rv64: regenerate softlib.rv64 for the co_splice cursor rewrite

Browse source

color_ops.c is compiled into the freestanding blob (mux/rv64/Makefile:
src/color_ops.o depends on ../lib/color_ops.c), so rewriting co_splice
leaves the shipped artifact stale.  Regenerated from this branch's
source: 138,072 -> 138,576 bytes.

Built with riscv64-unknown-elf-gcc reporting riscv-none-elf 15.2.0,
which is what softlib.rv64.toolchain already records, so the stamp is
unchanged and the artifact stays reproducible on the machine that has
been generating it.  A different GCC would produce a different binary
and fail the hash check even with identical source.

tests/blob/run.sh now reports a match, which it could not before -- it
rebuilds from source and compares:

  ok: softlib.rv64 matches its source
      (5422afab1fc60bc0ede4ce3da468a42b21cd60bc652a9f694476ed1064cc24c6)
      verified against riscv-none-elf 15.2.0

Note the differential test in this PR (600,000 cases, 5 seeds) covers
the host build of co_splice.  The blob is a separate compilation of the
same source, by a different compiler, for a different architecture, so
it needs its own runtime check -- that is what the jit targets in
make test exercise.

...

This commit is contained in:

Stephen Dennis 2026-08-03 15:30:13 -06:00

parent be7da13809

commit 241c89bd80

1 changed files with 0 additions and 0 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download patch file Download diff file Expand all files Collapse all files

BIN

mux/rv64/softlib.rv64

View file

Binary file not shown.