diff --git a/src/moonlight-protocol/CMakeLists.txt b/src/moonlight-protocol/CMakeLists.txt
index 5e3ebc32..57f08cb3 100644
--- a/src/moonlight-protocol/CMakeLists.txt
+++ b/src/moonlight-protocol/CMakeLists.txt
@@ -45,18 +45,20 @@ FetchContent_GetProperties(nanors)
 if (NOT nanors_POPULATED)
     FetchContent_Populate(nanors)
 
-    add_library(nanors STATIC ${nanors_SOURCE_DIR}/rs.c)
+    add_library(nanors)
     add_library(nanors::nanors ALIAS nanors)
-    target_include_directories(nanors PUBLIC ${nanors_SOURCE_DIR} ${nanors_SOURCE_DIR}/deps/obl/)
+    target_include_directories(nanors
+            PUBLIC
+            ./nanors
+            ${nanors_SOURCE_DIR}
+            ${nanors_SOURCE_DIR}/deps/obl/)
 
     target_sources(nanors
-            PRIVATE ${nanors_SOURCE_DIR}/rs.c
-            PUBLIC ${nanors_SOURCE_DIR}/rs.h)
+            PRIVATE ./nanors/rswrapper.c
+            PUBLIC ./nanors/rswrapper.h)
 
-    # TODO: There's a more advanced version of this with proper support for SSSE3,
-    # see: https://github.com/LizardByte/Sunshine/pull/2828
-    set_source_files_properties(${nanors_SOURCE_DIR}/rs.c
-            PROPERTIES COMPILE_FLAGS "-include deps/obl/autoshim.h -ftree-vectorize")
+    set_source_files_properties(./nanors/rswrapper.c
+            PROPERTIES COMPILE_FLAGS "-ftree-vectorize -funroll-loops")
 
     target_link_libraries_system(moonlight PUBLIC nanors::nanors)
 endif ()
diff --git a/src/moonlight-protocol/moonlight/fec.hpp b/src/moonlight-protocol/moonlight/fec.hpp
index ec15a67c..66338963 100644
--- a/src/moonlight-protocol/moonlight/fec.hpp
+++ b/src/moonlight-protocol/moonlight/fec.hpp
@@ -3,7 +3,7 @@
 #include <memory>
 
 extern "C" {
-#include <rs.h>
+#include "rswrapper.h"
 }
 
 /**
@@ -32,7 +32,7 @@ inline void init() {
 /**
  * A smart pointer to the reed_solomon data structure, it will release the memory when going out of scope
  */
-using rs_ptr = std::unique_ptr<reed_solomon, decltype(&reed_solomon_release)>;
+using rs_ptr = std::shared_ptr<reed_solomon>;
 
 /**
  * Creates and allocates the required Reed Solomon data structure.
@@ -43,8 +43,8 @@ using rs_ptr = std::unique_ptr<reed_solomon, decltype(&reed_solomon_release)>;
  * @return A smart pointer, it will release the memory when going out of scope
  */
 inline rs_ptr create(int data_shards, int parity_shards) {
-  auto rs = reed_solomon_new(data_shards, parity_shards);
-  return {rs, ::reed_solomon_release};
+  auto rs = reed_solomon_new_fn(data_shards, parity_shards);
+  return std::shared_ptr<reed_solomon>(rs, reed_solomon_release_fn);
 }
 
 /**
@@ -63,7 +63,7 @@ inline rs_ptr create(int data_shards, int parity_shards) {
  * @return zero on success or an error code if failing.
  */
 inline int encode(reed_solomon *rs, uint8_t **shards, int nr_shards, int block_size) {
-  return reed_solomon_encode(rs, shards, nr_shards, block_size);
+  return reed_solomon_encode_fn(rs, shards, nr_shards, block_size);
 }
 
 /**
@@ -82,7 +82,7 @@ inline int encode(reed_solomon *rs, uint8_t **shards, int nr_shards, int block_s
  * @return zero on success or an error code if failing
  */
 inline int decode(reed_solomon *rs, uint8_t **shards, uint8_t *marks, int nr_shards, int block_size) {
-  return reed_solomon_decode(rs, shards, marks, nr_shards, block_size);
+  return reed_solomon_decode_fn(rs, shards, marks, nr_shards, block_size);
 }
 
 } // namespace moonlight::fec
\ No newline at end of file
diff --git a/src/moonlight-protocol/nanors/rswrapper.c b/src/moonlight-protocol/nanors/rswrapper.c
new file mode 100644
index 00000000..dd3c08e5
--- /dev/null
+++ b/src/moonlight-protocol/nanors/rswrapper.c
@@ -0,0 +1,153 @@
+/**
+ * @file src/rswrapper.c
+ * @brief Wrappers for nanors vectorization with different ISA options
+ */
+
+// _FORTIY_SOURCE can cause some versions of GCC to try to inline
+// memset() with incompatible target options when compiling rs.c
+#ifdef _FORTIFY_SOURCE
+#undef _FORTIFY_SOURCE
+#endif
+
+// The assert() function is decorated with __cold on macOS which
+// is incompatible with Clang's target multiversioning feature
+#ifndef NDEBUG
+#define NDEBUG
+#endif
+
+#define DECORATE_FUNC_I(a, b) a##b
+#define DECORATE_FUNC(a, b) DECORATE_FUNC_I(a, b)
+
+// Append an ISA suffix to the public RS API
+#define reed_solomon_init DECORATE_FUNC(reed_solomon_init, ISA_SUFFIX)
+#define reed_solomon_new DECORATE_FUNC(reed_solomon_new, ISA_SUFFIX)
+#define reed_solomon_new_static DECORATE_FUNC(reed_solomon_new_static, ISA_SUFFIX)
+#define reed_solomon_release DECORATE_FUNC(reed_solomon_release, ISA_SUFFIX)
+#define reed_solomon_decode DECORATE_FUNC(reed_solomon_decode, ISA_SUFFIX)
+#define reed_solomon_encode DECORATE_FUNC(reed_solomon_encode, ISA_SUFFIX)
+
+// Append an ISA suffix to internal functions to prevent multiple definition errors
+#define obl_axpy_ref DECORATE_FUNC(obl_axpy_ref, ISA_SUFFIX)
+#define obl_scal_ref DECORATE_FUNC(obl_scal_ref, ISA_SUFFIX)
+#define obl_axpyb32_ref DECORATE_FUNC(obl_axpyb32_ref, ISA_SUFFIX)
+#define obl_axpy DECORATE_FUNC(obl_axpy, ISA_SUFFIX)
+#define obl_scal DECORATE_FUNC(obl_scal, ISA_SUFFIX)
+#define obl_swap DECORATE_FUNC(obl_swap, ISA_SUFFIX)
+#define obl_axpyb32 DECORATE_FUNC(obl_axpyb32, ISA_SUFFIX)
+#define axpy DECORATE_FUNC(axpy, ISA_SUFFIX)
+#define scal DECORATE_FUNC(scal, ISA_SUFFIX)
+#define gemm DECORATE_FUNC(gemm, ISA_SUFFIX)
+#define invert_mat DECORATE_FUNC(invert_mat, ISA_SUFFIX)
+
+#if defined(__x86_64__) || defined(__i386__)
+
+// Compile a variant for SSSE3
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((target("ssse3"))), apply_to = function)
+#else
+#pragma GCC push_options
+#pragma GCC target("ssse3")
+#endif
+#define ISA_SUFFIX _ssse3
+#define OBLAS_SSE3
+#include "./rs.c"
+#undef OBLAS_SSE3
+#undef ISA_SUFFIX
+#if defined(__clang__)
+#pragma clang attribute pop
+#else
+#pragma GCC pop_options
+#endif
+
+// Compile a variant for AVX2
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((target("avx2"))), apply_to = function)
+#else
+#pragma GCC push_options
+#pragma GCC target("avx2")
+#endif
+#define ISA_SUFFIX _avx2
+#define OBLAS_AVX2
+#include "./rs.c"
+#undef OBLAS_AVX2
+#undef ISA_SUFFIX
+#if defined(__clang__)
+#pragma clang attribute pop
+#else
+#pragma GCC pop_options
+#endif
+
+// Compile a variant for AVX512BW
+#if defined(__clang__)
+#pragma clang attribute push(__attribute__((target("avx512f,avx512bw"))), apply_to = function)
+#else
+#pragma GCC push_options
+#pragma GCC target("avx512f,avx512bw")
+#endif
+#define ISA_SUFFIX _avx512
+#define OBLAS_AVX512
+#include "./rs.c"
+#undef OBLAS_AVX512
+#undef ISA_SUFFIX
+#if defined(__clang__)
+#pragma clang attribute pop
+#else
+#pragma GCC pop_options
+#endif
+
+#endif
+
+// Compile a default variant
+#define ISA_SUFFIX _def
+#include "./autoshim.h"
+#include "./rs.c"
+#undef ISA_SUFFIX
+
+#undef reed_solomon_init
+#undef reed_solomon_new
+#undef reed_solomon_new_static
+#undef reed_solomon_release
+#undef reed_solomon_decode
+#undef reed_solomon_encode
+
+#include "rswrapper.h"
+
+reed_solomon_new_t reed_solomon_new_fn;
+reed_solomon_release_t reed_solomon_release_fn;
+reed_solomon_encode_t reed_solomon_encode_fn;
+reed_solomon_decode_t reed_solomon_decode_fn;
+
+/**
+ * @brief This initializes the RS function pointers to the best vectorized version available.
+ * @details The streaming code will directly invoke these function pointers during encoding.
+ */
+void reed_solomon_init(void) {
+#if defined(__x86_64__) || defined(__i386__)
+  if (__builtin_cpu_supports("avx512f") && __builtin_cpu_supports("avx512bw")) {
+    reed_solomon_new_fn = reed_solomon_new_avx512;
+    reed_solomon_release_fn = reed_solomon_release_avx512;
+    reed_solomon_encode_fn = reed_solomon_encode_avx512;
+    reed_solomon_decode_fn = reed_solomon_decode_avx512;
+    reed_solomon_init_avx512();
+  } else if (__builtin_cpu_supports("avx2")) {
+    reed_solomon_new_fn = reed_solomon_new_avx2;
+    reed_solomon_release_fn = reed_solomon_release_avx2;
+    reed_solomon_encode_fn = reed_solomon_encode_avx2;
+    reed_solomon_decode_fn = reed_solomon_decode_avx2;
+    reed_solomon_init_avx2();
+  } else if (__builtin_cpu_supports("ssse3")) {
+    reed_solomon_new_fn = reed_solomon_new_ssse3;
+    reed_solomon_release_fn = reed_solomon_release_ssse3;
+    reed_solomon_encode_fn = reed_solomon_encode_ssse3;
+    reed_solomon_decode_fn = reed_solomon_decode_ssse3;
+    reed_solomon_init_ssse3();
+  } else
+#endif
+  {
+    reed_solomon_new_fn = reed_solomon_new_def;
+    reed_solomon_release_fn = reed_solomon_release_def;
+    reed_solomon_encode_fn = reed_solomon_encode_def;
+    reed_solomon_decode_fn = reed_solomon_decode_def;
+    reed_solomon_init_def();
+  }
+}
diff --git a/src/moonlight-protocol/nanors/rswrapper.h b/src/moonlight-protocol/nanors/rswrapper.h
new file mode 100644
index 00000000..e705e3b3
--- /dev/null
+++ b/src/moonlight-protocol/nanors/rswrapper.h
@@ -0,0 +1,21 @@
+/**
+* @file src/rswrapper.h
+* @brief Wrappers for nanors vectorization
+* @details This is a drop-in replacement for nanors rs.h
+*/
+#pragma once
+
+#include "rs.h"
+#include <stdint.h>
+
+typedef struct _reed_solomon reed_solomon;
+
+typedef reed_solomon *(*reed_solomon_new_t)(int data_shards, int parity_shards);
+typedef void (*reed_solomon_release_t)(reed_solomon *rs);
+typedef int (*reed_solomon_encode_t)(reed_solomon *rs, uint8_t **shards, int nr_shards, int bs);
+typedef int (*reed_solomon_decode_t)(reed_solomon *rs, uint8_t **shards, uint8_t *marks, int nr_shards, int bs);
+
+extern reed_solomon_new_t reed_solomon_new_fn;
+extern reed_solomon_release_t reed_solomon_release_fn;
+extern reed_solomon_encode_t reed_solomon_encode_fn;
+extern reed_solomon_decode_t reed_solomon_decode_fn;
\ No newline at end of file