Re: [PATCH 2/4] wined3d: Implement vertex blending in glsl_vertex_pipe.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi, I like this patchset. I can't find anything wrong in it, but Matteo knows the fixed function pipeline replacement better than I do, so I'll wait for his comments. I tested it with various DirectX SDK samples (bend from dx7, SkinnedMesh and VertexBlend from dx8). It works very well, but I noticed a lighting oddity in SkinnedMesh. The lighting differs slightly between your code ("D3D non-indexed" option) and the test program's vertex shader or d3dx9's software vertex blending ("D3D indexed (VS)" / "Software Skinning"). This difference cannot be seen on Windows. I have attached screenshots. Cheers, Stefan -----BEGIN PGP SIGNATURE----- Version: GnuPG v2 iQIcBAEBAgAGBQJVlPPNAAoJEN0/YqbEcdMwUcIP/jwJsie6/93Ui3RQTqjT8oOP q0ips3E83dHpi0ceL8xXujr0pyO1x7cVuUX+hZ7Tf08bZu86FpieKZkTYZxRKbKj V5+x/bVlWGM+ogB06E7tHx2oM8tBFlh10wwh9M/QnitjtzCM+E3WKfE01W5NsRRL csUMxE1mBHvnp08Me2CZyzgtVWwQuTWH6IMUtmqQPfOmmOgzIoofjjRowjhZovFi D9cufKD/D+CBsuv/Py5AMYWfjv7gTZqB09sotaIzYWuyI/e6GRRFT2wzX1yOIl+I fhOr3AsHeFLWOAYd41DjMuqmqvQ5e/UigYxkDtssOCHVig7VtaQ0RKTEv6mMfDF4 6Ljgyi7RdGrgle41cDo6DBXTr0WZw6Y2eTLs0G+WCNkJ8/OMR9hLAt/YBg1mPEPo QUNTsqwNQ1Poxuxmo5fWqD0MhLAao7Wgm6dZTiECNPNNBsC91XRYR9jXF29xaB3h 3ekVCTdtI1+wUQolGw1/HZ7whaHkWGrxOY4qqiZsgYRYz1yl52Bt4sJDJYUHknbd GZ4JRkVQXD86D5txRLf500e03QCmDTXxjqEKSsNBNCtcpER1mp+3Hs/SImp/adRr 3tNuEMJduA/5jm27Nt+K13DwsazQXCJMDv5dTYlggBI9c3dhFDmjXNNGSvrvkBkf xW9FbSiHRRDoLR3UMuoR =Ghq/ -----END PGP SIGNATURE-----
Hi, On Thu, Jul 2, 2015 at 10:18 AM, Stefan Dösinger <stefandoesinger(a)gmail.com> wrote:
I tested it with various DirectX SDK samples (bend from dx7, SkinnedMesh and VertexBlend from dx8). It works very well, but I noticed a lighting oddity in SkinnedMesh. The lighting differs slightly between your code ("D3D non-indexed" option) and the test program's vertex shader or d3dx9's software vertex blending ("D3D indexed (VS)" / "Software Skinning"). This difference cannot be seen on Windows. I have attached screenshots.
Thanks for testing. It seems that normals are simply transformed by modelview matrices. I've fixed this in try 2. Cheers, Józef
participants (2)
-
Józef Kucia -
Stefan Dösinger