[x]); + dbpl /= 4; + sbpl /= 2; - const uint8x16_t src8 = vreinterpretq_u8_u32(src32); - const uint8x16_t dst8 = vreinterpretq_u8_u32(dst32); + quint32 *dst = (quint32 *) destPixels; + quint16 *src = (quint16 *) srcPixels; - const uint8x8_t src8_low = vget_low_u8(src8); - const uint8x8_t dst8_low = vget_low_u8(dst8); + if (const_alpha != 256) { + quint8 a = (255 * const_alpha) >> 8; + quint8 ia = 255 - a; + + while (h--) { + for (int x=0; x<w; ++x) + dst[x] = INTERPOLATE_PIXEL_255(qt_colorConvert(src[x], dst[x]), a, dst[x], ia); + dst += dbpl; + src += sbpl; + } + return; + } - const uint8x8_t src8_high = vget_high_u8(src8); - const uint8x8_t dst8_high = vget_high_u8(dst8); + pixman_composite_src_0565_8888_asm_neon(w, h, dst, dbpl, src, sbpl); +} - const uint16x8_t src16_low = vmovl_u8(src8_low); - const uint16x8_t dst16_low = vmovl_u8(dst8_low); +extern "C" void blend_8_pixels_argb32_on_rgb16_neon(quint16 *dst, const quint32 *src, int const_alpha); - const uint16x8_t src16_high = vmovl_u8(src8_high); - const uint16x8_t dst16_high = vmovl_u8(dst8_high); +void qt_blend_argb32_on_rgb16_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + int w, int h, + int const_alpha) +{ + quint16 *dst = (quint16 *) destPixels; + quint32 *src = (quint32 *) srcPixels; - const uint16x8_t result16_low = qvsource_over_u16(src16_low, dst16_low, half, full); - const uint16x8_t result16_high = qvsource_over_u16(src16_high, dst16_high, half, full); + if (const_alpha != 256) { + for (int y=0; y<h; ++y) { + int i = 0; + for (; i < w-7; i += 8) + blend_8_pixels_argb32_on_rgb16_neon(&dst[i], &src[i], const_alpha); - const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low)); - const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high)); + if (i < w) { + int tail = w - i; - vst1q_u32((uint32_t *)&dst[x], vcombine_u32(result32_low, result32_high)); + quint16 dstBuffer[8]; + quint32 srcBuffer[8]; + + for (int j = 0; j < tail; ++j) { + dstBuffer[j] = dst[i + j]; + srcBuffer[j] = src[i + j]; + } + + blend_8_pixels_argb32_on_rgb16_neon(dstBuffer, srcBuffer, const_alpha); + + for (int j = 0; j < tail; ++j) { + dst[i + j] = dstBuffer[j]; + src[i + j] = srcBuffer[j]; } } - for (; x<w; ++x) { - uint s = src[x]; - if (s >= 0xff000000) - dst[x] = s; - else if (s != 0) - dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); - } - dst = (quint32 *)(((uchar *) dst) + dbpl); - src = (const quint32 *)(((const uchar *) src) + sbpl); + + dst = (quint16 *)(((uchar *) dst) + dbpl); + src = (quint32 *)(((uchar *) src) + sbpl); } + return; + } + + pixman_composite_over_8888_0565_asm_neon(w, h, dst, dbpl / 2, src, sbpl / 4); +} + +void qt_blend_argb32_on_argb32_scanline_neon(uint *dest, const uint *src, int length, uint const_alpha) +{ + if (const_alpha == 255) { + pixman_composite_scanline_over_asm_neon(length, dest, src); + } else { + qt_blend_argb32_on_argb32_neon((uchar *)dest, 4 * length, (uchar *)src, 4 * length, length, 1, (const_alpha * 256) / 255); + } +} + +void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + int w, int h, + int const_alpha) +{ + const uint *src = (const uint *) srcPixels; + uint *dst = (uint *) destPixels; + uint16x8_t half = vdupq_n_u16(0x80); + uint16x8_t full = vdupq_n_u16(0xff); + if (const_alpha == 256) { + pixman_composite_over_8888_8888_asm_neon(w, h, (uint32_t *)destPixels, dbpl / 4, (uint32_t *)srcPixels, sbpl / 4); } else if (const_alpha != 0) { const_alpha = (const_alpha * 255) >> 8; uint16x8_t const_alpha16 = vdupq_n_u16(const_alpha); @@ -254,6 +339,246 @@ void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl, } } +void qt_alphamapblit_quint16_neon(QRasterBuffer *rasterBuffer, + int x, int y, quint32 color, + const uchar *bitmap, + int mapWidth, int mapHeight, int mapStride, + const QClipData *) +{ + quint16 *dest = reinterpret_cast<quint16*>(rasterBuffer->scanLine(y)) + x; + const int destStride = rasterBuffer->bytesPerLine() / sizeof(quint16); + + uchar *mask = const_cast<uchar *>(bitmap); + + pixman_composite_over_n_8_0565_asm_neon(mapWidth, mapHeight, dest, destStride, color, 0, mask, mapStride); +} + +extern "C" void blend_8_pixels_rgb16_on_rgb16_neon(quint16 *dst, const quint16 *src, int const_alpha); + +template <typename SRC, typename BlendFunc> +struct Blend_on_RGB16_SourceAndConstAlpha_Neon { + Blend_on_RGB16_SourceAndConstAlpha_Neon(BlendFunc blender, int const_alpha) + : m_index(0) + , m_blender(blender) + , m_const_alpha(const_alpha) + { + } + + inline void write(quint16 *dst, quint32 src) + { + srcBuffer[m_index++] = src; + + if (m_index == 8) { + m_blender(dst - 7, srcBuffer, m_const_alpha); + m_index = 0; + } + } + + inline void flush(quint16 *dst) + { + if (m_index > 0) { + quint16 dstBuffer[8]; + for (int i = 0; i < m_index; ++i) + dstBuffer[i] = dst[i - m_index]; + + m_blender(dstBuffer, srcBuffer, m_const_alpha); + + for (int i = 0; i < m_index; ++i) + dst[i - m_index] = dstBuffer[i]; + + m_index = 0; + } + } + + SRC srcBuffer[8]; + + int m_index; + BlendFunc m_blender; + int m_const_alpha; +}; + +template <typename SRC, typename BlendFunc> +Blend_on_RGB16_SourceAndConstAlpha_Neon<SRC, BlendFunc> +Blend_on_RGB16_SourceAndConstAlpha_Neon_create(BlendFunc blender, int const_alpha) +{ + return Blend_on_RGB16_SourceAndConstAlpha_Neon<SRC, BlendFunc>(blender, const_alpha); +} + +void qt_scale_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + const QRectF &targetRect, + const QRectF &sourceRect, + const QRect &clip, + int const_alpha) +{ + if (const_alpha == 0) + return; + + qt_scale_image_16bit<quint32>(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, + Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint32>(blend_8_pixels_argb32_on_rgb16_neon, const_alpha)); +} + +void qt_scale_image_rgb16_on_rgb16(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + const QRectF &targetRect, + const QRectF &sourceRect, + const QRect &clip, + int const_alpha); + +void qt_scale_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + const QRectF &targetRect, + const QRectF &sourceRect, + const QRect &clip, + int const_alpha) +{ + if (const_alpha == 0) + return; + + if (const_alpha == 256) { + qt_scale_image_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, const_alpha); + return; + } + + qt_scale_image_16bit<quint16>(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, + Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint16>(blend_8_pixels_rgb16_on_rgb16_neon, const_alpha)); +} + +extern void qt_transform_image_rgb16_on_rgb16(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + const QRectF &targetRect, + const QRectF &sourceRect, + const QRect &clip, + const QTransform &targetRectTransform, + int const_alpha); + +void qt_transform_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + const QRectF &targetRect, + const QRectF &sourceRect, + const QRect &clip, + const QTransform &targetRectTransform, + int const_alpha) +{ + if (const_alpha == 0) + return; + + if (const_alpha == 256) { + qt_transform_image_rgb16_on_rgb16(destPixels, dbpl, srcPixels, sbpl, targetRect, sourceRect, clip, targetRectTransform, const_alpha); + return; + } + + qt_transform_image(reinterpret_cast<quint16 *>(destPixels), dbpl, + reinterpret_cast<const quint16 *>(srcPixels), sbpl, targetRect, sourceRect, clip, targetRectTransform, + Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint16>(blend_8_pixels_rgb16_on_rgb16_neon, const_alpha)); +} + +void qt_transform_image_argb32_on_rgb16_neon(uchar *destPixels, int dbpl, + const uchar *srcPixels, int sbpl, + const QRectF &targetRect, + const QRectF &sourceRect, + const QRect &clip, + const QTransform &targetRectTransform, + int const_alpha) +{ + if (const_alpha == 0) + return; + + qt_transform_image(reinterpret_cast<quint16 *>(destPixels), dbpl, + reinterpret_cast<const quint32 *>(srcPixels), sbpl, targetRect, sourceRect, clip, targetRectTransform, + Blend_on_RGB16_SourceAndConstAlpha_Neon_create<quint32>(blend_8_pixels_argb32_on_rgb16_neon, const_alpha)); +} + +static inline void convert_8_pixels_rgb16_to_argb32(quint32 *dst, const quint16 *src) +{ + asm volatile ( + "vld1.16 { d0, d1 }, [%[SRC]]\n\t" + + /* convert 8 r5g6b5 pixel data from {d0, d1} to planar 8-bit format + and put data into d4 - red, d3 - green, d2 - blue */ + "vshrn.u16 d4, q0, #8\n\t" + "vshrn.u16 d3, q0, #3\n\t" + "vsli.u16 q0, q0, #5\n\t" + "vsri.u8 d4, d4, #5\n\t" + "vsri.u8 d3, d3, #6\n\t" + "vshrn.u16 d2, q0, #2\n\t" + + /* fill d5 - alpha with 0xff */ + "mov r2, #255\n\t" + "vdup.8 d5, r2\n\t" + + "vst4.8 { d2, d3, d4, d5 }, [%[DST]]" + : : [DST]"r" (dst), [SRC]"r" (src) + : "memory", "r2", "d0", "d1", "d2", "d3", "d4", "d5" + ); +} + +uint * QT_FASTCALL qt_destFetchRGB16_neon(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length) +{ + const ushort *data = (const ushort *)rasterBuffer->scanLine(y) + x; + + int i = 0; + for (; i < length - 7; i += 8) + convert_8_pixels_rgb16_to_argb32(&buffer[i], &data[i]); + + if (i < length) { + quint16 srcBuffer[8]; + quint32 dstBuffer[8]; + + int tail = length - i; + for (int j = 0; j < tail; ++j) + srcBuffer[j] = data[i + j]; + + convert_8_pixels_rgb16_to_argb32(dstBuffer, srcBuffer); + + for (int j = 0; j < tail; ++j) + buffer[i + j] = dstBuffer[j]; + } + + return buffer; +} + +static inline void convert_8_pixels_argb32_to_rgb16(quint16 *dst, const quint32 *src) +{ + asm volatile ( + "vld4.8 { d0, d1, d2, d3 }, [%[SRC]]\n\t" + + /* convert to r5g6b5 and store it into {d28, d29} */ + "vshll.u8 q14, d2, #8\n\t" + "vshll.u8 q8, d1, #8\n\t" + "vshll.u8 q9, d0, #8\n\t" + "vsri.u16 q14, q8, #5\n\t" + "vsri.u16 q14, q9, #11\n\t" + + "vst1.16 { d28, d29 }, [%[DST]]" + : : [DST]"r" (dst), [SRC]"r" (src) + : "memory", "d0", "d1", "d2", "d3", "d16", "d17", "d18", "d19", "d28", "d29" + ); +} + +void QT_FASTCALL qt_destStoreRGB16_neon(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length) +{ + quint16 *data = (quint16*)rasterBuffer->scanLine(y) + x; + + int i = 0; + for (; i < length - 7; i += 8) + convert_8_pixels_argb32_to_rgb16(&data[i], &buffer[i]); + + if (i < length) { + quint32 srcBuffer[8]; + quint16 dstBuffer[8]; + + int tail = length - i; + for (int j = 0; j < tail; ++j) + srcBuffer[j] = buffer[i + j]; + + convert_8_pixels_argb32_to_rgb16(dstBuffer, srcBuffer); + + for (int j = 0; j < tail; ++j) + data[i + j] = dstBuffer[j]; + } +} + QT_END_NAMESPACE #endif // QT_HAVE_NEON |