mirror-github-bspeice/fractorium
1
0
Fork 0
mirror of https://bitbucket.org/mfeemster/fractorium.git synced 2025-02-01 18:40:12 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity

--User changes

Browse Source 
-crackle is now 5-10 times faster as a result of using cached values as is done on the CPU. These were previously omitted due to a bug elsewhere in the code that was erroneously attributed to the crackle caching mechanism.
 -Report precision in command line programs.
This commit is contained in:
mfeemster 2016-04-14 19:32:42 -07:00
parent 0fbea60026
commit d639921d68
4 changed files with 30 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
Source/Ember/Variations06.h
View File

@ -3768,24 +3768,29 @@ public:
virtual string OpenCLFuncsString() const override virtual string OpenCLFuncsString() const override
{ {
//CPU version uses a cache of points if the abs() values are <= 10. However, this crashes on Nvidia GPUs. ostringstream os;
//The problem was traced to the usage of the cache array. os <<
//No possible solution was found, so it is unused here. "static void Position(__constant real2* cache, __global real_t* p, __global real_t* grad, int x, int y, real_t z, real_t s, real_t d, real2* v)\n"
//The full calculation is recomputed for every point. "{\n"
return " if (abs(x) <= " << CACHE_NUM << " && abs(y) <= " << CACHE_NUM << ")\n"
"static void Position(__global real_t* p, __global real_t* grad, int x, int y, real_t z, real_t s, real_t d, real2* v)\n" " {\n"
"{\n" " *v = cache[((x + " << CACHE_NUM << ") * " << CACHE_WIDTH << ") + (y + " << CACHE_NUM << ")];\n"
" real3 e, f;\n" " }\n"
" e.x = x * 2.5;\n" " else\n"
" e.y = y * 2.5;\n" " {\n"
" e.z = z * 2.5;\n" " real3 e, f;\n"
" f.x = y * 2.5 + 30.2;\n" " e.x = x * 2.5;\n"
" f.y = x * 2.5 - 12.1;\n" " e.y = y * 2.5;\n"
" f.z = z * 2.5 + 19.8;\n" " e.z = z * 2.5;\n"
" (*v).x = (x + d * SimplexNoise3D(&e, p, grad)) * s;\n" " f.x = y * 2.5 + 30.2;\n"
" (*v).y = (y + d * SimplexNoise3D(&f, p, grad)) * s;\n" " f.y = x * 2.5 - 12.1;\n"
"}\n" " f.z = z * 2.5 + 19.8;\n"
"\n"; " (*v).x = (x + d * SimplexNoise3D(&e, p, grad)) * s;\n"
" (*v).y = (y + d * SimplexNoise3D(&f, p, grad)) * s;\n"
" }\n"
"}\n"
"\n";
return os.str();
} }
virtual string OpenCLString() const override virtual string OpenCLString() const override
@ -3800,6 +3805,7 @@ public:
string scale = "parVars[" + ToUpper(m_Params[i++].Name()) + index; string scale = "parVars[" + ToUpper(m_Params[i++].Name()) + index;
string z = "parVars[" + ToUpper(m_Params[i++].Name()) + index; string z = "parVars[" + ToUpper(m_Params[i++].Name()) + index;
string halfCellSize = "parVars[" + ToUpper(m_Params[i++].Name()) + index; string halfCellSize = "parVars[" + ToUpper(m_Params[i++].Name()) + index;
string cache = "parVars[" + ToUpper(m_Params[i++].Name()) + index;
ss << "\t{\n" ss << "\t{\n"
<< "\t\tint di = -1, dj = -1;\n" << "\t\tint di = -1, dj = -1;\n"
<< "\t\tint i = 0;\n" << "\t\tint i = 0;\n"
@ -3822,7 +3828,7 @@ public:
<< "\t\t{\n" << "\t\t{\n"
<< "\t\t for (dj = -1; dj < 2; dj++)\n" << "\t\t for (dj = -1; dj < 2; dj++)\n"
<< "\t\t {\n" << "\t\t {\n"
<< "\t\t Position(globalShared + NOISE_INDEX, globalShared + NOISE_POINTS, cv.x + di, cv.y + dj, " << z << ", " << halfCellSize << ", " << distort << ", &p[i]);\n" << "\t\t Position((__constant real2*)(&" << cache << "), globalShared + NOISE_INDEX, globalShared + NOISE_POINTS, cv.x + di, cv.y + dj, " << z << ", " << halfCellSize << ", " << distort << ", &p[i]); \n"
<< "\t\t i++;\n" << "\t\t i++;\n"
<< "\t\t }\n" << "\t\t }\n"
<< "\t\t}\n" << "\t\t}\n"
@ -3838,7 +3844,7 @@ public:
<< "\t\t{\n" << "\t\t{\n"
<< "\t\t for (dj = -1; dj < 2; dj++)\n" << "\t\t for (dj = -1; dj < 2; dj++)\n"
<< "\t\t {\n" << "\t\t {\n"
<< "\t\t Position(globalShared + NOISE_INDEX, globalShared + NOISE_POINTS, cv.x + di, cv.y + dj, " << z << ", " << halfCellSize << ", " << distort << ", &p[i]);\n" << "\t\t Position((__constant real2*)(&" << cache << "), globalShared + NOISE_INDEX, globalShared + NOISE_POINTS, cv.x + di, cv.y + dj, " << z << ", " << halfCellSize << ", " << distort << ", &p[i]);\n"
<< "\t\t i++;\n" << "\t\t i++;\n"
<< "\t\t }\n" << "\t\t }\n"
<< "\t\t}\n" << "\t\t}\n"
@ -3877,6 +3883,7 @@ protected:
m_Params.push_back(ParamWithName<T>(&m_Scale, prefix + "crackle_scale", 1)); m_Params.push_back(ParamWithName<T>(&m_Scale, prefix + "crackle_scale", 1));
m_Params.push_back(ParamWithName<T>(&m_Z, prefix + "crackle_z")); m_Params.push_back(ParamWithName<T>(&m_Z, prefix + "crackle_z"));
m_Params.push_back(ParamWithName<T>(true, &m_HalfCellSize, prefix + "crackle_half_cellsize")); m_Params.push_back(ParamWithName<T>(true, &m_HalfCellSize, prefix + "crackle_half_cellsize"));
m_Params.push_back(ParamWithName<T>(true, &(m_C[0][0].x), prefix + "crackle_cache", sizeof(m_C)));
} }
private: private:

1
Source/EmberAnimate/EmberAnimate.cpp
View File

@ -24,6 +24,7 @@ bool EmberAnimate(EmberOptions& opt)
return true; return true;
} }
VerbosePrint("Using " << (sizeof(T) == sizeof(float) ? "single" : "double") << " precision.");
//Regular variables. //Regular variables.
Timing t; Timing t;
bool unsorted = false; bool unsorted = false;

1
Source/EmberGenome/EmberGenome.cpp
View File

@ -97,6 +97,7 @@ bool EmberGenome(EmberOptions& opt)
return true; return true;
} }
VerbosePrint("Using " << (sizeof(T) == sizeof(float) ? "single" : "double") << " precision.");
//Regular variables. //Regular variables.
Timing t; Timing t;
bool exactTimeMatch, randomMode, didColor, seqFlag; bool exactTimeMatch, randomMode, didColor, seqFlag;

1
Source/EmberRender/EmberRender.cpp
View File

@ -26,6 +26,7 @@ bool EmberRender(EmberOptions& opt)
return true; return true;
} }
VerbosePrint("Using " << (sizeof(T) == sizeof(float) ? "single" : "double") << " precision.");
Timing t; Timing t;
bool writeSuccess = false; bool writeSuccess = false;
byte* finalImagep; byte* finalImagep;