mirror-github-bspeice/fractorium
1
0
Fork 0
mirror of https://bitbucket.org/mfeemster/fractorium.git synced 2025-01-21 21:20:07 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity
47b6614c8a
fractorium/Source/EmberCL/FinalAccumOpenCLKernelCreator.cpp

317 lines
13 KiB
C++
Raw Normal View History

Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
#include "EmberCLPch.h"
#include "FinalAccumOpenCLKernelCreator.h"
namespace EmberCLns
{
/// <summary>
/// Constructor that creates all kernel strings.
/// The caller will access these strings through the accessor functions.
/// </summary>
--User changes -Remove Hue as a saved parameter, as well as animation parameters associated with it. It's now a GUI-only field that is never saved. -Make histogram, density filter buffer, and all associated fields always float, even when using double. In that case, only the iteration calculations are now double. Suggested by Thomas Ludwig. -Print all three kernels in EmberRender when the --dump_kernel option is specified. -Apply variations filter to randoms. --Bug fixes -Fix bug where hue was not being preserved when switching controllers and embers. Very hard to repro bug, but mostly overcome by eliminating hue as a saved parameter. --Code changes -De-templatized DEOpenCLKernelCreator and FinalAccumOpenCLKernelCreator. They now just take a bool as a parameter to specify double precision. -To accommodate the buffers being float, introduce a new #define types in EmberCL called real4_bucket, and real4reals_bucket. -Density and spatial filtering structs now use this type. -ConvertDensityFilter() and ConvertSpatialFilter() no longer return a value, they just assign to the member.
2015-08-10 23:10:23 -04:00
FinalAccumOpenCLKernelCreator::FinalAccumOpenCLKernelCreator(bool doublePrecision)
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
--User changes -Remove Hue as a saved parameter, as well as animation parameters associated with it. It's now a GUI-only field that is never saved. -Make histogram, density filter buffer, and all associated fields always float, even when using double. In that case, only the iteration calculations are now double. Suggested by Thomas Ludwig. -Print all three kernels in EmberRender when the --dump_kernel option is specified. -Apply variations filter to randoms. --Bug fixes -Fix bug where hue was not being preserved when switching controllers and embers. Very hard to repro bug, but mostly overcome by eliminating hue as a saved parameter. --Code changes -De-templatized DEOpenCLKernelCreator and FinalAccumOpenCLKernelCreator. They now just take a bool as a parameter to specify double precision. -To accommodate the buffers being float, introduce a new #define types in EmberCL called real4_bucket, and real4reals_bucket. -Density and spatial filtering structs now use this type. -ConvertDensityFilter() and ConvertSpatialFilter() no longer return a value, they just assign to the member.
2015-08-10 23:10:23 -04:00
m_DoublePrecision = doublePrecision;
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
m_GammaCorrectionWithoutAlphaCalcKernel = CreateGammaCorrectionKernelString();
m_FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumKernel = CreateFinalAccumKernelString(true);
m_FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumKernel = CreateFinalAccumKernelString(false);
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
/// <summary>
/// Kernel source and entry point properties, getters only.
/// </summary>
--User changes -Add Simon Detheridge's name to the About Box. --Bug fixes -Fix bug in OpenCL atomic string, which is never used. -Wrong hist and accum allocation size in RendererCL when using float-only buffers now. -Move some kernel initialization to a place where it's done once per render, rather than on every interactive iter chunk. --Code changes -Make ConvertCarToRas() just assign to the member rather than return a struct. -Make kernel string accessor functions in IterOpenCLKernelCreator, FinalAccumOpenCLKernelCreator and DEOpenCLKernelCreator be const and return a const string reference. -Don't include atomic string unless locking on the GPU, which is never.
2015-08-12 21:51:07 -04:00
const string& FinalAccumOpenCLKernelCreator::FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumKernel() const { return m_FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumKernel; }
const string& FinalAccumOpenCLKernelCreator::FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumEntryPoint() const { return m_FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumEntryPoint; }
const string& FinalAccumOpenCLKernelCreator::FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumKernel() const { return m_FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumKernel; }
const string& FinalAccumOpenCLKernelCreator::FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumEntryPoint() const { return m_FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumEntryPoint; }
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
const string& FinalAccumOpenCLKernelCreator::GammaCorrectionEntryPoint() const { return m_GammaCorrectionWithoutAlphaCalcEntryPoint; }
const string& FinalAccumOpenCLKernelCreator::GammaCorrectionKernel() const { return m_GammaCorrectionWithoutAlphaCalcKernel; }
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
/// <summary>
/// Get the final accumulation entry point.
/// </summary>
/// <param name="earlyClip">True if early clip is desired, else false.</param>
/// <returns>The name of the final accumulation entry point kernel function</returns>
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
const string& FinalAccumOpenCLKernelCreator::FinalAccumEntryPoint(bool earlyClip) const
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
if (earlyClip)
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
return FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumEntryPoint();
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
else
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
return FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumEntryPoint();
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
/// <summary>
/// Get the final accumulation kernel string.
/// </summary>
/// <param name="earlyClip">True if early clip is desired, else false.</param>
/// <returns>The final accumulation kernel string</returns>
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
const string& FinalAccumOpenCLKernelCreator::FinalAccumKernel(bool earlyClip) const
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
if (earlyClip)
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
return FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumKernel();
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
else
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
return FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumKernel();
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
/// <summary>
/// Create the final accumulation kernel string
/// </summary>
/// <param name="earlyClip">True if early clip is desired, else false.</param>
/// <returns>The final accumulation kernel string</returns>
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
string FinalAccumOpenCLKernelCreator::CreateFinalAccumKernelString(bool earlyClip)
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
ostringstream os;
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
ConstantDefinesString(m_DoublePrecision) <<
UnionCLStructString <<
RgbToHsvFunctionString <<
HsvToRgbFunctionString <<
CalcAlphaFunctionString <<
CurveAdjustFunctionString <<
SpatialFilterCLStructString;
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
if (earlyClip)
{
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
os << "__kernel void " << m_FinalAccumEarlyClipWithoutAlphaCalcWithAlphaAccumEntryPoint << "(\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
else
{
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
CreateCalcNewRgbFunctionString(false) <<
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
CreateGammaCorrectionFunctionString(false, true) <<
"__kernel void " << m_FinalAccumLateClipWithoutAlphaCalcWithAlphaAccumEntryPoint << "(\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" const __global real4reals_bucket* accumulator,\n"
" __write_only image2d_t pixels,\n"
" __constant SpatialFilterCL* spatialFilter,\n"
" __constant real_bucket_t* filterCoefs,\n"
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
" __global real4reals_bucket* csa,\n"
" const uint doCurves\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"\t)\n"
"{\n"
"\n"
" if ((GLOBAL_ID_Y >= spatialFilter->m_FinalRasH) || (GLOBAL_ID_X >= spatialFilter->m_FinalRasW))\n"
" return;\n"
"\n"
" uint accumX = spatialFilter->m_DensityFilterOffset + (GLOBAL_ID_X * spatialFilter->m_Supersample);\n"
" uint accumY = spatialFilter->m_DensityFilterOffset + (GLOBAL_ID_Y * spatialFilter->m_Supersample);\n"
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
" uint clampedFilterH = min((uint)spatialFilter->m_FilterWidth, spatialFilter->m_SuperRasH - accumY);"
" uint clampedFilterW = min((uint)spatialFilter->m_FilterWidth, spatialFilter->m_SuperRasW - accumX);"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" int2 finalCoord;\n"
" finalCoord.x = GLOBAL_ID_X;\n"
" finalCoord.y = (int)((spatialFilter->m_YAxisUp == 1) ? ((spatialFilter->m_FinalRasH - GLOBAL_ID_Y) - 1) : GLOBAL_ID_Y);\n"
" float4floats finalColor;\n"
" int ii, jj;\n"
" uint filterKRowIndex;\n"
" const __global real4reals_bucket* accumBucket;\n"
" real4reals_bucket newBucket;\n"
" newBucket.m_Real4 = 0;\n"
"\n"
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
" for (jj = 0; jj < clampedFilterH; jj++)\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" {\n"
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
" filterKRowIndex = jj * spatialFilter->m_FilterWidth;\n"//Use the full, non-clamped width to get the filter value.
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"\n"
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
" for (ii = 0; ii < clampedFilterW; ii++)\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" {\n"
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
" real_bucket_t k = filterCoefs[filterKRowIndex + ii];\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"\n"
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
" accumBucket = accumulator + ((accumY + jj) * spatialFilter->m_SuperRasW) + (accumX + ii);\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" newBucket.m_Real4 += (k * accumBucket->m_Real4);\n"
" }\n"
" }\n"
"\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
Remove ReadMe.txt from all project files. Add Curves.h, and CurvesGraphicsView.h/cpp to support bezier color curves. Add Curves member to Ember. Add curves capability to EmberCL. Remove some unused variables in the kernel created in RendererCL::CreateFinalAccumKernelString(). Use glm namespace for vec classes if GLM_VERSION >= 96, else use glm::detail. As a result of using glm namespace, all instances of min and max had to be qualified with std:: Split ComputeCamera into that and ComputeQuality(). Reduce the amount of ComputeCamera() and MakeDmap() calls on each incremental iter that doesn't use temporal samples. Fix clamping bug with DE filter widths. Provide functions to return the kernels from RendererCL to assist with diagnostics and debugging. Prevent extra newline in EmberRender when only rendering a single image. Add the ability to delete an ember at a given index in EmberFile. Allow deleting/focusing ember in library tab with delete and enter keys. Reorder some code in Fractorium.h to match the tabs order. Add and call ClearFinalImages() to clear buffers in controller to fix bug where previous CPU render would be shown for a split second when switching from OpenCL back to CPU. Refactor ember library pointer syncing to a function SyncPointers(). Add the ability to save ember Xmls to an unique automatically generated name after the first time the user has specified a name.
2015-03-21 18:27:37 -04:00
if (earlyClip)//If early clip, simply assign values directly to the temp float4 since they've been gamma corrected already, then write it straight to the output image below.
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" finalColor.m_Float4.x = (float)newBucket.m_Real4.x;\n"//CPU side clamps, skip here because write_imagef() does the clamping for us.
" finalColor.m_Float4.y = (float)newBucket.m_Real4.y;\n"
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
" finalColor.m_Float4.z = (float)newBucket.m_Real4.z;\n"
" finalColor.m_Float4.w = (float)newBucket.m_Real4.w;\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
else
{
--User changes -Better handling of crossed min/max DE filter values on the GUI. -Changes to spatial and density filter values will no longer restart the entire render. -Support reading cam_zoom from xml as zoom since JWildfire uses cam_zoom. -Change drag n drop behavior: the default is now to append, hold Ctrl to overwrite. -Set max value of zoom to 25 because it will crash with values around 28 or 29 due to numeric overflow when scaling quality. -Update links in About Dialog. -Rename some controls in Options Dialog. -Move the color curves control to the Palette tab. --Bug fixes -Remove extra space in info tree text when dealing with a linked xform. -Update summary xform name field whenever xform name changes. -Get rid of selection border on summary tab xform tree cells. -Only add variations from Xml that have a non-zero weight. There seemed to be spurious flattens coming from Apo with a weight of 0. --Code changes -Gutters are now a fixed size of 8 * supersample. -Use stl data() member to get pointer to first element instead of &vec[0]. -Eliminate extra calls in renderer to ComputeBounds(), CreateSpatialFilter() and CreateTemporalFilter() to only be done at the start of a render. -Another attempt at vectorizing LogScaleDensityFilter(). Vectorizes, but not sure if it helps. -Some other loop optimizations in Renderer. -No longer check temporal samples in response to some control changes, they are always 1 in the interactive renderer.
2016-03-28 21:49:10 -04:00
//Late clip, so must gamma correct from the temp newBucket to temp finalColor float4.
--User changes -Remove Hue as a saved parameter, as well as animation parameters associated with it. It's now a GUI-only field that is never saved. -Make histogram, density filter buffer, and all associated fields always float, even when using double. In that case, only the iteration calculations are now double. Suggested by Thomas Ludwig. -Print all three kernels in EmberRender when the --dump_kernel option is specified. -Apply variations filter to randoms. --Bug fixes -Fix bug where hue was not being preserved when switching controllers and embers. Very hard to repro bug, but mostly overcome by eliminating hue as a saved parameter. --Code changes -De-templatized DEOpenCLKernelCreator and FinalAccumOpenCLKernelCreator. They now just take a bool as a parameter to specify double precision. -To accommodate the buffers being float, introduce a new #define types in EmberCL called real4_bucket, and real4reals_bucket. -Density and spatial filtering structs now use this type. -ConvertDensityFilter() and ConvertSpatialFilter() no longer return a value, they just assign to the member.
2015-08-10 23:10:23 -04:00
if (m_DoublePrecision)
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" real4reals_bucket realFinal;\n"
"\n"
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
" GammaCorrectionFloats(&newBucket, &(spatialFilter->m_Background[0]), spatialFilter->m_Gamma, spatialFilter->m_LinRange, spatialFilter->m_Vibrancy, spatialFilter->m_HighlightPower, &(realFinal.m_Reals[0]));\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" finalColor.m_Float4.x = (float)realFinal.m_Real4.x;\n"
" finalColor.m_Float4.y = (float)realFinal.m_Real4.y;\n"
" finalColor.m_Float4.z = (float)realFinal.m_Real4.z;\n"
" finalColor.m_Float4.w = (float)realFinal.m_Real4.w;\n"
;
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
else
{
os <<
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
" GammaCorrectionFloats(&newBucket, &(spatialFilter->m_Background[0]), spatialFilter->m_Gamma, spatialFilter->m_LinRange, spatialFilter->m_Vibrancy, spatialFilter->m_HighlightPower, &(finalColor.m_Floats[0]));\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
}
}
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"\n"
" if (doCurves)\n"
" {\n"
" CurveAdjust(csa, &(finalColor.m_Floats[0]), 1);\n"
" CurveAdjust(csa, &(finalColor.m_Floats[1]), 2);\n"
" CurveAdjust(csa, &(finalColor.m_Floats[2]), 3);\n"
" }\n"
"\n"
" write_imagef(pixels, finalCoord, finalColor.m_Float4);\n"//Use write_imagef instead of write_imageui because only the former works when sharing with an OpenGL texture.
" barrier(CLK_GLOBAL_MEM_FENCE);\n"//Required, or else page tearing will occur during interactive rendering.
"}\n"
;
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
return os.str();
}
/// <summary>
/// Creates the gamma correction function string.
/// This is not a full kernel, just a function that is used in the kernels.
/// </summary>
/// <param name="globalBucket">True if writing to a global buffer (early clip), else false (late clip).</param>
/// <param name="finalOut">True if writing to global buffer (late clip), else false (early clip).</param>
/// <returns>The gamma correction function string</returns>
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
string FinalAccumOpenCLKernelCreator::CreateGammaCorrectionFunctionString(bool globalBucket, bool finalOut)
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
ostringstream os;
string dataType;
string unionMember;
--User changes -Remove Hue as a saved parameter, as well as animation parameters associated with it. It's now a GUI-only field that is never saved. -Make histogram, density filter buffer, and all associated fields always float, even when using double. In that case, only the iteration calculations are now double. Suggested by Thomas Ludwig. -Print all three kernels in EmberRender when the --dump_kernel option is specified. -Apply variations filter to randoms. --Bug fixes -Fix bug where hue was not being preserved when switching controllers and embers. Very hard to repro bug, but mostly overcome by eliminating hue as a saved parameter. --Code changes -De-templatized DEOpenCLKernelCreator and FinalAccumOpenCLKernelCreator. They now just take a bool as a parameter to specify double precision. -To accommodate the buffers being float, introduce a new #define types in EmberCL called real4_bucket, and real4reals_bucket. -Density and spatial filtering structs now use this type. -ConvertDensityFilter() and ConvertSpatialFilter() no longer return a value, they just assign to the member.
2015-08-10 23:10:23 -04:00
dataType = "real_bucket_t";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
//Use real_t for all cases, early clip and final accum.
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
os << "void GammaCorrectionFloats(" << (globalBucket ? "__global " : "") << "real4reals_bucket* bucket, __constant real_bucket_t* background, real_bucket_t g, real_bucket_t linRange, real_bucket_t vibrancy, real_bucket_t highlightPower, " << (finalOut ? "" : "__global") << " real_bucket_t* correctedChannels)\n";
os << "{\n"
<< " real_bucket_t alpha, ls, tmp, a;\n"
<< " real4reals_bucket newRgb;\n"
<< "\n"
<< " if (bucket->m_Reals[3] <= 0)\n"
<< " {\n"
<< " alpha = 0;\n"
<< " ls = 0;\n"
<< " }\n"
<< " else\n"
<< " {\n"
<< " tmp = bucket->m_Reals[3];\n"
<< " alpha = CalcAlpha(tmp, g, linRange);\n"
<< " ls = vibrancy * alpha / tmp;\n"
<< " alpha = clamp(alpha, (real_bucket_t)0.0, (real_bucket_t)1.0);\n"
<< " }\n"
<< "\n"
<< " CalcNewRgb(bucket, ls, highlightPower, &newRgb);\n"
<< "\n"
<< " for (uint rgbi = 0; rgbi < 3; rgbi++)\n"
<< " {\n"
<< " a = newRgb.m_Reals[rgbi] + ((1.0 - vibrancy) * pow(fabs(bucket->m_Reals[rgbi]), g));\n"
<< " a += ((1.0 - alpha) * background[rgbi]);\n"
<< " correctedChannels[rgbi] = (" << dataType << ")clamp(a, (real_bucket_t)0.0, (real_bucket_t)1.0);\n"
<< " }\n"
<< "\n"
<< " correctedChannels[3] = (" << dataType << ")alpha;\n"
<< "}\n"
<< "\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
return os.str();
}
/// <summary>
/// OpenCL equivalent of Palette::CalcNewRgb().
/// </summary>
/// <param name="globalBucket">True if writing the corrected value to a global buffer (early clip), else false (late clip).</param>
/// <returns>The CalcNewRgb function string</returns>
--User changes -Remove Hue as a saved parameter, as well as animation parameters associated with it. It's now a GUI-only field that is never saved. -Make histogram, density filter buffer, and all associated fields always float, even when using double. In that case, only the iteration calculations are now double. Suggested by Thomas Ludwig. -Print all three kernels in EmberRender when the --dump_kernel option is specified. -Apply variations filter to randoms. --Bug fixes -Fix bug where hue was not being preserved when switching controllers and embers. Very hard to repro bug, but mostly overcome by eliminating hue as a saved parameter. --Code changes -De-templatized DEOpenCLKernelCreator and FinalAccumOpenCLKernelCreator. They now just take a bool as a parameter to specify double precision. -To accommodate the buffers being float, introduce a new #define types in EmberCL called real4_bucket, and real4reals_bucket. -Density and spatial filtering structs now use this type. -ConvertDensityFilter() and ConvertSpatialFilter() no longer return a value, they just assign to the member.
2015-08-10 23:10:23 -04:00
string FinalAccumOpenCLKernelCreator::CreateCalcNewRgbFunctionString(bool globalBucket)
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
ostringstream os;
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"static void CalcNewRgb(" << (globalBucket ? "__global " : "") << "real4reals_bucket* oldRgb, real_bucket_t ls, real_bucket_t highPow, real4reals_bucket* newRgb)\n"
"{\n"
" int rgbi;\n"
--Bug fixes -Attempt to fix several OpenCL build errors that were occurring on Mac. --Code changes -Improve the coloring code during final accum by getting rid of the last remnants of unnecessary scaling to 255 from flam3. -Begin work of supporting 16bpc on the GPU.
2017-07-05 02:08:06 -04:00
" real_bucket_t lsratio;\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" real4reals_bucket newHsv;\n"
--Bug fixes -Attempt to fix several OpenCL build errors that were occurring on Mac. --Code changes -Improve the coloring code during final accum by getting rid of the last remnants of unnecessary scaling to 255 from flam3. -Begin work of supporting 16bpc on the GPU.
2017-07-05 02:08:06 -04:00
" real_bucket_t maxa, maxc, newls;\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" real_bucket_t adjhlp;\n"
"\n"
" if (ls == 0 || (oldRgb->m_Real4.x == 0 && oldRgb->m_Real4.y == 0 && oldRgb->m_Real4.z == 0))\n"//Can't do a vector compare to zero.
" {\n"
" newRgb->m_Real4 = 0;\n"
" return;\n"
" }\n"
"\n"
//Identify the most saturated channel.
" maxc = max(max(oldRgb->m_Reals[0], oldRgb->m_Reals[1]), oldRgb->m_Reals[2]);\n"
" maxa = ls * maxc;\n"
--Bug fixes -Attempt to fix several OpenCL build errors that were occurring on Mac. --Code changes -Improve the coloring code during final accum by getting rid of the last remnants of unnecessary scaling to 255 from flam3. -Begin work of supporting 16bpc on the GPU.
2017-07-05 02:08:06 -04:00
" newls = 1 / maxc;\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"\n"
//If a channel is saturated and highlight power is non-negative
//modify the color to prevent hue shift.
--Bug fixes -Attempt to fix several OpenCL build errors that were occurring on Mac. --Code changes -Improve the coloring code during final accum by getting rid of the last remnants of unnecessary scaling to 255 from flam3. -Begin work of supporting 16bpc on the GPU.
2017-07-05 02:08:06 -04:00
" if (maxa > 1 && highPow >= 0)\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" {\n"
" lsratio = pow(newls / ls, highPow);\n"
"\n"
//Calculate the max-value color (ranged 0 - 1).
" for (rgbi = 0; rgbi < 3; rgbi++)\n"
--Bug fixes -Attempt to fix several OpenCL build errors that were occurring on Mac. --Code changes -Improve the coloring code during final accum by getting rid of the last remnants of unnecessary scaling to 255 from flam3. -Begin work of supporting 16bpc on the GPU.
2017-07-05 02:08:06 -04:00
" newRgb->m_Reals[rgbi] = newls * oldRgb->m_Reals[rgbi];\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"\n"
//Reduce saturation by the lsratio.
" RgbToHsv(&(newRgb->m_Real4), &(newHsv.m_Real4));\n"
" newHsv.m_Real4.y *= lsratio;\n"
" HsvToRgb(&(newHsv.m_Real4), &(newRgb->m_Real4));\n"
" }\n"
" else\n"
" {\n"
" adjhlp = -highPow;\n"
"\n"
" if (adjhlp > 1)\n"
" adjhlp = 1;\n"
"\n"
--Bug fixes -Attempt to fix several OpenCL build errors that were occurring on Mac. --Code changes -Improve the coloring code during final accum by getting rid of the last remnants of unnecessary scaling to 255 from flam3. -Begin work of supporting 16bpc on the GPU.
2017-07-05 02:08:06 -04:00
" if (maxa <= 1)\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" adjhlp = 1;\n"
"\n"
//Calculate the max-value color (ranged 0 - 1) interpolated with the old behavior.
" for (rgbi = 0; rgbi < 3; rgbi++)\n"//Unrolling, caching and vectorizing makes no difference.
" newRgb->m_Reals[rgbi] = ((1.0 - adjhlp) * newls + adjhlp * ls) * oldRgb->m_Reals[rgbi];\n"
" }\n"
"}\n"
"\n";
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
return os.str();
}
/// <summary>
/// Create the gamma correction kernel string used for early clipping.
/// </summary>
/// <returns>The gamma correction kernel string used for early clipping</returns>
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
string FinalAccumOpenCLKernelCreator::CreateGammaCorrectionKernelString()
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
{
ostringstream os;
string dataType;
os <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
ConstantDefinesString(m_DoublePrecision) <<
UnionCLStructString <<
RgbToHsvFunctionString <<
HsvToRgbFunctionString <<
CalcAlphaFunctionString <<
CreateCalcNewRgbFunctionString(true) <<
SpatialFilterCLStructString <<
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
CreateGammaCorrectionFunctionString(true, false);//Will only be used with float in this case, early clip. Will always alpha accum.
os << "__kernel void " << m_GammaCorrectionWithoutAlphaCalcEntryPoint << "(\n" <<
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
" __global real4reals_bucket* accumulator,\n"
" __constant SpatialFilterCL* spatialFilter\n"
")\n"
"{\n"
" int testGutter = 0;\n"
"\n"
" if (GLOBAL_ID_Y >= (spatialFilter->m_SuperRasH - testGutter) || GLOBAL_ID_X >= (spatialFilter->m_SuperRasW - testGutter))\n"
" return;\n"
"\n"
" uint superIndex = (GLOBAL_ID_Y * spatialFilter->m_SuperRasW) + GLOBAL_ID_X;\n"
" __global real4reals_bucket* bucket = accumulator + superIndex;\n"
Features: --Add support for Exr files which use 32-bit floats for each RGBA channel. Seems to come out too washed out. --Allow for clearing an individual color curve. --Allow for saving multiple image types in EmberRender and EmberAnimate. All writes are threaded. --Remove --bpc command line argument. Add format png16 as a replacement. --Remove --enable_jpg_comments and --enable_png_comments command line arguments, and replace them with --enable_comments which applies to jpg, png and exr. --Add menu items to variations and affine spinners which allow for easy entry of specific numeric values like pi. --Make final render dialog be wider rather than so tall. Bug fixes: --Fix some OpenCL compile errors on Mac. --Remove ability to save bitmap files on all platforms but Windows. Code changes: --New dependency on OpenEXR. --Allow Curves class to interact with objects of a different template type. --Make m_Curves member of Ember always use float as template type. --Set the length of the curves array to always be 2^17 which should offer enough precision with new 32-bit float pixel types. --Set pixel types to always be 32-bit float. This results in a major reduction of code in the final accumulation part of Renderer.h/cpp. --Remove corresponding code from RendererCL and FinalAccumOpenCLKernelCreator. --Remove Transparency, NumChannels and BytesPerPixel setters from Renderer.h/cpp. --Add new global functions to format final image buffers and place all alpha calculation and scaling code in them. --Blending is no longer needed in OpenGLWidget because of the new pixel type. --Make new class, AffineDoubleSpinBox. --Attempt to make file save dialog code work the same on all OSes. --Remove some unused functions.
2017-07-22 16:43:35 -04:00
" GammaCorrectionFloats(bucket, &(spatialFilter->m_Background[0]), spatialFilter->m_Gamma, spatialFilter->m_LinRange, spatialFilter->m_Vibrancy, spatialFilter->m_HighlightPower, &(bucket->m_Reals[0]));\n"
--Bug fixes -Fix crash when using Lanczos2 filter with color curves due to negative numbers. Fix by passing abs() of first argument to pow() in Renderer::GammaCorrection(). -Fix crash in hexes with SP. Rounding error caused out of bounds index.
2016-02-24 00:01:02 -05:00
"}\n"
;
Initial source commit Initial source commit
2014-07-08 03:11:14 -04:00
return os.str();
}
More Linux work. This has Render, Animate and Genome building and running on Linux.
2014-12-05 21:30:46 -05:00
}
Reference in New Issue Copy Permalink