lakeroad

Lakeroad is a synthesizer for FPGAs. Given a description of an FPGA's resources, it can synthesize FPGA configurations (e.g. LUT programmings) to implement high-level programs (e.g. 8-bit add with carry).

Dependencies

• LLVM
  • set LLVM_CONFIG to the location of your llvm-config binary, e.g. /usr/local/Cellar/llvm/15.0.3/bin/llvm-config
• lit test runner, which can be installed with pip install lit
• Racket, plus Racket packages (installable via raco pkg install):
  • rosette
  • json
  • yaml
• Verilator, ideally 4.228 or later
• Boolector, which can be built from source from https://github.com/Boolector/boolector
• Yosys

Please see the Dockerfile for a complete listing of all dependencies.

Environment Variables

• LAKEROAD_DIR
• LAKEROAD_PRIVATE_DIR
• VERILATOR_INCLUDE_DIR
• LLVM_CONFIG

Testing in Lakeroad

run-tests.sh is the primary script for running tests. We have three sources of tests at the moment:

• Rust tests within the (dormant) Lakeroad Rust crate.
• Racket tests within the Lakeroad Racket implementation.
• Integration tests in ./integration_tests/.

Ideally, we should aim to do most of our testing via integration tests: tests which test Lakeroad from the outside, only running commands that users can run. We use the LLVM testing tool FileCheck to build our integration tests, and another LLVM tool, lit, to run them. FileCheck tests take the form of a file to be compiled using Lakeroad or some other tool within Lakeroad. FileCheck tests can be written in any language; in Lakeroad's case, they're generally writte in Verilog, representing a Verilog file to be compiled with Lakeroad. At the top of the file are generally commands for how to compile the file and how to execute FileCheck on the result. At the bottom of the file are syntactic checks to perform on the results.

An example of what Lakeroad integration tests will often look like:

// RUN: racket $LAKEROAD_DIR/bin/main.rkt \
// RUN:  --verilog-module-filepath %s \
// RUN:  --architecture xilinx-ultrascale-plus \
// RUN:  --template dsp \
// RUN:  --out-format verilog \
// RUN:  --top-module-name three_stage_multiplier \
// RUN:  --verilog-module-out-signal p:16 \
// RUN:  --initiation-interval 3 \
// RUN:  --clock-name clk \
// RUN:  --module-name out \
// RUN:  --input-signal a:16 \
// RUN:  --input-signal b:16 \
// RUN: | FileCheck %s

module three_stage_multiplier(input clk, input [15:0] a, b, output [15:0] p);

  reg [15:0] tmp0, tmp1, out;

  always @ (posedge clk) begin
    ...

endmodule

// CHECK: module out(a, b, clk, p);
// CHECK:   wire [47:0] P_0;
// CHECK:   input [15:0] a;
// CHECK:   wire [15:0] a;
// CHECK: ...

When generating tests, it can often be useful to replace the following line:

// RUN: | FileCheck %s

With this:

// RUN: | awk '{print "// CHECK: " $0}' >> %s

Which will automatically take the output of the tool and append it to the end of the file with ``// CHECK: ` as a prefix.

Notes for Mac

Mac users with an old version of gcc/g++ (i.e. clang/clang++) as their default compiler will need to install a newer clang++ via brew install llvm, and then ensure that clang++ is used when running tests. This can be done by setting CXX as follows:

CXX=/usr/local/Cellar/llvm/16.0.3/bin/clang++ lit -v integration_tests/lakeroad/presubaddor_2_stage_unsigned_10_bit.sv

TODO(@gussmith23): this is unintuitive and not user-friendly.

Use of GNU sed

The ./import_all_primitives.sh script uses sed, which it expects to be GNU sed. On Mac, you can install GNU sed via Homebrew and add it to your PATH temporarily or permanently.

TODO(@gussmith23): Find better cross-platform regex. I thought Perl might be the answer.