TorqueGameEngines/Torque3D
1
0
Fork 0
mirror of https://github.com/TorqueGameEngines/Torque3D.git synced 2026-07-11 22:54:34 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Hardware Skinning Support

Browse source 
- Supports GL, D3D9 & D3D11
- Extends vertex formats & shadergen to support blend indices and weights
- Adds basic support for using 4x3 matrices for shader constants
- Supports software fallback
This commit is contained in:
James Urquhart 2015-01-10 19:41:25 +00:00
parent 507c239a87
commit 3496c549b5
72 changed files with 2533 additions and 1327 deletions
Download patch file Download diff file Expand all files Collapse all files

80
Engine/source/ts/arch/tsMeshIntrinsics.sse4.cpp
View file

@ -25,84 +25,4 @@
#include "ts/tsMeshIntrinsics.h"
#include <smmintrin.h>
void m_matF_x_BatchedVertWeightList_SSE4(const MatrixF &mat,
const dsize_t count,
const TSSkinMesh::BatchData::BatchedVertWeight * __restrict batch,
U8 * const __restrict outPtr,
const dsize_t outStride)
{
const char * __restrict iPtr = reinterpret_cast<const char *>(batch);
const dsize_t inStride = sizeof(TSSkinMesh::BatchData::BatchedVertWeight);
__m128 sseMat[3];
sseMat[0] = _mm_loadu_ps(&mat[0]);
sseMat[1] = _mm_loadu_ps(&mat[4]);
sseMat[2] = _mm_loadu_ps(&mat[8]);
// temp registers
__m128 inPos, tempPos;
__m128 inNrm, tempNrm;
__m128 temp0, temp1, temp2, temp3;
// pre-populate cache
const TSSkinMesh::BatchData::BatchedVertWeight &firstElem = batch[0];
for(S32 i = 0; i < 8; i++)
{
_mm_prefetch(reinterpret_cast<const char *>(iPtr + inStride * i), _MM_HINT_T0);
_mm_prefetch(reinterpret_cast<const char *>(outPtr + outStride * (i + firstElem.vidx)), _MM_HINT_T0);
}
for(S32 i = 0; i < count; i++)
{
const TSSkinMesh::BatchData::BatchedVertWeight &inElem = batch[i];
TSMesh::__TSMeshVertexBase *outElem = reinterpret_cast<TSMesh::__TSMeshVertexBase *>(outPtr + inElem.vidx * outStride);
// process x (hiding the prefetches in the delays)
inPos = _mm_load_ps(inElem.vert);
inNrm = _mm_load_ps(inElem.normal);
// prefetch input
#define INPUT_PREFETCH_LOOKAHEAD 64
const char *prefetchInput = reinterpret_cast<const char *>(batch) + inStride * (i + INPUT_PREFETCH_LOOKAHEAD);
_mm_prefetch(prefetchInput, _MM_HINT_T0);
// prefetch ouput with half the lookahead distance of the input
#define OUTPUT_PREFETCH_LOOKAHEAD (INPUT_PREFETCH_LOOKAHEAD >> 1)
const char *outPrefetch = reinterpret_cast<const char*>(outPtr) + outStride * (inElem.vidx + OUTPUT_PREFETCH_LOOKAHEAD);
_mm_prefetch(outPrefetch, _MM_HINT_T0);
// Multiply position
tempPos = _mm_dp_ps(inPos, sseMat[0], 0xF1);
temp0 = _mm_dp_ps(inPos, sseMat[1], 0xF2);
temp1 = _mm_dp_ps(inPos, sseMat[2], 0xF4);
temp0 = _mm_or_ps(temp0, temp1);
tempPos = _mm_or_ps(tempPos, temp0);
// Multiply normal
tempNrm = _mm_dp_ps(inNrm, sseMat[0], 0x71);
temp2 = _mm_dp_ps(inNrm, sseMat[1], 0x72);
temp3 = _mm_dp_ps(inNrm, sseMat[2], 0x74);
temp2 = _mm_or_ps(temp2, temp3);
tempNrm = _mm_or_ps(tempNrm, temp2);
// Load bone weight and multiply
temp3 = _mm_shuffle_ps(inPos, inPos, _MM_SHUFFLE(3, 3, 3, 3));
tempPos = _mm_mul_ps(tempPos, temp3);
tempNrm = _mm_mul_ps(tempNrm, temp3);
inPos = _mm_load_ps(outElem->_vert); //< load position for accumulation
inNrm = _mm_load_ps(outElem->_normal); //< load normal for accumulation
// accumulate with previous values
tempNrm = _mm_add_ps(tempNrm, inNrm);
tempPos = _mm_add_ps(tempPos, inPos);
_mm_store_ps(outElem->_vert, tempPos); //< output position
_mm_store_ps(outElem->_normal, tempNrm); //< output normal
}
}
#endif // TORQUE_CPU_X86