optimized the fallback code for uint320 multiplication

mrdcvlsc · Apr 6, 2022 · 5163822 · 5163822
1 parent 8bff93f
commit 5163822
Show file tree

Hide file tree

Showing 2 changed files with 25 additions and 27 deletions.
diff --git a/tests/multiplication.cpp b/tests/multiplication.cpp
@@ -74,7 +74,6 @@ int main() {
     ASSERT_UINT512(MAXMAX,CMAXMAX,"MAX*MAX");
     ASSERT_UINT512(NUM1NUM2,CNUM1NUM2,"NUM1*NUM2");
 
-
     // SUMMARY OF RESULTS  
     size_t failed_cnt = 0; 
     for(auto e : TEST_RESULTS) {     

diff --git a/uint320.cpp b/uint320.cpp
@@ -349,30 +349,31 @@ uint320 uint320::operator*(const uint320& mr) const {
 
 #if(__x86_64 || __x86_64__ || __amd64 || __amd64__ || __aarch64__ || __aarch64)
 #if(_MSC_VER || _PURE_CPP)
+
     __uint128_t __uint128_product[UINT320LIMBS+1] = {0,0,0,0,0,0};
 
     for(size_t i=0; i<UINT320LIMBS; ++i) {
         for(size_t j=0; j<UINT320LIMBS-i; ++j) {
-
+
+            // index product
             __uint128_t prd = (__uint128_t)limbs[j] * (__uint128_t)mr.limbs[i];
+
+            // low part add
             __uint128_product[i+j] += ((prd << UINT64BITS) >> UINT64BITS);
-            __uint128_product[i+j+1] += (prd >> UINT64BITS); // carry
-
-            for(size_t k=j+i; k<UINT320LIMBS; ++k) {
-                if((ulongint)(__uint128_product[k] >> UINT64BITS)) { // if carry again in carry
-                    __uint128_product[k+1] += __uint128_product[k] >> UINT64BITS;
-                    __uint128_product[k] = (__uint128_product[k] << UINT64BITS) >> UINT64BITS;
-                }
-                else if(k>j+i){
-                    break;
-                }
-            }
+
+            // high part add
+            __uint128_product[i+j+1] += (prd >> UINT64BITS); // high-carry
+
+            // last carry
+            __uint128_product[i+j+1] += __uint128_product[i+j] >> UINT64BITS;
+            __uint128_product[i+j]    = (__uint128_product[i+j] << 64) >> 64;
         }
     }
 
     for(size_t i=0; i<UINT320LIMBS; ++i) {
         pd.limbs[i] = __uint128_product[i];
     }
+
 #elif ((__clang__ || __GNUC__ || __GNUG__ || __MINGW64__) && (__aarch64__ || __aarch64))
 #ifndef _HIDE_WARNING
 #warning using GCC inline asm, please enable optimization flag, recomended : -O2, to enable use C++ implementation instead, enable the -D_PURE_CPP flag.
@@ -382,20 +383,19 @@ uint320 uint320::operator*(const uint320& mr) const {
 
     for(size_t i=0; i<UINT320LIMBS; ++i) {
         for(size_t j=0; j<UINT320LIMBS-i; ++j) {
-
+
+            // index product
             __uint128_t prd = (__uint128_t)limbs[j] * (__uint128_t)mr.limbs[i];
+
+            // low part add
             __uint128_product[i+j] += ((prd << UINT64BITS) >> UINT64BITS);
-            __uint128_product[i+j+1] += (prd >> UINT64BITS); // carry
-
-            for(size_t k=j+i; k<UINT320LIMBS; ++k) {
-                if((ulongint)(__uint128_product[k] >> UINT64BITS)) { // if carry again in carry
-                    __uint128_product[k+1] += __uint128_product[k] >> UINT64BITS;
-                    __uint128_product[k] = (__uint128_product[k] << UINT64BITS) >> UINT64BITS;
-                }
-                else if(k>j+i){
-                    break;
-                }
-            }
+
+            // high part add
+            __uint128_product[i+j+1] += (prd >> UINT64BITS); // high-carry
+
+            // last carry
+            __uint128_product[i+j+1] += __uint128_product[i+j] >> UINT64BITS;
+            __uint128_product[i+j]    = (__uint128_product[i+j] << 64) >> 64;
         }
     }
 
@@ -511,9 +511,8 @@ uint320 uint320::operator*(const uint320& mr) const {
         [mc4]"m"(limbs[4])
         : "rax","rdx","cc"
     );
-
 #else
-    #error uint320 is not supported on this compiler
+#error uint320 is not supported on this compiler
 #endif
 #else
 #error uint320 is not supported on 32-bit computers (x86 architectures)