RawImageDataU16.cpp

Go to the documentation of this file.

/*
    RawSpeed - RAW file decoder.
 
    Copyright (C) 2009-2014 Klaus Post
 
    This library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
    License as published by the Free Software Foundation; either
    version 2 of the License, or (at your option) any later version.
 
    This library is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    Lesser General Public License for more details.
 
    You should have received a copy of the GNU Lesser General Public
    License along with this library; if not, write to the Free Software
    Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
 
#include "rawspeedconfig.h"
#include "common/RawImage.h"
#include "adt/Array1DRef.h"
#include "adt/Array2DRef.h"
#include "adt/Bit.h"
#include "adt/Casts.h"
#include "adt/CroppedArray2DRef.h"
#include "adt/Point.h"
#include "common/Common.h"
#include "common/TableLookUp.h"
#include "decoders/RawDecoderException.h"
#include "metadata/BlackArea.h"
#include <algorithm>
#include <array>
#include <cassert>
#include <cstdint>
#include <memory>
#include <vector>
 
#ifdef WITH_SSE2
#include "common/CpuFeatures.h"
#include <emmintrin.h>
#endif
 
using std::array;
using std::max;
using std::min;
using std::vector;
 
namespace rawspeed {
 
RawImageDataU16::RawImageDataU16() {
  dataType = RawImageType::UINT16;
  bpp = sizeof(uint16_t);
}
 
RawImageDataU16::RawImageDataU16(const iPoint2D& _dim, uint32_t _cpp)
    : RawImageData(RawImageType::UINT16, _dim, sizeof(uint16_t), _cpp) {}
 
void RawImageDataU16::calculateBlackAreas() {
  const Array2DRef<uint16_t> img = getU16DataAsUncroppedArray2DRef();
 
  std::vector<uint16_t> histogramStorage;
  auto histogram = Array2DRef<uint16_t>::create(histogramStorage, 65536, 4);
 
  for (int row = 0; row != histogram.height(); ++row) {
    for (int col = 0; col != histogram.width(); ++col) {
      histogram(row, col) = 0;
    }
  }
 
  int totalpixels = 0;
 
  for (auto area : blackAreas) {
    /* Make sure area sizes are multiple of two,
       so we have the same amount of pixels for each CFA group */
    area.size = area.size - (area.size & 1);
 
    /* Process horizontal area */
    if (!area.isVertical) {
      if (static_cast<int>(area.offset) + static_cast<int>(area.size) >
          uncropped_dim.y)
        ThrowRDE("Offset + size is larger than height of image");
      for (uint32_t y = area.offset; y < area.offset + area.size; y++) {
        for (int x = mOffset.x; x < dim.x + mOffset.x; x++) {
          // FIXME: this only samples a single row, not an area.
          const auto localhist = histogram[(2 * (y & 1)) + (x & 1)];
          const auto hBin = img(y, mOffset.x);
          localhist(hBin)++;
        }
      }
      totalpixels += area.size * dim.x;
    }
 
    /* Process vertical area */
    if (area.isVertical) {
      if (static_cast<int>(area.offset) + static_cast<int>(area.size) >
          uncropped_dim.x)
        ThrowRDE("Offset + size is larger than width of image");
      for (int y = mOffset.y; y < dim.y + mOffset.y; y++) {
        for (uint32_t x = area.offset; x < area.size + area.offset; x++) {
          // FIXME: this only samples a single row, not an area.
          const auto localhist = histogram[(2 * (y & 1)) + (x & 1)];
          const auto hBin = img(y, area.offset);
          localhist(hBin)++;
        }
      }
      totalpixels += area.size * dim.y;
    }
  }
 
  blackLevelSeparate = Array2DRef(blackLevelSeparateStorage.data(), 2, 2);
  auto blackLevelSeparate1D = *blackLevelSeparate->getAsArray1DRef();
 
  if (!totalpixels) {
    for (int& i : blackLevelSeparate1D)
      i = blackLevel;
    return;
  }
 
  /* Calculate median value of black areas for each component */
  /* Adjust the number of total pixels so it is the same as the median of each
   * histogram */
  totalpixels /= 4 * 2;
 
  for (int i = 0; i < 4; i++) {
    const auto localhist = histogram[i];
    int acc_pixels = localhist(0);
    int pixel_value = 0;
    while (acc_pixels <= totalpixels && pixel_value < 65535) {
      pixel_value++;
      acc_pixels += localhist(pixel_value);
    }
    blackLevelSeparate1D(i) = pixel_value;
  }
 
  /* If this is not a CFA image, we do not use separate blacklevels, use average
   */
  if (!isCFA) {
    int total = 0;
    for (int i : blackLevelSeparate1D)
      total += i;
    for (int& i : blackLevelSeparate1D)
      i = (total + 2) >> 2;
  }
}
 
void RawImageDataU16::scaleBlackWhite() {
  const int skipBorder = 250;
  int gw = (dim.x - skipBorder) * cpp;
  if ((blackAreas.empty() && !blackLevelSeparate && blackLevel < 0) ||
      !whitePoint) { // Estimate
    int b = 65536;
    int m = 0;
    auto img = getU16DataAsCroppedArray2DRef();
    for (int row = skipBorder; row < (dim.y - skipBorder); row++) {
      for (int col = skipBorder; col < gw; col++) {
        uint16_t pixel = img(row, skipBorder + col);
        b = min(static_cast<int>(pixel), b);
        m = max(static_cast<int>(pixel), m);
      }
    }
    if (blackLevel < 0)
      blackLevel = b;
    if (!whitePoint)
      whitePoint = m;
    writeLog(DEBUG_PRIO::INFO,
             "ISO:%d, Estimated black:%d, Estimated white: %d",
             metadata.isoSpeed, blackLevel, *whitePoint);
  }
 
  /* Skip, if not needed */
  if ((blackAreas.empty() && blackLevel == 0 && whitePoint == 65535 &&
       !blackLevelSeparate) ||
      dim.area() <= 0)
    return;
 
  /* If filter has not set separate blacklevel, compute or fetch it */
  if (!blackLevelSeparate)
    calculateBlackAreas();
 
  startWorker(RawImageWorker::RawImageWorkerTask::SCALE_VALUES, true);
}
 
void RawImageDataU16::scaleValues(int start_y, int end_y) {
#ifndef WITH_SSE2
 
  return scaleValues_plain(start_y, end_y);
 
#else
 
  int depth_values = *whitePoint - (*blackLevelSeparate)(0, 0);
  float app_scale = 65535.0F / implicit_cast<float>(depth_values);
 
  // Check SSE2
  if (Cpuid::SSE2() && app_scale < 63) {
    scaleValues_SSE2(start_y, end_y);
  } else {
    scaleValues_plain(start_y, end_y);
  }
 
#endif
}
 
#ifdef WITH_SSE2
void RawImageDataU16::scaleValues_SSE2(int start_y, int end_y) {
  assert(blackLevelSeparate->width() == 2 && blackLevelSeparate->height() == 2);
  auto blackLevelSeparate1D = *blackLevelSeparate->getAsArray1DRef();
 
  int depth_values = *whitePoint - blackLevelSeparate1D(0);
  float app_scale = 65535.0F / implicit_cast<float>(depth_values);
 
  // Scale in 30.2 fp
  auto full_scale_fp = static_cast<int>(app_scale * 4.0F);
  // Half Scale in 18.14 fp
  auto half_scale_fp = static_cast<int>(app_scale * 4095.0F);
 
  __m128i sseround;
  __m128i ssesub2;
  __m128i ssesign;
  __m128i rand_mul;
  __m128i rand_mask;
  __m128i sse_full_scale_fp;
  __m128i sse_half_scale_fp;
 
  std::array<uint32_t, 4 * 4> sub_mul;
 
  // 10 bit fraction
  uint32_t mul = static_cast<int>(
      1024.0F * 65535.0F /
      static_cast<float>(*whitePoint - blackLevelSeparate1D(mOffset.x & 1)));
  mul |= (static_cast<int>(
             1024.0F * 65535.0F /
             static_cast<float>(*whitePoint -
                                blackLevelSeparate1D((mOffset.x + 1) & 1))))
         << 16;
  uint32_t b = blackLevelSeparate1D(mOffset.x & 1) |
               (blackLevelSeparate1D((mOffset.x + 1) & 1) << 16);
 
  for (int i = 0; i < 4; i++) {
    sub_mul[i] = b;       // Subtract even lines
    sub_mul[4 + i] = mul; // Multiply even lines
  }
 
  mul = static_cast<int>(
      1024.0F * 65535.0F /
      static_cast<float>(*whitePoint -
                         blackLevelSeparate1D(2 + (mOffset.x & 1))));
  mul |= (static_cast<int>(
             1024.0F * 65535.0F /
             static_cast<float>(*whitePoint - blackLevelSeparate1D(
                                                  2 + ((mOffset.x + 1) & 1)))))
         << 16;
  b = blackLevelSeparate1D(2 + (mOffset.x & 1)) |
      (blackLevelSeparate1D(2 + ((mOffset.x + 1) & 1)) << 16);
 
  for (int i = 0; i < 4; i++) {
    sub_mul[8 + i] = b;    // Subtract odd lines
    sub_mul[12 + i] = mul; // Multiply odd lines
  }
 
  sseround = _mm_set_epi32(512, 512, 512, 512);
  ssesub2 = _mm_set_epi32(32768, 32768, 32768, 32768);
  ssesign = _mm_set_epi32(0x80008000, 0x80008000, 0x80008000, 0x80008000);
  sse_full_scale_fp = _mm_set1_epi32(full_scale_fp | (full_scale_fp << 16));
  sse_half_scale_fp = _mm_set1_epi32(half_scale_fp >> 4);
 
  if (mDitherScale) {
    rand_mul = _mm_set1_epi32(0x4d9f1d32);
  } else {
    rand_mul = _mm_set1_epi32(0);
  }
  rand_mask = _mm_set1_epi32(0x00ff00ff); // 8 random bits
 
  Array2DRef<uint16_t> out(getU16DataAsUncroppedArray2DRef());
 
  for (int y = start_y; y < end_y; y++) {
    __m128i sserandom;
    if (mDitherScale) {
      sserandom = _mm_set_epi32(
          (dim.x * 1676) + (y * 18000), (dim.x * 2342) + (y * 34311),
          (dim.x * 4272) + (y * 12123), (dim.x * 1234) + (y * 23464));
    } else {
      sserandom = _mm_setzero_si128();
    }
    __m128i ssescale;
    __m128i ssesub;
    if (((y + mOffset.y) & 1) == 0) {
      ssesub = _mm_loadu_si128(reinterpret_cast<__m128i*>(&sub_mul[0]));
      ssescale = _mm_loadu_si128(reinterpret_cast<__m128i*>(&sub_mul[4]));
    } else {
      ssesub = _mm_loadu_si128(reinterpret_cast<__m128i*>(&sub_mul[8]));
      ssescale = _mm_loadu_si128(reinterpret_cast<__m128i*>(&sub_mul[12]));
    }
 
    for (int x = 0; x < static_cast<int>(roundDown(uncropped_dim.x, 8));
         x += 8) {
      __m128i pix_high;
      __m128i temp;
      __m128i pix_low =
          _mm_load_si128(reinterpret_cast<__m128i*>(&out(mOffset.y + y, x)));
      // Subtract black
      pix_low = _mm_subs_epu16(pix_low, ssesub);
      // Multiply the two unsigned shorts and combine it to 32 bit result
      pix_high = _mm_mulhi_epu16(pix_low, ssescale);
      temp = _mm_mullo_epi16(pix_low, ssescale);
      pix_low = _mm_unpacklo_epi16(temp, pix_high);
      pix_high = _mm_unpackhi_epi16(temp, pix_high);
      // Add rounder
      pix_low = _mm_add_epi32(pix_low, sseround);
      pix_high = _mm_add_epi32(pix_high, sseround);
 
      sserandom = _mm_xor_si128(_mm_mulhi_epi16(sserandom, rand_mul),
                                _mm_mullo_epi16(sserandom, rand_mul));
      __m128i rand_masked =
          _mm_and_si128(sserandom, rand_mask); // Get 8 random bits
      rand_masked = _mm_mullo_epi16(rand_masked, sse_full_scale_fp);
 
      __m128i zero = _mm_setzero_si128();
      __m128i rand_lo = _mm_sub_epi32(sse_half_scale_fp,
                                      _mm_unpacklo_epi16(rand_masked, zero));
      __m128i rand_hi = _mm_sub_epi32(sse_half_scale_fp,
                                      _mm_unpackhi_epi16(rand_masked, zero));
 
      pix_low = _mm_add_epi32(pix_low, rand_lo);
      pix_high = _mm_add_epi32(pix_high, rand_hi);
 
      // Shift down
      pix_low = _mm_srai_epi32(pix_low, 10);
      pix_high = _mm_srai_epi32(pix_high, 10);
      // Subtract to avoid clipping
      pix_low = _mm_sub_epi32(pix_low, ssesub2);
      pix_high = _mm_sub_epi32(pix_high, ssesub2);
      // Pack
      pix_low = _mm_packs_epi32(pix_low, pix_high);
      // Shift sign off
      pix_low = _mm_xor_si128(pix_low, ssesign);
      _mm_store_si128(reinterpret_cast<__m128i*>(&out(mOffset.y + y, x)),
                      pix_low);
    }
  }
}
#endif
 
void RawImageDataU16::scaleValues_plain(int start_y, int end_y) {
  const CroppedArray2DRef<uint16_t> img(getU16DataAsCroppedArray2DRef());
 
  assert(blackLevelSeparate->width() == 2 && blackLevelSeparate->height() == 2);
  auto blackLevelSeparate1D = *blackLevelSeparate->getAsArray1DRef();
 
  int depth_values = *whitePoint - blackLevelSeparate1D(0);
  float app_scale = 65535.0F / implicit_cast<float>(depth_values);
 
  // Scale in 30.2 fp
  auto full_scale_fp = static_cast<int>(app_scale * 4.0F);
  // Half Scale in 18.14 fp
  auto half_scale_fp = static_cast<int>(app_scale * 4095.0F);
 
  // Not SSE2
  int gw = dim.x * cpp;
  std::array<int, 4> mul;
  std::array<int, 4> sub;
  for (int i = 0; i < 4; i++) {
    int v = i;
    if ((mOffset.x & 1) != 0)
      v ^= 1;
    if ((mOffset.y & 1) != 0)
      v ^= 2;
    mul[i] = static_cast<int>(
        16384.0F * 65535.0F /
        static_cast<float>(*whitePoint - blackLevelSeparate1D(v)));
    sub[i] = blackLevelSeparate1D(v);
  }
  for (int y = start_y; y < end_y; y++) {
    int v = dim.x + (y * 36969);
    for (int x = 0; x < gw; x++) {
      int rand;
      if (mDitherScale) {
        v = 18000 * (v & 65535) + (v >> 16);
        rand = half_scale_fp - (full_scale_fp * (v & 2047));
      } else {
        rand = 0;
      }
      uint16_t& pixel = img(y, x);
      pixel = clampBits(((pixel - sub[(2 * (y & 1)) + (x & 1)]) *
                             mul[(2 * (y & 1)) + (x & 1)] +
                         8192 + rand) >>
                            14,
                        16);
    }
  }
}
 
/* This performs a 4 way interpolated pixel */
/* The value is interpolated from the 4 closest valid pixels in */
/* the horizontal and vertical direction. Pixels found further away */
/* are weighed less */
 
void RawImageDataU16::fixBadPixel(uint32_t x, uint32_t y, int component) {
  const Array2DRef<uint16_t> img = getU16DataAsUncroppedArray2DRef();
 
  array<int, 4> values;
  array<int, 4> dist;
  array<int, 4> weight;
 
  values.fill(-1);
  dist.fill(0);
  weight.fill(0);
 
  const auto bad =
      Array2DRef(mBadPixelMap.data(), mBadPixelMapPitch, uncropped_dim.y);
  int step = isCFA ? 2 : 1;
 
  // Find pixel to the left
  int x_find = static_cast<int>(x) - step;
  int curr = 0;
  while (x_find >= 0 && values[curr] < 0) {
    if (0 == ((bad(y, x_find >> 3) >> (x_find & 7)) & 1)) {
      values[curr] = img(y, x_find + component);
      dist[curr] = static_cast<int>(x) - x_find;
    }
    x_find -= step;
  }
  // Find pixel to the right
  x_find = static_cast<int>(x) + step;
  curr = 1;
  while (x_find < uncropped_dim.x && values[curr] < 0) {
    if (0 == ((bad(y, x_find >> 3) >> (x_find & 7)) & 1)) {
      values[curr] = img(y, x_find + component);
      dist[curr] = x_find - static_cast<int>(x);
    }
    x_find += step;
  }
 
  // Find pixel upwards
  int y_find = static_cast<int>(y) - step;
  curr = 2;
  while (y_find >= 0 && values[curr] < 0) {
    if (0 == ((bad(y_find, x >> 3) >> (x & 7)) & 1)) {
      values[curr] = img(y_find, x + component);
      dist[curr] = static_cast<int>(y) - y_find;
    }
    y_find -= step;
  }
  // Find pixel downwards
  y_find = static_cast<int>(y) + step;
  curr = 3;
  while (y_find < uncropped_dim.y && values[curr] < 0) {
    if (0 == ((bad(y_find, x >> 3) >> (x & 7)) & 1)) {
      values[curr] = img(y_find, x + component);
      dist[curr] = y_find - static_cast<int>(y);
    }
    y_find += step;
  }
 
  // Find x weights
  int total_dist_x = dist[0] + dist[1];
 
  int total_shifts = 7;
  if (total_dist_x) {
    weight[0] = dist[0] ? (total_dist_x - dist[0]) * 256 / total_dist_x : 0;
    weight[1] = 256 - weight[0];
    total_shifts++;
  }
 
  // Find y weights
  if (int total_dist_y = dist[2] + dist[3]; total_dist_y) {
    weight[2] = dist[2] ? (total_dist_y - dist[2]) * 256 / total_dist_y : 0;
    weight[3] = 256 - weight[2];
    total_shifts++;
  }
 
  int total_pixel = 0;
  for (int i = 0; i < 4; i++)
    if (values[i] >= 0)
      total_pixel += values[i] * weight[i];
 
  total_pixel >>= total_shifts;
  img(y, x + component) = clampBits(total_pixel, 16);
 
  /* Process other pixels - could be done inline, since we have the weights */
  if (cpp > 1 && component == 0)
    for (int i = 1; i < cpp; i++)
      fixBadPixel(x, y, i);
}
 
// TODO: Could be done with SSE2
void RawImageDataU16::doLookup(int start_y, int end_y) {
  const Array2DRef<uint16_t> img = getU16DataAsUncroppedArray2DRef();
 
  if (table->ntables == 1) {
    if (table->dither) {
      int gw = uncropped_dim.x * cpp;
      const auto t = table->getTable(0);
      for (int y = start_y; y < end_y; y++) {
        uint32_t v = (uncropped_dim.x + y * 13) ^ 0x45694584;
        for (int x = 0; x < gw; x++) {
          uint16_t p = img(y, x);
          uint32_t base = t((2 * p) + 0);
          uint32_t delta = t((2 * p) + 1);
          v = 15700 * (v & 65535) + (v >> 16);
          uint32_t pix = base + ((delta * (v & 2047) + 1024) >> 12);
          img(y, x) = clampBits(pix, 16);
        }
      }
      return;
    }
 
    int gw = uncropped_dim.x * cpp;
    const auto t = table->getTable(0);
    for (int y = start_y; y < end_y; y++) {
      for (int x = 0; x < gw; x++) {
        img(y, x) = t(img(y, x));
      }
    }
    return;
  }
  ThrowRDE("Table lookup with multiple components not implemented");
}
 
} // namespace rawspeed