luke/verilator
1
0
Fork 0
mirror of https://github.com/verilator/verilator.git synced 2026-10-06 01:54:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

Perform VCD tracing in parallel when using --threads (#3449)

Browse Source 
VCD tracing is now parallelized using the same thread pool as the model.
We achieve this by breaking the top level trace functions into multiple
top level functions (as many as --threads), and after emitting the time
stamp to the VCD file on the main thread, we execute the tracing
functions in parallel on the same thread pool as the model (which we
pass to the trace file during registration), tracing into a secondary
per thread buffer. The main thread will then stitch (memcpy) the buffers
together into the output file.

This makes the `--trace-threads` option redundant with `--trace`, which
now only affects `--trace-fst`. FST tracing uses the previous offloading
scheme.

This obviously helps a lot in VCD tracing performance, and I have seen
better than Amdahl speedup, namely I get 3.9x on XiangShan 4T (2.7x on
OpenTitan 4T).
This commit is contained in:
Geza Lore
2022-05-29 19:08:39 +01:00
committed by GitHub
parent c4b8675d77
commit b51f887567
19 changed files with 811 additions and 315 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/V3EmitCImp.cpp
+8 -8
View File
@@ -751,26 +751,26 @@ class EmitCTrace final : EmitCFunc {
const string func = nodep->full() ? "full" : "chg";
bool emitWidth = true;
if (nodep->dtypep()->basicp()->isDouble()) {
puts("tracep->" + func + "Double");
puts("bufp->" + func + "Double");
emitWidth = false;
} else if (nodep->isWide() || emitTraceIsScBv(nodep) || emitTraceIsScBigUint(nodep)) {
puts("tracep->" + func + "WData");
puts("bufp->" + func + "WData");
} else if (nodep->isQuad()) {
puts("tracep->" + func + "QData");
puts("bufp->" + func + "QData");
} else if (nodep->declp()->widthMin() > 16) {
puts("tracep->" + func + "IData");
puts("bufp->" + func + "IData");
} else if (nodep->declp()->widthMin() > 8) {
puts("tracep->" + func + "SData");
puts("bufp->" + func + "SData");
} else if (nodep->declp()->widthMin() > 1) {
puts("tracep->" + func + "CData");
puts("bufp->" + func + "CData");
} else {
puts("tracep->" + func + "Bit");
puts("bufp->" + func + "Bit");
emitWidth = false;
}
const uint32_t offset = (arrayindex < 0) ? 0 : (arrayindex * nodep->declp()->widthWords());
const uint32_t code = nodep->declp()->code() + offset;
puts(v3Global.opt.useTraceOffloadThread() && !nodep->full() ? "(base+" : "(oldp+");
puts(v3Global.opt.useTraceOffload() && !nodep->full() ? "(base+" : "(oldp+");
puts(cvtToStr(code - nodep->baseCode()));
puts(",");
emitTraceValue(nodep, arrayindex);