Merge branch 'dev'

include/mcl/vint.hpp

@@ -90,35 +90,19 @@ inline uint32_t mulUnit(uint32_t *pH, uint32_t x, uint32_t y)
 inline uint64_t mulUnit(uint64_t *pH, uint64_t x, uint64_t y)
 {
 #ifdef MCL_VINT_64BIT_PORTABLE
-	uint32_t a = uint32_t(x >> 32);
+	const uint64_t mask = 0xffffffff;
-	uint32_t b = uint32_t(x);
+	uint64_t v = (x & mask) * (y & mask);
-	uint32_t c = uint32_t(y >> 32);
+	uint64_t L = uint32_t(v);
-	uint32_t d = uint32_t(y);
+	uint64_t H = v >> 32;
-
+	uint64_t ad = (x & mask) * uint32_t(y >> 32);
-	uint64_t ad = uint64_t(d) * a;
+	uint64_t bc = uint32_t(x >> 32) * (y & mask);
-	uint64_t bd = uint64_t(d) * b;
+	H += uint32_t(ad);
-	uint64_t L = uint32_t(bd);
+	H += uint32_t(bc);
-	ad += bd >> 32; // [ad:L]
-
-	uint64_t ac = uint64_t(c) * a;
-	uint64_t bc = uint64_t(c) * b;
-	uint64_t H = uint32_t(bc);
-	ac += bc >> 32; // [ac:H]
-	/*
-		  adL
-		 acH
-	*/
-	uint64_t t = (ac << 32) | H;
-	ac >>= 32;
-	H = t + ad;
-	if (H < t) {
-		ac++;
-	}
-	/*
-		ac:H:L
-	*/
 	L |= H << 32;
-	H = (ac << 32) | uint32_t(H >> 32);
+	H >>= 32;
+	H += ad >> 32;
+	H += bc >> 32;
+	H += (x >> 32) * (y >> 32);
 	*pH = H;
 	return L;
 #elif defined(_WIN64) && !defined(__INTEL_COMPILER)

misc/Makefile

@@ -2,5 +2,5 @@ all: low_test
 
 CFLAGS=-I ../include/ -m32 -Ofast -Wall -Wextra -DNDEBUG
 
-low_test: low_test.cpp ../src/low_funct.hpp
+low_test: low_test.cpp ../src/low_func_wasm.hpp
 	$(CXX) -o low_test low_test.cpp $(CFLAGS)

misc/low_test.cpp

@@ -9,7 +9,7 @@ void dump(const char *msg, const uint32_t *x, size_t n)
 	}
 	printf("\n");
 }
-#include "../src/low_funct.hpp"
+#include "../src/low_func_wasm.hpp"
 
 #define MCL_USE_VINT
 #define MCL_VINT_FIXED_BUFFER

src/fp.cpp

@@ -4,8 +4,8 @@
 #include <cybozu/endian.hpp>
 #include <mcl/conversion.hpp>
 #if defined(__EMSCRIPTEN__) && MCL_SIZEOF_UNIT == 4
-#define FOR_WASM
+#define USE_WASM
-#include "low_funct.hpp"
+#include "low_func_wasm.hpp"
 #endif
 
 #if defined(MCL_STATIC_CODE) || defined(MCL_USE_XBYAK) || (defined(MCL_USE_LLVM) && (CYBOZU_HOST == CYBOZU_HOST_INTEL))
@@ -411,7 +411,7 @@ static bool initForMont(Op& op, const Unit *p, Mode mode)
 	return true;
 }
 
-#ifdef FOR_WASM
+#ifdef USE_WASM
 template<size_t N>
 void setWasmOp(Op& op)
 {
@@ -570,7 +570,7 @@ bool Op::init(const mpz_class& _p, size_t maxBitSize, int _xi_a, Mode mode, size
 	default:
 		return false;
 	}
-#ifdef FOR_WASM
+#ifdef USE_WASM
 	if (N == 8) {
 		setWasmOp<8>(*this);
 	} else if (N == 12) {

src/low_func_wasm.hpp

@@ -118,56 +118,6 @@ void mulT(uint32_t z[N * 2], const uint32_t x[N], const uint32_t y[N])
 	}
 }
 
-#if 0
-// slower than mulT
-template<size_t N>
-uint32_t mulUnitWithTblT(uint32_t z[N], const uint64_t *tbl_j)
-{
-	uint32_t H = 0;
-	for (size_t i = 0; i < N; i++) {
-		uint64_t v = tbl_j[i];
-		v += H;
-		z[i] = uint32_t(v);
-		H = uint32_t(v >> 32);
-	}
-	return H;
-}
-
-template<size_t N>
-uint32_t addMulUnitWithTblT(uint32_t z[N], const uint64_t *tbl_j)
-{
-	uint32_t H = 0;
-	for (size_t i = 0; i < N; i++) {
-		uint64_t v = tbl_j[i];
-		v += H;
-		v += z[i];
-		z[i] = uint32_t(v);
-		H = uint32_t(v >> 32);
-	}
-	return H;
-}
-
-// y[N * 2] = x[N] * x[N]
-template<size_t N>
-void sqrT(uint32_t y[N * 2], const uint32_t x[N])
-{
-	uint64_t tbl[N * N]; // x[i]x[j]
-	for (size_t i = 0; i < N; i++) {
-		uint64_t xi = x[i];
-		tbl[i * N + i] = xi * xi;
-		for (size_t j = i + 1; j < N; j++) {
-			uint64_t v = xi * x[j];
-			tbl[i * N + j] = v;
-			tbl[j * N + i] = v;
-		}
-	}
-	y[N] = mulUnitWithTblT<N>(y, tbl);
-	for (size_t i = 1; i < N; i++) {
-		y[N + i] = addMulUnitWithTblT<N>(&y[i], tbl + N * i);
-	}
-}
-#endif
-
 /*
 	z[N * 2] = x[N] * y[N]
 	H = N/2