piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-21 13:20:27 -06:00
Code Issues Packages Projects Releases Wiki Activity

Vertex Loader: SSE implementations of more position/texcoord/normal formats

Browse Source 
~35-45% faster NFS:HP2, possibly other vertex-bound games.
This commit is contained in:
Fiora
2014-11-11 01:48:38 -08:00
parent da962a3d2b
commit 3ddf82a318
7 changed files with 299 additions and 106 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
Source/Core/VideoBackends/Software/SWVertexLoader.cpp
View File

@ -42,15 +42,15 @@ void SWVertexLoader::SetFormat(u8 attributeIndex, u8 primitiveType)
{
m_CurrentVat = &g_main_cp_state.vtx_attr[attributeIndex];
posScale = 1.0f / float(1 << m_CurrentVat->g0.PosFrac);
tcScale[0] = 1.0f / float(1 << m_CurrentVat->g0.Tex0Frac);
tcScale[1] = 1.0f / float(1 << m_CurrentVat->g1.Tex1Frac);
tcScale[2] = 1.0f / float(1 << m_CurrentVat->g1.Tex2Frac);
tcScale[3] = 1.0f / float(1 << m_CurrentVat->g1.Tex3Frac);
tcScale[4] = 1.0f / float(1 << m_CurrentVat->g2.Tex4Frac);
tcScale[5] = 1.0f / float(1 << m_CurrentVat->g2.Tex5Frac);
tcScale[6] = 1.0f / float(1 << m_CurrentVat->g2.Tex6Frac);
tcScale[7] = 1.0f / float(1 << m_CurrentVat->g2.Tex7Frac);
posScale[0] = posScale[1] = posScale[2] = posScale[3] = 1.0f / float(1 << m_CurrentVat->g0.PosFrac);
tcScale[0][0] = tcScale[0][1] = 1.0f / float(1 << m_CurrentVat->g0.Tex0Frac);
tcScale[1][0] = tcScale[1][1] = 1.0f / float(1 << m_CurrentVat->g1.Tex1Frac);
tcScale[2][0] = tcScale[2][1] = 1.0f / float(1 << m_CurrentVat->g1.Tex2Frac);
tcScale[3][0] = tcScale[3][1] = 1.0f / float(1 << m_CurrentVat->g1.Tex3Frac);
tcScale[4][0] = tcScale[4][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex4Frac);
tcScale[5][0] = tcScale[5][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex5Frac);
tcScale[6][0] = tcScale[6][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex6Frac);
tcScale[7][0] = tcScale[7][1] = 1.0f / float(1 << m_CurrentVat->g2.Tex7Frac);
//TexMtx
const u64 tmDesc[8] = {