Use buffer pool on avs+ MT.

also, disable AVX2 always on Avisynth2.6.

README.txt

@@ -27,7 +27,7 @@ GENERAL INFO:
 
    Syntax=>
 
-	TMM2(clip, int mode, int order, int field, int length, int mtype,
+    TMM2(clip, int mode, int order, int field, int length, int mtype,
            int ttype, int mtqL, int mthL, int mtqC, int mthC, int nt,
            int minthresh, int maxthresh, int cstr, int opt)
 
@@ -173,20 +173,35 @@ PARAMETERS:
 
          Controls which cpu optimizations are used for create motion masks.
 
-		 0 - Use C++ routine.
-		 1 - Use SSE2 routine if possible. When SSE2 can't be used, fallback to 0.
-		 others - Use AVX2 routine if possible. When AVX2 can't be used, fallback to 1.
-
-		 Default: -1  (int)
+         0 - Use C++ routine.
+         1 - Use SSE2 routine if possible. When SSE2 can't be used, fallback to 0.
+         others - Use AVX2 routine if possible. When AVX2 can't be used, fallback to 1.
 
+         Default: -1  (int)
 
 
+NOTE:
+    - TMM2_avx2.dll is compiled with /arch:AVX2.
+
+    - On Avisynth2.6, AVX2 is always disabled even if you use TMM2_avx2.dll.
+
+    - On Avisynth+ MT, TMM2 is set as MT_NICE_FILTER automatically.
+      You don't have to set SetFilterMTMode() yourself for this filter.
+
+    - This filter requires appropriate memory alignments.
+      Thus, if you want to crop the left side of your source clip before this filter,
+      you have to set crop(align=true).
+
 
 CHANGE LIST:
 
 
-   v0.0 - (2015-05-20)
+   v0.1 - (2016-05-25)
+      + Use buffer pool on avs+ MT.
+      + Disable AVX2 always on Avisynth2.6.
+
 
+   v0.0 - (2016-05-20)
       + initial release
 
 

src/CreateMM.cpp

@@ -171,8 +171,9 @@ combine_masks_simd(uint8_t* dstp, const uint8_t* sqp, const uint8_t* shp,
 }
 
 
-CreateMM::CreateMM(PClip mm1, PClip mm2, int _cstr, arch_t arch) :
-    GVFmod(mm1, arch), mmask2(mm2), cstr(_cstr), simd(arch != NO_SIMD)
+CreateMM::CreateMM(PClip mm1, PClip mm2, int _cstr, arch_t arch, bool ip) :
+    GVFmod(mm1, arch), mmask2(mm2), cstr(_cstr), simd(arch != NO_SIMD),
+    isPlus(ip)
 {
     vi.height /= 2;
 
@@ -197,21 +198,33 @@ CreateMM::CreateMM(PClip mm1, PClip mm2, int _cstr, arch_t arch) :
 
 
 struct AndBuff {
+    ise_t* env;
+    bool isPlus;
     void* orig;
     uint8_t* am0;
     uint8_t* am1;
     const int pitch;
 
-    AndBuff(int width, int height, size_t align) :
-        pitch((width + 2 + align - 1) & ~(align - 1))
+    AndBuff(int width, int height, size_t align, bool is_plus, ise_t* e) :
+        pitch((width + 2 + align - 1) & ~(align - 1)), env(e), isPlus(is_plus)
     {
-        orig = _mm_malloc(pitch * (height * 2 + 1), align);
+        size_t size = pitch * (height * 2 + 1);
+        if (isPlus) {
+            orig = static_cast<IScriptEnvironment2*>(
+                env)->Allocate(size, align, AVS_POOLED_ALLOC);
+        } else {
+            orig = _mm_malloc(size, align);
+        }
         am0 = reinterpret_cast<uint8_t*>(orig) + pitch;
         am1 = am0 + pitch * height;
     }
     ~AndBuff()
     {
-        _mm_free(orig);
+        if (isPlus) {
+            static_cast<IScriptEnvironment2*>(env)->Free(orig);
+        } else {
+            _mm_free(orig);
+        }
         orig = nullptr;
     }
 };
@@ -224,7 +237,7 @@ PVideoFrame __stdcall CreateMM::GetFrame(int n, ise_t* env)
     auto src0 = child->GetFrame(n, env);
     auto dst = env->NewVideoFrame(vi, align);
 
-    auto buff = AndBuff(vi.width, vi.height, align);
+    auto buff = AndBuff(vi.width, vi.height, align, isPlus, env);
 
     if (!buff.orig) {
         env->ThrowError("TMM: failed to allocate AndBuff.");

src/TMM2.h

@@ -35,6 +35,9 @@ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 #include <avisynth.h>
 
 
+#define TMM2_VERSION "0.1"
+
+
 
 typedef IScriptEnvironment ise_t;
 
@@ -104,6 +107,7 @@ class CreateMM : public GVFmod {
     PClip mmask2;
     const int cstr;
     const bool simd;
+    const bool isPlus;
 
     void(__stdcall *and_masks)(
         uint8_t* dstp0, uint8_t* dstp1, const uint8_t* srcp0,
@@ -117,7 +121,7 @@ class CreateMM : public GVFmod {
         const int cstr);
 
 public:
-    CreateMM(PClip mm1, PClip mm2, int cstr, arch_t arch);
+    CreateMM(PClip mm1, PClip mm2, int cstr, arch_t arch, bool is_avsplus);
     ~CreateMM() {}
     PVideoFrame __stdcall GetFrame(int n, ise_t* env);
 };

src/plugin.cpp

@@ -28,23 +28,23 @@ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 extern bool has_sse2();
 extern bool has_avx2();
 
-static inline arch_t get_arch(int opt)
+static inline arch_t get_arch(int opt, bool is_avsplus)
 {
     if (opt == 0 || !has_sse2()) {
         return NO_SIMD;
     }
 #if !defined(__AVX2__)
     return USE_SSE2;
 #else
-    if (opt == 1 || !has_avx2()) {
+    if (opt == 1 || !has_avx2() || !is_avsplus) {
         return USE_SSE2;
     }
     return USE_AVX2;
 #endif
 }
 
 
-static AVSValue __cdecl create_tmm(AVSValue args, void*, ise_t* env)
+static AVSValue __cdecl create_tmm(AVSValue args, void* user_data, ise_t* env)
 {
     try {
         PClip orig = args[0].AsClip();
@@ -93,7 +93,8 @@ static AVSValue __cdecl create_tmm(AVSValue args, void*, ise_t* env)
         int minth = clamp(args[12].AsInt(4), 0, 255);
         int maxth = clamp(args[13].AsInt(75), 0, 255);
         int cstr = clamp(args[14].AsInt(4), 0, 8);
-        arch_t arch = get_arch(args[15].AsInt(-1));
+        bool is_avsplus = user_data != nullptr;
+        arch_t arch = get_arch(args[15].AsInt(-1), is_avsplus);
 
         orig = env->Invoke("SeparateFields", orig).AsClip();
         const char* filter[] = { "SelectEven", "SelectOdd" };
@@ -116,10 +117,10 @@ static AVSValue __cdecl create_tmm(AVSValue args, void*, ise_t* env)
         btmf0 = env->Invoke("InternalCache", btmf0).AsClip();
         PClip btmf1 = new MotionMask(btmf, minth, maxth, nt, 2, arch);
 
-        topf = new CreateMM(topf0, topf1, cstr, arch);
+        topf = new CreateMM(topf0, topf1, cstr, arch, is_avsplus);
         topf = env->Invoke("InternalCache", topf).AsClip();
 
-        btmf = new CreateMM(btmf0, btmf1, cstr, arch);
+        btmf = new CreateMM(btmf0, btmf1, cstr, arch, is_avsplus);
         btmf = env->Invoke("InternalCache", btmf).AsClip();
 
         return new BuildMM(topf, btmf, mode, order, field, length, mtype, arch, env);
@@ -138,6 +139,9 @@ extern "C" __declspec(dllexport) const char* __stdcall
 AvisynthPluginInit3(ise_t* env, const AVS_Linkage* const vectors)
 {
     AVS_linkage = vectors;
+
+    void* is_avsplus = env->FunctionExists("SetFilterMTMode") ? "true" : nullptr;
+
     const char* args =
         "c"             // 0
         "[mode]i"       // 1
@@ -156,12 +160,12 @@ AvisynthPluginInit3(ise_t* env, const AVS_Linkage* const vectors)
         "[cstr]i"       //14
         "[opt]i";       //15
 
-    env->AddFunction("TMM2", args, create_tmm, nullptr);
+    env->AddFunction("TMM2", args, create_tmm, is_avsplus);
 
-    if (env->FunctionExists("SetFilterMTMode")) {
+    if (is_avsplus != nullptr) {
         static_cast<IScriptEnvironment2*>(
             env)->SetFilterMTMode("TMM2", MT_NICE_FILTER, true);
     }
 
-    return "TMM for avs2.6/avs+";
+    return "TMM for avs2.6/avs+ ver. " TMM2_VERSION;
 }

src/proc_thmask.h

@@ -139,7 +139,7 @@ proc_c(uint8_t* dqp, const uint8_t* srcp, const int pitch,
         }
         s0 = s1;
         s1 = s2;
-        s2 += (y < height - 1) ? pitch : -pitch;
+        s2 += (y < height - 2) ? pitch : -pitch;
         dqp += pitch;
         dhp += pitch;
     }
@@ -257,7 +257,7 @@ proc_simd(uint8_t* dqp, const uint8_t* srcp, const int pitch,
         }
         s0 = s1;
         s1 = s2;
-        s2 += (y < height - 1) ? pitch : -pitch;
+        s2 += (y < height - 2) ? pitch : -pitch;
         dqp += pitch;
         dhp += pitch;
     }

src/simd.h

@@ -275,7 +275,7 @@ static F_INLINE V cmpge(const V& x, const V& y)
 template <typename V>
 static F_INLINE V cmpneq(const V& x, const V& y)
 {
-    return or_reg(cmpeq(x, y), cmpeq(x, x));
+    return xor_reg(cmpeq(x, y), cmpeq(x, x));
 }
 
 template <typename V>