Context model + mixing for orders 2 + 4 + 6 + sse for other stages

int N = fin.len;
const int CH_BITS = 25;

FSM fsm;
memset(&fsm, 0, sizeof(FSM));
memset(fsm.index, -1, sizeof(fsm.index));
init_fsm(fsm);

{
	byte* t2 = new byte[1 << 24];
	uint16* t4 = new uint16[1 << CH_BITS];
	//uint16* t6 = new uint16[1 << CH_BITS];

	memset(t2, 0, 1 << 24);
	memset(t4, 0, 2 << CH_BITS);
	//memset(t6, 0, 2 << CH_BITS);

	uint32 table_mix2[64], table_mix24[64][64], table_mix246[32][32][32];
	for (int i = 0; i < 64; i++) {
		table_mix2[i] = (1 << (19 - 6)) + (i << (20 - 6));

		for (int j = 0; j < 64; j++) {
			int v = j + 3 * i;
			v <<= 20 - 6 - 2;
			v += 1 << (19 - 6);

			table_mix24[j][i] = v << 8;
		}
	}

	for (int k = 0; k < 32; k++) {
		for (int j = 0; j < 32; j++) {
			for (int i = 0; i < 32; i++) {
				int v = k + 3 * j + 12 * i;
				v <<= 20 - 5 - 4;
				v += 1 << (19 - 6);

				table_mix246[k][j][i] = v << 8;
			}
		}
	}

	uint32 table_sse[64];
	for (int i = 0; i < 64; i++) {
		int v = i << (20 - 6);
		v += 1 << (19 - 6);

		table_sse[i] = (v << 8) + 23;
	}

	Common::Benchmark b("forward");

	Entropy::BitCoder bc(&fpacked);
	bc.Init();

	uint32 ctx = 0, ctx1 = 0;
	for (int pos = 0; pos < N; pos++) {
		uint32 h4 = hash4(ctx);
		uint32 h6 = hash4(ctx) ^ hash4(ctx1 & 0xFFFF);

		byte* o2 = t2 + ((ctx & 0xFFFF) << 8);
		uint16* o4 = t4 + ((h4 >> (32 - CH_BITS)) & ~0x100);
		//uint16* o6 = t6 + ((h6 >> (32 - CH_BITS)) & ~0x100);

		byte h4b = h4 >> 24;
		byte h6b = h6 >> 24;

		int y = fin.buf[pos];
		int p = 1;
		for (int i = 7; i >= 0; i--) {
			int b = (y >> i) & 1;
			auto& c2 = o2[p];
			auto& c4 = o4[p];
			//auto& c6 = o6[p];

			bool has_c4 = (o4[p] >> 8) == h4b ? true : (o4[p] = h4b << 8, false);
			//bool has_c6 = (o6[p] >> 8) == h6b ? true : (o6[p] = h6b << 8, false);

			auto& mix_c =
				//has_c6 ? table_mix246[fsm.pr5[c2]][fsm.pr5[byte(c4)]][fsm.pr5[byte(c6)]] :
				(has_c4 ?
					table_mix24[fsm.pr6[c2]][fsm.pr6[byte(c4)]] :
					table_mix2[fsm.pr6[c2]]);

			auto& sse = table_sse[mix_c >> 22];

			//uint16 pr = mix_c >> 16;
			uint16 pr = sse >> 16;
			pr += pr == 0;

			bc.Encode12P(pr, b);

			c2 = fsm.next[b][c2];
			c4 = fsm.next[b][byte(c4)] + (h4b << 8);
			//c6 = fsm.next[b][byte(c6)] + (h6b << 8);

			int mix_upd = (b << 19) - (mix_c >> 9);
			mix_upd >>= 8;
			mix_c += (mix_upd << 8) + (byte(mix_c) < 64);

			int sse_upd = (b << 19) - (sse >> 9);
			sse_upd >>= byte(sse) >> 3;
			sse += (sse_upd << 8) + (byte(sse) < 64);

			p += p + b;
		}

		ctx1 = (ctx1 << 8) + (ctx >> 24);
		ctx = (ctx << 8) + byte(p);
	}

	bc.Flush();
	delete[] t2;
	delete[] t4;
	//delete[] t6;
}

printf("%d -> %d (%.2f KB)\n", N, fpacked.Position(), fpacked.Position() / 1024.0);
fpacked.Rewind();

{
	byte* t2 = new byte[1 << 24];
	uint16* t4 = new uint16[1 << CH_BITS];
	//uint16* t6 = new uint16[1 << CH_BITS];

	memset(t2, 0, 1 << 24);
	memset(t4, 0, 2 << CH_BITS);
	//memset(t6, 0, 2 << CH_BITS);

	uint32 table_mix2[64], table_mix24[64][64], table_mix246[32][32][32];
	for (int i = 0; i < 64; i++) {
		table_mix2[i] = (1 << (19 - 6)) + (i << (20 - 6));

		for (int j = 0; j < 64; j++) {
			int v = j + 3 * i;
			v <<= 20 - 6 - 2;
			v += 1 << (19 - 6);

			table_mix24[j][i] = v << 8;
		}
	}

	for (int k = 0; k < 32; k++) {
		for (int j = 0; j < 32; j++) {
			for (int i = 0; i < 32; i++) {
				int v = k + 3 * j + 12 * i;
				v <<= 20 - 5 - 4;
				v += 1 << (19 - 6);

				table_mix246[k][j][i] = v << 8;
			}
		}
	}

	uint32 table_sse[64];
	for (int i = 0; i < 64; i++) {
		int v = i << (20 - 6);
		v += 1 << (19 - 6);

		table_sse[i] = (v << 8) + 23;
	}

	Common::Benchmark b("reverse");

	Entropy::BitDecoder bc(&fpacked);
	bc.Init();

	uint32 ctx = 0, ctx1 = 0;
	for (int pos = 0; pos < N; pos++) {
		uint32 h4 = hash4(ctx);
		uint32 h6 = hash4(ctx) ^ hash4(ctx1 & 0xFFFF);

		byte* o2 = t2 + ((ctx & 0xFFFF) << 8);
		uint16* o4 = t4 + ((h4 >> (32 - CH_BITS)) & ~0x100);
		//uint16* o6 = t6 + ((h6 >> (32 - CH_BITS)) & ~0x100);

		byte h4b = h4 >> 24;
		byte h6b = h6 >> 24;

		int y = fin.buf[pos];
		int p = 1;
		for (int i = 7; i >= 0; i--) {
			auto& c2 = o2[p];
			auto& c4 = o4[p];
			//auto& c6 = o6[p];

			bool has_c4 = (o4[p] >> 8) == h4b ? true : (o4[p] = h4b << 8, false);
			//bool has_c6 = (o6[p] >> 8) == h6b ? true : (o6[p] = h6b << 8, false);

			auto& mix_c =
				//has_c6 ? table_mix246[fsm.pr5[c2]][fsm.pr5[byte(c4)]][fsm.pr5[byte(c6)]] :
				(has_c4 ?
					table_mix24[fsm.pr6[c2]][fsm.pr6[byte(c4)]] :
					table_mix2[fsm.pr6[c2]]);

			auto& sse = table_sse[mix_c >> 22];

			//uint16 pr = mix_c >> 16;
			uint16 pr = sse >> 16;
			pr += pr == 0;

			int b = bc.Decode12P(pr);

			c2 = fsm.next[b][c2];
			c4 = fsm.next[b][byte(c4)] + (h4b << 8);
			//c6 = fsm.next[b][byte(c6)] + (h6b << 8);

			int mix_upd = (b << 19) - (mix_c >> 9);
			mix_upd >>= 8;
			mix_c += (mix_upd << 8) + (byte(mix_c) < 64);

			int sse_upd = (b << 19) - (sse >> 9);
			sse_upd >>= byte(sse) >> 3;
			sse += (sse_upd << 8) + (byte(sse) < 64);

			p += p + b;
		}

		ctx1 = (ctx1 << 8) + (ctx >> 24);
		ctx = (ctx << 8) + byte(p);
		funpacked.Put(p - 0x100);
	}

	delete[] t2;
	delete[] t4;
	//delete[] t6;
}

ASSERT(!memcmp(fin.buf, funpacked.buf, N));

delete[] fin.buf;
delete[] fpacked.buf;
delete[] funpacked.buf;