Apply inline asm only to vle in indexed load tests

When data width is different from index width, lmul is also different.
If written in intrinsics, we hit a compiler bug in llvm and gcc that
incorrectly combines the vset, causing the vle to see bad lmul.

A previous workaround rewrites the tests in inline asm.

This new workaround rewrites only the vle part in inline asm, passing
the results out as intrinsics-compatible vector data types, allowing
the indexed loads to happen in intrinsics. This reduces the number of
constraints we need on the asm blocks and reduces the chance of
bugs.

A bug within indexed load tests' expected output generation is fixed.

Change-Id: I7bf1e8a7e9208753e6428a19c178f6f0e40e43e6
5 files changed
tree: 6963d6dd0538ea661b78759757588fe625762b76
  1. doc/
  2. examples/
  3. external/
  4. fpga/
  5. hdl/
  6. hw_sim/
  7. kelvin_test_utils/
  8. lib/
  9. platforms/
  10. rules/
  11. tests/
  12. third_party/
  13. toolchain/
  14. utils/
  15. .bazelrc
  16. .bazelversion
  17. .gitignore
  18. CONTRIBUTING.md
  19. LICENSE
  20. PREUPLOAD.cfg
  21. README.md
  22. WORKSPACE
README.md

Kelvin

Kelvin is a RISC-V32IM core with a custom instruction set.

Kelvin block diagram

More information on the design can be found in the overview.

Getting Started

  • If you are hardware engineer looking to integrate Kelvin into your design, check out our integration guide.
  • If you are a software engineer looking to write code for Kelvin, start with this tutorial.

Building

Kelvin uses bazel as it's build system. The Verilated simulator for Kelvin can be generated using:

bazel build //tests/verilator_sim:core_sim

The verilog source for the Kelvin core can be generated using:

bazel build //hdl/chisel/src/kelvin:core_cc_library_emit_verilog

Verilog source for the Matcha SoC can be generated using:

bazel clean --expunge  # To generate the ToT sha
bazel build //hdl/chisel:matcha_kelvin_verilog