def synthesize(self, code):
old_code = spu.get_active_code()
spu.set_active_code(code)
self._load_parameters(code)
log = spu_log.SPULog()
log.setup(code)
if self.renderer is not None:
self.renderer.setup(code)
self.renderer.set_one(log.consts['ONE'])
r1_inc = var.SingleFloat()
r2_inc = var.SingleFloat()
r1 = var.SingleFloat()
r2 = var.SingleFloat()
result = var.SingleFloat()
pattern = var.Word(0)
self.ly_point.set_pattern_reg(pattern)
self.ly_point.set_result_reg(result)
self.ly_point.set_r_regs(r1, r2)
self.ly_point.set_log(log)
self.ly_point.setup(code)
spu.lqa(r1, 0)
spu.lqa(r2, 4)
spu.lqa(r1_inc, 8)
spu.lqa(r2_inc, 12)
spu.lqa(pattern, 16)
for y in spuiter.syn_iter(code, self.h):
spu.lqa(r1, 0)
for x in spuiter.syn_iter(code, self.w / 4):
self.ly_point.synthesize(code)
r1.v = spu.fa.ex(r1, r1_inc)
if self.renderer is not None:
# result.v = spu.fm.ex(r1, r2)
self.renderer.set_result_reg(result)
self.renderer.synthesize(code)
if self.renderer is not None:
self.renderer.row_complete(code)
r2.v = spu.fa.ex(r2, r2_inc)
# return Numeric.where(Numeric.less(results, 0), results, 0)
spu.set_active_code(old_code)
return
def TestMFC():
size = 32
#data_array = array.array('I', range(size))
#data = synspu.aligned_memory(size, typecode = 'I')
#data.copy_to(data_array.buffer_info()[0], len(data_array))
data = extarray.extarray('I', range(size))
code = synspu.InstructionStream()
r_zero = code.acquire_register()
r_ea_data = code.acquire_register()
r_ls_data = code.acquire_register()
r_size = code.acquire_register()
r_tag = code.acquire_register()
# Load zero
util.load_word(code, r_zero, 0)
print 'array ea: %X' % (data.buffer_info()[0])
print 'r_zero = %s, ea_data = %s, ls_data = %s, r_size = %s, r_tag = %s' % (
str(r_zero), str(r_ea_data), str(r_ls_data), str(r_size), str(r_tag))
# Load the effective address
print 'test ea: %X' % data.buffer_info()[0]
util.load_word(code, r_ea_data, data.buffer_info()[0])
# Load the size
code.add(spu.ai(r_size, r_zero, size * 4))
# Load the tag
code.add(spu.ai(r_tag, r_zero, 2))
# Load the lsa
code.add(spu.ai(r_ls_data, r_zero, 0))
# Load the data into address 0
mfc_get(code, r_ls_data, r_ea_data, r_size, r_tag)
# Set the tag bit to 2
mfc_write_tag_mask(code, 1 << 2)
# Wait for the transfer to complete
mfc_read_tag_status_all(code)
# Increment the data values by 1 using an unrolled loop (no branches)
r_current = code.acquire_register()
for lsa in range(0, size * 4, 16):
code.add(spu.lqa(r_current, (lsa >> 2)))
code.add(spu.ai(r_current, r_current, 1))
code.add(spu.stqa(r_current, (lsa >> 2)))
code.release_register(r_current)
# Store the values back to main memory
# Load the data into address 0
mfc_put(code, r_ls_data, r_ea_data, r_size, r_tag)
# Set the tag bit to 2
mfc_write_tag_mask(code, 1 << 2)
# Wait for the transfer to complete
mfc_read_tag_status_all(code)
# Cleanup
code.release_register(r_zero)
code.release_register(r_ea_data)
code.release_register(r_ls_data)
code.release_register(r_size)
code.release_register(r_tag)
# Stop for debugging
# code.add(spu.stop(0xA))
# Execute the code
proc = synspu.Processor()
# code.print_code()
#print data_array
proc.execute(code)
#data.copy_from(data_array.buffer_info()[0], len(data_array))
for i in range(size):
assert (data[i] == i + 1)
return
def synthesize(self, code):
old_code = spu.get_active_code()
spu.set_active_code(code)
self._load_parameters(code)
log = spu_log.SPULog()
log.setup(code)
if self.renderer is not None:
self.renderer.setup(code)
self.renderer.set_one(log.consts['ONE'])
r1_inc = var.SingleFloat()
r2_inc = var.SingleFloat()
r1 = var.SingleFloat()
r2 = var.SingleFloat()
result = var.SingleFloat()
pattern = var.Word(0)
self.ly_point.set_pattern_reg(pattern)
self.ly_point.set_result_reg(result)
self.ly_point.set_r_regs(r1, r2)
self.ly_point.set_log(log)
self.ly_point.setup(code)
spu.lqa(r1, 0)
spu.lqa(r2, 4)
spu.lqa(r1_inc, 8)
spu.lqa(r2_inc, 12)
spu.lqa(pattern, 16)
for y in spuiter.syn_iter(code, self.h):
spu.lqa(r1, 0)
for x in spuiter.syn_iter(code, self.w / 4):
self.ly_point.synthesize(code)
r1.v = spu.fa.ex(r1, r1_inc)
if self.renderer is not None:
# result.v = spu.fm.ex(r1, r2)
self.renderer.set_result_reg(result)
self.renderer.synthesize(code)
if self.renderer is not None:
self.renderer.row_complete(code)
r2.v = spu.fa.ex(r2, r2_inc)
# return Numeric.where(Numeric.less(results, 0), results, 0)
spu.set_active_code(old_code)
return
def TestMFC():
import corepy.lib.extarray as extarray
import corepy.arch.spu.platform as synspu
size = 32
#data_array = array.array('I', range(size))
#data = synspu.aligned_memory(size, typecode = 'I')
#data.copy_to(data_array.buffer_info()[0], len(data_array))
data = extarray.extarray('I', range(size))
code = synspu.InstructionStream()
r_zero = code.acquire_register()
r_ea_data = code.acquire_register()
r_ls_data = code.acquire_register()
r_size = code.acquire_register()
r_tag = code.acquire_register()
# Load zero
util.load_word(code, r_zero, 0)
print 'array ea: %X' % (data.buffer_info()[0])
print 'r_zero = %s, ea_data = %s, ls_data = %s, r_size = %s, r_tag = %s' % (
str(r_zero), str(r_ea_data), str(r_ls_data), str(r_size), str(r_tag))
# Load the effective address
print 'test ea: %X' % data.buffer_info()[0]
util.load_word(code, r_ea_data, data.buffer_info()[0])
# Load the size
code.add(spu.ai(r_size, r_zero, size * 4))
# Load the tag
code.add(spu.ai(r_tag, r_zero, 2))
# Load the lsa
code.add(spu.ai(r_ls_data, r_zero, 0))
# Load the data into address 0
mfc_get(code, r_ls_data, r_ea_data, r_size, r_tag)
# Set the tag bit to 2
mfc_write_tag_mask(code, 1<<2);
# Wait for the transfer to complete
mfc_read_tag_status_all(code);
# Increment the data values by 1 using an unrolled loop (no branches)
r_current = code.acquire_register()
for lsa in range(0, size * 4, 16):
code.add(spu.lqa(r_current, (lsa >> 2)))
code.add(spu.ai(r_current, r_current, 1))
code.add(spu.stqa(r_current, (lsa >> 2)))
code.release_register(r_current)
# Store the values back to main memory
# Load the data into address 0
mfc_put(code, r_ls_data, r_ea_data, r_size, r_tag)
# Set the tag bit to 2
mfc_write_tag_mask(code, 1<<2);
# Wait for the transfer to complete
mfc_read_tag_status_all(code);
# Cleanup
code.release_register(r_zero)
code.release_register(r_ea_data)
code.release_register(r_ls_data)
code.release_register(r_size)
code.release_register(r_tag)
# Stop for debugging
# code.add(spu.stop(0xA))
# Execute the code
proc = synspu.Processor()
# code.print_code()
#print data_array
proc.execute(code)
#data.copy_from(data_array.buffer_info()[0], len(data_array))
for i in range(size):
assert(data[i] == i + 1)
return
