git-svn-id: https://casparcg.svn.sourceforge.net/svnroot/casparcg/server/branches...

[casparcg] / accelerator / cpu / image / image_mixer.cpp

diff --git a/accelerator/cpu/image/image_mixer.cpp b/accelerator/cpu/image/image_mixer.cpp
index 21ddadada0a8eac3947bf9059537032cfa5288df..054f7f59dcded928ac448bdb44a98c6d96661739 100644 (file)
--- a/accelerator/cpu/image/image_mixer.cpp
+++ b/accelerator/cpu/image/image_mixer.cpp
@@ -24,7 +24,7 @@
 #include "image_mixer.h"\r
 \r
 #include "../util/write_frame.h"\r
-#include "../util/simd.h"\r
+#include "../util/xmm.h"\r
 \r
 #include <common/assert.h>\r
 #include <common/gl/gl_check.h>\r
@@ -94,50 +94,54 @@ bool operator!=(const item& lhs, const item& rhs)
        return !(lhs == rhs);\r
 }\r
        \r
-inline xmm_epi8 blend(xmm_epi8 dest, xmm_epi8 source)\r
+inline xmm::s8_x blend(xmm::s8_x dest, xmm::s8_x source)\r
 {      \r
-       auto s = xmm_cast<xmm_epi16>(source);\r
+       using namespace xmm;\r
+\r
+       auto s = s16_x(source);\r
        auto d = dest;\r
 \r
-       const xmm_epi16 round   = 128;\r
-       const xmm_epi16 lomask  = 0x00FF;\r
+       const s16_x round       = 128;\r
+       const s16_x lomask      = 0x00FF;\r
 \r
        // T(S, D) = S * D[A] + 0x80\r
-       auto aaaa   = xmm_epi8::shuffle(d, xmm_epi8(15, 15, 15, 15, 11, 11, 11, 11, 7, 7, 7, 7, 3, 3, 3, 3));\r
-       d                       = xmm_epi8::umin(d, aaaa); // overflow guard\r
+       auto aaaa   = s8_x::shuffle(d, s8_x(15, 15, 15, 15, 11, 11, 11, 11, 7, 7, 7, 7, 3, 3, 3, 3));\r
+       d                       = s8_x(u8_x::min(u8_x(d), u8_x(aaaa))); // overflow guard\r
 \r
-       auto xaxa       = xmm_cast<xmm_epi16>(aaaa) & lomask;           \r
+       auto xaxa       = s16_x(aaaa) & lomask;         \r
                              \r
        auto xrxb       = s & lomask;\r
-       auto t1         = xmm_epi16::multiply_low(xrxb, xaxa) + round;    \r
+       auto t1         = s16_x::multiply_low(xrxb, xaxa) + round;    \r
                        \r
        auto xaxg       = s >> 8;\r
-       auto t2         = xmm_epi16::multiply_low(xaxg, xaxa) + round;\r
+       auto t2         = s16_x::multiply_low(xaxg, xaxa) + round;\r
                \r
        // C(S, D) = S + D - (((T >> 8) + T) >> 8);\r
-       auto rxbx       = xmm_cast<xmm_epi8>(((t1 >> 8) + t1) >> 8);      \r
-       auto axgx       = xmm_cast<xmm_epi8>((t2 >> 8) + t2);    \r
-       auto argb   = xmm_epi8::blend(rxbx, axgx, xmm_epi8(-1, 0, -1, 0));\r
+       auto rxbx       = s8_x(((t1 >> 8) + t1) >> 8);      \r
+       auto axgx       = s8_x((t2 >> 8) + t2);    \r
+       auto argb   = s8_x::blend(rxbx, axgx, s8_x(-1, 0, -1, 0));\r
 \r
-       return xmm_cast<xmm_epi8>(s) + (d - argb);\r
+       return s8_x(s) + (d - argb);\r
 }\r
        \r
-template<typename write_op>\r
+template<typename write_tag>\r
 static void kernel(uint8_t* dest, const uint8_t* source, size_t count, const core::frame_transform& transform)\r
-{                              \r
+{                      \r
+       using namespace xmm;\r
+\r
        for(auto n = 0; n < count; n += 32)    \r
        {\r
-               auto s0 = xmm_epi8::load(dest+n+0);\r
-               auto s1 = xmm_epi8::load(dest+n+16);\r
+               auto s0 = s8_x::load(dest+n+0);\r
+               auto s1 = s8_x::load(dest+n+16);\r
 \r
-               auto d0 = xmm_epi8::load(source+n+0);\r
-               auto d1 = xmm_epi8::load(source+n+16);\r
+               auto d0 = s8_x::load(source+n+0);\r
+               auto d1 = s8_x::load(source+n+16);\r
                \r
                auto argb0 = blend(d0, s0);\r
                auto argb1 = blend(d1, s1);\r
 \r
-               xmm_epi8::write<write_op>(argb0, dest+n+0);\r
-               xmm_epi8::write<write_op>(argb1, dest+n+16);\r
+               s8_x::write(argb0, dest+n+0 , write_tag());\r
+               s8_x::write(argb1, dest+n+16, write_tag());\r
        } \r
 }\r
 \r
@@ -208,9 +212,9 @@ private:
 \r
                                auto it = items.begin();\r
                                for(; it != items.end()-1; ++it)                        \r
-                                       kernel<store_write>(dest + y*width*4, it->buffers.at(0)->data() + y*width*4, width*4, it->transform);\r
+                                       kernel<xmm::store_tag>(dest + y*width*4, it->buffers.at(0)->data() + y*width*4, width*4, it->transform);\r
 \r
-                               kernel<stream_write>(dest + y*width*4, it->buffers.at(0)->data() + y*width*4, width*4, it->transform);\r
+                               kernel<xmm::stream_tag>(dest + y*width*4, it->buffers.at(0)->data() + y*width*4, width*4, it->transform);\r
                        }\r
                });\r
        }\r