|
|
@@ -1,9 +1,14 @@
|
|
|
from c import sizeof
|
|
|
|
|
|
+# https://www.cnblogs.com/liuchanglc/p/14203783.html
|
|
|
if sizeof('void_p') == 4:
|
|
|
- PyLong_SHIFT = 28//2
|
|
|
+ PyLong_SHIFT = 28//2 - 1
|
|
|
+ PyLong_NTT_P = 12289
|
|
|
+ PyLong_NTT_PR = 11
|
|
|
elif sizeof('void_p') == 8:
|
|
|
- PyLong_SHIFT = 60//2
|
|
|
+ PyLong_SHIFT = 60//2 - 1
|
|
|
+ PyLong_NTT_P = 1004535809 # PyLong_NTT_P**2 should not overflow
|
|
|
+ PyLong_NTT_PR = 3
|
|
|
else:
|
|
|
raise NotImplementedError
|
|
|
|
|
|
@@ -12,6 +17,64 @@ PyLong_MASK = PyLong_BASE - 1
|
|
|
PyLong_DECIMAL_SHIFT = 4
|
|
|
PyLong_DECIMAL_BASE = 10 ** PyLong_DECIMAL_SHIFT
|
|
|
|
|
|
+assert PyLong_NTT_P > PyLong_BASE
|
|
|
+
|
|
|
+#----------------------------------------------------------------------------#
|
|
|
+# #
|
|
|
+# Number Theoretic Transform #
|
|
|
+# #
|
|
|
+#----------------------------------------------------------------------------#
|
|
|
+
|
|
|
+def ibin(n, bits):
|
|
|
+ assert type(bits) is int and bits >= 0
|
|
|
+ return bin(n)[2:].rjust(bits, "0")
|
|
|
+
|
|
|
+def _number_theoretic_transform(a: list, p, pr, inverse):
|
|
|
+ n = len(a)
|
|
|
+ assert n&(n - 1) == 0
|
|
|
+
|
|
|
+ a = [x % p for x in a]
|
|
|
+ b = n.bit_length() - 1
|
|
|
+
|
|
|
+ for i in range(1, n):
|
|
|
+ j = int(ibin(i, b)[::-1], 2)
|
|
|
+ if i < j:
|
|
|
+ a[i], a[j] = a[j], a[i]
|
|
|
+
|
|
|
+ rt = pow(pr, (p - 1) // n, p)
|
|
|
+ if inverse:
|
|
|
+ rt = pow(rt, p - 2, p)
|
|
|
+
|
|
|
+ w = [1]*(n // 2)
|
|
|
+ for i in range(1, n // 2):
|
|
|
+ w[i] = w[i - 1]*rt % p
|
|
|
+
|
|
|
+ h = 2
|
|
|
+ while h <= n:
|
|
|
+ hf, ut = h // 2, n // h
|
|
|
+ for i in range(0, n, h):
|
|
|
+ for j in range(hf):
|
|
|
+ u = a[i + j]
|
|
|
+ v = a[i + j + hf] * w[ut * j] % p
|
|
|
+ a[i + j] = (u + v) % p
|
|
|
+ a[i + j + hf] = (u - v + p) % p
|
|
|
+ h *= 2
|
|
|
+
|
|
|
+ if inverse:
|
|
|
+ rv = pow(n, p - 2, p)
|
|
|
+ a = [x*rv % p for x in a]
|
|
|
+
|
|
|
+ return a
|
|
|
+
|
|
|
+
|
|
|
+def ntt(a, p, pr):
|
|
|
+ return _number_theoretic_transform(a, p, pr, False)
|
|
|
+
|
|
|
+def intt(a, p, pr):
|
|
|
+ return _number_theoretic_transform(a, p, pr, True)
|
|
|
+
|
|
|
+##############################################################
|
|
|
+
|
|
|
def ulong_fromint(x: int):
|
|
|
# return a list of digits and sign
|
|
|
if x == 0: return [0], 1
|
|
|
@@ -109,16 +172,50 @@ def ulong_muli(a: list, b: int):
|
|
|
return res
|
|
|
|
|
|
def ulong_mul(a: list, b: list):
|
|
|
- res = [0] * (len(a) + len(b))
|
|
|
- for i in range(len(a)):
|
|
|
+ N = len(a) + len(b)
|
|
|
+ if False:
|
|
|
+ # use grade-school multiplication
|
|
|
+ res = [0] * N
|
|
|
+ for i in range(len(a)):
|
|
|
+ carry = 0
|
|
|
+ for j in range(len(b)):
|
|
|
+ carry += res[i+j] + a[i] * b[j]
|
|
|
+ res[i+j] = carry & PyLong_MASK
|
|
|
+ carry >>= PyLong_SHIFT
|
|
|
+ res[i+len(b)] = carry
|
|
|
+ ulong_unpad_(res)
|
|
|
+ return res
|
|
|
+ else:
|
|
|
+ # use fast number-theoretic transform
|
|
|
+ limit = 1
|
|
|
+ while limit < N:
|
|
|
+ limit <<= 1
|
|
|
+ a += [0]*(limit - len(a))
|
|
|
+ b += [0]*(limit - len(b))
|
|
|
+ # print(a, b)
|
|
|
+ a = ntt(a, PyLong_NTT_P, PyLong_NTT_PR)
|
|
|
+ b = ntt(b, PyLong_NTT_P, PyLong_NTT_PR)
|
|
|
+ # print(a, b)
|
|
|
+ c = [0] * limit
|
|
|
+ for i in range(limit):
|
|
|
+ c[i] = (a[i] * b[i]) % PyLong_NTT_P
|
|
|
+
|
|
|
+ # print(c)
|
|
|
+ c = intt(c, PyLong_NTT_P, PyLong_NTT_PR)
|
|
|
+ # print(c)
|
|
|
+
|
|
|
+ # handle carry
|
|
|
carry = 0
|
|
|
- for j in range(len(b)):
|
|
|
- carry += res[i+j] + a[i] * b[j]
|
|
|
- res[i+j] = carry & PyLong_MASK
|
|
|
+ for i in range(limit-1):
|
|
|
+ carry += c[i]
|
|
|
+ c[i] = carry & PyLong_MASK
|
|
|
carry >>= PyLong_SHIFT
|
|
|
- res[i+len(b)] = carry
|
|
|
- ulong_unpad_(res)
|
|
|
- return res
|
|
|
+ if carry > 0:
|
|
|
+ c[limit-1] = carry
|
|
|
+ # print(c)
|
|
|
+ ulong_unpad_(c) # should we use this?
|
|
|
+ # print(c)
|
|
|
+ return c
|
|
|
|
|
|
def ulong_powi(a: list, b: int):
|
|
|
# b >= 0
|