luke
/
verilator
mirror of https://github.com/verilator/verilator.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Commentary

This commit is contained in:
Wilson Snyder 2021-09-17 18:52:12 -04:00
parent 08c8b0d7d6
commit 97d8d32049
2 changed files with 30 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
bin/verilator_gantt
View File

@ -58,7 +58,7 @@ def read_data(filename):
if 'elapsed' not in Mtasks[mtask]:
Mtasks[mtask] = {'end': 0, 'elapsed': 0}
Mtasks[mtask]['elapsed'] += elapsed_time
Mtasks[mtask]['predict'] = predict_time
Mtasks[mtask]['predict_cost'] = predict_time
Mtasks[mtask]['end'] = max(Mtasks[mtask]['end'], end)
elif re.match(r'^VLPROFTHREAD', line):
None # pylint: disable=pointless-statement
@ -166,9 +166,9 @@ def report():
for mtask in sorted(Mtasks.keys()):
if Mtasks[mtask]['elapsed'] > 0:
if Mtasks[mtask]['predict'] == 0:
Mtasks[mtask]['predict'] = 1 # don't log(0) below
p2e_ratio = math.log(Mtasks[mtask]['predict'] /
if Mtasks[mtask]['predict_cost'] == 0:
Mtasks[mtask]['predict_cost'] = 1 # don't log(0) below
p2e_ratio = math.log(Mtasks[mtask]['predict_cost'] /
Mtasks[mtask]['elapsed'])
p2e_ratios.append(p2e_ratio)
@ -182,12 +182,12 @@ def report():
print("\nStatistics:")
print(" min log(p2e) = %0.3f" % min_p2e, end="")
print(" from mtask %d (predict %d," %
(min_mtask, Mtasks[min_mtask]['predict']),
(min_mtask, Mtasks[min_mtask]['predict_cost']),
end="")
print(" elapsed %d)" % Mtasks[min_mtask]['elapsed'])
print(" max log(p2e) = %0.3f" % max_p2e, end="")
print(" from mtask %d (predict %d," %
(max_mtask, Mtasks[max_mtask]['predict']),
(max_mtask, Mtasks[max_mtask]['predict_cost']),
end="")
print(" elapsed %d)" % Mtasks[max_mtask]['elapsed'])

48
src/V3EmitCModel.cpp
View File

@ -407,45 +407,45 @@ class EmitCModel final : public EmitCFunc {
}
if (v3Global.opt.mtasks() && v3Global.opt.profThreads()) {
puts("if (VL_UNLIKELY((vlSymsp->_vm_contextp__->profThreadsStart() != "
"vlSymsp->__Vm_profile_time_finished)\n");
puts("if (VL_UNLIKELY((vlSymsp->_vm_contextp__->profThreadsStart()"
" != vlSymsp->__Vm_profile_time_finished)\n");
puts(" && (VL_TIME_Q() > vlSymsp->_vm_contextp__->profThreadsStart())\n");
puts(" && (vlSymsp->_vm_contextp__->profThreadsWindow() >= 1))) {\n");
// Within a profile (either starting, middle, or end)
puts("if (vlSymsp->__Vm_profile_window_ct == 0) {\n"); // Opening file?
puts(/**/ "if (vlSymsp->__Vm_profile_window_ct == 0) {\n"); // Opening file?
// Start profile on this cycle. We'll capture a window worth, then
// only analyze the next window worth. The idea is that the first window
// capture will hit some cache-cold stuff (eg printf) but it'll be warm
// by the time we hit the second window, we hope.
puts("vlSymsp->__Vm_profile_cycle_start = VL_RDTSC_Q();\n");
puts(/****/ "vlSymsp->__Vm_profile_cycle_start = VL_RDTSC_Q();\n");
// "* 2" as first half is warmup, second half is collection
puts("vlSymsp->__Vm_profile_window_ct = vlSymsp->_vm_contextp__->profThreadsWindow() "
"* 2 "
"+ "
"1;\n");
puts("}\n");
puts("--(vlSymsp->__Vm_profile_window_ct);\n");
puts("if (vlSymsp->__Vm_profile_window_ct == "
"vlSymsp->_vm_contextp__->profThreadsWindow()) {\n");
puts(/****/ "vlSymsp->__Vm_profile_window_ct"
" = vlSymsp->_vm_contextp__->profThreadsWindow()"
" * 2 + 1;\n");
puts(/**/ "}\n");
puts(/**/ "--(vlSymsp->__Vm_profile_window_ct);\n");
puts(/**/ "if (vlSymsp->__Vm_profile_window_ct"
" == vlSymsp->_vm_contextp__->profThreadsWindow()) {\n");
// This barrier record in every threads' profile demarcates the
// cache-warm-up cycles before the barrier from the actual profile
// cycles afterward.
puts("vlSymsp->__Vm_threadPoolp->profileAppendAll(");
puts("VlProfileRec(VlProfileRec::Barrier()));\n");
puts("vlSymsp->__Vm_profile_cycle_start = VL_RDTSC_Q();\n");
puts("}\n");
puts("else if (vlSymsp->__Vm_profile_window_ct == 0) {\n");
puts(/****/ "vlSymsp->__Vm_threadPoolp->profileAppendAll(");
puts(/****/ "VlProfileRec(VlProfileRec::Barrier()));\n");
puts(/****/ "vlSymsp->__Vm_profile_cycle_start = VL_RDTSC_Q();\n");
puts(/**/ "}\n");
puts(/**/ "else if (vlSymsp->__Vm_profile_window_ct == 0) {\n");
// Ending file.
puts("vluint64_t elapsed = VL_RDTSC_Q() - vlSymsp->__Vm_profile_cycle_start;\n");
puts("vlSymsp->__Vm_threadPoolp->profileDump(vlSymsp->_vm_contextp__->"
"profThreadsFilename().c_str(), elapsed);\n");
puts(/****/ "vluint64_t elapsed"
" = VL_RDTSC_Q() - vlSymsp->__Vm_profile_cycle_start;\n");
puts(/****/ "vlSymsp->__Vm_threadPoolp->profileDump(vlSymsp->_vm_contextp__->"
"profThreadsFilename().c_str(), elapsed);\n");
// This turns off the test to enter the profiling code, but still
// allows the user to collect another profile by changing
// profThreadsStart
puts("vlSymsp->__Vm_profile_time_finished = "
"vlSymsp->_vm_contextp__->profThreadsStart();\n");
puts("vlSymsp->__Vm_profile_cycle_start = 0;\n");
puts("}\n");
puts(/****/ "vlSymsp->__Vm_profile_time_finished"
" = vlSymsp->_vm_contextp__->profThreadsStart();\n");
puts(/****/ "vlSymsp->__Vm_profile_cycle_start = 0;\n");
puts(/**/ "}\n");
puts("}\n");
}