Stream: git-wasmtime

Topic: wasmtime / issue #10331 Differential fuzzing failure betw...

Wasmtime GitHub notifications bot (Mar 05 2025 at 00:38):

fitzgen added the fuzz-bug label to Issue #10331.

Wasmtime GitHub notifications bot (Mar 05 2025 at 00:38):

fitzgen added the bug label to Issue #10331.

Wasmtime GitHub notifications bot (Mar 05 2025 at 00:38):

fitzgen opened issue #10331:

(module
  (func (export "test") (result v128)
    v128.const i64x2 32768 0
    call 1
  )
  (func (param v128) (result v128)
    local.get 0
    i16x8.extadd_pairwise_i8x16_s
  )
)

$ target/debug/wasmtime -Ccompiler=cranelift --invoke test test.wat
65408

$ target/debug/wasmtime -Ccompiler=winch --invoke test test.wat
0

cc @saulecabrera @jeffcharles

Wasmtime GitHub notifications bot (Mar 05 2025 at 17:09):

jeffcharles commented on issue #10331:

The use of self.v128_extend(src, writable!(tmp), low_kind)?; in the macroassembler implementation for extadd looks incorrect to me since that extends the values in low half of the vector and puts that extended lower half of the vector in the scratch register and my understanding of extadd is it should be extending the adjacent lane and then adding the result. I'll work on an implementation that avoids that.

Wasmtime GitHub notifications bot (Mar 06 2025 at 15:45):

saulecabrera closed issue #10331:

(module
  (func (export "test") (result v128)
    v128.const i64x2 32768 0
    call 1
  )
  (func (param v128) (result v128)
    local.get 0
    i16x8.extadd_pairwise_i8x16_s
  )
)

$ target/debug/wasmtime -Ccompiler=cranelift --invoke test test.wat
65408

$ target/debug/wasmtime -Ccompiler=winch --invoke test test.wat
0

cc @saulecabrera @jeffcharles

Last updated: Apr 18 2025 at 03:16 UTC