2.1.0: -cpu/image_mixer: Added support for interlacing.

[casparcg] / accelerator / cpu / util / blend.h

diff --git a/accelerator/cpu/util/blend.h b/accelerator/cpu/util/blend.h
index 3e749ac2c646f58de6ec736a521918fafe88c8df..d5855e4643664346ad0a5030a37e4f22ba7b3eca 100644 (file)
--- a/accelerator/cpu/util/blend.h
+++ b/accelerator/cpu/util/blend.h
@@ -8,8 +8,6 @@
 \r
 #include "simd.h"\r
 \r
-#include <tbb/parallel_for.h>\r
-\r
 #include <intrin.h>\r
 #include <stdint.h>\r
 \r
@@ -35,31 +33,28 @@ static void blend(uint8_t* dest, const uint8_t* source1, const uint8_t* source2,
     const xmm_epi16 round  = 128;\r
     const xmm_epi16 lomask = 0x00FF;\r
                \r
-       tbb::parallel_for(tbb::blocked_range<size_t>(0, count/sizeof(xmm_epi8)), [&](const tbb::blocked_range<size_t>& r)\r
+       for(auto n = 0; n < count; n += 16)    \r
        {\r
-               for(auto n = r.begin(); n != r.end(); ++n)    \r
-               {\r
-                       auto s = xmm_epi16::load(source1+n*16);\r
-                       auto d = xmm_epi16::load(source2+n*16);\r
+               auto s = xmm_epi16::load(source1+n);\r
+               auto d = xmm_epi16::load(source2+n);\r
 \r
-                       // T(S, D) = S * D[A] + 0x80\r
-                       auto xxxa       = xmm_cast<xmm_epi32>(d) >> 24;\r
-                       auto xaxa       = xmm_cast<xmm_epi16>((xxxa << 16) | xxxa);\r
+               // T(S, D) = S * D[A] + 0x80\r
+               auto xxxa       = xmm_cast<xmm_epi32>(d) >> 24;\r
+               auto xaxa       = xmm_cast<xmm_epi16>((xxxa << 16) | xxxa);\r
                              \r
-                       auto xbxr       = s & lomask;\r
-                       auto t1         = xmm_epi16::multiply_low(xbxr, xaxa) + round;    \r
+               auto xbxr       = s & lomask;\r
+               auto t1         = xmm_epi16::multiply_low(xbxr, xaxa) + round;    \r
                        \r
-                       auto xaxg       = s >> 8;\r
-                       auto t2         = xmm_epi16::multiply_low(xaxg, xaxa) + round;\r
+               auto xaxg       = s >> 8;\r
+               auto t2         = xmm_epi16::multiply_low(xaxg, xaxa) + round;\r
                \r
-                       // C(S, D) = S + D - (((T >> 8) + T) >> 8);\r
-                       auto bxrx       = (t1 >> 8) + t1;      \r
-                       auto axgx       = (t2 >> 8) + t2;    \r
-                       auto bgra       = xmm_cast<xmm_epi8>((bxrx >> 8) | xmm_epi16::and_not(axgx, lomask));\r
+               // C(S, D) = S + D - (((T >> 8) + T) >> 8);\r
+               auto bxrx       = (t1 >> 8) + t1;      \r
+               auto axgx       = (t2 >> 8) + t2;    \r
+               auto bgra       = xmm_cast<xmm_epi8>((bxrx >> 8) | xmm_epi16::and_not(axgx, lomask));\r
                                \r
-                       xmm_epi8::stream(xmm_cast<xmm_epi8>(s) + (xmm_cast<xmm_epi8>(d) - bgra), dest + n*16);\r
-               }     \r
-       });\r
+               xmm_epi8::stream(xmm_cast<xmm_epi8>(s) + (xmm_cast<xmm_epi8>(d) - bgra), dest+n);\r
+       } \r
 }\r
 \r
 }}}
\ No newline at end of file