p(".text")
p(".hidden poly_mod_3_Phi_n")
p(".global poly_mod_3_Phi_n")
p(".att_syntax prefix")
p("poly_mod_3_Phi_n:")
# rdi holds r
N_min_1 = 0
t = 1
# NTRU_N is in 509th element; 13th word of 32nd register
p("vmovdqa {}(%rdi), %ymm{}".format(31*32, N_min_1))
p("vpermq ${}, %ymm{}, %ymm{}".format(int('00000011', 2), N_min_1, N_min_1))
# move into high 16 in doubleword (to clear high 16) and multiply by two
p("vpslld $17, %ymm{}, %ymm{}".format(N_min_1, N_min_1))
# clone into bottom 16
p("vpsrld $16, %ymm{}, %ymm{}".format(N_min_1, t))
p("vpor %ymm{}, %ymm{}, %ymm{}".format(N_min_1, t, N_min_1))
# and now it's everywhere in N_min_1
p("vbroadcastss %xmm{}, %ymm{}".format(N_min_1, N_min_1))
retval = 2
for i in range(NTRU_N32//16):
p("vpaddw {}(%rdi), %ymm{}, %ymm{}".format(i*32, N_min_1, t))
mod3(t, retval)
p("vmovdqa %ymm{}, {}(%rdi)".format(retval, i*32))
for i in range(NTRU_N, NTRU_N32):
p("movw $0, {}(%rdi)".format(2*i))
p("ret")
a = 1
threes = 3
last = 4
retval = 5
modq = 6
p("vmovdqa const_3_repeating(%rip), %ymm{}".format(threes))
p("vmovdqa mask_modq(%rip), %ymm{}".format(modq))
p("vmovdqa {}(%rsi), %ymm{}".format((NTRU_N32 // 16 - 1) * 32, last))
p("vpand %ymm{}, %ymm{}, %ymm{}".format(modq, last, last))
p("vpsrlw ${}, %ymm{}, %ymm{}".format(LOGQ - 1, last, r))
p("vpxor %ymm{}, %ymm{}, %ymm{}".format(threes, r, r))
p("vpsllw ${}, %ymm{}, %ymm{}".format(LOGQ, r, r))
p("vpaddw %ymm{}, %ymm{}, %ymm{}".format(last, r, last))
mod3(last, retval)
p("vpsllw $1, %ymm{}, %ymm{}".format(retval, last))
p("vextracti128 $1, %ymm{}, %xmm{}".format(last, last))
p("vpshufb shuf_b8_to_low_doubleword(%rip), %ymm{}, %ymm{}".format(
last, last))
p("vinserti128 $1, %xmm{}, %ymm{}, %ymm{}".format(last, last, last))
for i in range(NTRU_N32 // 16):
p("vmovdqa {}(%rsi), %ymm{}".format(i * 32, a))
p("vpand %ymm{}, %ymm{}, %ymm{}".format(modq, a, a))
p("vpsrlw ${}, %ymm{}, %ymm{}".format(LOGQ - 1, a, r))
p("vpxor %ymm{}, %ymm{}, %ymm{}".format(threes, r, r))
p("vpsllw ${}, %ymm{}, %ymm{}".format(LOGQ, r, r))
p("vpaddw %ymm{}, %ymm{}, %ymm{}".format(a, r, r))
p("vpaddw %ymm{}, %ymm{}, %ymm{}".format(last, r, r))
mod3(r, retval)
import mod3
def mod1():
print('Func from mod1.py')
def go():
print('Go from mod1.py')
# go() will not be imported in lesson.py
__all__ = ['mod1']
print('This is mod1.py')
print(__name__)
mod3.mod3()
p("vpsrlq ${}, %ymm{}, %ymm{}".format(16 * word_of_low, last, last))
# Copy the last coefficient to high 16 of low double word
p("vpslld $16, %ymm{}, %ymm{}".format(last, r))
p("vpsrld $16, %ymm{}, %ymm{}".format(r, last))
p("vpor %ymm{}, %ymm{}, %ymm{}".format(last, r, last))
# Broadcast the last coefficient to all 16 words
p("vbroadcastss %xmm{}, %ymm{}".format(last, last))
# Add (-q) mod 3 to last if last is >= q/2.
# Note (-q) mod 3=(-2^k) mod 3=1<<(1-(k&1))
p("vpsrlw ${}, %ymm{}, %ymm{}".format(LOGQ - 1, last, r))
if LOGQ % 2 == 0: p("vpsllw ${}, %ymm{}, %ymm{}".format(1, r, r))
p("vpaddw %ymm{}, %ymm{}, %ymm{}".format(last, r, last))
# map last to -last mod 3
mod3(last, r)
p("vpsllw $1, %ymm{}, %ymm{}".format(r, last))
for i in range(NTRU_N32 // 16):
p("vmovdqa {}(%rsi), %ymm{}".format(i * 32, a))
p("vpand %ymm{}, %ymm{}, %ymm{}".format(modq, a, a))
p("vpsrlw ${}, %ymm{}, %ymm{}".format(LOGQ - 1, a, r))
if LOGQ % 2 == 0: p("vpsllw ${}, %ymm{}, %ymm{}".format(1, r, r))
p("vpaddw %ymm{}, %ymm{}, %ymm{}".format(a, r, a))
p("vpaddw %ymm{}, %ymm{}, %ymm{}".format(a, last, a))
mod3(a, r)
p("vmovdqa %ymm{}, {}(%rdi)".format(r, i * 32))
p("ret")
import mod3 mod3.mod3(10, 12)