Multithreaded Verilator #654
Conversation
|
This is untested. Additionally, someone may want to double-check the |
| @@ -165,15 +169,15 @@ verilog: $(sim_vsrcs) | |||
|
|
|||
| # run normal binary with hardware-logged insn dissassembly | |||
| run-binary: $(output_dir) $(sim) | |||
| (set -o pipefail && $(sim) $(PERMISSIVE_ON) $(SIM_FLAGS) $(EXTRA_SIM_FLAGS) $(SEED_FLAG) $(VERBOSE_FLAGS) $(PERMISSIVE_OFF) $(BINARY) </dev/null 2> >(spike-dasm > $(sim_out_name).out) | tee $(sim_out_name).log) | |||
| (set -o pipefail && $(NUMA_PREFIX) $(sim) $(PERMISSIVE_ON) $(SIM_FLAGS) $(EXTRA_SIM_FLAGS) $(SEED_FLAG) $(VERBOSE_FLAGS) $(PERMISSIVE_OFF) $(BINARY) </dev/null 2> >(spike-dasm > $(sim_out_name).out) | tee $(sim_out_name).log) | |||
There was a problem hiding this comment.
What happens when you set this but use vcs?
There was a problem hiding this comment.
From testing on a5, nothing since the script returns nothing. In general, I don't think you want to enable NUMA_PREFIX, when using VCS though.
There was a problem hiding this comment.
Did you do any tests where the script does return something?
There was a problem hiding this comment.
No I didn't. It shouldn't really be used with VCS but even if it was I think numactl is fairly understandable, it just controls where the program is placed (what CPU/Socket) and nothing more.
|
|
||
| // RocketChip currently only supports RBB port 0, so this needs to stay here |
There was a problem hiding this comment.
The new <widget>_t should be set by the simulator widget code when first ticked. The main difference between this and the orig. PR is that you need to have the jtag widget struct initialized again otherwise the rbb_port will not be assigned to it.
|
Ok this was tested with the |
abejgonzalez
force-pushed
the
multithreaded-verilator
branch
from
9162279
to
2168813
Compare
|
Can you verify that |
I can confirm that |
scripts/numa_prefix
Outdated
| # loop through available nodes, selecting the node with the most free mem | ||
| for i in avail_nodes: | ||
| cpu_line = lines[line_idx] | ||
| mem_size_line = lines[line_idx + 1] |
There was a problem hiding this comment.
This isn't used anywhere. I think we should just remove it and comment on mem_free_line that we skip the size line.
Related issue: Split up from PR #562
Type of change: new feature
Impact: other
Release Notes
This splits up #562 to just include the multi-threaded Verilator improvements. The main differences from the prior PR is that
numa_prefixis moved toscripts/,numa_prefixis directly pointed to incommon.mk, andNUMA_PREFIXis added in more places (only where Verilator might be called).Finally, I caught extra flags being passed to
LDFLAGSthat were unneeded and used theTRACE_OPTSthat was missed in PR #650 .