chore: update deps, properly feature gate simd

Signed-off-by: Henry Gressmann <mail@henrygressmann.de>

17 files changed, 282 insertions, 259 deletions

diff --git a/Cargo.lock b/Cargo.lock
index 48bfcf9..acdc41b 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -114,18 +114,18 @@ dependencies = [
 
 [[package]]
 name = "clap"
-version = "4.5.35"
+version = "4.5.36"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d8aa86934b44c19c50f87cc2790e19f54f7a67aedb64101c2e1a2e5ecfb73944"
+checksum = "2df961d8c8a0d08aa9945718ccf584145eee3f3aa06cddbeac12933781102e04"
 dependencies = [
  "clap_builder",
 ]
 
 [[package]]
 name = "clap_builder"
-version = "4.5.35"
+version = "4.5.36"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2414dbb2dd0695280da6ea9261e327479e9d37b0630f6b53ba2a11c60c679fd9"
+checksum = "132dbda40fb6753878316a489d5a1242a8ef2f0d9e47ba01c951ea8aa7d013a5"
 dependencies = [
  "anstyle",
  "clap_lex",
@@ -258,9 +258,9 @@ dependencies = [
 
 [[package]]
 name = "half"
-version = "2.5.0"
+version = "2.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7db2ff139bba50379da6aa0766b52fdcb62cb5b263009b09ed58ba604e14bbd1"
+checksum = "459196ed295495a68f7d7fe1d84f6c4b7ff0e21fe3017b2f283c6fac3ad803c9"
 dependencies = [
  "cfg-if",
  "crunchy",
@@ -311,9 +311,9 @@ checksum = "ce23b50ad8242c51a442f3ff322d56b02f08852c77e4c0b4d3fd684abc89c683"
 
 [[package]]
 name = "indexmap"
-version = "2.8.0"
+version = "2.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3954d50fe15b02142bf25d3b8bdadb634ec3948f103d04ffe3031bc8fe9d7058"
+checksum = "cea70ddb795996207ad57735b50c5982d8844f38ba9ee5f1aedcfb708a2aa11e"
 dependencies = [
  "equivalent",
  "hashbrown",
@@ -664,9 +664,9 @@ dependencies = [
 
 [[package]]
 name = "wasm-encoder"
-version = "0.228.0"
+version = "0.229.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "05d30290541f2d4242a162bbda76b8f2d8b1ac59eab3568ed6f2327d52c9b2c4"
+checksum = "38ba1d491ecacb085a2552025c10a675a6fddcbd03b1fc9b36c536010ce265d2"
 dependencies = [
  "leb128fmt",
  "wasmparser",
@@ -674,9 +674,9 @@ dependencies = [
 
 [[package]]
 name = "wasm-testsuite"
-version = "0.5.2"
+version = "0.5.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ec81df5b65371f919b41f3682861d98f1c90d081cfd8cb68784eb5d942dc1e36"
+checksum = "735333508bfc70a3cc7a7ed86b3cd53e4a775cbf78511eaec8ac4926ed3a0bf7"
 dependencies = [
  "include_dir",
  "wast",
@@ -684,9 +684,9 @@ dependencies = [
 
 [[package]]
 name = "wasmparser"
-version = "0.228.0"
+version = "0.229.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4abf1132c1fdf747d56bbc1bb52152400c70f336870f968b85e89ea422198ae3"
+checksum = "0cc3b1f053f5d41aa55640a1fa9b6d1b8a9e4418d118ce308d20e24ff3575a8c"
 dependencies = [
  "bitflags",
  "indexmap",
@@ -695,9 +695,9 @@ dependencies = [
 
 [[package]]
 name = "wast"
-version = "228.0.0"
+version = "229.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9e5aae124478cb51439f6587f074a3a5e835afd22751c59a87b2e2a882727c97"
+checksum = "63fcaff613c12225696bb163f79ca38ffb40e9300eff0ff4b8aa8b2f7eadf0d9"
 dependencies = [
  "bumpalo",
  "leb128fmt",
@@ -708,9 +708,9 @@ dependencies = [
 
 [[package]]
 name = "wat"
-version = "1.228.0"
+version = "1.229.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7ec29c89a8d055df988de7236483bf569988ac3d6905899f6af5ef920f9385ad"
+checksum = "4189bad08b70455a9e9e67dc126d2dcf91fac143a80f1046747a5dde6d4c33e0"
 dependencies = [
  "wast",
 ]
diff --git a/Cargo.toml b/Cargo.toml
index 56e0be1..1f645cf 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -4,9 +4,9 @@ default-members=[".", "crates/tinywasm", "crates/types", "crates/parser"]
 resolver="2"
 
 [workspace.dependencies]
-wast="228"
-wat="1.228"
-wasmparser={version="0.228", default-features=false}
+wast="229"
+wat="1.229"
+wasmparser={version="0.229", default-features=false}
 eyre="0.6"
 log="0.4"
 pretty_env_logger="0.5"
diff --git a/crates/parser/src/lib.rs b/crates/parser/src/lib.rs
index 1292fa6..aa33995 100644
--- a/crates/parser/src/lib.rs
+++ b/crates/parser/src/lib.rs
@@ -82,6 +82,7 @@ impl Parser {
             cm_async_stackful: false,
             cm_nested_names: false,
             cm_values: false,
+            cm_error_context: false,
         };
         Validator::new_with_features(features.into())
     }
diff --git a/crates/parser/src/module.rs b/crates/parser/src/module.rs
index ff6b22b..0e17ad9 100644
--- a/crates/parser/src/module.rs
+++ b/crates/parser/src/module.rs
@@ -130,7 +130,7 @@ impl ModuleReader {
                     return Err(ParseError::DuplicateSection("Code section".into()));
                 }
                 self.code.reserve(count as usize);
-                validator.code_section_start(count, &range)?;
+                validator.code_section_start(&range)?;
             }
             CodeSectionEntry(function) => {
                 debug!("Found code section entry");
diff --git a/crates/parser/src/visit.rs b/crates/parser/src/visit.rs
index 3e6fa4b..10005c0 100644
--- a/crates/parser/src/visit.rs
+++ b/crates/parser/src/visit.rs
@@ -46,7 +46,7 @@ pub(crate) fn process_operators_and_validate<R: WasmModuleResources>(
         reader.visit_operator(&mut ValidateThenVisit(reader.original_position(), &mut builder))??;
     }
 
-    builder.validator_finish(reader.original_position())?;
+    reader.finish()?;
     if !builder.errors.is_empty() {
         return Err(builder.errors.remove(0));
     }
@@ -124,10 +124,6 @@ impl<R: WasmModuleResources> FunctionBuilder<R> {
     ) -> impl VisitOperator<'_, Output = Result<(), wasmparser::BinaryReaderError>> + VisitSimdOperator<'_> {
         self.validator.simd_visitor(offset)
     }
-
-    pub(crate) fn validator_finish(&mut self, offset: usize) -> Result<(), wasmparser::BinaryReaderError> {
-        self.validator.finish(offset)
-    }
 }
 
 impl<R: WasmModuleResources> FunctionBuilder<R> {
diff --git a/crates/tinywasm/Cargo.toml b/crates/tinywasm/Cargo.toml
index 5810f0e..3281476 100644
--- a/crates/tinywasm/Cargo.toml
+++ b/crates/tinywasm/Cargo.toml
@@ -20,7 +20,7 @@ tinywasm-types={version="0.9.0-alpha.0", path="../types", default-features=false
 libm={version="0.2", default-features=false}
 
 [dev-dependencies]
-wasm-testsuite={version="0.5.2"}
+wasm-testsuite={version="0.5.3"}
 indexmap="2.7"
 wast={workspace=true}
 wat={workspace=true}
@@ -32,12 +32,14 @@ serde_json={version="1.0"}
 serde={version="1.0", features=["derive"]}
 
 [features]
-default=["std", "parser", "logging", "archive", "simd"]
+default=["std", "parser", "logging", "archive"]
 logging=["log", "tinywasm-parser?/logging", "tinywasm-types/logging"]
 std=["tinywasm-parser?/std", "tinywasm-types/std"]
 parser=["dep:tinywasm-parser"]
 archive=["tinywasm-types/archive"]
-simd=[]
+
+# enable simd support (unstable / unfinished)
+__simd=[]
 
 [[test]]
 name="test-wasm-1"
diff --git a/crates/tinywasm/src/interpreter/executor.rs b/crates/tinywasm/src/interpreter/executor.rs
index a0153af..4bf4f5d 100644
--- a/crates/tinywasm/src/interpreter/executor.rs
+++ b/crates/tinywasm/src/interpreter/executor.rs
@@ -3,16 +3,19 @@
 use super::no_std_floats::NoStdFloatExt;
 
 use alloc::{format, rc::Rc, string::ToString};
-use core::ops::{ControlFlow, Index, IndexMut, Shl, Shr};
+use core::ops::ControlFlow;
 
 use interpreter::stack::CallFrame;
 use tinywasm_types::*;
 
-#[cfg(all(feature = "std", feature = "simd"))]
+#[cfg(all(feature = "std", feature = "__simd"))]
 use crate::std::simd::StdFloat;
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 use core::simd::{cmp::*, num::*, *};
 
+#[cfg(feature = "__simd")]
+use core::ops::{Index, IndexMut, Shl, Shr};
+
 use super::num_helpers::*;
 use super::stack::{BlockFrame, BlockType, Stack};
 use super::values::*;
@@ -313,234 +316,235 @@ impl<'store, 'stack> Executor<'store, 'stack> {
             LocalCopy128(from, to) => self.exec_local_copy::<Value128>(*from, *to),
             LocalCopyRef(from, to) => self.exec_local_copy::<ValueRef>(*from, *to),
 
-            V128Not => self.stack.values.replace_top_same::<Value128>(|v| Ok(!v)).to_cf()?,
-            V128And => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a & b)).to_cf()?,
-            V128AndNot => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a & (!b))).to_cf()?,
-            V128Or => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a | b)).to_cf()?,
-            V128Xor => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a ^ b)).to_cf()?,
-            V128Bitselect => self.stack.values.calculate_same_3::<Value128>(|v1, v2, c| Ok((v1 & c) | (v2 & !c))).to_cf()?,
-            V128AnyTrue => self.stack.values.replace_top::<Value128, i32>(|v| Ok((v.reduce_or() != 0) as i32)).to_cf()?,
-            I8x16Swizzle => self.stack.values.calculate_same::<Value128>(|a, s| Ok(a.swizzle_dyn(s))).to_cf()?,
-            V128Load(arg) => self.exec_mem_load::<Value128, 16, _>(arg.mem_addr(), arg.offset(), |v| v)?,
-            V128Store(arg) => self.exec_mem_store::<Value128, Value128, 16>(arg.mem_addr(), arg.offset(), |v| v)?,
+            #[cfg(feature = "__simd")] V128Not => self.stack.values.replace_top_same::<Value128>(|v| Ok(!v)).to_cf()?,
+            #[cfg(feature = "__simd")] V128And => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a & b)).to_cf()?,
+            #[cfg(feature = "__simd")] V128AndNot => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a & (!b))).to_cf()?,
+            #[cfg(feature = "__simd")] V128Or => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a | b)).to_cf()?,
+            #[cfg(feature = "__simd")] V128Xor => self.stack.values.calculate_same::<Value128>(|a, b| Ok(a ^ b)).to_cf()?,
+            #[cfg(feature = "__simd")] V128Bitselect => self.stack.values.calculate_same_3::<Value128>(|v1, v2, c| Ok((v1 & c) | (v2 & !c))).to_cf()?,
+            #[cfg(feature = "__simd")] V128AnyTrue => self.stack.values.replace_top::<Value128, i32>(|v| Ok((v.reduce_or() != 0) as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Swizzle => self.stack.values.calculate_same::<Value128>(|a, s| Ok(a.swizzle_dyn(s))).to_cf()?,
+            #[cfg(feature = "__simd")] V128Load(arg) => self.exec_mem_load::<Value128, 16, _>(arg.mem_addr(), arg.offset(), |v| v)?,
+            #[cfg(feature = "__simd")] V128Store(arg) => self.exec_mem_store::<Value128, Value128, 16>(arg.mem_addr(), arg.offset(), |v| v)?,
 
-            V128Store8Lane(arg, lane) => self.exec_mem_store_lane::<i8x16, i8, 1>(arg.mem_addr(), arg.offset(), *lane)?,
-            V128Store16Lane(arg, lane) => self.exec_mem_store_lane::<i16x8, i16, 2>(arg.mem_addr(), arg.offset(), *lane)?,
-            V128Store32Lane(arg, lane) => self.exec_mem_store_lane::<i32x4, i32, 4>(arg.mem_addr(), arg.offset(), *lane)?,
-            V128Store64Lane(arg, lane) => self.exec_mem_store_lane::<i64x2, i64, 8>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Store8Lane(arg, lane) => self.exec_mem_store_lane::<i8x16, i8, 1>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Store16Lane(arg, lane) => self.exec_mem_store_lane::<i16x8, i16, 2>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Store32Lane(arg, lane) => self.exec_mem_store_lane::<i32x4, i32, 4>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Store64Lane(arg, lane) => self.exec_mem_store_lane::<i64x2, i64, 8>(arg.mem_addr(), arg.offset(), *lane)?,
 
             // Load a single 32-bit or 64-bit element into the lowest bits of a v128 vector, and initialize all other bits of the v128 vector to zero.
-            V128Load32Zero(arg) => self.exec_mem_load::<i32, 4, Value128>(arg.mem_addr(), arg.offset(), |v|  {
+            #[cfg(feature = "__simd")] V128Load32Zero(arg) => self.exec_mem_load::<i32, 4, Value128>(arg.mem_addr(), arg.offset(), |v|  {
                 let bytes = v.to_le_bytes();
                 u8x16::from_array([bytes[0], bytes[1], bytes[2], bytes[3], 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])
             })?,
-            V128Load64Zero(arg) => self.exec_mem_load::<i64, 8, Value128>(arg.mem_addr(), arg.offset(), |v|  {
+            #[cfg(feature = "__simd")] V128Load64Zero(arg) => self.exec_mem_load::<i64, 8, Value128>(arg.mem_addr(), arg.offset(), |v|  {
                 let bytes = v.to_le_bytes();
                 u8x16::from_array([bytes[0], bytes[1], bytes[2], bytes[3], bytes[4], bytes[5], bytes[6], bytes[7], 0, 0, 0, 0, 0, 0, 0, 0])
             })?,
 
-            V128Const(arg) => self.exec_const::<Value128>( self.cf.data().v128_constants[*arg as usize].to_le_bytes().into()),
+            #[cfg(feature = "__simd")] V128Const(arg) => self.exec_const::<Value128>( self.cf.data().v128_constants[*arg as usize].to_le_bytes().into()),
 
-            I8x16ExtractLaneS(lane) => self.stack.values.replace_top::<i8x16, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
-            I8x16ExtractLaneU(lane) => self.stack.values.replace_top::<u8x16, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
-            I16x8ExtractLaneS(lane) => self.stack.values.replace_top::<i16x8, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
-            I16x8ExtractLaneU(lane) => self.stack.values.replace_top::<u16x8, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
-            I32x4ExtractLane(lane) => self.stack.values.replace_top::<i32x4, i32>(|v| Ok(v[*lane as usize])).to_cf()?,
-            I64x2ExtractLane(lane) => self.stack.values.replace_top::<i64x2, i64>(|v| Ok(v[*lane as usize])).to_cf()?,
-            F32x4ExtractLane(lane) => self.stack.values.replace_top::<f32x4, f32>(|v| Ok(v[*lane as usize])).to_cf()?,
-            F64x2ExtractLane(lane) => self.stack.values.replace_top::<f64x2, f64>(|v| Ok(v[*lane as usize])).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16ExtractLaneS(lane) => self.stack.values.replace_top::<i8x16, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16ExtractLaneU(lane) => self.stack.values.replace_top::<u8x16, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8ExtractLaneS(lane) => self.stack.values.replace_top::<i16x8, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8ExtractLaneU(lane) => self.stack.values.replace_top::<u16x8, i32>(|v| Ok(v[*lane as usize] as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4ExtractLane(lane) => self.stack.values.replace_top::<i32x4, i32>(|v| Ok(v[*lane as usize])).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2ExtractLane(lane) => self.stack.values.replace_top::<i64x2, i64>(|v| Ok(v[*lane as usize])).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4ExtractLane(lane) => self.stack.values.replace_top::<f32x4, f32>(|v| Ok(v[*lane as usize])).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2ExtractLane(lane) => self.stack.values.replace_top::<f64x2, f64>(|v| Ok(v[*lane as usize])).to_cf()?,
 
-            V128Load8Lane(arg, lane) => self.exec_mem_load_lane::<i8, i8x16, 1>(arg.mem_addr(), arg.offset(), *lane)?,
-            V128Load16Lane(arg, lane) => self.exec_mem_load_lane::<i16, i16x8, 2>(arg.mem_addr(), arg.offset(), *lane)?,
-            V128Load32Lane(arg, lane) => self.exec_mem_load_lane::<i32, i32x4, 4>(arg.mem_addr(), arg.offset(), *lane)?,
-            V128Load64Lane(arg, lane) => self.exec_mem_load_lane::<i64, i64x2, 8>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Load8Lane(arg, lane) => self.exec_mem_load_lane::<i8, i8x16, 1>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Load16Lane(arg, lane) => self.exec_mem_load_lane::<i16, i16x8, 2>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Load32Lane(arg, lane) => self.exec_mem_load_lane::<i32, i32x4, 4>(arg.mem_addr(), arg.offset(), *lane)?,
+            #[cfg(feature = "__simd")] V128Load64Lane(arg, lane) => self.exec_mem_load_lane::<i64, i64x2, 8>(arg.mem_addr(), arg.offset(), *lane)?,
 
-            I8x16Splat => self.stack.values.replace_top::<i32, i8x16>(|v| Ok(Simd::<i8, 16>::splat(v as i8))).to_cf()?,
-            I16x8Splat => self.stack.values.replace_top::<i32, i16x8>(|v| Ok(Simd::<i16, 8>::splat(v as i16))).to_cf()?,
-            I32x4Splat => self.stack.values.replace_top::<i32, i32x4>(|v| Ok(Simd::<i32, 4>::splat(v))).to_cf()?,
-            I64x2Splat => self.stack.values.replace_top::<i64, i64x2>(|v| Ok(Simd::<i64, 2>::splat(v))).to_cf()?,
-            F32x4Splat => self.stack.values.replace_top::<f32, f32x4>(|v| Ok(Simd::<f32, 4>::splat(v))).to_cf()?,
-            F64x2Splat => self.stack.values.replace_top::<f64, f64x2>(|v| Ok(Simd::<f64, 2>::splat(v))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Splat => self.stack.values.replace_top::<i32, i8x16>(|v| Ok(Simd::<i8, 16>::splat(v as i8))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Splat => self.stack.values.replace_top::<i32, i16x8>(|v| Ok(Simd::<i16, 8>::splat(v as i16))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Splat => self.stack.values.replace_top::<i32, i32x4>(|v| Ok(Simd::<i32, 4>::splat(v))).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Splat => self.stack.values.replace_top::<i64, i64x2>(|v| Ok(Simd::<i64, 2>::splat(v))).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Splat => self.stack.values.replace_top::<f32, f32x4>(|v| Ok(Simd::<f32, 4>::splat(v))).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Splat => self.stack.values.replace_top::<f64, f64x2>(|v| Ok(Simd::<f64, 2>::splat(v))).to_cf()?,
 
-            I8x16Eq => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
-            I16x8Eq => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
-            I32x4Eq => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
-            F32x4Eq => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
-            F64x2Eq => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Eq => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Eq => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Eq => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Eq => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Eq => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_eq(b).to_int())).to_cf()?,
 
-            I8x16Ne => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
-            I16x8Ne => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
-            I32x4Ne => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
-            F32x4Ne => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
-            F64x2Ne => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Ne => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Ne => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Ne => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Ne => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Ne => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_ne(b).to_int())).to_cf()?,
 
-            I8x16LtS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            I16x8LtS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            I32x4LtS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            I64x2LtS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            I8x16LtU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            I16x8LtU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            I32x4LtU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            F32x4Lt => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
-            F64x2Lt => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16LtS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8LtS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4LtS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2LtS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16LtU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8LtU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4LtU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Lt => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Lt => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_lt(b).to_int())).to_cf()?,
 
-            I64x2GtS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            F32x4Gt => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            F64x2Gt => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2GtS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Gt => self.stack.values.calculate::<f32x4, _>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Gt => self.stack.values.calculate::<f64x2, _>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
 
-            I8x16GtS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            I16x8GtS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            I32x4GtS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            I64x2LeS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
-            F32x4Le => self.stack.values.calculate::<f32x4,_>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
-            F64x2Le => self.stack.values.calculate::<f64x2,_>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16GtS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8GtS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4GtS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2LeS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Le => self.stack.values.calculate::<f32x4,_>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Le => self.stack.values.calculate::<f64x2,_>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
 
-            I8x16GtU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            I16x8GtU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            I32x4GtU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
-            I64x2GeS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
-            F32x4Ge => self.stack.values.calculate::<f32x4,_>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
-            F64x2Ge => self.stack.values.calculate::<f64x2,_>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16GtU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8GtU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4GtU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_gt(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2GeS => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Ge => self.stack.values.calculate::<f32x4,_>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Ge => self.stack.values.calculate::<f64x2,_>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
 
-            I8x16LeS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
-            I16x8LeS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
-            I32x4LeS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16LeS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8LeS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4LeS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
 
-            I8x16LeU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
-            I16x8LeU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
-            I32x4LeU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16LeU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8LeU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4LeU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_le(b).to_int())).to_cf()?,
 
-            I8x16GeS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
-            I16x8GeS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
-            I32x4GeS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16GeS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8GeS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4GeS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
 
-            I8x16GeU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
-            I16x8GeU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
-            I32x4GeU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16GeU => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8GeU => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4GeU => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_ge(b).to_int())).to_cf()?,
 
-            I8x16Abs => self.stack.values.replace_top_same::<i8x16>(|a| Ok(a.abs())).to_cf()?,
-            I16x8Abs => self.stack.values.replace_top_same::<i16x8>(|a| Ok(a.abs())).to_cf()?,
-            I32x4Abs => self.stack.values.replace_top_same::<i32x4>(|a| Ok(a.abs())).to_cf()?,
-            I64x2Abs => self.stack.values.replace_top_same::<i64x2>(|a| Ok(a.abs())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Abs => self.stack.values.replace_top_same::<i8x16>(|a| Ok(a.abs())).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Abs => self.stack.values.replace_top_same::<i16x8>(|a| Ok(a.abs())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Abs => self.stack.values.replace_top_same::<i32x4>(|a| Ok(a.abs())).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Abs => self.stack.values.replace_top_same::<i64x2>(|a| Ok(a.abs())).to_cf()?,
 
-            I8x16Neg => self.stack.values.replace_top_same::<i8x16>(|a| Ok(-a)).to_cf()?,
-            I16x8Neg => self.stack.values.replace_top_same::<i16x8>(|a| Ok(-a)).to_cf()?,
-            I32x4Neg => self.stack.values.replace_top_same::<i32x4>(|a| Ok(-a)).to_cf()?,
-            I64x2Neg => self.stack.values.replace_top_same::<i64x2>(|a| Ok(-a)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Neg => self.stack.values.replace_top_same::<i8x16>(|a| Ok(-a)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Neg => self.stack.values.replace_top_same::<i16x8>(|a| Ok(-a)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Neg => self.stack.values.replace_top_same::<i32x4>(|a| Ok(-a)).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Neg => self.stack.values.replace_top_same::<i64x2>(|a| Ok(-a)).to_cf()?,
 
-            I8x16AllTrue => self.stack.values.replace_top::<i8x16, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
-            I16x8AllTrue => self.stack.values.replace_top::<i16x8, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
-            I32x4AllTrue => self.stack.values.replace_top::<i32x4, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
-            I64x2AllTrue => self.stack.values.replace_top::<i64x2, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16AllTrue => self.stack.values.replace_top::<i8x16, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8AllTrue => self.stack.values.replace_top::<i16x8, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4AllTrue => self.stack.values.replace_top::<i32x4, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2AllTrue => self.stack.values.replace_top::<i64x2, i32>(|v| Ok((v.simd_ne(Simd::splat(0)).all()) as i32)).to_cf()?,
 
-            I8x16Bitmask => self.stack.values.replace_top::<i8x16, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
-            I16x8Bitmask => self.stack.values.replace_top::<i16x8, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
-            I32x4Bitmask => self.stack.values.replace_top::<i32x4, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
-            I64x2Bitmask => self.stack.values.replace_top::<i64x2, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Bitmask => self.stack.values.replace_top::<i8x16, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Bitmask => self.stack.values.replace_top::<i16x8, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Bitmask => self.stack.values.replace_top::<i32x4, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Bitmask => self.stack.values.replace_top::<i64x2, i32>(|v| Ok(v.simd_lt(Simd::splat(0)).to_bitmask() as i32)).to_cf()?,
 
-            I8x16Shl => self.stack.values.calculate_diff::<i32, i8x16, i8x16>(|a, b| Ok(b.shl(a as i8))).to_cf()?,
-            I16x8Shl => self.stack.values.calculate_diff::<i32, i16x8, i16x8>(|a, b| Ok(b.shl(a as i16))).to_cf()?,
-            I32x4Shl => self.stack.values.calculate_diff::<i32, i32x4, i32x4>(|a, b| Ok(b.shl(a))).to_cf()?,
-            I64x2Shl => self.stack.values.calculate_diff::<i32, i64x2, i64x2>(|a, b| Ok(b.shl(a as i64))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Shl => self.stack.values.calculate_diff::<i32, i8x16, i8x16>(|a, b| Ok(b.shl(a as i8))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Shl => self.stack.values.calculate_diff::<i32, i16x8, i16x8>(|a, b| Ok(b.shl(a as i16))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Shl => self.stack.values.calculate_diff::<i32, i32x4, i32x4>(|a, b| Ok(b.shl(a))).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Shl => self.stack.values.calculate_diff::<i32, i64x2, i64x2>(|a, b| Ok(b.shl(a as i64))).to_cf()?,
 
-            I8x16ShrS => self.stack.values.calculate_diff::<i32, i8x16, i8x16>(|a, b| Ok(b.shr(a as i8))).to_cf()?,
-            I16x8ShrS => self.stack.values.calculate_diff::<i32, i16x8, i16x8>(|a, b| Ok(b.shr(a as i16))).to_cf()?,
-            I32x4ShrS => self.stack.values.calculate_diff::<i32, i32x4, i32x4>(|a, b| Ok(b.shr(a))).to_cf()?,
-            I64x2ShrS => self.stack.values.calculate_diff::<i32, i64x2, i64x2>(|a, b| Ok(b.shr(a as i64))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16ShrS => self.stack.values.calculate_diff::<i32, i8x16, i8x16>(|a, b| Ok(b.shr(a as i8))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8ShrS => self.stack.values.calculate_diff::<i32, i16x8, i16x8>(|a, b| Ok(b.shr(a as i16))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4ShrS => self.stack.values.calculate_diff::<i32, i32x4, i32x4>(|a, b| Ok(b.shr(a))).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2ShrS => self.stack.values.calculate_diff::<i32, i64x2, i64x2>(|a, b| Ok(b.shr(a as i64))).to_cf()?,
 
-            I8x16ShrU => self.stack.values.calculate_diff::<i32, u8x16, u8x16>(|a, b| Ok(b.shr(a as u8))).to_cf()?,
-            I16x8ShrU => self.stack.values.calculate_diff::<i32, u16x8, u16x8>(|a, b| Ok(b.shr(a as u16))).to_cf()?,
-            I32x4ShrU => self.stack.values.calculate_diff::<i32, u32x4, u32x4>(|a, b| Ok(b.shr(a as u32))).to_cf()?,
-            I64x2ShrU => self.stack.values.calculate_diff::<i32, u64x2, u64x2>(|a, b| Ok(b.shr(a as u64))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16ShrU => self.stack.values.calculate_diff::<i32, u8x16, u8x16>(|a, b| Ok(b.shr(a as u8))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8ShrU => self.stack.values.calculate_diff::<i32, u16x8, u16x8>(|a, b| Ok(b.shr(a as u16))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4ShrU => self.stack.values.calculate_diff::<i32, u32x4, u32x4>(|a, b| Ok(b.shr(a as u32))).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2ShrU => self.stack.values.calculate_diff::<i32, u64x2, u64x2>(|a, b| Ok(b.shr(a as u64))).to_cf()?,
 
-            I8x16Add => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a + b)).to_cf()?,
-            I16x8Add => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a + b)).to_cf()?,
-            I32x4Add => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a + b)).to_cf()?,
-            I64x2Add => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a + b)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Add => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a + b)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Add => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a + b)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Add => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a + b)).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Add => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a + b)).to_cf()?,
 
-            I8x16Sub => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a - b)).to_cf()?,
-            I16x8Sub => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a - b)).to_cf()?,
-            I32x4Sub => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a - b)).to_cf()?,
-            I64x2Sub => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a - b)).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Sub => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a - b)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Sub => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a - b)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Sub => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a - b)).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Sub => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a - b)).to_cf()?,
 
-            I8x16MinS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_min(b))).to_cf()?,
-            I16x8MinS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_min(b))).to_cf()?,
-            I32x4MinS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_min(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16MinS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_min(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8MinS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_min(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4MinS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_min(b))).to_cf()?,
 
-            I8x16MinU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.simd_min(b))).to_cf()?,
-            I16x8MinU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.simd_min(b))).to_cf()?,
-            I32x4MinU => self.stack.values.calculate_same::<u32x4>(|a, b| Ok(a.simd_min(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16MinU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.simd_min(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8MinU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.simd_min(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4MinU => self.stack.values.calculate_same::<u32x4>(|a, b| Ok(a.simd_min(b))).to_cf()?,
 
-            I8x16MaxS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_max(b))).to_cf()?,
-            I16x8MaxS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_max(b))).to_cf()?,
-            I32x4MaxS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_max(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16MaxS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.simd_max(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8MaxS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.simd_max(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4MaxS => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a.simd_max(b))).to_cf()?,
 
-            I8x16MaxU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.simd_max(b))).to_cf()?,
-            I16x8MaxU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.simd_max(b))).to_cf()?,
-            I32x4MaxU => self.stack.values.calculate_same::<u32x4>(|a, b| Ok(a.simd_max(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16MaxU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.simd_max(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8MaxU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.simd_max(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4MaxU => self.stack.values.calculate_same::<u32x4>(|a, b| Ok(a.simd_max(b))).to_cf()?,
 
-            I64x2Mul => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a * b)).to_cf()?,
-            I16x8Mul => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a * b)).to_cf()?,
-            I32x4Mul => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a * b)).to_cf()?,
+            #[cfg(feature = "__simd")] I64x2Mul => self.stack.values.calculate_same::<i64x2>(|a, b| Ok(a * b)).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8Mul => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a * b)).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4Mul => self.stack.values.calculate_same::<i32x4>(|a, b| Ok(a * b)).to_cf()?,
 
-            I8x16NarrowI16x8S => unimplemented!(),
-            I8x16NarrowI16x8U => unimplemented!(),
-            I16x8NarrowI32x4S => unimplemented!(),
-            I16x8NarrowI32x4U => unimplemented!(),
+            #[cfg(feature = "__simd")] I8x16NarrowI16x8S => unimplemented!(),
+            #[cfg(feature = "__simd")] I8x16NarrowI16x8U => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8NarrowI32x4S => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8NarrowI32x4U => unimplemented!(),
 
-            I8x16AddSatS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
-            I16x8AddSatS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
-            I8x16AddSatU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
-            I16x8AddSatU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
-            I8x16SubSatS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
-            I16x8SubSatS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
-            I8x16SubSatU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
-            I16x8SubSatU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16AddSatS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8AddSatS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16AddSatU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8AddSatU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.saturating_add(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16SubSatS => self.stack.values.calculate_same::<i8x16>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8SubSatS => self.stack.values.calculate_same::<i16x8>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16SubSatU => self.stack.values.calculate_same::<u8x16>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
+            #[cfg(feature = "__simd")] I16x8SubSatU => self.stack.values.calculate_same::<u16x8>(|a, b| Ok(a.saturating_sub(b))).to_cf()?,
 
-            I16x8ExtAddPairwiseI8x16S => unimplemented!(),
-            I16x8ExtAddPairwiseI8x16U => unimplemented!(),
-            I32x4ExtAddPairwiseI16x8S => unimplemented!(),
-            I32x4ExtAddPairwiseI16x8U => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtAddPairwiseI8x16S => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtAddPairwiseI8x16U => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtAddPairwiseI16x8S => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtAddPairwiseI16x8U => unimplemented!(),
 
-            I16x8ExtMulLowI8x16S => unimplemented!(),
-            I16x8ExtMulLowI8x16U => unimplemented!(),
-            I16x8ExtMulHighI8x16S => unimplemented!(),
-            I16x8ExtMulHighI8x16U => unimplemented!(),
-            I32x4ExtMulLowI16x8S => unimplemented!(),
-            I32x4ExtMulLowI16x8U => unimplemented!(),
-            I32x4ExtMulHighI16x8S => unimplemented!(),
-            I32x4ExtMulHighI16x8U => unimplemented!(),
-            I64x2ExtMulLowI32x4S => unimplemented!(),
-            I64x2ExtMulLowI32x4U => unimplemented!(),
-            I64x2ExtMulHighI32x4S => unimplemented!(),
-            I64x2ExtMulHighI32x4U => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtMulLowI8x16S => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtMulLowI8x16U => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtMulHighI8x16S => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtMulHighI8x16U => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtMulLowI16x8S => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtMulLowI16x8U => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtMulHighI16x8S => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtMulHighI16x8U => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtMulLowI32x4S => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtMulLowI32x4U => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtMulHighI32x4S => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtMulHighI32x4U => unimplemented!(),
 
-            I16x8ExtendLowI8x16S => unimplemented!(),
-            I16x8ExtendLowI8x16U => unimplemented!(),
-            I16x8ExtendHighI8x16S => unimplemented!(),
-            I16x8ExtendHighI8x16U => unimplemented!(),
-            I32x4ExtendLowI16x8S => unimplemented!(),
-            I32x4ExtendLowI16x8U => unimplemented!(),
-            I32x4ExtendHighI16x8S => unimplemented!(),
-            I32x4ExtendHighI16x8U => unimplemented!(),
-            I64x2ExtendLowI32x4S => unimplemented!(),
-            I64x2ExtendLowI32x4U => unimplemented!(),
-            I64x2ExtendHighI32x4S => unimplemented!(),
-            I64x2ExtendHighI32x4U => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtendLowI8x16S => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtendLowI8x16U => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtendHighI8x16S => unimplemented!(),
+            #[cfg(feature = "__simd")] I16x8ExtendHighI8x16U => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtendLowI16x8S => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtendLowI16x8U => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtendHighI16x8S => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4ExtendHighI16x8U => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtendLowI32x4S => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtendLowI32x4U => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtendHighI32x4S => unimplemented!(),
+            #[cfg(feature = "__simd")] I64x2ExtendHighI32x4U => unimplemented!(),
 
-            I8x16Popcnt => self.stack.values.replace_top::<i8x16, _>(|v| Ok(v.count_ones())).to_cf()?,
+            #[cfg(feature = "__simd")] I8x16Popcnt => self.stack.values.replace_top::<i8x16, _>(|v| Ok(v.count_ones())).to_cf()?,
 
+            #[cfg(feature = "__simd")]
             I16x8Q15MulrSatS =>  self.stack.values.calculate_same::<i16x8>(|a, b| {
                 let subq15mulr = |a,b| {
                     let a = a as i32;
                     let b = b as i32;
                     let r = (a * b + 0x4000) >> 15;
                     if r > i16::MAX as i32 {
-                        i16::MAX
+                         i16::MAX
                     } else if r < i16::MIN as i32 {
-                        i16::MIN
+                         i16::MIN
                     } else {
-                        r as i16
+                         r as i16
                     }
                 };
                 Ok(Simd::<i16, 8>::from_array([
@@ -555,6 +559,7 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                 ]))
             }).to_cf()?,
 
+            #[cfg(feature = "__simd")]
             I32x4DotI16x8S => self.stack.values.calculate::<i16x8, i32x4>(|a, b| {
                 Ok(Simd::<i32, 4>::from_array([
                     i32::from(a[0] * b[0] + a[1] * b[1]),
@@ -564,26 +569,27 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                 ]))
             }).to_cf()?,
 
-            F32x4Ceil => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.ceil())).to_cf()?,
-            F64x2Ceil => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.ceil())).to_cf()?,
-            F32x4Floor => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.floor())).to_cf()?,
-            F64x2Floor => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.floor())).to_cf()?,
-            F32x4Trunc => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.trunc())).to_cf()?,
-            F64x2Trunc => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.trunc())).to_cf()?,
-            F32x4Abs => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.abs())).to_cf()?,
-            F64x2Abs => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.abs())).to_cf()?,
-            F32x4Neg => self.stack.values.replace_top_same::<f32x4>(|v| Ok(-v)).to_cf()?,
-            F64x2Neg => self.stack.values.replace_top_same::<f64x2>(|v| Ok(-v)).to_cf()?,
-            F32x4Sqrt => self.stack.values.replace_top_same::<f32x4>(|v| Ok(canonicalize_f32x4(v.sqrt()))).to_cf()?,
-            F64x2Sqrt => self.stack.values.replace_top_same::<f64x2>(|v| Ok(canonicalize_f64x2(v.sqrt()))).to_cf()?,
-            F32x4Add => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a + b))).to_cf()?,
-            F64x2Add => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a + b))).to_cf()?,
-            F32x4Sub => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a - b))).to_cf()?,
-            F64x2Sub => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a - b))).to_cf()?,
-            F32x4Mul => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a * b))).to_cf()?,
-            F64x2Mul => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a * b))).to_cf()?,
-            F32x4Div => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a / b))).to_cf()?,
-            F64x2Div => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a / b))).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Ceil => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.ceil())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Ceil => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.ceil())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Floor => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.floor())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Floor => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.floor())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Trunc => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.trunc())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Trunc => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.trunc())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Abs => self.stack.values.replace_top_same::<f32x4>(|v| Ok(v.abs())).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Abs => self.stack.values.replace_top_same::<f64x2>(|v| Ok(v.abs())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Neg => self.stack.values.replace_top_same::<f32x4>(|v| Ok(-v)).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Neg => self.stack.values.replace_top_same::<f64x2>(|v| Ok(-v)).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Sqrt => self.stack.values.replace_top_same::<f32x4>(|v| Ok(canonicalize_f32x4(v.sqrt()))).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Sqrt => self.stack.values.replace_top_same::<f64x2>(|v| Ok(canonicalize_f64x2(v.sqrt()))).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Add => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a + b))).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Add => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a + b))).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Sub => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a - b))).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Sub => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a - b))).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Mul => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a * b))).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Mul => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a * b))).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4Div => self.stack.values.calculate_same::<f32x4>(|a, b| Ok(canonicalize_f32x4(a / b))).to_cf()?,
+            #[cfg(feature = "__simd")] F64x2Div => self.stack.values.calculate_same::<f64x2>(|a, b| Ok(canonicalize_f64x2(a / b))).to_cf()?,
+            #[cfg(feature = "__simd")]
             F32x4Min => self.stack.values.calculate_same::<f32x4>(|a, b| {
                 Ok(Simd::<f32, 4>::from_array([
                     b[0].tw_minimum(a[0]),
@@ -592,12 +598,16 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                     b[3].tw_minimum(a[3]),
                 ]))
             }).to_cf()?,
+
+            #[cfg(feature = "__simd")]
             F64x2Min => self.stack.values.calculate_same::<f64x2>(|a, b| {
                 Ok(Simd::<f64, 2>::from_array([
                     b[0].tw_minimum(a[0]),
                     b[1].tw_minimum(a[1]),
                 ]))
             }).to_cf()?,
+
+            #[cfg(feature = "__simd")]
             F32x4Max => self.stack.values.calculate_same::<f32x4>(|a, b| {
                 Ok(Simd::<f32, 4>::from_array([
                     b[0].tw_maximum(a[0]),
@@ -606,6 +616,8 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                     b[3].tw_maximum(a[3]),
                 ]))
             }).to_cf()?,
+
+            #[cfg(feature = "__simd")]
             F64x2Max => self.stack.values.calculate_same::<f64x2>(|a, b| {
                 Ok(Simd::<f64, 2>::from_array([
                     b[0].tw_maximum(a[0]),
@@ -613,6 +625,7 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                 ]))
             }).to_cf()?,
 
+            #[cfg(feature = "__simd")]
             F32x4PMin => self.stack.values.calculate_same::<f32x4>(|a, b| {
                 Ok(Simd::<f32, 4>::from_array([
                     if b[0] < a[0] { b[0] } else { a[0]},
@@ -621,6 +634,8 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                     if b[3] < a[3] { b[3] } else { a[3]},
                 ]))
             }).to_cf()?,
+
+            #[cfg(feature = "__simd")]
             F32x4PMax => self.stack.values.calculate_same::<f32x4>(|a, b| {
                 Ok(Simd::<f32, 4>::from_array([
                     if b[0] > a[0] { b[0] } else { a[0]},
@@ -629,12 +644,16 @@ impl<'store, 'stack> Executor<'store, 'stack> {
                     if b[3] > a[3] { b[3] } else { a[3]},
                 ]))
             }).to_cf()?,
+
+            #[cfg(feature = "__simd")]
             F64x2PMin => self.stack.values.calculate_same::<f64x2>(|a, b| {
                 Ok(Simd::<f64, 2>::from_array([
                     if b[0] < a[0] { b[0] } else { a[0]},
                     if b[1] < a[1] { b[1] } else { a[1]},
                 ]))
             }).to_cf()?,
+
+            #[cfg(feature = "__simd")]
             F64x2PMax => self.stack.values.calculate_same::<f64x2>(|a, b| {
                 Ok(Simd::<f64, 2>::from_array([
                     if b[0] > a[0] { b[0] } else { a[0]},
@@ -643,16 +662,16 @@ impl<'store, 'stack> Executor<'store, 'stack> {
             }).to_cf()?,
 
             // not correct
-            I32x4TruncSatF32x4S => self.stack.values.replace_top::<f32x4, f32x4>(|v| Ok(v.trunc())).to_cf()?,
-            I32x4TruncSatF32x4U => self.stack.values.replace_top::<f32x4, f32x4>(|v| Ok(v.trunc())).to_cf()?,
-            F32x4ConvertI32x4S => {},
-            F32x4ConvertI32x4U => {},
-            F64x2ConvertLowI32x4S => {},
-            F64x2ConvertLowI32x4U => {},
-            F32x4DemoteF64x2Zero => {},
-            F64x2PromoteLowF32x4 => {},
-            I32x4TruncSatF64x2SZero => unimplemented!(),
-            I32x4TruncSatF64x2UZero => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4TruncSatF32x4S => self.stack.values.replace_top::<f32x4, f32x4>(|v| Ok(v.trunc())).to_cf()?,
+            #[cfg(feature = "__simd")] I32x4TruncSatF32x4U => self.stack.values.replace_top::<f32x4, f32x4>(|v| Ok(v.trunc())).to_cf()?,
+            #[cfg(feature = "__simd")] F32x4ConvertI32x4S => {},
+            #[cfg(feature = "__simd")] F32x4ConvertI32x4U => {},
+            #[cfg(feature = "__simd")] F64x2ConvertLowI32x4S => {},
+            #[cfg(feature = "__simd")] F64x2ConvertLowI32x4U => {},
+            #[cfg(feature = "__simd")] F32x4DemoteF64x2Zero => {},
+            #[cfg(feature = "__simd")] F64x2PromoteLowF32x4 => {},
+            #[cfg(feature = "__simd")] I32x4TruncSatF64x2SZero => unimplemented!(),
+            #[cfg(feature = "__simd")] I32x4TruncSatF64x2UZero => unimplemented!(),
 
             i => return ControlFlow::Break(Some(Error::UnsupportedFeature(format!("unimplemented opcode: {i:?}")))),
         };
@@ -978,6 +997,7 @@ impl<'store, 'stack> Executor<'store, 'stack> {
         }
     }
 
+    #[cfg(feature = "__simd")]
     fn exec_mem_load_lane<
         LOAD: MemLoadable<LOAD_SIZE>,
         INTO: InternalValue + IndexMut<usize, Output = LOAD>,
@@ -1018,7 +1038,7 @@ impl<'store, 'stack> Executor<'store, 'stack> {
             false => self.stack.values.pop::<i32>() as u32 as u64,
         };
 
-        let Some(Ok(addr)) = offset.checked_add(addr).map(TryInto::try_into) else {
+        let Some(Ok(addr)) = offset.checked_add(addr).map(|a| a.try_into()) else {
             cold();
             return ControlFlow::Break(Some(Error::Trap(Trap::MemoryOutOfBounds {
                 offset: addr as usize,
@@ -1031,6 +1051,7 @@ impl<'store, 'stack> Executor<'store, 'stack> {
         ControlFlow::Continue(())
     }
 
+    #[cfg(feature = "__simd")]
     fn exec_mem_store_lane<T: InternalValue + Index<usize, Output = U>, U: MemStorable<N> + Copy, const N: usize>(
         &mut self,
         mem_addr: tinywasm_types::MemAddr,
diff --git a/crates/tinywasm/src/interpreter/num_helpers.rs b/crates/tinywasm/src/interpreter/num_helpers.rs
index bba35cb..678c32e 100644
--- a/crates/tinywasm/src/interpreter/num_helpers.rs
+++ b/crates/tinywasm/src/interpreter/num_helpers.rs
@@ -184,7 +184,7 @@ macro_rules! impl_checked_wrapping_rem {
 
 impl_checked_wrapping_rem! { i32 i64 u32 u64 }
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 /// replace all NaNs in a f32x4 with f32::NAN
 pub(crate) fn canonicalize_f32x4(x: core::simd::f32x4) -> core::simd::f32x4 {
     use core::simd::{Simd, num::SimdFloat};
@@ -193,7 +193,7 @@ pub(crate) fn canonicalize_f32x4(x: core::simd::f32x4) -> core::simd::f32x4 {
     mask.select(nan, x)
 }
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 /// replace all NaNs in a f64x2 with f64::NAN
 pub(crate) fn canonicalize_f64x2(x: core::simd::f64x2) -> core::simd::f64x2 {
     use core::simd::{Simd, num::SimdFloat};
diff --git a/crates/tinywasm/src/interpreter/stack/call_stack.rs b/crates/tinywasm/src/interpreter/stack/call_stack.rs
index e94e401..c7f02b6 100644
--- a/crates/tinywasm/src/interpreter/stack/call_stack.rs
+++ b/crates/tinywasm/src/interpreter/stack/call_stack.rs
@@ -71,6 +71,7 @@ impl CallFrame {
     }
 
     #[inline]
+    #[allow(dead_code)]
     pub(crate) fn data(&self) -> &WasmFunctionData {
         &self.func_instance.data
     }
diff --git a/crates/tinywasm/src/interpreter/stack/value_stack.rs b/crates/tinywasm/src/interpreter/stack/value_stack.rs
index 01cea1c..481a88c 100644
--- a/crates/tinywasm/src/interpreter/stack/value_stack.rs
+++ b/crates/tinywasm/src/interpreter/stack/value_stack.rs
@@ -72,6 +72,7 @@ impl ValueStack {
     }
 
     #[inline]
+    #[allow(dead_code)]
     pub(crate) fn calculate_same_3<T: InternalValue>(&mut self, func: impl FnOnce(T, T, T) -> Result<T>) -> Result<()> {
         T::stack_calculate3(self, func)
     }
@@ -88,6 +89,7 @@ impl ValueStack {
     }
 
     #[inline]
+    #[allow(dead_code)]
     pub(crate) fn calculate_diff<A: InternalValue, B: InternalValue, RES: InternalValue>(
         &mut self,
         func: impl FnOnce(A, B) -> Result<RES>,
@@ -191,10 +193,10 @@ impl ValueStack {
             ValType::RefExtern => WasmValue::RefExtern(ExternRef::new(self.pop())),
             ValType::RefFunc => WasmValue::RefFunc(FuncRef::new(self.pop())),
 
-            #[cfg(not(feature = "simd"))]
+            #[cfg(not(feature = "__simd"))]
             ValType::V128 => WasmValue::V128(self.pop()),
 
-            #[cfg(feature = "simd")]
+            #[cfg(feature = "__simd")]
             ValType::V128 => WasmValue::V128(i128::from_le_bytes(self.pop::<Value128>().to_array())),
         }
     }
diff --git a/crates/tinywasm/src/interpreter/values.rs b/crates/tinywasm/src/interpreter/values.rs
index 335da59..779fc60 100644
--- a/crates/tinywasm/src/interpreter/values.rs
+++ b/crates/tinywasm/src/interpreter/values.rs
@@ -7,9 +7,9 @@ pub(crate) type Value32 = u32;
 pub(crate) type Value64 = u64;
 pub(crate) type ValueRef = Option<u32>;
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 pub(crate) type Value128 = core::simd::u8x16;
-#[cfg(not(feature = "simd"))]
+#[cfg(not(feature = "__simd"))]
 pub(crate) type Value128 = i128;
 
 #[derive(Debug, Clone, Copy, PartialEq)]
@@ -113,10 +113,10 @@ impl TinyWasmValue {
             ValType::RefExtern => WasmValue::RefExtern(ExternRef::new(self.unwrap_ref())),
             ValType::RefFunc => WasmValue::RefFunc(FuncRef::new(self.unwrap_ref())),
 
-            #[cfg(feature = "simd")]
+            #[cfg(feature = "__simd")]
             ValType::V128 => WasmValue::V128(i128::from_le_bytes(self.unwrap_128().to_array())),
 
-            #[cfg(not(feature = "simd"))]
+            #[cfg(not(feature = "__simd"))]
             ValType::V128 => WasmValue::V128(self.unwrap_128()),
         }
     }
@@ -132,10 +132,10 @@ impl From<&WasmValue> for TinyWasmValue {
             WasmValue::RefExtern(v) => TinyWasmValue::ValueRef(v.addr()),
             WasmValue::RefFunc(v) => TinyWasmValue::ValueRef(v.addr()),
 
-            #[cfg(not(feature = "simd"))]
+            #[cfg(not(feature = "__simd"))]
             WasmValue::V128(v) => TinyWasmValue::Value128(*v),
 
-            #[cfg(feature = "simd")]
+            #[cfg(feature = "__simd")]
             WasmValue::V128(v) => TinyWasmValue::Value128(v.to_le_bytes().into()),
         }
     }
@@ -160,6 +160,7 @@ pub(crate) trait InternalValue: sealed::Sealed + Into<TinyWasmValue> {
     fn stack_calculate(stack: &mut ValueStack, func: impl FnOnce(Self, Self) -> Result<Self>) -> Result<()>
     where
         Self: Sized;
+    #[allow(dead_code)]
     fn stack_calculate3(stack: &mut ValueStack, func: impl FnOnce(Self, Self, Self) -> Result<Self>) -> Result<()>
     where
         Self: Sized;
@@ -273,10 +274,10 @@ impl_internalvalue! {
     Value128, stack_128, locals_128, Value128, Value128, |v| v, |v| v
 }
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 use core::simd::{num::SimdUint, *};
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 impl_internalvalue! {
     Value128, stack_128, locals_128, u8x16, i128, |v: i128| v.to_le_bytes().into(), |v: u8x16| i128::from_le_bytes(v.into())
     Value128, stack_128, locals_128, u8x16, u128, |v: u128| v.to_le_bytes().into(), |v: u8x16| u128::from_le_bytes(v.into())
diff --git a/crates/tinywasm/src/lib.rs b/crates/tinywasm/src/lib.rs
index 72c38cd..9cb6c0a 100644
--- a/crates/tinywasm/src/lib.rs
+++ b/crates/tinywasm/src/lib.rs
@@ -5,7 +5,7 @@
 ))]
 #![warn(missing_docs, missing_debug_implementations, rust_2018_idioms, unreachable_pub)]
 #![forbid(unsafe_code)]
-#![cfg_attr(feature = "simd", feature(portable_simd))]
+#![cfg_attr(feature = "__simd", feature(portable_simd))]
 
 //! A tiny WebAssembly Runtime written in Rust
 //!
diff --git a/crates/tinywasm/src/store/memory.rs b/crates/tinywasm/src/store/memory.rs
index df3d49e..7abd7b9 100644
--- a/crates/tinywasm/src/store/memory.rs
+++ b/crates/tinywasm/src/store/memory.rs
@@ -4,7 +4,7 @@ use tinywasm_types::{MemoryArch, MemoryType, ModuleInstanceAddr};
 
 use crate::{Error, Result, cold, interpreter::Value128, log};
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 use core::simd::ToBytes;
 
 /// A WebAssembly Memory Instance
@@ -189,7 +189,7 @@ macro_rules! impl_mem_traits {
 
 impl_mem_traits!(u8, 1, i8, 1, u16, 2, i16, 2, u32, 4, i32, 4, f32, 4, u64, 8, i64, 8, f64, 8, Value128, 16);
 
-#[cfg(feature = "simd")]
+#[cfg(feature = "__simd")]
 impl_mem_traits!(
     core::simd::i8x16,
     16,
diff --git a/crates/tinywasm/tests/generated/wasm-1.csv b/crates/tinywasm/tests/generated/wasm-1.csv
index d403e3b..ebb511b 100644
--- a/crates/tinywasm/tests/generated/wasm-1.csv
+++ b/crates/tinywasm/tests/generated/wasm-1.csv
@@ -5,4 +5,4 @@
 0.6.1,27572,335,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":162,"failed":0},{"name":"binary-leb128.wast","passed":91,"failed":0},{"name":"binary.wast","passed":112,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"bulk.wast","passed":75,"failed":42},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":170,"failed":0},{"name":"comments.wast","passed":8,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"elem.wast","passed":98,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":179,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":110,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":241,"failed":0},{"name":"imports.wast","passed":186,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":79,"failed":0},{"name":"memory_copy.wast","passed":4450,"failed":0},{"name":"memory_fill.wast","passed":100,"failed":0},{"name":"memory_grow.wast","passed":96,"failed":0},{"name":"memory_init.wast","passed":240,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"ref_func.wast","passed":9,"failed":8},{"name":"ref_is_null.wast","passed":4,"failed":12},{"name":"ref_null.wast","passed":1,"failed":2},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":148,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table-sub.wast","passed":2,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"table_copy.wast","passed":1613,"failed":115},{"name":"table_fill.wast","passed":23,"failed":22},{"name":"table_get.wast","passed":10,"failed":6},{"name":"table_grow.wast","passed":21,"failed":29},{"name":"table_init.wast","passed":719,"failed":61},{"name":"table_set.wast","passed":19,"failed":7},{"name":"table_size.wast","passed":8,"failed":31},{"name":"token.wast","passed":58,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":7,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
 0.7.0,27572,335,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":162,"failed":0},{"name":"binary-leb128.wast","passed":91,"failed":0},{"name":"binary.wast","passed":112,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"bulk.wast","passed":75,"failed":42},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":170,"failed":0},{"name":"comments.wast","passed":8,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"elem.wast","passed":98,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":179,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":110,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":241,"failed":0},{"name":"imports.wast","passed":186,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":79,"failed":0},{"name":"memory_copy.wast","passed":4450,"failed":0},{"name":"memory_fill.wast","passed":100,"failed":0},{"name":"memory_grow.wast","passed":96,"failed":0},{"name":"memory_init.wast","passed":240,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"ref_func.wast","passed":9,"failed":8},{"name":"ref_is_null.wast","passed":4,"failed":12},{"name":"ref_null.wast","passed":1,"failed":2},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":148,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table-sub.wast","passed":2,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"table_copy.wast","passed":1613,"failed":115},{"name":"table_fill.wast","passed":23,"failed":22},{"name":"table_get.wast","passed":10,"failed":6},{"name":"table_grow.wast","passed":21,"failed":29},{"name":"table_init.wast","passed":719,"failed":61},{"name":"table_set.wast","passed":19,"failed":7},{"name":"table_size.wast","passed":8,"failed":31},{"name":"token.wast","passed":58,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":7,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
 0.8.0,20360,0,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":162,"failed":0},{"name":"binary-leb128.wast","passed":91,"failed":0},{"name":"binary.wast","passed":128,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":170,"failed":0},{"name":"comments.wast","passed":8,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"elem.wast","passed":98,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":927,"failed":0},{"name":"float_literals.wast","passed":179,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":471,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":110,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":241,"failed":0},{"name":"imports.wast","passed":178,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":88,"failed":0},{"name":"memory_grow.wast","passed":104,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":148,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"token.wast","passed":58,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":7,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
-0.9.0-alpha.0,19243,0,[{"name":"address.wast","passed":243,"failed":0},{"name":"align.wast","passed":156,"failed":0},{"name":"binary-leb128.wast","passed":79,"failed":0},{"name":"binary.wast","passed":67,"failed":0},{"name":"block.wast","passed":171,"failed":0},{"name":"br.wast","passed":84,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":168,"failed":0},{"name":"break-drop.wast","passed":4,"failed":0},{"name":"call.wast","passed":82,"failed":0},{"name":"call_indirect.wast","passed":152,"failed":0},{"name":"comments.wast","passed":4,"failed":0},{"name":"const.wast","passed":668,"failed":0},{"name":"conversions.wast","passed":435,"failed":0},{"name":"custom.wast","passed":10,"failed":0},{"name":"data.wast","passed":45,"failed":0},{"name":"elem.wast","passed":55,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":82,"failed":0},{"name":"f32.wast","passed":2512,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2512,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":7,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":161,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":121,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"globals.wast","passed":78,"failed":0},{"name":"i32.wast","passed":443,"failed":0},{"name":"i64.wast","passed":389,"failed":0},{"name":"if.wast","passed":151,"failed":0},{"name":"imports.wast","passed":146,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":116,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":81,"failed":0},{"name":"memory.wast","passed":71,"failed":0},{"name":"memory_grow.wast","passed":94,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":173,"failed":0},{"name":"names.wast","passed":483,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":111,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":5,"failed":0},{"name":"start.wast","passed":19,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"token.wast","passed":2,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":62,"failed":0},{"name":"unreached-invalid.wast","passed":110,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
+0.9.0-alpha.0,19241,0,[{"name":"address.wast","passed":243,"failed":0},{"name":"align.wast","passed":156,"failed":0},{"name":"binary-leb128.wast","passed":77,"failed":0},{"name":"binary.wast","passed":67,"failed":0},{"name":"block.wast","passed":171,"failed":0},{"name":"br.wast","passed":84,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":168,"failed":0},{"name":"break-drop.wast","passed":4,"failed":0},{"name":"call.wast","passed":82,"failed":0},{"name":"call_indirect.wast","passed":152,"failed":0},{"name":"comments.wast","passed":4,"failed":0},{"name":"const.wast","passed":668,"failed":0},{"name":"conversions.wast","passed":435,"failed":0},{"name":"custom.wast","passed":10,"failed":0},{"name":"data.wast","passed":45,"failed":0},{"name":"elem.wast","passed":55,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":82,"failed":0},{"name":"f32.wast","passed":2512,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2512,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":7,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":161,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":121,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"globals.wast","passed":78,"failed":0},{"name":"i32.wast","passed":443,"failed":0},{"name":"i64.wast","passed":389,"failed":0},{"name":"if.wast","passed":151,"failed":0},{"name":"imports.wast","passed":146,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":116,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":81,"failed":0},{"name":"memory.wast","passed":71,"failed":0},{"name":"memory_grow.wast","passed":94,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":173,"failed":0},{"name":"names.wast","passed":483,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":111,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":5,"failed":0},{"name":"start.wast","passed":19,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"token.wast","passed":2,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":62,"failed":0},{"name":"unreached-invalid.wast","passed":110,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
diff --git a/crates/tinywasm/tests/generated/wasm-2.csv b/crates/tinywasm/tests/generated/wasm-2.csv
index b0949ec..8e228b4 100644
--- a/crates/tinywasm/tests/generated/wasm-2.csv
+++ b/crates/tinywasm/tests/generated/wasm-2.csv
@@ -10,4 +10,4 @@
 0.6.1,20278,0,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":162,"failed":0},{"name":"binary-leb128.wast","passed":91,"failed":0},{"name":"binary.wast","passed":112,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":170,"failed":0},{"name":"comments.wast","passed":8,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"elem.wast","passed":98,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":179,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":110,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":241,"failed":0},{"name":"imports.wast","passed":186,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":79,"failed":0},{"name":"memory_grow.wast","passed":96,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":148,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"token.wast","passed":58,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":7,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
 0.7.0,20278,0,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":162,"failed":0},{"name":"binary-leb128.wast","passed":91,"failed":0},{"name":"binary.wast","passed":112,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":170,"failed":0},{"name":"comments.wast","passed":8,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"elem.wast","passed":98,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":179,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":110,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":241,"failed":0},{"name":"imports.wast","passed":186,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":79,"failed":0},{"name":"memory_grow.wast","passed":96,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":148,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"token.wast","passed":58,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":7,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
 0.8.0,28008,0,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":162,"failed":0},{"name":"binary-leb128.wast","passed":91,"failed":0},{"name":"binary.wast","passed":128,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"bulk.wast","passed":117,"failed":0},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":170,"failed":0},{"name":"comments.wast","passed":8,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"elem.wast","passed":98,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":927,"failed":0},{"name":"float_literals.wast","passed":179,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":471,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":110,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":241,"failed":0},{"name":"imports.wast","passed":178,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":88,"failed":0},{"name":"memory_copy.wast","passed":4450,"failed":0},{"name":"memory_fill.wast","passed":100,"failed":0},{"name":"memory_grow.wast","passed":104,"failed":0},{"name":"memory_init.wast","passed":240,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"obsolete-keywords.wast","passed":11,"failed":0},{"name":"ref_func.wast","passed":17,"failed":0},{"name":"ref_is_null.wast","passed":16,"failed":0},{"name":"ref_null.wast","passed":3,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":148,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table-sub.wast","passed":2,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"table_copy.wast","passed":1728,"failed":0},{"name":"table_fill.wast","passed":45,"failed":0},{"name":"table_get.wast","passed":16,"failed":0},{"name":"table_grow.wast","passed":58,"failed":0},{"name":"table_init.wast","passed":780,"failed":0},{"name":"table_set.wast","passed":26,"failed":0},{"name":"table_size.wast","passed":39,"failed":0},{"name":"token.wast","passed":58,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":7,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
-0.9.0-alpha.0,27826,0,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":156,"failed":0},{"name":"binary-leb128.wast","passed":81,"failed":0},{"name":"binary.wast","passed":162,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"bulk.wast","passed":117,"failed":0},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":169,"failed":0},{"name":"comments.wast","passed":4,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":58,"failed":0},{"name":"elem.wast","passed":74,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":161,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":108,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":239,"failed":0},{"name":"imports.wast","passed":183,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":79,"failed":0},{"name":"memory_copy.wast","passed":4450,"failed":0},{"name":"memory_fill.wast","passed":100,"failed":0},{"name":"memory_grow.wast","passed":96,"failed":0},{"name":"memory_init.wast","passed":240,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"ref_func.wast","passed":17,"failed":0},{"name":"ref_is_null.wast","passed":16,"failed":0},{"name":"ref_null.wast","passed":3,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":147,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table-sub.wast","passed":2,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"table_copy.wast","passed":1728,"failed":0},{"name":"table_fill.wast","passed":45,"failed":0},{"name":"table_get.wast","passed":16,"failed":0},{"name":"table_grow.wast","passed":50,"failed":0},{"name":"table_init.wast","passed":780,"failed":0},{"name":"table_set.wast","passed":26,"failed":0},{"name":"table_size.wast","passed":39,"failed":0},{"name":"token.wast","passed":2,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":6,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
+0.9.0-alpha.0,27822,0,[{"name":"address.wast","passed":260,"failed":0},{"name":"align.wast","passed":156,"failed":0},{"name":"binary-leb128.wast","passed":79,"failed":0},{"name":"binary.wast","passed":160,"failed":0},{"name":"block.wast","passed":223,"failed":0},{"name":"br.wast","passed":97,"failed":0},{"name":"br_if.wast","passed":118,"failed":0},{"name":"br_table.wast","passed":174,"failed":0},{"name":"bulk.wast","passed":117,"failed":0},{"name":"call.wast","passed":91,"failed":0},{"name":"call_indirect.wast","passed":169,"failed":0},{"name":"comments.wast","passed":4,"failed":0},{"name":"const.wast","passed":778,"failed":0},{"name":"conversions.wast","passed":619,"failed":0},{"name":"custom.wast","passed":11,"failed":0},{"name":"data.wast","passed":58,"failed":0},{"name":"elem.wast","passed":74,"failed":0},{"name":"endianness.wast","passed":69,"failed":0},{"name":"exports.wast","passed":96,"failed":0},{"name":"f32.wast","passed":2514,"failed":0},{"name":"f32_bitwise.wast","passed":364,"failed":0},{"name":"f32_cmp.wast","passed":2407,"failed":0},{"name":"f64.wast","passed":2514,"failed":0},{"name":"f64_bitwise.wast","passed":364,"failed":0},{"name":"f64_cmp.wast","passed":2407,"failed":0},{"name":"fac.wast","passed":8,"failed":0},{"name":"float_exprs.wast","passed":900,"failed":0},{"name":"float_literals.wast","passed":161,"failed":0},{"name":"float_memory.wast","passed":90,"failed":0},{"name":"float_misc.wast","passed":441,"failed":0},{"name":"forward.wast","passed":5,"failed":0},{"name":"func.wast","passed":172,"failed":0},{"name":"func_ptrs.wast","passed":36,"failed":0},{"name":"global.wast","passed":108,"failed":0},{"name":"i32.wast","passed":460,"failed":0},{"name":"i64.wast","passed":416,"failed":0},{"name":"if.wast","passed":239,"failed":0},{"name":"imports.wast","passed":183,"failed":0},{"name":"inline-module.wast","passed":1,"failed":0},{"name":"int_exprs.wast","passed":108,"failed":0},{"name":"int_literals.wast","passed":51,"failed":0},{"name":"labels.wast","passed":29,"failed":0},{"name":"left-to-right.wast","passed":96,"failed":0},{"name":"linking.wast","passed":132,"failed":0},{"name":"load.wast","passed":97,"failed":0},{"name":"local_get.wast","passed":36,"failed":0},{"name":"local_set.wast","passed":53,"failed":0},{"name":"local_tee.wast","passed":97,"failed":0},{"name":"loop.wast","passed":120,"failed":0},{"name":"memory.wast","passed":79,"failed":0},{"name":"memory_copy.wast","passed":4450,"failed":0},{"name":"memory_fill.wast","passed":100,"failed":0},{"name":"memory_grow.wast","passed":96,"failed":0},{"name":"memory_init.wast","passed":240,"failed":0},{"name":"memory_redundancy.wast","passed":8,"failed":0},{"name":"memory_size.wast","passed":42,"failed":0},{"name":"memory_trap.wast","passed":182,"failed":0},{"name":"names.wast","passed":486,"failed":0},{"name":"nop.wast","passed":88,"failed":0},{"name":"ref_func.wast","passed":17,"failed":0},{"name":"ref_is_null.wast","passed":16,"failed":0},{"name":"ref_null.wast","passed":3,"failed":0},{"name":"return.wast","passed":84,"failed":0},{"name":"select.wast","passed":147,"failed":0},{"name":"skip-stack-guard-page.wast","passed":11,"failed":0},{"name":"stack.wast","passed":7,"failed":0},{"name":"start.wast","passed":20,"failed":0},{"name":"store.wast","passed":68,"failed":0},{"name":"switch.wast","passed":28,"failed":0},{"name":"table-sub.wast","passed":2,"failed":0},{"name":"table.wast","passed":19,"failed":0},{"name":"table_copy.wast","passed":1728,"failed":0},{"name":"table_fill.wast","passed":45,"failed":0},{"name":"table_get.wast","passed":16,"failed":0},{"name":"table_grow.wast","passed":50,"failed":0},{"name":"table_init.wast","passed":780,"failed":0},{"name":"table_set.wast","passed":26,"failed":0},{"name":"table_size.wast","passed":39,"failed":0},{"name":"token.wast","passed":2,"failed":0},{"name":"traps.wast","passed":36,"failed":0},{"name":"type.wast","passed":3,"failed":0},{"name":"unreachable.wast","passed":64,"failed":0},{"name":"unreached-invalid.wast","passed":118,"failed":0},{"name":"unreached-valid.wast","passed":6,"failed":0},{"name":"unwind.wast","passed":50,"failed":0},{"name":"utf8-custom-section-id.wast","passed":176,"failed":0},{"name":"utf8-import-field.wast","passed":176,"failed":0},{"name":"utf8-import-module.wast","passed":176,"failed":0},{"name":"utf8-invalid-encoding.wast","passed":176,"failed":0}]
diff --git a/crates/tinywasm/tests/generated/wasm-multi-memory.csv b/crates/tinywasm/tests/generated/wasm-multi-memory.csv
index ae9ba03..33debd6 100644
--- a/crates/tinywasm/tests/generated/wasm-multi-memory.csv
+++ b/crates/tinywasm/tests/generated/wasm-multi-memory.csv
@@ -1,2 +1,2 @@
 0.8.0,1872,0,[{"name":"address0.wast","passed":92,"failed":0},{"name":"address1.wast","passed":127,"failed":0},{"name":"align.wast","passed":160,"failed":0},{"name":"align0.wast","passed":5,"failed":0},{"name":"binary.wast","passed":126,"failed":0},{"name":"binary0.wast","passed":7,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"data0.wast","passed":7,"failed":0},{"name":"data1.wast","passed":14,"failed":0},{"name":"data_drop0.wast","passed":11,"failed":0},{"name":"exports0.wast","passed":8,"failed":0},{"name":"float_exprs0.wast","passed":14,"failed":0},{"name":"float_exprs1.wast","passed":3,"failed":0},{"name":"float_memory0.wast","passed":30,"failed":0},{"name":"imports.wast","passed":175,"failed":0},{"name":"imports0.wast","passed":8,"failed":0},{"name":"imports1.wast","passed":5,"failed":0},{"name":"imports2.wast","passed":20,"failed":0},{"name":"imports3.wast","passed":10,"failed":0},{"name":"imports4.wast","passed":16,"failed":0},{"name":"linking0.wast","passed":6,"failed":0},{"name":"linking1.wast","passed":14,"failed":0},{"name":"linking2.wast","passed":11,"failed":0},{"name":"linking3.wast","passed":14,"failed":0},{"name":"load.wast","passed":118,"failed":0},{"name":"load0.wast","passed":3,"failed":0},{"name":"load1.wast","passed":18,"failed":0},{"name":"load2.wast","passed":38,"failed":0},{"name":"memory-multi.wast","passed":6,"failed":0},{"name":"memory.wast","passed":86,"failed":0},{"name":"memory_copy0.wast","passed":29,"failed":0},{"name":"memory_copy1.wast","passed":14,"failed":0},{"name":"memory_fill0.wast","passed":16,"failed":0},{"name":"memory_grow.wast","passed":157,"failed":0},{"name":"memory_init0.wast","passed":13,"failed":0},{"name":"memory_size.wast","passed":49,"failed":0},{"name":"memory_size0.wast","passed":8,"failed":0},{"name":"memory_size1.wast","passed":15,"failed":0},{"name":"memory_size2.wast","passed":21,"failed":0},{"name":"memory_size3.wast","passed":2,"failed":0},{"name":"memory_trap0.wast","passed":14,"failed":0},{"name":"memory_trap1.wast","passed":168,"failed":0},{"name":"multi-memory/simd_memory-multi.wast (skipped)","passed":0,"failed":0},{"name":"start0.wast","passed":9,"failed":0},{"name":"store.wast","passed":111,"failed":0},{"name":"store0.wast","passed":5,"failed":0},{"name":"store1.wast","passed":13,"failed":0},{"name":"traps0.wast","passed":15,"failed":0}]
-0.9.0-alpha.0,1872,0,[{"name":"address0.wast","passed":92,"failed":0},{"name":"address1.wast","passed":127,"failed":0},{"name":"align.wast","passed":160,"failed":0},{"name":"align0.wast","passed":5,"failed":0},{"name":"binary.wast","passed":126,"failed":0},{"name":"binary0.wast","passed":7,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"data0.wast","passed":7,"failed":0},{"name":"data1.wast","passed":14,"failed":0},{"name":"data_drop0.wast","passed":11,"failed":0},{"name":"exports0.wast","passed":8,"failed":0},{"name":"float_exprs0.wast","passed":14,"failed":0},{"name":"float_exprs1.wast","passed":3,"failed":0},{"name":"float_memory0.wast","passed":30,"failed":0},{"name":"imports.wast","passed":175,"failed":0},{"name":"imports0.wast","passed":8,"failed":0},{"name":"imports1.wast","passed":5,"failed":0},{"name":"imports2.wast","passed":20,"failed":0},{"name":"imports3.wast","passed":10,"failed":0},{"name":"imports4.wast","passed":16,"failed":0},{"name":"linking0.wast","passed":6,"failed":0},{"name":"linking1.wast","passed":14,"failed":0},{"name":"linking2.wast","passed":11,"failed":0},{"name":"linking3.wast","passed":14,"failed":0},{"name":"load.wast","passed":118,"failed":0},{"name":"load0.wast","passed":3,"failed":0},{"name":"load1.wast","passed":18,"failed":0},{"name":"load2.wast","passed":38,"failed":0},{"name":"memory-multi.wast","passed":6,"failed":0},{"name":"memory.wast","passed":86,"failed":0},{"name":"memory_copy0.wast","passed":29,"failed":0},{"name":"memory_copy1.wast","passed":14,"failed":0},{"name":"memory_fill0.wast","passed":16,"failed":0},{"name":"memory_grow.wast","passed":157,"failed":0},{"name":"memory_init0.wast","passed":13,"failed":0},{"name":"memory_size.wast","passed":49,"failed":0},{"name":"memory_size0.wast","passed":8,"failed":0},{"name":"memory_size1.wast","passed":15,"failed":0},{"name":"memory_size2.wast","passed":21,"failed":0},{"name":"memory_size3.wast","passed":2,"failed":0},{"name":"memory_trap0.wast","passed":14,"failed":0},{"name":"memory_trap1.wast","passed":168,"failed":0},{"name":"simd_memory-multi.wast (skipped)","passed":0,"failed":0},{"name":"start0.wast","passed":9,"failed":0},{"name":"store.wast","passed":111,"failed":0},{"name":"store0.wast","passed":5,"failed":0},{"name":"store1.wast","passed":13,"failed":0},{"name":"traps0.wast","passed":15,"failed":0}]
+0.9.0-alpha.0,1871,0,[{"name":"address0.wast","passed":92,"failed":0},{"name":"address1.wast","passed":127,"failed":0},{"name":"align.wast","passed":160,"failed":0},{"name":"align0.wast","passed":5,"failed":0},{"name":"binary.wast","passed":126,"failed":0},{"name":"binary0.wast","passed":6,"failed":0},{"name":"data.wast","passed":61,"failed":0},{"name":"data0.wast","passed":7,"failed":0},{"name":"data1.wast","passed":14,"failed":0},{"name":"data_drop0.wast","passed":11,"failed":0},{"name":"exports0.wast","passed":8,"failed":0},{"name":"float_exprs0.wast","passed":14,"failed":0},{"name":"float_exprs1.wast","passed":3,"failed":0},{"name":"float_memory0.wast","passed":30,"failed":0},{"name":"imports.wast","passed":175,"failed":0},{"name":"imports0.wast","passed":8,"failed":0},{"name":"imports1.wast","passed":5,"failed":0},{"name":"imports2.wast","passed":20,"failed":0},{"name":"imports3.wast","passed":10,"failed":0},{"name":"imports4.wast","passed":16,"failed":0},{"name":"linking0.wast","passed":6,"failed":0},{"name":"linking1.wast","passed":14,"failed":0},{"name":"linking2.wast","passed":11,"failed":0},{"name":"linking3.wast","passed":14,"failed":0},{"name":"load.wast","passed":118,"failed":0},{"name":"load0.wast","passed":3,"failed":0},{"name":"load1.wast","passed":18,"failed":0},{"name":"load2.wast","passed":38,"failed":0},{"name":"memory-multi.wast","passed":6,"failed":0},{"name":"memory.wast","passed":86,"failed":0},{"name":"memory_copy0.wast","passed":29,"failed":0},{"name":"memory_copy1.wast","passed":14,"failed":0},{"name":"memory_fill0.wast","passed":16,"failed":0},{"name":"memory_grow.wast","passed":157,"failed":0},{"name":"memory_init0.wast","passed":13,"failed":0},{"name":"memory_size.wast","passed":49,"failed":0},{"name":"memory_size0.wast","passed":8,"failed":0},{"name":"memory_size1.wast","passed":15,"failed":0},{"name":"memory_size2.wast","passed":21,"failed":0},{"name":"memory_size3.wast","passed":2,"failed":0},{"name":"memory_trap0.wast","passed":14,"failed":0},{"name":"memory_trap1.wast","passed":168,"failed":0},{"name":"simd_memory-multi.wast (skipped)","passed":0,"failed":0},{"name":"start0.wast","passed":9,"failed":0},{"name":"store.wast","passed":111,"failed":0},{"name":"store0.wast","passed":5,"failed":0},{"name":"store1.wast","passed":13,"failed":0},{"name":"traps0.wast","passed":15,"failed":0}]
diff --git a/crates/types/src/instructions.rs b/crates/types/src/instructions.rs
index bf4f6fb..9db5ed5 100644
--- a/crates/types/src/instructions.rs
+++ b/crates/types/src/instructions.rs
@@ -4,7 +4,6 @@ use crate::{ConstIdx, DataAddr, ElemAddr, ExternAddr, MemAddr};
 /// Represents a memory immediate in a WebAssembly memory instruction.
 #[derive(Debug, Copy, Clone, PartialEq)]
 #[cfg_attr(feature = "archive", derive(serde::Serialize, serde::Deserialize))]
-
 pub struct MemoryArg([u8; 12]);
 
 impl MemoryArg {