RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-22 18:53:28 +01:00
Code Issues Releases Wiki Activity

fp32 precision on GPUs is embarassing

Browse Source 
- Division seems to suffer from drift easily on GPUs due to limited precision
This commit is contained in:
kd-11 2017-11-18 13:02:10 +03:00
parent ccc0383f75
commit 5cf2d774f3
1 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
rpcs3/Emu/RSX/GL/GLVertexProgram.cpp
View File

@ -306,13 +306,15 @@ void GLVertexDecompilerThread::insertMainEnd(std::stringstream & OS)
//It is therefore critical that this step is done post-transform and the result re-scaled by w
//SEE Naruto: UNS
OS << " float ndc_z = gl_Position.z / gl_Position.w;\n";
OS << " ndc_z = (ndc_z * 2.) - 1.;\n";
OS << " gl_Position.z = ndc_z * gl_Position.w;\n";
//NOTE: On GPUs, poor fp32 precision means dividing z by w, then multiplying by w again gives slightly incorrect results
//This equation is simplified algebraically to an addition and subreaction which gives more accurate results (Fixes flickering skybox in Dark Souls 2)
//OS << " float ndc_z = gl_Position.z / gl_Position.w;\n";
//OS << " ndc_z = (ndc_z * 2.) - 1.;\n";
//OS << " gl_Position.z = ndc_z * gl_Position.w;\n";
OS << " gl_Position.z = (gl_Position.z + gl_Position.z) - gl_Position.w;\n";
OS << "}\n";
}
void GLVertexDecompilerThread::Task()
{
m_shader = Decompile();