diff --git a/build.rs b/build.rs
index 47852cf61815..773b38e6a828 100644
--- a/build.rs
+++ b/build.rs
@@ -32,6 +32,7 @@ fn main() -> anyhow::Result<()> {
             test_directory_module(out, "tests/misc_testsuite/reference-types", strategy)?;
             test_directory_module(out, "tests/misc_testsuite/multi-memory", strategy)?;
             test_directory_module(out, "tests/misc_testsuite/module-linking", strategy)?;
+            test_directory_module(out, "tests/misc_testsuite/simd", strategy)?;
             test_directory_module(out, "tests/misc_testsuite/threads", strategy)?;
             Ok(())
         })?;
diff --git a/cranelift/codegen/src/ir/instructions.rs b/cranelift/codegen/src/ir/instructions.rs
index 94ce878a9753..4981ad8ddc8f 100644
--- a/cranelift/codegen/src/ir/instructions.rs
+++ b/cranelift/codegen/src/ir/instructions.rs
@@ -16,13 +16,16 @@ use core::str::FromStr;
 #[cfg(feature = "enable-serde")]
 use serde::{Deserialize, Serialize};
 
-use crate::ir::{self, trapcode::TrapCode, types, Block, FuncRef, JumpTable, SigRef, Type, Value};
-use crate::isa;
-
 use crate::bitset::BitSet;
 use crate::data_value::DataValue;
 use crate::entity;
-use ir::condcodes::{FloatCC, IntCC};
+use crate::ir::{
+    self,
+    condcodes::{FloatCC, IntCC},
+    trapcode::TrapCode,
+    types, Block, FuncRef, JumpTable, MemFlags, SigRef, Type, Value,
+};
+use crate::isa;
 
 /// Some instructions use an external list of argument values because there is not enough space in
 /// the 16-byte `InstructionData` struct. These value lists are stored in a memory pool in
@@ -395,6 +398,19 @@ impl InstructionData {
         }
     }
 
+    /// If this is a load/store instruction, return its memory flags.
+    pub fn memflags(&self) -> Option<MemFlags> {
+        match self {
+            &InstructionData::Load { flags, .. }
+            | &InstructionData::LoadComplex { flags, .. }
+            | &InstructionData::LoadNoOffset { flags, .. }
+            | &InstructionData::Store { flags, .. }
+            | &InstructionData::StoreComplex { flags, .. }
+            | &InstructionData::StoreNoOffset { flags, .. } => Some(flags),
+            _ => None,
+        }
+    }
+
     /// Return information about a call instruction.
     ///
     /// Any instruction that can call another function reveals its call signature here.
diff --git a/cranelift/codegen/src/isa/x64/lower.rs b/cranelift/codegen/src/isa/x64/lower.rs
index fcb5e71355c8..9bdecdfe764f 100644
--- a/cranelift/codegen/src/isa/x64/lower.rs
+++ b/cranelift/codegen/src/isa/x64/lower.rs
@@ -153,6 +153,12 @@ fn is_mergeable_load<C: LowerCtx<I = Inst>>(
         return None;
     }
 
+    // SIMD instructions can only be load-coalesced when the loaded value comes
+    // from an aligned address.
+    if load_ty.is_vector() && !insn_data.memflags().map_or(false, |f| f.aligned()) {
+        return None;
+    }
+
     // Just testing the opcode is enough, because the width will always match if
     // the type does (and the type should match if the CLIF is properly
     // constructed).
diff --git a/tests/misc_testsuite/simd/unaligned-load.wast b/tests/misc_testsuite/simd/unaligned-load.wast
new file mode 100644
index 000000000000..359dbe0a6faf
--- /dev/null
+++ b/tests/misc_testsuite/simd/unaligned-load.wast
@@ -0,0 +1,13 @@
+(; See discussion at https://github.com/bytecodealliance/wasmtime/issues/2943 ;)
+(module
+  (memory 1)
+  (data (i32.const 1) "\01\00\00\00\01\00\00\00")
+
+  (func $unaligned_load (export "unaligned_load") (result v128)
+    v128.const i32x4 0 0 1 1
+    i32.const 1
+    v128.load
+    v128.xor)
+)
+
+(assert_return (invoke "unaligned_load") (v128.const i32x4 1 1 1 1))