Skip to content

Ajr/pldi benchmarks fixes - #213

Merged
rootjalex merged 3 commits into
ajr/pldi-benchmarksfrom
ajr/pldi-benchmarks-fixes
Aug 30, 2026
Merged

rootjalex merged 3 commits into
ajr/pldi-benchmarksfrom
ajr/pldi-benchmarks-fixes

Conversation

@rootjalex

Copy link
Copy Markdown
Owner

No description provided.

printf is variadic, so the default argument promotions apply: a float
arrives as a double and that is what "%f" reads. Pushing an f32 straight
through left the upper half of the argument undefined, so every float this
branch printed came out as 0.000000 -- correctness/llvm/print,
correctness/llvm/exported-function, correctness/llvm/defer-sum and
correctness/llvm/map all printed zeroes instead of their values.

Ported from the same fix on main.
generic_fib, mismatched_types and struct_duplicate_default record the source
line of the internal_assert that fires, and edits to Expr.cpp and Parser.cpp
moved those asserts without the goldens being updated:

  Expr.cpp    340 -> 341   (BinOp of non-number or boolean types)
  Expr.cpp    276 -> 277   (implicit casting of types)
  Parser.cpp  451 -> 469   (duplicate default value for field)

All three asserts have their condition on a single line, so the reported
number does not depend on the compiler -- GCC and Clang only disagree about
__LINE__ for a macro invocation that spans lines. The new values are byte
identical to main's goldens, which pass under both.
The backend now emits 'using float3 = vector<float, 3>'; these still said
vec3_float and would not compile.
@rootjalex
rootjalex merged commit ea1b861 into ajr/pldi-benchmarks Aug 30, 2026
0 of 2 checks passed
@rootjalex
rootjalex deleted the ajr/pldi-benchmarks-fixes branch August 30, 2026 18:48
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.

1 participant