diff options
| -rw-r--r-- | src/wasm-interpreter.h | 6 | ||||
| m--------- | test/spec | 0 |
2 files changed, 6 insertions, 0 deletions
diff --git a/src/wasm-interpreter.h b/src/wasm-interpreter.h index adaee0b6f..fcf03c591 100644 --- a/src/wasm-interpreter.h +++ b/src/wasm-interpreter.h @@ -288,6 +288,12 @@ public: double val = value.getFloat(); if (std::isnan(val)) return Literal(float(val)); if (std::isinf(val)) return Literal(float(val)); + // when close to the limit, but still truncatable to a valid value, do that + // see https://github.com/WebAssembly/sexpr-wasm-prototype/blob/2d375e8d502327e814d62a08f22da9d9b6b675dc/src/wasm-interpreter.c#L247 + uint64_t bits = value.reinterpreti64(); + if (bits > 0x47efffffe0000000ULL && bits < 0x47effffff0000000ULL) return Literal(std::numeric_limits<float>::max()); + if (bits > 0xc7efffffe0000000ULL && bits < 0xc7effffff0000000ULL) return Literal(-std::numeric_limits<float>::max()); + // when we must convert to infinity, do that if (val < -std::numeric_limits<float>::max()) return Literal(-std::numeric_limits<float>::infinity()); if (val > std::numeric_limits<float>::max()) return Literal(std::numeric_limits<float>::infinity()); return value.truncateToF32(); diff --git a/test/spec b/test/spec -Subproject 88274ef4376b5a679a2cd0a1d630bf7e71f96cf +Subproject d4b150ce7268f903757ebdf1c7c001c99607da7 |