simd-128: The input lanes of integer-to-integer narrowing ops should be interpreted as signed (#2850)

https://github.com/WebAssembly/spec/blob/main/proposals/simd/SIMD.md#integer-to-integer-narrowing

core/iwasm/compilation/simd/simd_conversions.c

@@ -158,11 +158,15 @@ simd_integer_narrow_common(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
         return false;
     }
 
-    /* sat */
+    /* Refer to:
+     * https://github.com/WebAssembly/spec/blob/main/proposals/simd/SIMD.md#integer-to-integer-narrowing
+     * Regardless of the whether the operation is signed or unsigned, the input
+     * lanes are interpreted as signed integers.
+     */
     if (!(vec1 = simd_saturate(comp_ctx, func_ctx, e_sat_i16x8, vec1, min, max,
-                               is_signed))
+                               true))
         || !(vec2 = simd_saturate(comp_ctx, func_ctx, e_sat_i16x8, vec2, min,
-                                  max, is_signed))) {
+                                  max, true))) {
         return false;
     }
 
@@ -740,4 +744,4 @@ aot_compile_simd_i64x2_extmul_i32x4(AOTCompContext *comp_ctx,
 {
     return simd_integer_extmul(comp_ctx, func_ctx, lower_half, is_signed,
                                e_i64x2_extmul_i32x4);
-}
+}