def TestCompileExec():
import time
SIZE = 1024
kernel = (
"il_ps_2_0\n" +
"dcl_input_position_interp(linear_noperspective) v0\n" +
"dcl_output_generic o0\n" + "dcl_output_generic o1\n" +
#"dcl_output_generic o2\n" +
"dcl_resource_id(0)_type(2d,unnorm)_fmtx(float)_fmty(float)_fmtz(float)_fmtw(float)\n"
+
#"mov r0, g[0]\n" +
"sample_resource(0)_sampler(0) o0, v0.xyxx\n" + "mov g[0], r0\n" +
"end\n")
t1 = time.time()
image = cal_exec.compile(kernel)
t2 = time.time()
print "compile time", t2 - t1
input = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, SIZE, SIZE, 0)
output = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, SIZE, SIZE, 0)
#glob = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, 4096, 4096, cal_exec.GLOBAL_BUFFER)
print "input", input
print "output", output
remote = {"o0": output, "i0": input}
local = {
"o1": (SIZE, SIZE, cal_exec.FMT_FLOAT32_4),
"g[]": (4096, 4096, cal_exec.FMT_FLOAT32_4)
}
domain = (0, 0, SIZE, SIZE)
print "remote bindings", remote
print "local bindings", local
# image, dev num, (x, y, w, h)
t1 = time.time()
cal_exec.run_stream(image, 0, domain, local, remote)
t2 = time.time()
print "run time", t2 - t1
cal_exec.free_remote(input)
cal_exec.free_remote(output)
#cal_exec.free_remote(glob)
cal_exec.free_image(image)
return
def TestCompileExec():
import time
SIZE = 1024
kernel = ("il_ps_2_0\n" +
"dcl_input_position_interp(linear_noperspective) v0\n" +
"dcl_output_generic o0\n" +
"dcl_output_generic o1\n" +
#"dcl_output_generic o2\n" +
"dcl_resource_id(0)_type(2d,unnorm)_fmtx(float)_fmty(float)_fmtz(float)_fmtw(float)\n" +
#"mov r0, g[0]\n" +
"sample_resource(0)_sampler(0) o0, v0.xyxx\n" +
"mov g[0], r0\n" +
"end\n")
t1 = time.time()
image = cal_exec.compile(kernel)
t2 = time.time()
print "compile time", t2 - t1
input = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, SIZE, SIZE, 0)
output = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, SIZE, SIZE, 0)
#glob = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, 4096, 4096, cal_exec.GLOBAL_BUFFER)
print "input", input
print "output", output
remote = {"o0": output, "i0": input}
local = {"o1": (SIZE, SIZE, cal_exec.FMT_FLOAT32_4),
"g[]": (4096, 4096, cal_exec.FMT_FLOAT32_4)}
domain = (0, 0, SIZE, SIZE)
print "remote bindings", remote
print "local bindings", local
# image, dev num, (x, y, w, h)
t1 = time.time()
cal_exec.run_stream(image, 0, domain, local, remote)
t2 = time.time()
print "run time", t2 - t1
cal_exec.free(input)
cal_exec.free(output)
#cal_exec.free(glob)
cal_exec.free_image(image)
return
def alloc_remote(self, typecode, comps, width, height=1, globl=False):
if typecode == 'f':
if comps == 1:
fmt = cal_exec.FMT_FLOAT32_1
elif comps == 2:
fmt = cal_exec.FMT_FLOAT32_2
elif comps == 4:
fmt = cal_exec.FMT_FLOAT32_4
else:
raise Exception("Number of components must be 1, 2, or 4")
elif typecode == 'i':
if comps == 1:
fmt = cal_exec.FMT_SIGNED_INT32_1
elif comps == 2:
fmt = cal_exec.FMT_SIGNED_INT32_2
elif comps == 4:
fmt = cal_exec.FMT_SIGNED_INT32_4
else:
raise Exception("Number of components must be 1, 2, or 4")
elif typecode == 'I':
if comps == 1:
fmt = cal_exec.FMT_UNSIGNED_INT32_1
elif comps == 2:
fmt = cal_exec.FMT_UNSIGNED_INT32_2
elif comps == 4:
fmt = cal_exec.FMT_UNSIGNED_INT32_4
else:
raise Exception("Number of components must be 1, 2, or 4")
else:
raise Exception("Unsupported data type: " + str(typecode))
if globl:
globl = cal_exec.GLOBAL_BUFFER
# Allocate and initialize the memory
# TODO - more operand error checking
mem = cal_exec.alloc_remote(self.device, fmt, width, height, globl)
arr = extarray.extarray(typecode, 0)
arr.data_len = mem[1] * height * comps
arr.set_memory(mem[0], arr.data_len * 4)
arr.gpu_mem_handle = mem
arr.gpu_device = self.device
arr.gpu_width = width
arr.gpu_pitch = mem[1]
return arr
def alloc_remote(self, typecode, comps, width, height = 1, globl = False):
if typecode == 'f':
if comps == 1:
fmt = cal_exec.FMT_FLOAT32_1
elif comps == 2:
fmt = cal_exec.FMT_FLOAT32_2
elif comps == 4:
fmt = cal_exec.FMT_FLOAT32_4
else:
raise Exception("Number of components must be 1, 2, or 4")
elif typecode == 'i':
if comps == 1:
fmt = cal_exec.FMT_SIGNED_INT32_1
elif comps == 2:
fmt = cal_exec.FMT_SIGNED_INT32_2
elif comps == 4:
fmt = cal_exec.FMT_SIGNED_INT32_4
else:
raise Exception("Number of components must be 1, 2, or 4")
elif typecode == 'I':
if comps == 1:
fmt = cal_exec.FMT_UNSIGNED_INT32_1
elif comps == 2:
fmt = cal_exec.FMT_UNSIGNED_INT32_2
elif comps == 4:
fmt = cal_exec.FMT_UNSIGNED_INT32_4
else:
raise Exception("Number of components must be 1, 2, or 4")
else:
raise Exception("Unsupported data type: " + str(typecode))
if globl:
globl = cal_exec.GLOBAL_BUFFER
# Allocate and initialize the memory
# TODO - more operand error checking
mem = cal_exec.alloc_remote(self.device, fmt, width, height, globl)
arr = extarray.extarray(typecode, 0)
arr.data_len = mem[1] * height * comps
arr.set_memory(mem[0], arr.data_len * 4)
arr.gpu_mem_handle = mem
arr.gpu_device = self.device
arr.gpu_width = width
arr.gpu_pitch = mem[1]
return arr
def alloc_remote(self, typecode, comps, width, height = 1, globl = False):
"""Allocate an ExtArray backed by remote (main) memory."""
fmt = self._get_fmt(typecode, comps)
if globl:
globl = cal_exec.GLOBAL_BUFFER
# Allocate and initialize the memory
# TODO - more operand error checking
mem = cal_exec.alloc_remote(self.device, fmt, width, height, globl)
arr = extarray.extarray(typecode, 0)
arr.data_len = mem[2] * height * comps
arr.set_memory(mem[1], arr.data_len * 4)
arr.gpu_mem_handle = mem
arr.gpu_device = self.device
arr.gpu_width = width
arr.gpu_pitch = mem[2]
arr.gpu_height = height
return arr
def TestRemoteAlloc():
mem_handle = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, 1024, 1024, 0)
print "mem handle", mem_handle
cal_exec.free_remote(mem_handle)
return
def TestRemoteAlloc(): mem_handle = cal_exec.alloc_remote(cal_exec.FMT_FLOAT32_4, 1024, 1024, 0) print "mem handle", mem_handle cal_exec.free(mem_handle) return
