1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
|
import NvRules
import nvanalysis
def get_name():
return "Kernel Bounds Analysis"
def get_description():
return "Kernel Bounds Analysis Rule"
def get_identifier():
return "kernel_instance_bounds_analysis"
def get_section_identifier():
return "kernel_instance_bounds_analysis"
#def evaluate(handle):
# List rules proposed by this rule here once they are ready
# return NvRules.require_rules(handle, [ "kernel_instance_latency_analysis" ])
def apply(handle):
ctx = NvRules.get_context(handle)
fe = ctx.frontend()
action = ctx.range_by_idx(0).action_by_idx(0)
inst_issued_slots = action.metric_by_name("smsp__inst_issued_slots").as_double()
inst_executed_lsu_pipe = action.metric_by_name("smsp__inst_executed_lsu_pipe").as_double()
inst_executed_tex_pipe = action.metric_by_name("smsp__inst_executed_tex_pipe").as_double()
inst_executed_bru_pipe = action.metric_by_name("smsp__inst_executed_bru_pipe").as_double()
utilization_issue = action.metric_by_name("smsp__utilization_issue").as_double()
m_issue_slots = inst_issued_slots
m_ldst_issued = inst_executed_lsu_pipe + inst_executed_tex_pipe
m_cf_issued = inst_executed_bru_pipe
m_issue_slot_utilization = utilization_issue
m_cc_major = action.metric_by_name("device__attribute_compute_capability_major").as_uint64()
m_cc_minor = action.metric_by_name("device__attribute_compute_capability_minor").as_uint64()
m_device_name = action.metric_by_name("device__attribute_display_name").as_string()
avail_mem_unit_enums = nvanalysis.units.get_memory_unit_enums(m_cc_major, m_cc_minor)
avail_func_unit_enums = nvanalysis.units.get_function_unit_enums(m_cc_major, m_cc_minor)
avail_mem_units = []
avail_func_units = []
for memUnitEnum in avail_mem_unit_enums:
avail_mem_units.append(nvanalysis.units.get_memory_unit(memUnitEnum))
for funcUnitEnum in avail_func_unit_enums:
avail_func_units.append(nvanalysis.units.get_function_unit(funcUnitEnum))
slot_utlization = m_issue_slot_utilization / 100.0
max_utilized_function_unit = nvanalysis.units.get_max_utilized_function_unit(avail_func_units, action)
fu_utilization = nvanalysis.metrics.get_utilization_percent(max_utilized_function_unit.value(action))
# TODO
# if (memMap.containsKey(MemoryUnit.SYSMEM)) ...
(mem_util, mem_bound) = nvanalysis.units.get_memory_utilization(avail_func_units, avail_mem_units, action)
if fu_utilization > slot_utlization:
bound = nvanalysis.bounds.get_kernel_bound(fu_utilization, mem_util)
else:
cf_util = slot_utlization * (m_cf_issued / m_issue_slots)
ldst_util = slot_utlization * (m_ldst_issued / m_issue_slots)
arith_util = slot_utlization - cf_util - ldst_util
sm_util = slot_utlization - ldst_util
bound = nvanalysis.bounds.get_kernel_bound(sm_util, mem_util)
fe.message(nvanalysis.messages.kernel_bounds_msg(bound, mem_bound, m_device_name))
|
