Create FP16 LUTs when needed

cc/output/color_lut_cache.cc

 // Copyright 2016 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "cc/output/color_lut_cache.h"
 
 #include <stdint.h>
 #include <cmath>
 #include <vector>
 
 #include "gpu/command_buffer/client/gles2_interface.h"
+#include "third_party/khronos/GLES2/gl2ext.h"
 #include "ui/gfx/color_transform.h"
 
 // After a LUT has not been used for this many frames, we release it.
 const uint32_t kMaxFramesUnused = 10;
 
-ColorLUTCache::ColorLUTCache(gpu::gles2::GLES2Interface* gl)
-    : lut_cache_(0), gl_(gl) {}
+ColorLUTCache::ColorLUTCache(gpu::gles2::GLES2Interface* gl,
+                             bool texture_half_float_linear)
+    : lut_cache_(0),
+      gl_(gl),
+      texture_half_float_linear_(texture_half_float_linear) {}
 
 ColorLUTCache::~ColorLUTCache() {
   GLuint textures[10];
   size_t n = 0;
   for (const auto& cache_entry : lut_cache_) {
-    textures[n++] = cache_entry.second.texture;
+    textures[n++] = cache_entry.second.lut.texture;
     if (n == arraysize(textures)) {
       gl_->DeleteTextures(n, textures);
       n = 0;
     }
   }
   if (n)
     gl_->DeleteTextures(n, textures);
 }
 
 namespace {
 
-unsigned char FloatToLUT(float f) {
+unsigned short baseTable[512] = {

[ccameron, 2017/01/21 19:49:20]

See the source of this block,
https://cs.chromium.org/chromium/src/third_party/qcms/src/halffloat.h?rcl=0&l=27

[hubbe, 2017/01/22 05:12:39]

That's not actually where I copied it from.
Anywho, gone now.

+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     0,     0,     0,     0,     0,     0,     0,
+    0,     0,     0,     0,     1,     2,     4,     8,     16,    32,    64,
+    128,   256,   512,   1024,  2048,  3072,  4096,  5120,  6144,  7168,  8192,
+    9216,  10240, 11264, 12288, 13312, 14336, 15360, 16384, 17408, 18432, 19456,
+    20480, 21504, 22528, 23552, 24576, 25600, 26624, 27648, 28672, 29696, 30720,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744, 31744,
+    31744, 31744, 31744, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768,
+    32768, 32768, 32768, 32768, 32768, 32768, 32768, 32769, 32770, 32772, 32776,
+    32784, 32800, 32832, 32896, 33024, 33280, 33792, 34816, 35840, 36864, 37888,
+    38912, 39936, 40960, 41984, 43008, 44032, 45056, 46080, 47104, 48128, 49152,
+    50176, 51200, 52224, 53248, 54272, 55296, 56320, 57344, 58368, 59392, 60416,
+    61440, 62464, 63488, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512, 64512,
+    64512, 64512, 64512, 64512, 64512, 64512};
+
+unsigned char shiftTable[512] = {
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13,
+    13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
+    13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 13, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 23, 22,
+    21, 20, 19, 18, 17, 16, 15, 14, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
+    13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
+    24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 13};
+
+template <typename T>
+T FloatToLUT(float f) {}
+
+template <>
+unsigned short FloatToLUT<unsigned short>(float f) {
+  unsigned temp = *(reinterpret_cast<unsigned*>(&f));
+  unsigned signexp = (temp >> 23) & 0x1ff;
+  return baseTable[signexp] + ((temp & 0x007fffff) >> shiftTable[signexp]);
+}
+
+template <>
+unsigned char FloatToLUT<unsigned char>(float f) {
   return std::min<int>(255, std::max<int>(0, floorf(f * 255.0f + 0.5f)));
 }
 };
 
+template <typename T>
 unsigned int ColorLUTCache::MakeLUT(const gfx::ColorSpace& from,
                                     gfx::ColorSpace to,
                                     int lut_samples) {
   if (to == gfx::ColorSpace()) {
     to = gfx::ColorSpace::CreateSRGB();
   }
   std::unique_ptr<gfx::ColorTransform> transform(
       gfx::ColorTransform::NewColorTransform(
           from, to, gfx::ColorTransform::Intent::INTENT_PERCEPTUAL));
 
   int lut_entries = lut_samples * lut_samples * lut_samples;
   float inverse = 1.0f / (lut_samples - 1);
-  std::vector<unsigned char> lut(lut_entries * 4);
+  std::vector<T> lut(lut_entries * 4);
   std::vector<gfx::ColorTransform::TriStim> samples(lut_samples);
-  unsigned char* lutp = lut.data();
+  T* lutp = lut.data();
   for (int v = 0; v < lut_samples; v++) {
     for (int u = 0; u < lut_samples; u++) {
       for (int y = 0; y < lut_samples; y++) {
         samples[y].set_x(y * inverse);
         samples[y].set_y(u * inverse);
         samples[y].set_z(v * inverse);
       }
       transform->transform(samples.data(), samples.size());
       for (int y = 0; y < lut_samples; y++) {
-        *(lutp++) = FloatToLUT(samples[y].x());
-        *(lutp++) = FloatToLUT(samples[y].y());
-        *(lutp++) = FloatToLUT(samples[y].z());
-        *(lutp++) = 255;  // alpha
+        *(lutp++) = FloatToLUT<T>(samples[y].x());
+        *(lutp++) = FloatToLUT<T>(samples[y].y());
+        *(lutp++) = FloatToLUT<T>(samples[y].z());
+        *(lutp++) = FloatToLUT<T>(1.0);  // alpha
       }
     }
   }
 
   unsigned int lut_texture;
   gl_->GenTextures(1, &lut_texture);
   gl_->BindTexture(GL_TEXTURE_2D, lut_texture);
   gl_->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
   gl_->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
   gl_->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
   gl_->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);

[ccameron, 2017/01/21 01:10:49]

To use 16-bit floating point, the upload data doesn't need to be 16-bit floating
point (in fact, I'm a bit surprised that this worked).

The easier way to do this is to do:

  // Create storage in the requested format. Note that the last 3
  // args are irrelevant, because the "data" ptr is nullptr
  TexImage2D(
      GL_TEXTURE_2D, 0, GL_RGBA16F, width, height, 0,
      GL_RGBA, GL_UNSIGNED_BYTE, nullptr);
  // Populate the storage with 32-bit floating point data. This will
  // convert from 32-bit floating point to 16-bit floating point in
  // the DMA engine.
  TexSubImage2D(
      GL_TEXTURE_2D, 0, 0, 0, width, height,
      GL_RGBA, GL_FLOAT, data)

I think you can also use TexStorage2DEXT instead of TexImage2D to create the
texture storage.

[hubbe, 2017/01/22 05:12:39]

Done.

   gl_->TexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, lut_samples,
-                  lut_samples * lut_samples, 0, GL_RGBA, GL_UNSIGNED_BYTE,
+                  lut_samples * lut_samples, 0, GL_RGBA,
+                  sizeof(T) == 1 ? GL_UNSIGNED_BYTE : GL_HALF_FLOAT_OES,
                   lut.data());
   return lut_texture;
 }
 
-unsigned int ColorLUTCache::GetLUT(const gfx::ColorSpace& from,
-                                   const gfx::ColorSpace& to,
-                                   int lut_samples) {
-  CacheKey key(from, std::make_pair(to, lut_samples));
+ColorLUTCache::LUT ColorLUTCache::GetLUT(const gfx::ColorSpace& from,
+                                         const gfx::ColorSpace& to) {
+  CacheKey key(from, to);
   auto iter = lut_cache_.Get(key);
   if (iter != lut_cache_.end()) {
     iter->second.last_used_frame = current_frame_;
-    return iter->second.texture;
+    return iter->second.lut;
   }
 
-  unsigned int lut = MakeLUT(from, to, lut_samples);
+  LUT lut;
+  // If input is HDR, and the output is scRGB, we're going to need
+  // to produce values outside of 0-1, so we'll need to make a half-float
+  // LUT. Also, we'll need to build a larger lut to maintain accuracy.
+  // All LUT sizes should be odd a some transforms hav a knee at 0.5.
+  if (to == gfx::ColorSpace::CreateSCRGBLinear() && from.IsHDR() &&
+      texture_half_float_linear_) {
+    lut.size = 37;
+    lut.texture = MakeLUT<unsigned short>(from, to, lut.size);
+  } else {
+    lut.size = 17;
+    lut.texture = MakeLUT<unsigned char>(from, to, lut.size);
+  }
   lut_cache_.Put(key, CacheVal(lut, current_frame_));
   return lut;
 }
 
 void ColorLUTCache::Swap() {
   current_frame_++;
   while (!lut_cache_.empty() &&
          current_frame_ - lut_cache_.rbegin()->second.last_used_frame >
              kMaxFramesUnused) {
-    gl_->DeleteTextures(1, &lut_cache_.rbegin()->second.texture);
+    gl_->DeleteTextures(1, &lut_cache_.rbegin()->second.lut.texture);
     lut_cache_.ShrinkToSize(lut_cache_.size() - 1);
   }
 }