def fdiv(code, d, x, y, one = None):
"""
Single-precision floating point division for x / y
"""
Y = code.acquire_registers(3)
t = code.acquire_register()
regs = Y[:]
regs.append(t)
if one is None:
one = code.acquire_register()
spu.xor(one, one, one)
spu.ai(one, one, 1)
spu.cuflt(one, one, 155)
regs.append(one)
# Compute 1/y (from SPU ISA 1.1, p208, Normal case)
spu.frest(Y[0], y)
spu.fi(Y[1], y, Y[0])
spu.fnms(t, y, Y[1], one)
spu.fma(Y[2], t, Y[1], Y[1])
# Compute x * (1/y)
spu.fm(d, x, Y[2])
code.release_registers(regs)
return
result = 118
ione = 110
fone = 111
insts = [
# Create fone = 1.0, fa = 2.0 and fb = 4.0
spu.ai(ione, 0, 1),
spu.ai(ia, 0, 2),
spu.ai(ib, 0, 4),
spu.cuflt(fone, ione, 155),
spu.cuflt(fa, ia, 155),
spu.cuflt(fb, ib, 155),
# Compute 1/fb
spu.frest(y0, fb),
spu.fi(y1, fb, y0),
spu.fnms(t1, fb, y1, fone),
spu.fma(y2, t1, y1, y1),
spu.fm(result, fa, y2)
]
for inst in insts:
cli.execute(inst)
regs = cli.get_regs()
for reg in (ione, fone, ia, ib, fa, fb, y0, y1, y2, t1, result):
print reg, '0x%08X 0x%08X 0x%08X 0x%08X' % regs[reg]
cli.stop()
result = 118 ione = 110 fone = 111 insts = [ # Create fone = 1.0, fa = 2.0 and fb = 4.0 spu.ai(ione, 0, 1), spu.ai(ia, 0, 2), spu.ai(ib, 0, 4), spu.cuflt(fone, ione, 155), spu.cuflt(fa, ia, 155), spu.cuflt(fb, ib, 155), # Compute 1/fb spu.frest(y0, fb), spu.fi(y1, fb, y0), spu.fnms(t1, fb, y1, fone), spu.fma(y2, t1, y1, y1), spu.fm(result, fa, y2) ] for inst in insts: cli.execute(inst) regs = cli.get_regs() for reg in (ione, fone, ia, ib, fa, fb, y0, y1, y2, t1, result): print reg, '0x%08X 0x%08X 0x%08X 0x%08X' % regs[reg]