okseby/Quake-III-Arena
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Improve IQM CPU vertex skinning performance

Browse source 
Only calculate vertex blend matrix for each unique bone indexes/weights
combination once per-surface instead of recalculating for each vertex.
For best performance the model surfaces needs to use few vertex bone
indexes and weights combinations.

Unroll loops so GCC better optimizes them.

In my tests drawing animated IQM may take 50% as long in opengl1 and
70% as long in opengl2. It will vary by model though and might not
help much at all.

Made unanimated IQM models skip matrix math altogether.
This commit is contained in:
Zack Middleton 2018-07-27 17:40:21 -05:00
parent fdc08e860e
commit cccd283be8
4 changed files with 601 additions and 254 deletions
Download patch file Download diff file Expand all files Collapse all files

14
code/renderergl2/tr_local.h
View file

@ -950,21 +950,26 @@ typedef struct {
struct srfIQModel_s *surfaces;
int *triangles;
// vertex arrays
float *positions;
float *texcoords;
float *normals;
float *tangents;
byte *blendIndexes;
byte *colors;
int *influences; // [num_vertexes] indexes into influenceBlendVertexes
// unique list of vertex blend indexes/weights for faster CPU vertex skinning
byte *influenceBlendIndexes; // [num_influences]
union {
float *f;
byte *b;
} blendWeights;
byte *colors;
} influenceBlendWeights; // [num_influences]
// depending upon the exporter, blend indices and weights might be int/float
// as opposed to the recommended byte/byte, for example Noesis exports
// int/float whereas the official IQM tool exports byte/byte
byte blendWeightsType; // IQM_UBYTE or IQM_FLOAT
int blendWeightsType; // IQM_UBYTE or IQM_FLOAT
char *jointNames;
int *jointParents;
@ -981,6 +986,7 @@ typedef struct srfIQModel_s {
iqmData_t *data;
int first_vertex, num_vertexes;
int first_triangle, num_triangles;
int first_influence, num_influences;
} srfIQModel_t;
typedef struct srfVaoMdvMesh_s