add mul2

include/mcl/op.hpp

@@ -232,12 +232,12 @@ struct Op {
 	void1u fp_clear;
 	void2u fp_copy;
 	void2u fp_shr1;
-	void3u fp_mul2;
 	void3u fp_neg;
 	void4u fp_add;
 	void4u fp_sub;
 	void4u fp_mul;
 	void3u fp_sqr;
+	void3u fp_mul2;
 	void2uOp fp_invOp;
 	void2uIu fp_mulUnit; // fpN1_mod + fp_mulUnitPre
 
@@ -319,12 +319,12 @@ struct Op {
 		fp_clear = 0;
 		fp_copy = 0;
 		fp_shr1 = 0;
-		fp_mul2 = 0;
 		fp_neg = 0;
 		fp_add = 0;
 		fp_sub = 0;
 		fp_mul = 0;
 		fp_sqr = 0;
+		fp_mul2 = 0;
 		fp_invOp = 0;
 		fp_mulUnit = 0;
 

src/fp_generator.hpp

@@ -379,6 +379,10 @@ private:
 		op.fp_shr1 = gen_shr1();
 		setFuncInfo(prof_, suf, "_shr1", op.fp_shr1, getCurr());
 
+		align(16);
+		op.fp_mul2A_ = gen_mul2();
+		setFuncInfo(prof_, suf, "_mul2", op.fp_mul2A_, getCurr());
+
 		align(16);
 		op.fp_negA_ = gen_fp_neg();
 		setFuncInfo(prof_, suf, "_neg", op.fp_negA_, getCurr());
@@ -915,6 +919,33 @@ private:
 		mov(ptr [pz + (pn_ - 1) * 8], *t0);
 		return func;
 	}
+	void2u gen_mul2()
+	{
+		if (isFullBit_) return 0;
+		if (!(pn_ == 4 || pn_ == 6)) return 0;
+		void2u func = getCurr<void2u>();
+		const int n = pn_ * 2 - 2;
+		StackFrame sf(this, 2, n);
+		Pack x = sf.t.sub(0, pn_);
+		load_rm(x, sf.p[1]);
+#if 0
+		add_rr(x, x);
+#else
+		for (int i = pn_ - 1; i > 0; i--) {
+			shld(x[i], x[i - 1], 1);
+		}
+		shl(x[0], 1);
+#endif
+		Pack t = sf.t.sub(pn_, n - pn_);
+		t.append(sf.p[1]);
+		t.append(rax); // destroy last
+		mov_rr(t, x);
+		lea(rax, ptr[rip + pL_]);
+		sub_rm(t, rax);
+		cmovc_rr(t, x);
+		store_mr(sf.p[0], t);
+		return func;
+	}
 	void3u gen_mul()
 	{
 		void3u func = getCurr<void3u>();

src/fp_static_code.hpp

@@ -23,6 +23,7 @@ void mclx_Fp_neg(Unit*, const Unit*);
 void mclx_FpDbl_mod(Unit*, const Unit*);
 void mclx_Fp_mul(Unit*, const Unit*, const Unit*);
 void mclx_Fp_sqr(Unit*, const Unit*);
+void mclx_Fp_mul2(Unit*, const Unit*);
 void mclx_FpDbl_add(Unit*, const Unit*, const Unit*);
 void mclx_FpDbl_sub(Unit*, const Unit*, const Unit*);
 int mclx_Fp_preInv(Unit*, const Unit*);
@@ -45,6 +46,7 @@ void mclx_Fr_shr1(Unit*, const Unit*);
 void mclx_Fr_neg(Unit*, const Unit*);
 void mclx_Fr_mul(Unit*, const Unit*, const Unit*);
 void mclx_Fr_sqr(Unit*, const Unit*);
+void mclx_Fr_mul2(Unit*, const Unit*);
 int mclx_Fr_preInv(Unit*, const Unit*);
 } // extern "C"
 
@@ -61,6 +63,7 @@ void setStaticCode(mcl::fp::Op& op)
 		op.fpDbl_modA_ = mclx_FpDbl_mod;
 		op.fp_mulA_ = mclx_Fp_mul;
 		op.fp_sqrA_ = mclx_Fp_sqr;
+		op.fp_mul2A_ = mclx_Fp_mul2;
 		op.fpDbl_addA_ = mclx_FpDbl_add;
 		op.fpDbl_subA_ = mclx_FpDbl_sub;
 		op.fpDbl_addPre = mclx_FpDbl_addPre;
@@ -85,6 +88,7 @@ void setStaticCode(mcl::fp::Op& op)
 		op.fp_negA_ = mclx_Fr_neg;
 		op.fp_mulA_ = mclx_Fr_mul;
 		op.fp_sqrA_ = mclx_Fr_sqr;
+		op.fp_mul2A_ = mclx_Fr_mul2;
 		op.fp_preInv = mclx_Fr_preInv;
 	}
 	op.fp_mul = fp::func_ptr_cast<void4u>(op.fp_mulA_);

test/bench.hpp

@@ -116,6 +116,8 @@ void testBench(const G1& P, const G2& Q)
 		CYBOZU_BENCH_C("Fr::add       ", C3, Fr::add, a, a, b);
 		CYBOZU_BENCH_C("Fr::sub       ", C3, Fr::sub, a, a, b);
 		CYBOZU_BENCH_C("Fr::neg       ", C3, Fr::neg, a, a);
+		CYBOZU_BENCH_C("Fr::add 2     ", C3, Fr::add, a, a, b);
+		CYBOZU_BENCH_C("Fr::mul2      ", C3, Fr::mul2, a, a);
 		CYBOZU_BENCH_C("Fr::mul       ", C3, Fr::mul, a, a, b);
 		CYBOZU_BENCH_C("Fr::sqr       ", C3, Fr::sqr, a, a);
 		CYBOZU_BENCH_C("Fr::inv       ", C3, invAdd, a, a, b);