diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index 575892f..b7c5964 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -7,8 +7,15 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v2
-    - run: make test_ci DEBUG=1 -j3
+    - run: lscpu
+    - run: wget https://software.intel.com/content/dam/develop/external/us/en/documents/downloads/sde-external-8.63.0-2021-01-18-lin.tar.bz2
+    - run: bzip2 -dc sde-external-8.63.0-2021-01-18-lin.tar.bz2 | tar xvf -
+    - run: make bin/bn_test.exe DEBUG=1 -j4
+    - run: sde-external-8.63.0-2021-01-18-lin/sde64 -hsw -- bin/bn_test.exe
+    - run: make test_ci DEBUG=1 -j4 || dmesg | tail
     - run: make clean
-    - run: make test_ci DEBUG=1 -j3 CXX=clang++
+    - run: make test_ci DEBUG=1 -j4 CXX=clang++ || dmesg | tail
     - run: make clean
     - run: make test_go
+    - run: sudo apt install openjdk-8-jdk
+    - run: make -C ffi/java test JAVA_INC=-I/usr/lib/jvm/java-8-openjdk-amd64/include
diff --git a/include/mcl/gmp_util.hpp b/include/mcl/gmp_util.hpp
index c5e9700..f3fcfa3 100644
--- a/include/mcl/gmp_util.hpp
+++ b/include/mcl/gmp_util.hpp
@@ -951,7 +951,7 @@ struct SmallModp {
 	static const size_t maxTblSize = (MCL_MAX_BIT_SIZE + unitBitSize - 1) / unitBitSize + 1;
 	static const size_t maxMulN = 9;
 	static const size_t pMulTblN = maxMulN + 1;
-	int N_;
+	uint32_t N_;
 	uint32_t shiftL_;
 	uint32_t shiftR_;
 	uint32_t maxIdx_;
@@ -997,7 +997,7 @@ struct SmallModp {
 	void init(const mpz_class& p)
 	{
 		size_t pBitSize = mcl::gmp::getBitSize(p);
-		N_ = (pBitSize + unitBitSize - 1) / unitBitSize;
+		N_ = uint32_t((pBitSize + unitBitSize - 1) / unitBitSize);
 		shiftR_ = (pBitSize - 1) % unitBitSize;
 		shiftL_ = unitBitSize - shiftR_;
 		mpz_class t = 0;
diff --git a/misc/snark-p.py b/misc/snark-p.py
index 8168f3b..cbb7f5a 100644
--- a/misc/snark-p.py
+++ b/misc/snark-p.py
@@ -11,3 +11,8 @@ print("maxarg")
 for i in range(16):
 	print(i, maxarg(i << 253))
 
+
+x=0x2c130429c1d4802eb8703197d038ebd5109f96aee333bd027963094f5bb33ad
+
+y = x * 9
+print(hex(y))
diff --git a/src/fp_generator.hpp b/src/fp_generator.hpp
index ecd022c..d420980 100644
--- a/src/fp_generator.hpp
+++ b/src/fp_generator.hpp
@@ -1095,7 +1095,7 @@ private:
 	*/
 	void gen_fpDbl_mod4(const Reg64& z, const Reg64& xy, const Pack& t)
 	{
-		if (!isFullBit_) {
+		if (!isFullBit_ && useMulx_ && useAdx_) {
 			gen_fpDbl_mod4NF(z, xy, t);
 			return;
 		}
@@ -2547,7 +2547,7 @@ private:
 	}
 	void gen_fpDbl_mulPre(void3u& f)
 	{
-		if (!(useMulx_ && useAdx_)) return;
+		if (!useMulx_ || (pn_ == 6 && !useAdx_)) return;
 		void3u func = getCurr<void3u>();
 		switch (pn_) {
 		case 2:
@@ -3656,25 +3656,25 @@ private:
 		}
 
 		{
-			Pack t = sf.t;
+			Pack t2 = sf.t;
 			if (pn_ == 4) {
-				t = t.sub(0, pn_ * 2);
+				t2 = t2.sub(0, pn_ * 2);
 			} else if (pn_ == 6) {
-				t.append(gp1);
-				t.append(gp2);
+				t2.append(gp1);
+				t2.append(gp2);
 			}
-			assert((int)t.size() == pn_ * 2);
+			assert((int)t2.size() == pn_ * 2);
 
 			mov(gp0, ptr [z]);
-			load_rm(t, gp0 + FpByte_ * 2);
-			sub_rm(t, gp0); // d1 -= d0
-			sub_rm(t, (RegExp)d2); // d1 -= d2
-			store_mr(gp0 + FpByte_ * 2, t);
+			load_rm(t2, gp0 + FpByte_ * 2);
+			sub_rm(t2, gp0); // d1 -= d0
+			sub_rm(t2, (RegExp)d2); // d1 -= d2
+			store_mr(gp0 + FpByte_ * 2, t2);
 
 			gen_raw_sub(gp0, gp0, d2, rax, pn_);
 			const RegExp& d0H = gp0 + pn_ * 8;
 			const RegExp& d2H = (RegExp)d2 + pn_ * 8;
-			gen_raw_fp_sub(d0H, d0H, d2H, t, true);
+			gen_raw_fp_sub(d0H, d0H, d2H, t2, true);
 		}
 		add(rsp, SS);
 		ret();
diff --git a/test/bench.hpp b/test/bench.hpp
index 11ced31..d407bc7 100644
--- a/test/bench.hpp
+++ b/test/bench.hpp
@@ -85,16 +85,18 @@ void testBench(const G1& P, const G2& Q)
 	const int C3 = 100000;
 #if 1
 	const int C2 = 3000;
-	mpz_class a = x.getMpz();
-	CYBOZU_BENCH_C("G1::mulCT     ", C, G1::mulCT, Pa, P, a);
-	CYBOZU_BENCH_C("G1::mul       ", C, G1::mul, Pa, Pa, a);
-	CYBOZU_BENCH_C("G1::add       ", C, G1::add, Pa, Pa, P);
-	CYBOZU_BENCH_C("G1::dbl       ", C, G1::dbl, Pa, Pa);
-	CYBOZU_BENCH_C("G2::mulCT     ", C, G2::mulCT, Qa, Q, a);
-	CYBOZU_BENCH_C("G2::mul       ", C, G2::mul, Qa, Qa, a);
-	CYBOZU_BENCH_C("G2::add       ", C, G2::add, Qa, Qa, Q);
-	CYBOZU_BENCH_C("G2::dbl       ", C, G2::dbl, Qa, Qa);
-	CYBOZU_BENCH_C("GT::pow       ", C, GT::pow, e1, e1, a);
+	{
+		mpz_class a = x.getMpz();
+		CYBOZU_BENCH_C("G1::mulCT     ", C, G1::mulCT, Pa, P, a);
+		CYBOZU_BENCH_C("G1::mul       ", C, G1::mul, Pa, Pa, a);
+		CYBOZU_BENCH_C("G1::add       ", C, G1::add, Pa, Pa, P);
+		CYBOZU_BENCH_C("G1::dbl       ", C, G1::dbl, Pa, Pa);
+		CYBOZU_BENCH_C("G2::mulCT     ", C, G2::mulCT, Qa, Q, a);
+		CYBOZU_BENCH_C("G2::mul       ", C, G2::mul, Qa, Qa, a);
+		CYBOZU_BENCH_C("G2::add       ", C, G2::add, Qa, Qa, Q);
+		CYBOZU_BENCH_C("G2::dbl       ", C, G2::dbl, Qa, Qa);
+		CYBOZU_BENCH_C("GT::pow       ", C, GT::pow, e1, e1, a);
+	}
 //	CYBOZU_BENCH_C("GT::powGLV    ", C, BN::param.glv2.pow, e1, e1, a);
 	G1 PP;
 	G2 QQ;