SSE 4.1 Skinning

This appears to offer no benefit on my Wolfdale CPU, but I expect it will
be useful on more recent ones.

Sources/Plasma/PubUtilLib/plPipeline/plDXPipeline.cpp

@@ -10699,6 +10699,38 @@ static inline void ISkinVertexSSE3(const hsMatrix44& xfm, float wgt,
 #endif // HS_SSE3
 }
 
+#ifdef HS_SSE41
+static inline void ISkinDpSSE41(const float* src, float* dst, const __m128& mc0,
+                                const __m128& mc1, const __m128& mc2, const __m128& mwt)
+{
+    enum { DP_F4_X = 0xF1, DP_F4_Y = 0xF2, DP_F4_Z = 0xF4 };
+
+    __m128 msr = _mm_load_ps(src);
+    __m128 _r =        _mm_dp_ps(msr, mc0, DP_F4_X);
+    _r = _mm_or_ps(_r, _mm_dp_ps(msr, mc1, DP_F4_Y));
+    _r = _mm_or_ps(_r, _mm_dp_ps(msr, mc2, DP_F4_Z));
+
+    __m128 _dst = _mm_load_ps(dst);
+    _dst = _mm_add_ps(_dst, _mm_mul_ps(_r, mwt));
+    _mm_store_ps(dst, _dst);
+}
+#endif // HS_SSE41
+
+static inline void ISkinVertexSSE41(const hsMatrix44& xfm, float wgt,
+                                    const float* pt_src, float* pt_dst,
+                                    const float* vec_src, float* vec_dst)
+{
+#ifdef HS_SSE41
+    __m128 mc0 = _mm_load_ps(xfm.fMap[0]);
+    __m128 mc1 = _mm_load_ps(xfm.fMap[1]);
+    __m128 mc2 = _mm_load_ps(xfm.fMap[2]);
+    __m128 mwt = _mm_set_ps1(wgt);
+
+    ISkinDpSSE41(pt_src, pt_dst, mc0, mc1, mc2, mwt);
+    ISkinDpSSE41(vec_src, vec_dst, mc0, mc1, mc2, mwt);
+#endif // HS_SSE41
+}
+
 typedef void(*skin_vert_ptr)(const hsMatrix44&, float, const float*, float*, const float*, float*);
 
 template<skin_vert_ptr T>
@@ -10763,7 +10795,8 @@ static void IBlendVertBuffer(plSpan* span, hsMatrix44* matrixPalette, int numMat
 
 // CPU-optimized functions requiring dispatch
 hsFunctionDispatcher<plDXPipeline::blend_vert_buffer_ptr> plDXPipeline::blend_vert_buffer(
-    IBlendVertBuffer<ISkinVertexFPU>, 0, 0, IBlendVertBuffer<ISkinVertexSSE3>);
+    IBlendVertBuffer<ISkinVertexFPU>, 0, 0, IBlendVertBuffer<ISkinVertexSSE3>, 0,
+    IBlendVertBuffer<ISkinVertexSSE41>);
 
 // ISetPipeConsts //////////////////////////////////////////////////////////////////
 // A shader can request that the pipeline fill in certain constants that are indeterminate