diff --git a/crap/mugi4.hpp b/crap/mugi4.hpp
index 94a256d..30ea8bc 100644
--- a/crap/mugi4.hpp
+++ b/crap/mugi4.hpp
@@ -18,7 +18,7 @@ https://aaltodoc.aalto.fi/bitstream/handle/123456789/14420/article6.pdf
 #include "Param.hpp"
 #include "Crap.hpp"
 #include "os2piir.hpp"
-#include "BufferOS2.hpp"
+#include "Buffer2OS2.hpp"
 
 #define VT 0.026
 #define N 4
@@ -123,7 +123,7 @@ struct mugi4 {
 };
 
 struct Crap_mugi4
-:public AdjustAll<BufferOS2<Crap>> {
+:public AdjustAll<Buffer2OS2<Crap>> {
 	static constexpr ulong id = 0xD8D0D8D0;
 	static constexpr char label[] = "crap_mugi4";
 	static constexpr char name[] = "crap mugi4 (moog-like)";
@@ -140,10 +140,11 @@ struct Crap_mugi4
 		memset(&filter, 0, sizeof(mugi4));
 	}
 
-	virtual inline v2df
-	process2(v2df in)
+	inline void
+	process2(v2df *buf, ulong rem)
 	{
-		return filter.process(in);
+		for (ulong i = 0; i < rem; i++)
+			buf[i] = filter.process(buf[i]);
 	}
 
 	static inline void
diff --git a/include/Buffer2OS2.hpp b/include/Buffer2OS2.hpp
new file mode 100644
index 0000000..b81d299
--- /dev/null
+++ b/include/Buffer2OS2.hpp
@@ -0,0 +1,68 @@
+template<class Mixin>
+struct Buffer2OS2 : public virtual Mixin {
+	virtual inline void
+	process2(v2df *buf, ulong rem) = 0;
+
+	halfband_t<v2df> hb_up, hb_down;
+
+	TEMPLATE void
+	_process(T *in_L, T *in_R, T *out_L, T *out_R, ulong count)
+	{
+		disable_denormals();
+		v2df buf[BLOCK_SIZE];
+		v2df over[FULL_SIZE];
+
+		for (ulong pos = 0; pos < count; pos += BLOCK_SIZE) {
+			ulong rem = BLOCK_SIZE;
+			if (pos + BLOCK_SIZE > count)
+				rem = count - pos;
+
+			ulong rem2 = rem*OVERSAMPLING;
+
+			for (ulong i = 0; i < rem; i++) {
+				buf[i][0] = in_L[i];
+				buf[i][1] = in_R[i];
+			}
+
+			for (ulong i = 0; i < rem; i++) {
+				over[i*2+0] = interpolate_a(&hb_up, buf[i]);
+				over[i*2+1] = interpolate_b(&hb_up, buf[i]);
+			}
+
+			process2(over, rem2);
+
+			for (ulong i = 0; i < rem; i++) {
+					 decimate_a(&hb_down, over[i*2+0]);
+				buf[i] = decimate_b(&hb_down, over[i*2+1]);
+			}
+
+			for (ulong i = 0; i < rem; i++) {
+				out_L[i] = buf[i][0];
+				out_R[i] = buf[i][1];
+			}
+
+			in_L += BLOCK_SIZE;
+			in_R += BLOCK_SIZE;
+			out_L += BLOCK_SIZE;
+			out_R += BLOCK_SIZE;
+		}
+	}
+
+	void
+	process(
+	    double *in_L, double *in_R,
+	    double *out_L, double *out_R,
+	    ulong count)
+	{
+		_process(in_L, in_R, out_L, out_R, count);
+	}
+
+	void
+	process(
+	    float *in_L, float *in_R,
+	    float *out_L, float *out_R,
+	    ulong count)
+	{
+		_process(in_L, in_R, out_L, out_R, count);
+	}
+};