Skip to content

Cmake rework - #35

Open
Rvidye wants to merge 9 commits into
Utah-Graphics-Lab:masterfrom
Rvidye:cmake-rework
Open

Cmake rework#35
Rvidye wants to merge 9 commits into
Utah-Graphics-Lab:masterfrom
Rvidye:cmake-rework

Conversation

@Rvidye

@Rvidye Rvidye commented Jun 25, 2026

Copy link
Copy Markdown

This PR aims to refactor the old cmake workflow which was functional but still confusing at times.

  • Remove TBB binaries
  • modern cmake with packagee-managed deps
  • unified CLI for simulator and native kernels
  • Startup guide for HWRT course

Rvidye added 8 commits June 22, 2026 13:39
LatencyFIFO::clock() both advances the latency counter and reopens the
issue slot, so clocking twice per cycle to get two issues also halved the
modeled latency.
Use one pipeline per issue slot, each clocked once, matching the existing
UnitSFU pattern. Throughput stays at 2 issues/cycle; latency and in-flight
capacity are restored. Also widens the DRAM return path, which drained only
one 32B sector per cycle and would have bottlenecked the request side.
Reported by Haocong Luo (SAFARI, ETH Zurich).
// number of parallel intersection pipelines.
// each pipeline is a separate unit that is clocked once per cycle, so windening this changes throughput without changing the modeled latency.
const static uint NUM_ISECT_PIPLINES = 2;

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's make this a config parameter: one for tri pipes, one for node pipes

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

Comment thread src/arches-v2/units/unit-dram.hpp Outdated
//each pipline is a seperate path that is clocked once per cycle, so the width sets throughput without scaling the modeled L2 to controller latency.

const static uint NUM_REQ_PIPLINES = 2;

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same thing; this should be a config parameter.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

done

Comment thread src/arches-v2/units/unit-dram.cpp Outdated

UnitDRAMRamulator::UnitDRAMRamulator(Configuration config) : UnitMainMemoryBase(config.size),
_request_network(config.num_ports, config.num_controllers * 2, 1 << 13), _return_network(config.num_controllers * 2, config.num_ports), _partition_mask(config.partition_stride)
_request_network(config.num_ports, config.num_controllers * MemoryController::NUM_REQ_PIPLINES, 1 << 13),

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should double-check that 2^13 actually matches the channel stride in the DRAM controller

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants