scripting/wamr_patches/0023-fast-interp-v128-register-marshalling.patch

--- a/core/iwasm/interpreter/wasm_interp_fast.c
+++ b/core/iwasm/interpreter/wasm_interp_fast.c
@@ -6264,18 +6264,20 @@
                 goto call_func_from_entry;
             }
 #if WASM_ENABLE_SIMDE != 0
-#define SIMD_V128_TO_SIMDE_V128(s_v)                                    \
-    ({                                                                  \
-        bh_assert(sizeof(V128) == sizeof(simde_v128_t));                \
-        simde_v128_t se_v;                                              \
-        bh_memcpy_s(&se_v, sizeof(simde_v128_t), &(s_v), sizeof(V128)); \
-        se_v;                                                           \
+/* rive patch: a constant sized memcpy folds to a register move, where the
+   checked bh_memcpy_s was an out of line call on every operand and result */
+#define SIMD_V128_TO_SIMDE_V128(s_v)                       \
+    ({                                                     \
+        bh_assert(sizeof(V128) == sizeof(simde_v128_t));   \
+        simde_v128_t se_v;                                 \
+        memcpy(&se_v, &(s_v), sizeof(simde_v128_t));       \
+        se_v;                                              \
     })
 
-#define SIMDE_V128_TO_SIMD_V128(sv, v)                                \
-    do {                                                              \
-        bh_assert(sizeof(V128) == sizeof(simde_v128_t));              \
-        bh_memcpy_s(&(v), sizeof(V128), &(sv), sizeof(simde_v128_t)); \
+#define SIMDE_V128_TO_SIMD_V128(sv, v)                     \
+    do {                                                   \
+        bh_assert(sizeof(V128) == sizeof(simde_v128_t));   \
+        memcpy(&(v), &(sv), sizeof(V128));                 \
     } while (0)
 
             HANDLE_OP(WASM_OP_SIMD_PREFIX)