diff --git a/sharpyuv/Makefile.am b/sharpyuv/Makefile.am
index eda07187..ac5862b6 100644
--- a/sharpyuv/Makefile.am
+++ b/sharpyuv/Makefile.am
@@ -12,6 +12,7 @@ libsharpyuvinclude_HEADERS += sharpyuv.h
 libsharpyuvinclude_HEADERS += sharpyuv_csp.h
 noinst_HEADERS =
 noinst_HEADERS += ../src/dsp/cpu.h
+noinst_HEADERS += ../src/webp/types.h
 
 libsharpyuv_sse2_la_SOURCES =
 libsharpyuv_sse2_la_SOURCES += sharpyuv_sse2.c
diff --git a/sharpyuv/sharpyuv.c b/sharpyuv/sharpyuv.c
index 6497fb02..5ff02225 100644
--- a/sharpyuv/sharpyuv.c
+++ b/sharpyuv/sharpyuv.c
@@ -15,7 +15,7 @@
 
 #include <assert.h>
 #include <limits.h>
-#include <math.h>
+#include <stddef.h>
 #include <stdlib.h>
 #include <string.h>
 
@@ -414,6 +414,22 @@ static int DoSharpArgbToYuv(const uint8_t* r_ptr, const uint8_t* g_ptr,
 }
 #undef SAFE_ALLOC
 
+#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
+#include <pthread.h>  // NOLINT
+
+#define LOCK_ACCESS \
+    static pthread_mutex_t sharpyuv_lock = PTHREAD_MUTEX_INITIALIZER; \
+    if (pthread_mutex_lock(&sharpyuv_lock)) return
+#define UNLOCK_ACCESS_AND_RETURN                  \
+    do {                                          \
+      (void)pthread_mutex_unlock(&sharpyuv_lock); \
+      return;                                     \
+    } while (0)
+#else  // !(defined(WEBP_USE_THREAD) && !defined(_WIN32))
+#define LOCK_ACCESS do {} while (0)
+#define UNLOCK_ACCESS_AND_RETURN return
+#endif  // defined(WEBP_USE_THREAD) && !defined(_WIN32)
+
 // Hidden exported init function.
 // By default SharpYuvConvert calls it with NULL. If needed, users can declare
 // it as extern and call it with a VP8CPUInfo function.
@@ -421,17 +437,21 @@ SHARPYUV_EXTERN void SharpYuvInit(VP8CPUInfo cpu_info_func);
 void SharpYuvInit(VP8CPUInfo cpu_info_func) {
   static volatile VP8CPUInfo sharpyuv_last_cpuinfo_used =
       (VP8CPUInfo)&sharpyuv_last_cpuinfo_used;
-  const int initialized =
+  LOCK_ACCESS;
+  {
+    const int initialized =
       (sharpyuv_last_cpuinfo_used != (VP8CPUInfo)&sharpyuv_last_cpuinfo_used);
-  if (cpu_info_func == NULL && initialized) return;
-  if (sharpyuv_last_cpuinfo_used == cpu_info_func) return;
+    if (cpu_info_func == NULL && initialized) UNLOCK_ACCESS_AND_RETURN;
+    if (sharpyuv_last_cpuinfo_used == cpu_info_func) UNLOCK_ACCESS_AND_RETURN;
 
-  SharpYuvInitDsp(cpu_info_func);
-  if (!initialized) {
-    SharpYuvInitGammaTables();
+    SharpYuvInitDsp(cpu_info_func);
+    if (!initialized) {
+      SharpYuvInitGammaTables();
+    }
+
+    sharpyuv_last_cpuinfo_used = cpu_info_func;
   }
-
-  sharpyuv_last_cpuinfo_used = cpu_info_func;
+  UNLOCK_ACCESS_AND_RETURN;
 }
 
 int SharpYuvConvert(const void* r_ptr, const void* g_ptr,
diff --git a/sharpyuv/sharpyuv.h b/sharpyuv/sharpyuv.h
index 2ea7564d..28040a1d 100644
--- a/sharpyuv/sharpyuv.h
+++ b/sharpyuv/sharpyuv.h
@@ -12,8 +12,6 @@
 #ifndef WEBP_SHARPYUV_SHARPYUV_H_
 #define WEBP_SHARPYUV_SHARPYUV_H_
 
-#include <inttypes.h>
-
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/sharpyuv/sharpyuv_csp.c b/sharpyuv/sharpyuv_csp.c
index 5334fa64..0ad22be9 100644
--- a/sharpyuv/sharpyuv_csp.c
+++ b/sharpyuv/sharpyuv_csp.c
@@ -13,7 +13,7 @@
 
 #include <assert.h>
 #include <math.h>
-#include <string.h>
+#include <stddef.h>
 
 static int ToFixed16(float f) { return (int)floor(f * (1 << 16) + 0.5f); }
 
diff --git a/sharpyuv/sharpyuv_dsp.h b/sharpyuv/sharpyuv_dsp.h
index e561d8d3..c7755b97 100644
--- a/sharpyuv/sharpyuv_dsp.h
+++ b/sharpyuv/sharpyuv_dsp.h
@@ -12,9 +12,8 @@
 #ifndef WEBP_SHARPYUV_SHARPYUV_DSP_H_
 #define WEBP_SHARPYUV_SHARPYUV_DSP_H_
 
-#include <stdint.h>
-
 #include "src/dsp/cpu.h"
+#include "src/webp/types.h"
 
 extern uint64_t (*SharpYuvUpdateY)(const uint16_t* src, const uint16_t* ref,
                                    uint16_t* dst, int len, int bit_depth);
diff --git a/sharpyuv/sharpyuv_gamma.c b/sharpyuv/sharpyuv_gamma.c
index 05b5436f..20ab2da6 100644
--- a/sharpyuv/sharpyuv_gamma.c
+++ b/sharpyuv/sharpyuv_gamma.c
@@ -13,7 +13,6 @@
 
 #include <assert.h>
 #include <math.h>
-#include <stdint.h>
 
 #include "src/webp/types.h"
 
diff --git a/sharpyuv/sharpyuv_gamma.h b/sharpyuv/sharpyuv_gamma.h
index 2f1a3ff4..d13aff59 100644
--- a/sharpyuv/sharpyuv_gamma.h
+++ b/sharpyuv/sharpyuv_gamma.h
@@ -12,7 +12,7 @@
 #ifndef WEBP_SHARPYUV_SHARPYUV_GAMMA_H_
 #define WEBP_SHARPYUV_SHARPYUV_GAMMA_H_
 
-#include <stdint.h>
+#include "src/webp/types.h"
 
 #ifdef __cplusplus
 extern "C" {
diff --git a/sharpyuv/sharpyuv_neon.c b/sharpyuv/sharpyuv_neon.c
index 5cf6aaff..58409148 100644
--- a/sharpyuv/sharpyuv_neon.c
+++ b/sharpyuv/sharpyuv_neon.c
@@ -17,11 +17,6 @@
 #include <assert.h>
 #include <stdlib.h>
 #include <arm_neon.h>
-#endif
-
-extern void InitSharpYuvNEON(void);
-
-#if defined(WEBP_USE_NEON)
 
 static uint16_t clip_NEON(int v, int max) {
   return (v < 0) ? 0 : (v > max) ? max : (uint16_t)v;
@@ -169,6 +164,8 @@ static void SharpYuvFilterRow_NEON(const int16_t* A, const int16_t* B, int len,
 
 //------------------------------------------------------------------------------
 
+extern void InitSharpYuvNEON(void);
+
 WEBP_TSAN_IGNORE_FUNCTION void InitSharpYuvNEON(void) {
   SharpYuvUpdateY = SharpYuvUpdateY_NEON;
   SharpYuvUpdateRGB = SharpYuvUpdateRGB_NEON;
@@ -177,6 +174,8 @@ WEBP_TSAN_IGNORE_FUNCTION void InitSharpYuvNEON(void) {
 
 #else  // !WEBP_USE_NEON
 
+extern void InitSharpYuvNEON(void);
+
 void InitSharpYuvNEON(void) {}
 
 #endif  // WEBP_USE_NEON
diff --git a/sharpyuv/sharpyuv_sse2.c b/sharpyuv/sharpyuv_sse2.c
index 19438737..9744d1bb 100644
--- a/sharpyuv/sharpyuv_sse2.c
+++ b/sharpyuv/sharpyuv_sse2.c
@@ -16,11 +16,6 @@
 #if defined(WEBP_USE_SSE2)
 #include <stdlib.h>
 #include <emmintrin.h>
-#endif
-
-extern void InitSharpYuvSSE2(void);
-
-#if defined(WEBP_USE_SSE2)
 
 static uint16_t clip_SSE2(int v, int max) {
   return (v < 0) ? 0 : (v > max) ? max : (uint16_t)v;
@@ -199,6 +194,8 @@ WEBP_TSAN_IGNORE_FUNCTION void InitSharpYuvSSE2(void) {
 }
 #else  // !WEBP_USE_SSE2
 
+extern void InitSharpYuvSSE2(void);
+
 void InitSharpYuvSSE2(void) {}
 
 #endif  // WEBP_USE_SSE2
diff --git a/src/dsp/cpu.h b/src/dsp/cpu.h
index 57a40d87..be80727c 100644
--- a/src/dsp/cpu.h
+++ b/src/dsp/cpu.h
@@ -14,6 +14,8 @@
 #ifndef WEBP_DSP_CPU_H_
 #define WEBP_DSP_CPU_H_
 
+#include <stddef.h>
+
 #ifdef HAVE_CONFIG_H
 #include "src/webp/config.h"
 #endif
diff --git a/src/enc/picture_csp_enc.c b/src/enc/picture_csp_enc.c
index 8a3a86c4..78c8ca47 100644
--- a/src/enc/picture_csp_enc.c
+++ b/src/enc/picture_csp_enc.c
@@ -172,21 +172,6 @@ static const int kMinDimensionIterativeConversion = 4;
 //------------------------------------------------------------------------------
 // Main function
 
-extern void SharpYuvInit(VP8CPUInfo cpu_info_func);
-
-static void SafeInitSharpYuv(void) {
-#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
-  static pthread_mutex_t initsharpyuv_lock = PTHREAD_MUTEX_INITIALIZER;
-  if (pthread_mutex_lock(&initsharpyuv_lock)) return;
-#endif
-
-  SharpYuvInit(VP8GetCPUInfo);
-
-#if defined(WEBP_USE_THREAD) && !defined(_WIN32)
-  (void)pthread_mutex_unlock(&initsharpyuv_lock);
-#endif
-}
-
 static int PreprocessARGB(const uint8_t* r_ptr,
                           const uint8_t* g_ptr,
                           const uint8_t* b_ptr,
@@ -483,6 +468,8 @@ static WEBP_INLINE void ConvertRowsToUV(const uint16_t* rgb,
   }
 }
 
+extern void SharpYuvInit(VP8CPUInfo cpu_info_func);
+
 static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
                               const uint8_t* g_ptr,
                               const uint8_t* b_ptr,
@@ -518,7 +505,7 @@ static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
   }
 
   if (use_iterative_conversion) {
-    SafeInitSharpYuv();
+    SharpYuvInit(VP8GetCPUInfo);
     if (!PreprocessARGB(r_ptr, g_ptr, b_ptr, step, rgb_stride, picture)) {
       return 0;
     }