def pow_cf_c(gen,t,srcs):
nonzero = gen.symbols.newLabel()
done = gen.symbols.newLabel()
dst_re = gen.newTemp(Float)
dst_im = gen.newTemp(Float)
# FIXME: this shortcut breaks 3damand01, not sure why
#gen.emit_cjump(srcs[0].im,nonzero)
# compute result if just real
#tdest = pow_ff_f(gen,t,[srcs[0].re,srcs[1]])
#gen.emit_move(tdest,dst_re)
#gen.emit_move(ConstFloatArg(0.0),dst_im)
#gen.emit_jump(done)
#gen.emit_label(nonzero)\
# result if real + imag
# temp = log(a+ib)
# polar(y * real(temp), y * imag(temp))
temp = log_c_c(gen,t,[srcs[0]])
t_re = gen.emit_binop('*',[temp.re, srcs[1]], Float)
t_re = gen.emit_func('exp',[t_re], Float)
t_im = gen.emit_binop('*',[temp.im, srcs[1]], Float)
temp2 = polar_ff_c(gen,t,[t_re, t_im])
gen.emit_move(temp2.re,dst_re)
gen.emit_move(temp2.im,dst_im)
gen.emit_label(done)
return ComplexArg(dst_re,dst_im)
def pow_cc_c(gen,t,srcs):
nonzero = gen.symbols.newLabel()
done = gen.symbols.newLabel()
dst_re = gen.newTemp(Float)
dst_im = gen.newTemp(Float)
gen.emit_cjump(srcs[0].re,nonzero)
gen.emit_cjump(srcs[0].im,nonzero)
# 0^foo = 0
gen.emit_move(ConstFloatArg(0.0),dst_re)
gen.emit_move(ConstFloatArg(0.0),dst_im)
gen.emit_jump(done)
gen.emit_label(nonzero)
# exp(y*log(x))
logx = log_c_c(gen,t,[srcs[0]])
ylogx = mul_cc_c(gen,t,[srcs[1],logx])
xtoy = exp_c_c(gen,t,[ylogx])
gen.emit_move(xtoy.re,dst_re)
gen.emit_move(xtoy.im,dst_im)
gen.emit_label(done)
return ComplexArg(dst_re,dst_im)
def genfunc(gen,t,srcs):
src = srcs[0]
ax = gen.emit_binop('-', [src.parts[0], src.parts[3]], Float)
ay = gen.emit_binop('+', [src.parts[1], src.parts[2]], Float)
bx = gen.emit_binop('+', [src.parts[0], src.parts[3]], Float)
by = gen.emit_binop('-', [src.parts[1], src.parts[2]], Float)
res_a = f(gen,t, [ComplexArg(ax,ay)])
res_b = f(gen,t, [ComplexArg(bx,by)])
outa = gen.emit_binop('+', [res_a.re, res_b.re], Float)
outb = gen.emit_binop('+', [res_a.im, res_b.im], Float)
outc = gen.emit_binop('-', [res_a.im, res_b.im], Float)
outd = gen.emit_binop('-', [res_b.re, res_a.re], Float)
# divide by 2
return mul_hf_h(gen,t,[HyperArg(outa, outb, outc, outd),
ConstFloatArg(0.5)])
def _read_lookup_aci_c(gen,t,srcs):
# lookup pair of floats
i1 = gen.emit_binop('*', [ConstIntArg(2), srcs[1]], Int)
i2 = gen.emit_binop('+', [ConstIntArg(1), i1], Int)
d1 = gen.emit_func2("read_float_array_1D", [srcs[0], i1], Float)
d2 = gen.emit_func2("read_float_array_1D", [srcs[0], i2], Float)
return ComplexArg(d1,d2)
def cos_c_c(gen,t,srcs):
# cos(a+ib) = (cos(a) * cosh(b), -(sin(a) * sinh(b)))
a = srcs[0].re ; b = srcs[0].im
re = gen.emit_binop('*', [ cos_f_f(gen,t,[a]), cosh_f_f(gen,t,[b])], Float)
im = gen.emit_binop('*', [ sin_f_f(gen,t,[a]), sinh_f_f(gen,t,[b])], Float)
nim = gen.emit_func('-',[im], Float)
return ComplexArg(re,nim)
def sqr_c_c(gen,t,srcs):
# sqr(a+ib) = a2 - b2 + i(2ab)
src = srcs[0]
a2 = gen.emit_binop('*', [src.re, src.re], Float)
b2 = gen.emit_binop('*', [src.im, src.im], Float)
ab = gen.emit_binop('*', [src.re, src.im], Float)
dst = ComplexArg(
gen.emit_binop('-', [a2, b2], Float),
gen.emit_binop('*', [ ConstFloatArg(2.0), ab], Float))
return dst
def mul_cc_c(gen,t,srcs):
# (a+ib) * (c+id) = ac - bd + i(bc + ad)
ac = gen.emit_binop('*', [srcs[0].re, srcs[1].re], Float)
bd = gen.emit_binop('*', [srcs[0].im, srcs[1].im], Float)
bc = gen.emit_binop('*', [srcs[0].im, srcs[1].re], Float)
ad = gen.emit_binop('*', [srcs[0].re, srcs[1].im], Float)
dst = ComplexArg(
gen.emit_binop('-', [ac, bd], Float),
gen.emit_binop('+', [bc, ad], Float))
return dst
def div_cc_c(gen,t,srcs):
# (a+ib)/(c+id) = (a+ib)*(c-id) / (c+id)*(c-id)
# = (ac + bd + i(bc - ad))/mag(c+id)
denom = cmag_c_f(gen,'mag', [srcs[1]])
ac = gen.emit_binop('*', [srcs[0].re, srcs[1].re], Float)
bd = gen.emit_binop('*', [srcs[0].im, srcs[1].im], Float)
bc = gen.emit_binop('*', [srcs[0].im, srcs[1].re], Float)
ad = gen.emit_binop('*', [srcs[0].re, srcs[1].im], Float)
dre = gen.emit_binop('+', [ac, bd], Float)
dim = gen.emit_binop('-', [bc, ad], Float)
return ComplexArg(
gen.emit_binop('/', [dre, denom], Float),
gen.emit_binop('/', [dim, denom], Float))
def trunc_c_c(gen,t,srcs):
return ComplexArg(trunc_f_i(gen,t,[srcs[0].re]),
trunc_f_i(gen,t,[srcs[0].im]))
def sinh_c_c(gen,t,srcs):
# sinh(a+ib) = sinh(a)*cos(b) + i (cosh(a) * sin(b))
a = [srcs[0].re]; b = [srcs[0].im]
re = gen.emit_binop('*', [ sinh_f_f(gen,t,a), cos_f_f(gen,t,b)], Float)
im = gen.emit_binop('*', [ cosh_f_f(gen,t,a), sin_f_f(gen,t,b)], Float)
return ComplexArg(re,im)
def acos_c_c(gen,t,srcs):
# acos(z) = pi/2 - asin(z)
pi_by_2 = ComplexArg(ConstFloatArg(math.pi/2.0),ConstFloatArg(0.0))
return sub_cc_c(gen,t,[pi_by_2, asin_c_c(gen,t,srcs)])
def sin_c_c(gen,t,srcs):
# sin(a+ib) = (sin(a) * cosh(b), cos(a) * sinh(b))
a = srcs[0].re ; b = srcs[0].im
re = gen.emit_binop('*', [ sin_f_f(gen,t,[a]), cosh_f_f(gen,t,[b])], Float)
im = gen.emit_binop('*', [ cos_f_f(gen,t,[a]), sinh_f_f(gen,t,[b])], Float)
return ComplexArg(re,im)
def log_c_c(gen,t,srcs):
# log(a+ib) = (log(mag(a+ib)), atan2(a+ib))
re = gen.emit_func('log', [cabs_c_f(gen,t,srcs)], Float)
im = atan2_c_f(gen,t,srcs)
return ComplexArg(re,im)
def mul_cf_c(gen,t,srcs):
# multiply a complex number by a real one
return ComplexArg(
gen.emit_binop('*',[srcs[0].re, srcs[1]], Float),
gen.emit_binop('*',[srcs[0].im, srcs[1]], Float))
def __init__(self):
self.i = ComplexArg(ConstFloatArg(0.0),ConstFloatArg(1.0))
self.iby2 = ComplexArg(ConstFloatArg(0.0),ConstFloatArg(0.5))
self.minus_i = ComplexArg(ConstFloatArg(0.0),ConstFloatArg(-1.0))
self.one = ComplexArg(ConstFloatArg(1.0),ConstFloatArg(0.0))
def ceil_c_c(gen,t,srcs):
return ComplexArg(ceil_f_i(gen,t,[srcs[0].re]),
ceil_f_i(gen,t,[srcs[0].im]))
def times_i(gen,t,srcs):
# multiply by i = (-im,re)
return ComplexArg(neg_f_f(gen,t,[srcs[0].im]),srcs[0].re)
def hyper_jk_h_c(gen,t,srcs):
return ComplexArg(srcs[0].parts[2], srcs[0].parts[3])
def recip_c_c(gen,t,srcs):
return div_cc_c(gen, None,
[ComplexArg(ConstFloatArg(1.0), ConstFloatArg(0.0)), srcs[0]])
def hyper_ri_h_c(gen,t,srcs):
return ComplexArg(srcs[0].parts[0], srcs[0].parts[1])
def flip_c_c(gen,t,srcs):
# flip(a+ib) = b+ia
return ComplexArg(srcs[0].im,srcs[0].re)
def conj_c_c(gen,t,srcs):
# conj (a+ib) = a-ib
b = gen.emit_binop('-', [ ConstFloatArg(0.0), srcs[0].im], Float)
return ComplexArg(srcs[0].re,b)
def round_c_c(gen,t,srcs):
return ComplexArg(round_f_i(gen,t,[srcs[0].re]),
round_f_i(gen,t,[srcs[0].im]))
def pow_ff_c(gen,t,srcs):
arg = ComplexArg(srcs[0], ConstFloatArg(0.0))
return pow_cf_c(gen,t,[arg,srcs[1]])
def floor_c_c(gen,t,srcs):
return ComplexArg(floor_f_i(gen,t,[srcs[0].re]),
floor_f_i(gen,t,[srcs[0].im]))
def neg_c_c(gen,t,srcs):
return ComplexArg(
gen.emit_func('-', [srcs[0].re], Float),
gen.emit_func('-', [srcs[0].im], Float))
def zero_c_c(gen,t,srcs):
return ComplexArg(ConstFloatArg(0.0),ConstFloatArg(0.0))
def sqrt_c_c(gen,t,srcs):
xnonzero = gen.symbols.newLabel()
done = gen.symbols.newLabel()
dst_re = gen.newTemp(Float)
dst_im = gen.newTemp(Float)
gen.emit_cjump(srcs[0].re,xnonzero)
# only an imaginary part :
# temp = sqrt(abs(z.im) / 2);
# return (temp, __y < 0 ? -__temp : __temp);
temp = sqrt_f_f(gen, t, [ abs_f_f(gen,t, [
gen.emit_binop('/',[srcs[0].im, ConstFloatArg(2.0)],Float)])])
gen.emit_move(temp,dst_re)
# y >= 0?
ypos = gen.emit_binop('>=',[srcs[0].im,ConstFloatArg(0.0)], Float)
ygtzero = gen.symbols.newLabel()
gen.emit_cjump(ypos,ygtzero)
nt = neg_f_f(gen,t, [temp])
gen.emit_move(nt,temp)
gen.emit_label(ygtzero)
gen.emit_move(temp,dst_im)
gen.emit_jump(done)
gen.emit_label(xnonzero)
# both real and imaginary
# temp = sqrt(2 * (cabs(z) + abs(z.re)));
# u = temp/2
temp = sqrt_f_f(
gen,t,
[ gen.emit_binop(
'*',
[ConstFloatArg(2.0),
gen.emit_binop(
'+',
[cabs_c_f(gen,t,[srcs[0]]),
abs_f_f(gen,t,[srcs[0].re])],
Float)
],
Float)
])
u = gen.emit_binop('/',[temp,ConstFloatArg(2.0)], Float)
#x > 0?
xpos = gen.emit_binop('>',[srcs[0].re,ConstFloatArg(0.0)], Float)
xgtzero = gen.symbols.newLabel()
gen.emit_cjump(xpos,xgtzero)
# x < 0:
# x = abs(im)/temp
gen.emit_move(gen.emit_binop(
'/',
[abs_f_f(gen,t,[srcs[0].im]), temp], Float) , dst_re)
# y < 0 ? -u : u
ypos2 = gen.emit_binop('>=',[srcs[0].im,ConstFloatArg(0.0)], Float)
ygtzero2 = gen.symbols.newLabel()
gen.emit_cjump(ypos2,ygtzero2)
gen.emit_move(neg_f_f(gen,t,[u]), dst_im)
gen.emit_jump(done)
gen.emit_label(ygtzero2)
gen.emit_move(u, dst_im)
gen.emit_jump(done)
# x > 0:
gen.emit_label(xgtzero)
# (u, im/temp)
gen.emit_move(u,dst_re)
gen.emit_move(gen.emit_binop('/',[srcs[0].im, temp], Float),dst_im)
gen.emit_label(done)
return ComplexArg(dst_re,dst_im)
def abs_c_c(gen,t,srcs):
return ComplexArg(abs_f_f(gen,t,[srcs[0].re]), abs_f_f(gen,t,[srcs[0].im]))
def polar_ff_c(gen,t,srcs):
# polar(r,theta) = (r * cos(theta), r * sin(theta))
re = gen.emit_binop('*',[srcs[0],cos_f_f(gen,t,[srcs[1]])], Float)
im = gen.emit_binop('*',[srcs[0],sin_f_f(gen,t,[srcs[1]])], Float)
return ComplexArg(re,im)
