diff --git a/cmake/compiler_flags.cmake b/cmake/compiler_flags.cmake
index b6cf7f760..bde986920 100644
--- a/cmake/compiler_flags.cmake
+++ b/cmake/compiler_flags.cmake
@@ -16,6 +16,11 @@ macro(fastgltf_compiler_flags TARGET)
 
             # https://github.com/simdjson/simdjson/blob/master/doc/basics.md#performance-tips
             target_compile_options(${TARGET} PRIVATE $<$<CONFIG:RELEASE>:-DNDEBUG>)
+
+            if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+                # For the conversion of ARM Neon vectors (say int16x8_t to int8x16_t)
+                target_compile_options(${TARGET} PRIVATE -flax-vector-conversions)
+            endif()
         endif()
     endif()
 endmacro()
diff --git a/include/fastgltf/types.hpp b/include/fastgltf/types.hpp
index e6030aebc..ada90d7b4 100644
--- a/include/fastgltf/types.hpp
+++ b/include/fastgltf/types.hpp
@@ -1336,6 +1336,12 @@ namespace fastgltf {
         FASTGLTF_STD_PMR_NS::string name;
     };
 
+    struct TRS {
+        std::array<num, 3> translation = {{ 0.f, 0.f, 0.f }};
+        std::array<num, 4> rotation = {{ 0.f, 0.f, 0.f, 1.f }};
+        std::array<num, 3> scale = {{ 1.f, 1.f, 1.f }};
+    };
+
     struct Node {
         Optional<std::size_t> meshIndex;
 	    Optional<std::size_t> skinIndex;
@@ -1349,11 +1355,6 @@ namespace fastgltf {
 	    FASTGLTF_FG_PMR_NS::MaybeSmallVector<std::size_t> children;
 	    FASTGLTF_FG_PMR_NS::MaybeSmallVector<num> weights;
 
-        struct TRS {
-            std::array<num, 3> translation = {{ 0.f, 0.f, 0.f }};
-            std::array<num, 4> rotation = {{ 0.f, 0.f, 0.f, 1.f }};
-            std::array<num, 3> scale = {{ 1.f, 1.f, 1.f }};
-        };
         using TransformMatrix = std::array<num, 16>;
 
         /**
diff --git a/src/fastgltf.cpp b/src/fastgltf.cpp
index fbf5681a1..bf9298ee9 100644
--- a/src/fastgltf.cpp
+++ b/src/fastgltf.cpp
@@ -998,7 +998,7 @@ fg::Error fg::validate(const fastgltf::Asset& asset) {
 		if (node.meshIndex.has_value() && asset.meshes.size() <= node.meshIndex.value())
 			return Error::InvalidGltf;
 
-		if (const auto* pTRS = std::get_if<Node::TRS>(&node.transform)) {
+		if (const auto* pTRS = std::get_if<TRS>(&node.transform)) {
 			for (const auto& x : pTRS->rotation)
 				if (x > 1.0 || x < -1.0)
 					return Error::InvalidGltf;
@@ -2897,14 +2897,14 @@ fg::Error fg::Parser::parseNodes(simdjson::dom::array& nodes, Asset& asset) {
             }
 
             if (hasBit(options, Options::DecomposeNodeMatrices)) {
-                Node::TRS trs = {};
+                TRS trs = {};
                 decomposeTransformMatrix(transformMatrix, trs.scale, trs.rotation, trs.translation);
                 node.transform = trs;
             } else {
                 node.transform = transformMatrix;
             }
         } else if (error == NO_SUCH_FIELD) {
-            Node::TRS trs = {};
+            TRS trs = {};
 
             // There's no matrix, let's see if there's scale, rotation, or rotation fields.
             if (auto error = nodeObject["scale"].get_array().get(array); error == SUCCESS) {
diff --git a/tests/basic_test.cpp b/tests/basic_test.cpp
index eff6f45b1..0c57ac5b9 100644
--- a/tests/basic_test.cpp
+++ b/tests/basic_test.cpp
@@ -153,7 +153,7 @@ TEST_CASE("Loading some basic glTF", "[gltf-loader]") {
 
         REQUIRE(cube->nodes.size() == 1);
         REQUIRE(cube->nodes.front().name == "Cube");
-        REQUIRE(std::holds_alternative<fastgltf::Node::TRS>(cube->nodes.front().transform));
+        REQUIRE(std::holds_alternative<fastgltf::TRS>(cube->nodes.front().transform));
 
         REQUIRE(cube->accessors.size() == 5);
         REQUIRE(cube->accessors[0].type == fastgltf::AccessorType::Scalar);
@@ -384,10 +384,10 @@ TEST_CASE("Test TRS parsing and optional decomposition", "[gltf-loader]") {
         REQUIRE(assetWithMatrix->nodes.size() == 2);
         REQUIRE(assetDecomposed->nodes.size() == 2);
         REQUIRE(std::holds_alternative<fastgltf::Node::TransformMatrix>(assetWithMatrix->nodes.back().transform));
-        REQUIRE(std::holds_alternative<fastgltf::Node::TRS>(assetDecomposed->nodes.back().transform));
+        REQUIRE(std::holds_alternative<fastgltf::TRS>(assetDecomposed->nodes.back().transform));
 
         // Get the TRS components from the first node and use them as the test data for decomposing.
-        const auto* pDefaultTRS = std::get_if<fastgltf::Node::TRS>(&assetWithMatrix->nodes.front().transform);
+        const auto* pDefaultTRS = std::get_if<fastgltf::TRS>(&assetWithMatrix->nodes.front().transform);
         REQUIRE(pDefaultTRS != nullptr);
         auto translation = glm::make_vec3(pDefaultTRS->translation.data());
         auto rotation = glm::make_quat(pDefaultTRS->rotation.data());
@@ -401,7 +401,7 @@ TEST_CASE("Test TRS parsing and optional decomposition", "[gltf-loader]") {
         REQUIRE(glm::make_mat4x4(pMatrix->data()) == transform);
 
         // Check if the decomposed components equal the original components.
-        const auto* pDecomposedTRS = std::get_if<fastgltf::Node::TRS>(&assetDecomposed->nodes.back().transform);
+        const auto* pDecomposedTRS = std::get_if<fastgltf::TRS>(&assetDecomposed->nodes.back().transform);
         REQUIRE(glm::make_vec3(pDecomposedTRS->translation.data()) == translation);
         REQUIRE(glm::make_quat(pDecomposedTRS->rotation.data()) == rotation);
         REQUIRE(glm::make_vec3(pDecomposedTRS->scale.data()) == scale);