DebugLog changes (#2073)

@tensorflow/micro

Change DebugLog to allow stdio style formatted output using either <cstdio> or a third party printf library.

Remove micro_string code.

bug=fixes #2083
diff --git a/tensorflow/lite/kernels/op_macros.h b/tensorflow/lite/kernels/op_macros.h
index 4255d25..39e3119 100644
--- a/tensorflow/lite/kernels/op_macros.h
+++ b/tensorflow/lite/kernels/op_macros.h
@@ -1,4 +1,4 @@
-/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -15,14 +15,13 @@
 #ifndef TENSORFLOW_LITE_KERNELS_OP_MACROS_H_
 #define TENSORFLOW_LITE_KERNELS_OP_MACROS_H_
 
-#include "tensorflow/lite/micro/debug_log.h"
-
 #if !defined(TF_LITE_MCU_DEBUG_LOG)
 #include <cstdlib>
 #define TFLITE_ABORT abort()
 #else
+#include "tensorflow/lite/micro/micro_log.h"
 inline void AbortImpl() {
-  DebugLog("HALTED\n");
+  MicroPrintf("HALTED");
   while (1) {
   }
 }
diff --git a/tensorflow/lite/micro/BUILD b/tensorflow/lite/micro/BUILD
index 72c23f3..475c541 100644
--- a/tensorflow/lite/micro/BUILD
+++ b/tensorflow/lite/micro/BUILD
@@ -269,7 +269,6 @@
     copts = micro_copts(),
     deps = [
         ":debug_log",
-        ":micro_string",
     ],
 )
 
@@ -292,17 +291,6 @@
 )
 
 cc_library(
-    name = "micro_string",
-    srcs = [
-        "micro_string.cc",
-    ],
-    hdrs = [
-        "micro_string.h",
-    ],
-    copts = micro_copts(),
-)
-
-cc_library(
     name = "micro_time",
     srcs = [
         "micro_time.cc",
@@ -550,17 +538,6 @@
 )
 
 cc_test(
-    name = "micro_string_test",
-    srcs = [
-        "micro_string_test.cc",
-    ],
-    deps = [
-        ":micro_string",
-        "//tensorflow/lite/micro/testing:micro_test",
-    ],
-)
-
-cc_test(
     name = "micro_time_test",
     srcs = [
         "micro_time_test.cc",
diff --git a/tensorflow/lite/micro/arc_emsdp/debug_log.cc b/tensorflow/lite/micro/arc_emsdp/debug_log.cc
index 1b4d641..55cf420 100644
--- a/tensorflow/lite/micro/arc_emsdp/debug_log.cc
+++ b/tensorflow/lite/micro/arc_emsdp/debug_log.cc
@@ -1,4 +1,4 @@
-/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -19,6 +19,12 @@
 #include <cstdio>
 #include <cstring>
 
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+#include "eyalroz_printf/src/printf/printf.h"
+#endif
+
+namespace {
+
 // Print to debug console by default. One can define next to extend destinations
 // set: EMSDP_LOG_TO_MEMORY
 //   : fill .debug_log memory region (data section) with passed chars.
@@ -89,9 +95,7 @@
   debug_log_mem[cursor] = '^';
 }
 
-extern "C" void DebugLog(const char* s) {
-#ifndef TF_LITE_STRIP_ERROR_STRINGS
-
+void LogDebugString(const char* s) {
 #if defined EMSDP_LOG_TO_UART
   DbgUartSendStr(s);
 #endif
@@ -106,6 +110,16 @@
 #warning "EMSDP_LOG_TO_HOST is defined. Ensure hostlib is linked."
   fprintf(stderr, "%s", s);
 #endif
-
-#endif  // TF_LITE_STRIP_ERROR_STRINGS
 }
+
+}  // namespace
+
+extern "C" void DebugLog(const char* format, va_list args) {
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+  constexpr int kMaxLogLen = 256;
+  char log_buffer[kMaxLogLen];
+
+  vsnprintf_(log_buffer, kMaxLogLen, format, args);
+  LogDebugString(log_buffer);
+#endif
+}
\ No newline at end of file
diff --git a/tensorflow/lite/micro/bluepill/debug_log.cc b/tensorflow/lite/micro/bluepill/debug_log.cc
index 3fd2d52..6d0d53b 100644
--- a/tensorflow/lite/micro/bluepill/debug_log.cc
+++ b/tensorflow/lite/micro/bluepill/debug_log.cc
@@ -1,4 +1,4 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -15,9 +15,16 @@
 
 #include "tensorflow/lite/micro/debug_log.h"
 
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+#include "eyalroz_printf/src/printf/printf.h"
+#endif
+
+namespace {
+
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
 // For Arm Cortex-M devices, calling SYS_WRITE0 will output the zero-terminated
 // string pointed to by R1 to any debug console that's attached to the system.
-extern "C" void DebugLog(const char* s) {
+void SysWriteDebugConsole(const char* s) {
   asm("mov r0, #0x04\n"  // SYS_WRITE0
       "mov r1, %[str]\n"
       "bkpt #0xAB\n"
@@ -25,3 +32,16 @@
       : [str] "r"(s)
       : "r0", "r1");
 }
+#endif  // TF_LITE_STRIP_ERROR_STRINGS
+
+}  // namespace
+
+extern "C" void DebugLog(const char* format, va_list args) {
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+  constexpr int kMaxLogLen = 256;
+  char log_buffer[kMaxLogLen];
+
+  vsnprintf_(log_buffer, kMaxLogLen, format, args);
+  SysWriteDebugConsole(log_buffer);
+#endif  // TF_LITE_STRIP_ERROR_STRINGS
+}
diff --git a/tensorflow/lite/micro/chre/debug_log.cc b/tensorflow/lite/micro/chre/debug_log.cc
index 23bb82e..b599ebe 100644
--- a/tensorflow/lite/micro/chre/debug_log.cc
+++ b/tensorflow/lite/micro/chre/debug_log.cc
@@ -1,4 +1,4 @@
-/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -17,6 +17,16 @@
 
 #include <chre.h>
 
-extern "C" void DebugLog(const char* s) {
-  chreLog(CHRE_LOG_DEBUG, "[TFL_MICRO] %s", s);
-}
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+#include "eyalroz_printf/src/printf/printf.h"
+#endif
+
+extern "C" void DebugLog(const char* format, va_list args) {
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+  constexpr int kMaxLogLen = 256;
+  char log_buffer[kMaxLogLen];
+
+  vsnprintf_(log_buffer, kMaxLogLen, format, args);
+  chreLog(CHRE_LOG_DEBUG, "[TFL_MICRO] %s", log_buffer);
+#endif
+}
\ No newline at end of file
diff --git a/tensorflow/lite/micro/cortex_m_generic/debug_log.cc b/tensorflow/lite/micro/cortex_m_generic/debug_log.cc
index bc79d43..2c237fe 100644
--- a/tensorflow/lite/micro/cortex_m_generic/debug_log.cc
+++ b/tensorflow/lite/micro/cortex_m_generic/debug_log.cc
@@ -1,4 +1,4 @@
-/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -24,17 +24,33 @@
 
 #include "tensorflow/lite/micro/cortex_m_generic/debug_log_callback.h"
 
+#ifndef TF_LITE_STRIP_ERROR_STRINGS
+#include "eyalroz_printf/src/printf/printf.h"
+#endif
+
 static DebugLogCallback debug_log_callback = nullptr;
 
+namespace {
+
+void InvokeDebugLogCallback(const char* s) {
+  if (debug_log_callback != nullptr) {
+    debug_log_callback(s);
+  }
+}
+
+}  // namespace
+
 void RegisterDebugLogCallback(void (*cb)(const char* s)) {
   debug_log_callback = cb;
 }
 
-void DebugLog(const char* s) {
+void DebugLog(const char* format, va_list args) {
 #ifndef TF_LITE_STRIP_ERROR_STRINGS
-  if (debug_log_callback != nullptr) {
-    debug_log_callback(s);
-  }
+  constexpr int kMaxLogLen = 256;
+  char log_buffer[kMaxLogLen];
+
+  vsnprintf_(log_buffer, kMaxLogLen, format, args);
+  InvokeDebugLogCallback(log_buffer);
 #endif
 }
 
diff --git a/tensorflow/lite/micro/debug_log.cc b/tensorflow/lite/micro/debug_log.cc
index 46ca253..a8ef36b 100644
--- a/tensorflow/lite/micro/debug_log.cc
+++ b/tensorflow/lite/micro/debug_log.cc
@@ -1,4 +1,4 @@
-/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -18,21 +18,17 @@
 // the only function that's absolutely required to be available on a target
 // device, since it's used for communicating test results back to the host so
 // that we can verify the implementation is working correctly.
-// It's designed to be as easy as possible to supply an implementation though.
-// On platforms that have a POSIX stack or C library, it can be written as a
-// single call to `fprintf(stderr, "%s", s)` to output a string to the error
-// stream of the console, but if there's no OS or C library available, there's
-// almost always an equivalent way to write out a string to some serial
-// interface that can be used instead. For example on Arm M-series MCUs, calling
-// the `bkpt #0xAB` assembler instruction will output the string in r1 to
-// whatever debug serial connection is available. If you're running mbed, you
-// can do the same by creating `Serial pc(USBTX, USBRX)` and then calling
-// `pc.printf("%s", s)`.
-// To add an equivalent function for your own platform, create your own
-// implementation file, and place it in a subfolder with named after the OS
-// you're targeting. For example, see the Cortex M bare metal version in
-// tensorflow/lite/micro/bluepill/debug_log.cc or the mbed one on
-// tensorflow/lite/micro/mbed/debug_log.cc.
+// This function should support standard C/C++ stdio style formatting
+// operations. It's designed to be as easy as possible to supply an
+// implementation though. On platforms that have a POSIX stack or C library, it
+// can be written as a single call to `vfprintf(stderr, format, args)` to output
+// a string to the error stream of the console, but if there's no OS or C
+// library available, there's almost always an equivalent way to write out a
+// string to some serial interface that can be used instead. To add an
+// equivalent function for your own platform, create your own implementation
+// file, and place it in a subfolder with named after the OS you're targeting.
+// For example, see the Cortex M bare metal version in the
+// tensorflow/lite/micro/bluepill/debug_log.cc file.
 
 #include "tensorflow/lite/micro/debug_log.h"
 
@@ -40,11 +36,11 @@
 #include <cstdio>
 #endif
 
-extern "C" void DebugLog(const char* s) {
+extern "C" void DebugLog(const char* format, va_list args) {
 #ifndef TF_LITE_STRIP_ERROR_STRINGS
   // Reusing TF_LITE_STRIP_ERROR_STRINGS to disable DebugLog completely to get
   // maximum reduction in binary size. This is because we have DebugLog calls
   // via TF_LITE_CHECK that are not stubbed out by TF_LITE_REPORT_ERROR.
-  fprintf(stderr, "%s", s);
+  vfprintf(stderr, format, args);
 #endif
 }
diff --git a/tensorflow/lite/micro/debug_log.h b/tensorflow/lite/micro/debug_log.h
index c2840d0..c0cf699 100644
--- a/tensorflow/lite/micro/debug_log.h
+++ b/tensorflow/lite/micro/debug_log.h
@@ -1,4 +1,4 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -16,13 +16,20 @@
 #define TENSORFLOW_LITE_MICRO_DEBUG_LOG_H_
 
 #ifdef __cplusplus
+#include <cstdarg>
+#else
+#include <stdarg.h>
+#endif  // __cplusplus
+
+#ifdef __cplusplus
 extern "C" {
 #endif  // __cplusplus
 
 // This function should be implemented by each target platform, and provide a
 // way for strings to be output to some text stream. For more information, see
-// tensorflow/lite/micro/debug_log.cc.
-void DebugLog(const char* s);
+// the tensorflow/lite/micro/debug_log.cc file.  This function should support
+// standard C/C++ stdio style formatting operations.
+void DebugLog(const char* format, va_list args);
 
 #ifdef __cplusplus
 }  // extern "C"
diff --git a/tensorflow/lite/micro/memory_planner/BUILD b/tensorflow/lite/micro/memory_planner/BUILD
index 0329e73..7111da4 100644
--- a/tensorflow/lite/micro/memory_planner/BUILD
+++ b/tensorflow/lite/micro/memory_planner/BUILD
@@ -52,7 +52,6 @@
         ":micro_memory_planner",
         "//tensorflow/lite/micro:micro_compatibility",
         "//tensorflow/lite/micro:micro_log",
-        "//tensorflow/lite/micro:micro_string",
     ],
 )
 
diff --git a/tensorflow/lite/micro/memory_planner/greedy_memory_planner.cc b/tensorflow/lite/micro/memory_planner/greedy_memory_planner.cc
index 471a5b2..a087b23 100644
--- a/tensorflow/lite/micro/memory_planner/greedy_memory_planner.cc
+++ b/tensorflow/lite/micro/memory_planner/greedy_memory_planner.cc
@@ -1,4 +1,4 @@
-/* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -16,7 +16,6 @@
 #include "tensorflow/lite/micro/memory_planner/greedy_memory_planner.h"
 
 #include "tensorflow/lite/micro/micro_log.h"
-#include "tensorflow/lite/micro/micro_string.h"
 
 namespace tflite {
 
diff --git a/tensorflow/lite/micro/micro_context.cc b/tensorflow/lite/micro/micro_context.cc
index b06252a..352d2c0 100644
--- a/tensorflow/lite/micro/micro_context.cc
+++ b/tensorflow/lite/micro/micro_context.cc
@@ -1,4 +1,4 @@
-/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -142,7 +142,7 @@
                                const char* format, ...) {
   va_list args;
   va_start(args, format);
-  Log(format, args);
+  VMicroPrintf(format, args);
   va_end(args);
 }
 
diff --git a/tensorflow/lite/micro/micro_log.cc b/tensorflow/lite/micro/micro_log.cc
index 9c8ccaa..45f7051 100644
--- a/tensorflow/lite/micro/micro_log.cc
+++ b/tensorflow/lite/micro/micro_log.cc
@@ -1,4 +1,4 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -17,31 +17,33 @@
 
 #include <cstdarg>
 #include <cstdint>
-#include <new>
 
 #if !defined(TF_LITE_STRIP_ERROR_STRINGS)
 #include "tensorflow/lite/micro/debug_log.h"
-#include "tensorflow/lite/micro/micro_string.h"
 #endif
 
-void Log(const char* format, va_list args) {
 #if !defined(TF_LITE_STRIP_ERROR_STRINGS)
-  // Only pulling in the implementation of this function for builds where we
-  // expect to make use of it to be extra cautious about not increasing the code
-  // size.
-  static constexpr int kMaxLogLen = 256;
-  char log_buffer[kMaxLogLen];
-  MicroVsnprintf(log_buffer, kMaxLogLen, format, args);
-  DebugLog(log_buffer);
-  DebugLog("\r\n");
-#endif
+namespace {
+
+void VDebugLog(const char* format, ...) {
+  va_list args;
+  va_start(args, format);
+  DebugLog(format, args);
+  va_end(args);
 }
 
-#if !defined(TF_LITE_STRIP_ERROR_STRINGS)
+}  // namespace
+
+void VMicroPrintf(const char* format, va_list args) {
+  DebugLog(format, args);
+  // TODO(b/290051015): remove "\r\n"
+  VDebugLog("\r\n");
+}
+
 void MicroPrintf(const char* format, ...) {
   va_list args;
   va_start(args, format);
-  Log(format, args);
+  VMicroPrintf(format, args);
   va_end(args);
 }
-#endif
+#endif  // !defined(TF_LITE_STRIP_ERROR_STRINGS)
diff --git a/tensorflow/lite/micro/micro_log.h b/tensorflow/lite/micro/micro_log.h
index d9cfbe8..669e9e6 100644
--- a/tensorflow/lite/micro/micro_log.h
+++ b/tensorflow/lite/micro/micro_log.h
@@ -1,4 +1,4 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -15,20 +15,17 @@
 #ifndef TENSORFLOW_LITE_MICRO_MICRO_LOG_H_
 #define TENSORFLOW_LITE_MICRO_MICRO_LOG_H_
 
-#include <cstdarg>
-
-// This is a free function used to perform the actual logging.
-// This function will be used by MicroPrintf and MicroErrorReporter::Report()
-void Log(const char* format, va_list args);
-
 #if !defined(TF_LITE_STRIP_ERROR_STRINGS)
-// This function can be used independent of the MicroErrorReporter to get
+#include <cstdarg>
+// These functions can be used independent of the MicroErrorReporter to get
 // printf-like functionalitys and are common to all target platforms.
 void MicroPrintf(const char* format, ...);
+void VMicroPrintf(const char* format, va_list args);
 #else
 // We use a #define to ensure that the strings are completely stripped, to
 // prevent an unnecessary increase in the binary size.
 #define MicroPrintf(...) tflite::Unused(__VA_ARGS__)
+#define VMicroPrintf(...) tflite::Unused(__VA_ARGS__)
 #endif
 
 namespace tflite {
@@ -39,6 +36,7 @@
 void Unused(Args&&... args) {
   (void)(sizeof...(args));
 }
+
 }  // namespace tflite
 
 #endif  // TENSORFLOW_LITE_MICRO_MICRO_LOG_H_
diff --git a/tensorflow/lite/micro/micro_string.cc b/tensorflow/lite/micro/micro_string.cc
deleted file mode 100644
index bb41a9e..0000000
--- a/tensorflow/lite/micro/micro_string.cc
+++ /dev/null
@@ -1,317 +0,0 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-==============================================================================*/
-
-// Implements debug logging for numbers by converting them into strings and then
-// calling the main DebugLog(char*) function. These are separated into a
-// different file so that platforms can just implement the string output version
-// of DebugLog() and then get the numerical variations without requiring any
-// more code.
-
-#include "tensorflow/lite/micro/micro_string.h"
-
-#include <cstdarg>
-#include <cstdint>
-#include <cstring>
-
-namespace {
-
-// Int formats can need up to 10 bytes for the value plus a single byte for the
-// sign.
-constexpr int kMaxIntCharsNeeded = 10 + 1;
-// Hex formats can need up to 8 bytes for the value plus two bytes for the "0x".
-constexpr int kMaxHexCharsNeeded = 8 + 2;
-
-// Float formats can need up to 7 bytes for the fraction plus 3 bytes for "x2^"
-// plus 3 bytes for the exponent and a single sign bit.
-constexpr float kMaxFloatCharsNeeded = 7 + 3 + 3 + 1;
-
-// All input buffers to the number conversion functions must be this long.
-const int kFastToBufferSize = 48;
-
-// Reverses a zero-terminated string in-place.
-char* ReverseStringInPlace(char* start, char* end) {
-  char* p1 = start;
-  char* p2 = end - 1;
-  while (p1 < p2) {
-    char tmp = *p1;
-    *p1++ = *p2;
-    *p2-- = tmp;
-  }
-  return start;
-}
-
-// Appends a string to a string, in-place. You need to pass in the maximum
-// string length as the second argument.
-char* StrCatStr(char* main, int main_max_length, const char* to_append) {
-  char* current = main;
-  while (*current != 0) {
-    ++current;
-  }
-  char* current_end = main + (main_max_length - 1);
-  while ((*to_append != 0) && (current < current_end)) {
-    *current = *to_append;
-    ++current;
-    ++to_append;
-  }
-  *current = 0;
-  return current;
-}
-
-// Populates the provided buffer with an ASCII representation of the number.
-char* FastUInt32ToBufferLeft(uint32_t i, char* buffer, int base) {
-  char* start = buffer;
-  do {
-    int32_t digit = i % base;
-    char character;
-    if (digit < 10) {
-      character = '0' + digit;
-    } else {
-      character = 'a' + (digit - 10);
-    }
-    *buffer++ = character;
-    i /= base;
-  } while (i > 0);
-  *buffer = 0;
-  ReverseStringInPlace(start, buffer);
-  return buffer;
-}
-
-// Populates the provided buffer with an ASCII representation of the number.
-char* FastInt32ToBufferLeft(int32_t i, char* buffer) {
-  uint32_t u = i;
-  if (i < 0) {
-    *buffer++ = '-';
-    u = -u;
-  }
-  return FastUInt32ToBufferLeft(u, buffer, 10);
-}
-
-// Converts a number to a string and appends it to another.
-char* StrCatInt32(char* main, int main_max_length, int32_t number) {
-  char number_string[kFastToBufferSize];
-  FastInt32ToBufferLeft(number, number_string);
-  return StrCatStr(main, main_max_length, number_string);
-}
-
-// Converts a number to a string and appends it to another.
-char* StrCatUInt32(char* main, int main_max_length, uint32_t number, int base) {
-  char number_string[kFastToBufferSize];
-  FastUInt32ToBufferLeft(number, number_string, base);
-  return StrCatStr(main, main_max_length, number_string);
-}
-
-// Populates the provided buffer with ASCII representation of the float number.
-// Avoids the use of any floating point instructions (since these aren't
-// supported on many microcontrollers) and as a consequence prints values with
-// power-of-two exponents.
-char* FastFloatToBufferLeft(float f, char* buffer) {
-  char* current = buffer;
-  char* current_end = buffer + (kFastToBufferSize - 1);
-  // Access the bit fields of the floating point value to avoid requiring any
-  // float instructions. These constants are derived from IEEE 754.
-  const uint32_t sign_mask = 0x80000000;
-  const uint32_t exponent_mask = 0x7f800000;
-  const int32_t exponent_shift = 23;
-  const int32_t exponent_bias = 127;
-  const uint32_t fraction_mask = 0x007fffff;
-  uint32_t u;
-  memcpy(&u, &f, sizeof(int32_t));
-  const int32_t exponent =
-      ((u & exponent_mask) >> exponent_shift) - exponent_bias;
-  const uint32_t fraction = (u & fraction_mask);
-  // Expect ~0x2B1B9D3 for fraction.
-  if (u & sign_mask) {
-    *current = '-';
-    current += 1;
-  }
-  *current = 0;
-  // These are special cases for infinities and not-a-numbers.
-  if (exponent == 128) {
-    if (fraction == 0) {
-      current = StrCatStr(current, (current_end - current), "Inf");
-      return current;
-    } else {
-      current = StrCatStr(current, (current_end - current), "NaN");
-      return current;
-    }
-  }
-  // 0x007fffff (8388607) represents 0.99... for the fraction, so to print the
-  // correct decimal digits we need to scale our value before passing it to the
-  // conversion function. This scale should be 10000000/8388608 = 1.1920928955.
-  // We can approximate this using multiply-adds and right-shifts using the
-  // values in this array. The 1. portion of the number string is printed out
-  // in a fixed way before the fraction, below.
-  const int32_t scale_shifts_size = 13;
-  const int8_t scale_shifts[13] = {3,  4,  8,  11, 13, 14, 17,
-                                   18, 19, 20, 21, 22, 23};
-  uint32_t scaled_fraction = fraction;
-  for (int i = 0; i < scale_shifts_size; ++i) {
-    scaled_fraction += (fraction >> scale_shifts[i]);
-  }
-  *current = '1';
-  current += 1;
-  *current = '.';
-  current += 1;
-  *current = 0;
-
-  // Prepend leading zeros to fill in all 7 bytes of the fraction. Truncate
-  // zeros off the end of the fraction. Every fractional value takes 7 bytes.
-  // For example, 2500 would be written into the buffer as 0002500 since it
-  // represents .00025.
-  constexpr int kMaxFractionalDigits = 7;
-
-  // Abort early if there is not enough space in the buffer.
-  if (current_end - current <= kMaxFractionalDigits) {
-    return current;
-  }
-
-  // Pre-fill buffer with zeros to ensure zero-truncation works properly.
-  for (int i = 1; i < kMaxFractionalDigits; i++) {
-    *(current + i) = '0';
-  }
-
-  // Track how large the fraction is to add leading zeros.
-  char* previous = current;
-  current = StrCatUInt32(current, (current_end - current), scaled_fraction, 10);
-  int fraction_digits = current - previous;
-  int leading_zeros = kMaxFractionalDigits - fraction_digits;
-
-  // Overwrite the null terminator from StrCatUInt32 to ensure zero-trunctaion
-  // works properly.
-  *current = '0';
-
-  // Shift fraction values and prepend zeros if necessary.
-  if (leading_zeros != 0) {
-    for (int i = 0; i < fraction_digits; i++) {
-      current--;
-      *(current + leading_zeros) = *current;
-      *current = '0';
-    }
-    current += kMaxFractionalDigits;
-  }
-
-  // Truncate trailing zeros for cleaner logs. Ensure we leave at least one
-  // fractional character for the case when scaled_fraction is 0.
-  while (*(current - 1) == '0' && (current - 1) > previous) {
-    current--;
-  }
-  *current = 0;
-  current = StrCatStr(current, (current_end - current), "*2^");
-  current = StrCatInt32(current, (current_end - current), exponent);
-  return current;
-}
-
-int FormatInt32(char* output, int32_t i) {
-  return static_cast<int>(FastInt32ToBufferLeft(i, output) - output);
-}
-
-int FormatUInt32(char* output, uint32_t i) {
-  return static_cast<int>(FastUInt32ToBufferLeft(i, output, 10) - output);
-}
-
-int FormatHex(char* output, uint32_t i) {
-  return static_cast<int>(FastUInt32ToBufferLeft(i, output, 16) - output);
-}
-
-int FormatFloat(char* output, float i) {
-  return static_cast<int>(FastFloatToBufferLeft(i, output) - output);
-}
-
-}  // namespace
-
-extern "C" int MicroVsnprintf(char* output, int len, const char* format,
-                              va_list args) {
-  int output_index = 0;
-  const char* current = format;
-  // One extra character must be left for the null terminator.
-  const int usable_length = len - 1;
-  while (*current != '\0' && output_index < usable_length) {
-    if (*current == '%') {
-      current++;
-      switch (*current) {
-        case 'd':
-          // Cut off log message if format could exceed log buffer length.
-          if (usable_length - output_index < kMaxIntCharsNeeded) {
-            output[output_index++] = '\0';
-            return output_index;
-          }
-          output_index +=
-              FormatInt32(&output[output_index], va_arg(args, int32_t));
-          current++;
-          break;
-        case 'u':
-          if (usable_length - output_index < kMaxIntCharsNeeded) {
-            output[output_index++] = '\0';
-            return output_index;
-          }
-          output_index +=
-              FormatUInt32(&output[output_index], va_arg(args, uint32_t));
-          current++;
-          break;
-        case 'x':
-          if (usable_length - output_index < kMaxHexCharsNeeded) {
-            output[output_index++] = '\0';
-            return output_index;
-          }
-          output[output_index++] = '0';
-          output[output_index++] = 'x';
-          output_index +=
-              FormatHex(&output[output_index], va_arg(args, uint32_t));
-          current++;
-          break;
-        case 'f':
-          if (usable_length - output_index < kMaxFloatCharsNeeded) {
-            output[output_index++] = '\0';
-            return output_index;
-          }
-          output_index +=
-              FormatFloat(&output[output_index], va_arg(args, double));
-          current++;
-          break;
-        case '%':
-          output[output_index++] = *current++;
-          break;
-        case 'c':
-          if (usable_length - output_index < 1) {
-            output[output_index++] = '\0';
-            return output_index;
-          }
-          output[output_index++] = va_arg(args, int32_t);
-          current++;
-          break;
-        case 's':
-          char* string = va_arg(args, char*);
-          int string_idx = 0;
-          while (string_idx + output_index < usable_length &&
-                 string[string_idx] != '\0') {
-            output[output_index++] = string[string_idx++];
-          }
-          current++;
-      }
-    } else {
-      output[output_index++] = *current++;
-    }
-  }
-  output[output_index++] = '\0';
-  return output_index;
-}
-
-extern "C" int MicroSnprintf(char* output, int len, const char* format, ...) {
-  va_list args;
-  va_start(args, format);
-  int bytes_written = MicroVsnprintf(output, len, format, args);
-  va_end(args);
-  return bytes_written;
-}
diff --git a/tensorflow/lite/micro/micro_string.h b/tensorflow/lite/micro/micro_string.h
deleted file mode 100644
index 59303e8..0000000
--- a/tensorflow/lite/micro/micro_string.h
+++ /dev/null
@@ -1,33 +0,0 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-==============================================================================*/
-#ifndef TENSORFLOW_LITE_MICRO_MICRO_STRING_H_
-#define TENSORFLOW_LITE_MICRO_MICRO_STRING_H_
-
-#include <cstdarg>
-
-// Implements simple string formatting for numeric types.  Returns the number of
-// bytes written to output.
-extern "C" {
-// Functionally equivalent to vsnprintf, trimmed down for TFLite Micro.
-// MicroSnprintf() is implemented using MicroVsnprintf().
-int MicroVsnprintf(char* output, int len, const char* format, va_list args);
-// Functionally equavalent to snprintf, trimmed down for TFLite Micro.
-// For example, MicroSnprintf(buffer, 10, "int %d", 10) will put the string
-// "int 10" in the buffer.
-// Floating point values are logged in exponent notation (1.XXX*2^N).
-int MicroSnprintf(char* output, int len, const char* format, ...);
-}
-
-#endif  // TENSORFLOW_LITE_MICRO_MICRO_STRING_H_
diff --git a/tensorflow/lite/micro/micro_string_test.cc b/tensorflow/lite/micro/micro_string_test.cc
deleted file mode 100644
index 3c1d8e9..0000000
--- a/tensorflow/lite/micro/micro_string_test.cc
+++ /dev/null
@@ -1,161 +0,0 @@
-/* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-==============================================================================*/
-
-#include "tensorflow/lite/micro/micro_string.h"
-
-#include "tensorflow/lite/micro/testing/micro_test.h"
-
-TF_LITE_MICRO_TESTS_BEGIN
-
-TF_LITE_MICRO_TEST(FormatPositiveIntShouldMatchExpected) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "Int: 55";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Int: %d", 55);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FormatNegativeIntShouldMatchExpected) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "Int: -55";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Int: %d", -55);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FormatUnsignedIntShouldMatchExpected) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "UInt: 12345";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "UInt: %u", 12345);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FormatHexShouldMatchExpected) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "Hex: 0x12345";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Hex: %x", 0x12345);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FormatFloatShouldMatchExpected) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "Float: 1.0*2^4";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Float: %f", 16.);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FormatCharShouldMatchExpected) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "Chars: @,Z";
-  int bytes_written =
-      MicroSnprintf(buffer, kBufferLen, "Chars: %c,%c", 64, 'Z');
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(BadlyFormattedStringShouldProduceReasonableString) {
-  const int kBufferLen = 32;
-  char buffer[kBufferLen];
-  const char golden[] = "Test Badly % formated % string";
-  int bytes_written =
-      MicroSnprintf(buffer, kBufferLen, "Test Badly %% formated %% string%");
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(IntFormatOverrunShouldTruncate) {
-  const int kBufferLen = 8;
-  char buffer[kBufferLen];
-  const char golden[] = "Int: ";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Int: %d", 12345);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(UnsignedIntFormatOverrunShouldTruncate) {
-  const int kBufferLen = 8;
-  char buffer[kBufferLen];
-  const char golden[] = "UInt: ";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "UInt: %u", 12345);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(HexFormatOverrunShouldTruncate) {
-  const int kBufferLen = 8;
-  char buffer[kBufferLen];
-  const char golden[] = "Hex: ";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Hex: %x", 0x12345);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FloatFormatOverrunShouldTruncate) {
-  const int kBufferLen = 12;
-  char buffer[kBufferLen];
-  const char golden[] = "Float: ";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Float: %x", 12345.);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FloatFormatShouldPrintFractionCorrectly) {
-  const int kBufferLen = 24;
-  char buffer[kBufferLen];
-  const char golden[] = "Float: 1.0625*2^0";
-  // Add small offset to float value to account for float rounding error.
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Float: %f", 1.0625001);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(FloatFormatShouldPrintFractionCorrectlyNoLeadingZeros) {
-  const int kBufferLen = 24;
-  char buffer[kBufferLen];
-  const char golden[] = "Float: 1.6332993*2^-1";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "Float: %f", 0.816650);
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(StringFormatOverrunShouldTruncate) {
-  const int kBufferLen = 10;
-  char buffer[kBufferLen];
-  const char golden[] = "String: h";
-  int bytes_written =
-      MicroSnprintf(buffer, kBufferLen, "String: %s", "hello world");
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TEST(StringFormatWithExactOutputSizeOverrunShouldTruncate) {
-  const int kBufferLen = 10;
-  char buffer[kBufferLen];
-  const char golden[] = "format st";
-  int bytes_written = MicroSnprintf(buffer, kBufferLen, "format str");
-  TF_LITE_MICRO_EXPECT_EQ(static_cast<int>(sizeof(golden)), bytes_written);
-  TF_LITE_MICRO_EXPECT_STRING_EQ(golden, buffer);
-}
-
-TF_LITE_MICRO_TESTS_END
diff --git a/tensorflow/lite/micro/tflite_bridge/micro_error_reporter.cc b/tensorflow/lite/micro/tflite_bridge/micro_error_reporter.cc
index 63cc42e..d5d77c3 100644
--- a/tensorflow/lite/micro/tflite_bridge/micro_error_reporter.cc
+++ b/tensorflow/lite/micro/tflite_bridge/micro_error_reporter.cc
@@ -1,4 +1,4 @@
-/* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
+/* Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -36,7 +36,7 @@
 }
 
 int MicroErrorReporter::Report(const char* format, va_list args) {
-  Log(format, args);
+  VMicroPrintf(format, args);
   return 0;
 }
 
diff --git a/tensorflow/lite/micro/tools/make/Makefile b/tensorflow/lite/micro/tools/make/Makefile
index 25ebe9d..e9b69dc 100644
--- a/tensorflow/lite/micro/tools/make/Makefile
+++ b/tensorflow/lite/micro/tools/make/Makefile
@@ -80,6 +80,7 @@
 
 INCLUDES := \
 -I. \
+-I$(DOWNLOADS_DIR) \
 -I$(DOWNLOADS_DIR)/gemmlowp \
 -I$(DOWNLOADS_DIR)/flatbuffers/include \
 -I$(DOWNLOADS_DIR)/kissfft \
@@ -298,7 +299,6 @@
 $(TENSORFLOW_ROOT)tensorflow/lite/micro/micro_interpreter_test.cc \
 $(TENSORFLOW_ROOT)tensorflow/lite/micro/micro_mutable_op_resolver_test.cc \
 $(TENSORFLOW_ROOT)tensorflow/lite/micro/micro_resource_variable_test.cc \
-$(TENSORFLOW_ROOT)tensorflow/lite/micro/micro_string_test.cc \
 $(TENSORFLOW_ROOT)tensorflow/lite/micro/micro_time_test.cc \
 $(TENSORFLOW_ROOT)tensorflow/lite/micro/micro_utils_test.cc \
 $(TENSORFLOW_ROOT)tensorflow/lite/micro/recording_micro_allocator_test.cc \
diff --git a/tensorflow/lite/micro/tools/make/ext_libs/eyalroz_printf.inc b/tensorflow/lite/micro/tools/make/ext_libs/eyalroz_printf.inc
new file mode 100644
index 0000000..63112c1
--- /dev/null
+++ b/tensorflow/lite/micro/tools/make/ext_libs/eyalroz_printf.inc
@@ -0,0 +1,28 @@
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+DOWNLOAD_RESULT := $(shell $(MAKEFILE_DIR)/ext_libs/eyalroz_printf_download.sh  $(DOWNLOADS_DIR) $(TENSORFLOW_ROOT))
+ifneq ($(DOWNLOAD_RESULT), SUCCESS)
+  $(error Something went wrong with the printf download: $(DOWNLOAD_RESULT))
+endif
+
+PRINTF_PATH := $(DOWNLOADS_DIR)/eyalroz_printf
+THIRD_PARTY_CC_SRCS += \
+  $(PRINTF_PATH)/src/printf/printf.c
+THIRD_PARTY_CC_HDRS += \
+  $(PRINTF_PATH)/src/printf/printf.h
+
+INCLUDES += \
+	-I$(PRINTF_PATH)/src
diff --git a/tensorflow/lite/micro/tools/make/ext_libs/eyalroz_printf_download.sh b/tensorflow/lite/micro/tools/make/ext_libs/eyalroz_printf_download.sh
new file mode 100755
index 0000000..79c5ba0
--- /dev/null
+++ b/tensorflow/lite/micro/tools/make/ext_libs/eyalroz_printf_download.sh
@@ -0,0 +1,63 @@
+#!/bin/bash
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+#
+# Called with following arguments:
+# 1 - Path to the downloads folder which is typically
+#     ${TENSORFLOW_ROOT}/tensorflow/lite/micro/tools/make/downloads
+# 2 - (optional) TENSORFLOW_ROOT: path to root of the TFLM tree (relative to directory from where the script is called).
+#
+# This script is called from the Makefile and uses the following convention to
+# enable determination of sucess/failure:
+#
+#   - If the script is successful, the only output on stdout should be SUCCESS.
+#     The makefile checks for this particular string.
+#
+#   - Any string on stdout that is not SUCCESS will be shown in the makefile as
+#     the cause for the script to have failed.
+#
+#   - Any other informational prints should be on stderr.
+
+set -e
+
+TENSORFLOW_ROOT=${2}
+source ${TENSORFLOW_ROOT}tensorflow/lite/micro/tools/make/bash_helpers.sh
+
+DOWNLOADS_DIR=${1}
+if [ ! -d ${DOWNLOADS_DIR} ]; then
+  echo "The top-level downloads directory: ${DOWNLOADS_DIR} does not exist."
+  exit 1
+fi
+
+DOWNLOADED_PRINTF_PATH=${DOWNLOADS_DIR}/eyalroz_printf
+
+if [ -d ${DOWNLOADED_PRINTF_PATH} ]; then
+  echo >&2 "${DOWNLOADED_PRINTF_PATH} already exists, skipping the download."
+else
+
+  ZIP_PREFIX="f8ed5a9bd9fa8384430973465e94aa14c925872d"
+  PRINTF_URL="https://github.com/eyalroz/printf/archive/${ZIP_PREFIX}.zip"
+  PRINTF_MD5="5772534c1d6f718301bca1fefaba28f3"
+
+  # wget is much faster than git clone of the entire repo. So we wget a specific
+  # version and can then apply a patch, as needed.
+  wget ${PRINTF_URL} -O /tmp/${ZIP_PREFIX}.zip >&2
+  check_md5 /tmp/${ZIP_PREFIX}.zip ${PRINTF_MD5}
+
+  unzip -qo /tmp/${ZIP_PREFIX}.zip -d /tmp >&2
+  mv /tmp/printf-${ZIP_PREFIX} ${DOWNLOADED_PRINTF_PATH}
+fi
+
+echo "SUCCESS"
\ No newline at end of file
diff --git a/tensorflow/lite/micro/tools/make/targets/arc_emsdp_makefile.inc b/tensorflow/lite/micro/tools/make/targets/arc_emsdp_makefile.inc
index f7c3066..7fce24b 100644
--- a/tensorflow/lite/micro/tools/make/targets/arc_emsdp_makefile.inc
+++ b/tensorflow/lite/micro/tools/make/targets/arc_emsdp_makefile.inc
@@ -1,4 +1,4 @@
-# Copyright 2021 The TensorFlow Authors. All Rights Reserved.
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -26,4 +26,6 @@
 
 TCF_FILE = $(PWD)/$(MAKEFILE_DIR)/downloads/$(MLI_LIB_DIR)/hw/emsdp_em11d_em9d_dfss.tcf
 
-include $(MAKEFILE_DIR)/targets/arc/arc_common.inc
\ No newline at end of file
+include $(MAKEFILE_DIR)/targets/arc/arc_common.inc
+
+include $(MAKEFILE_DIR)/ext_libs/eyalroz_printf.inc
diff --git a/tensorflow/lite/micro/tools/make/targets/bluepill_makefile.inc b/tensorflow/lite/micro/tools/make/targets/bluepill_makefile.inc
index c14bda4..532689c 100644
--- a/tensorflow/lite/micro/tools/make/targets/bluepill_makefile.inc
+++ b/tensorflow/lite/micro/tools/make/targets/bluepill_makefile.inc
@@ -1,3 +1,18 @@
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
 export PATH := $(DOWNLOADS_DIR)/gcc_embedded/bin/:$(PATH)
 TARGET_ARCH := cortex-m3
 TARGET_TOOLCHAIN_PREFIX := arm-none-eabi-
@@ -87,3 +102,5 @@
 
 test: build
 	$(TEST_SCRIPT) "$(TEST_TARGET_BINARIES)" $(TEST_PASS_STRING) $(TARGET)
+
+include $(MAKEFILE_DIR)/ext_libs/eyalroz_printf.inc
diff --git a/tensorflow/lite/micro/tools/make/targets/chre_makefile.inc b/tensorflow/lite/micro/tools/make/targets/chre_makefile.inc
index 3665b26..d2e5892 100644
--- a/tensorflow/lite/micro/tools/make/targets/chre_makefile.inc
+++ b/tensorflow/lite/micro/tools/make/targets/chre_makefile.inc
@@ -1,4 +1,4 @@
-# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -32,3 +32,5 @@
 MICROLITE_TEST_SRCS := $(filter-out $(EXCLUDED_TESTS),$(MICROLITE_TEST_SRCS))
 THIRD_PARTY_CC_HDRS := $(filter-out $(EXCLUDED_HDRS),$(THIRD_PARTY_CC_HDRS))
 MICROLITE_CC_HDRS := $(filter-out $(EXCLUDED_KERNEL_HDRS),$(MICROLITE_CC_HDRS))
+
+include $(MAKEFILE_DIR)/ext_libs/eyalroz_printf.inc
diff --git a/tensorflow/lite/micro/tools/make/targets/cortex_m_generic_makefile.inc b/tensorflow/lite/micro/tools/make/targets/cortex_m_generic_makefile.inc
index 0ed14fc..0e034ad 100644
--- a/tensorflow/lite/micro/tools/make/targets/cortex_m_generic_makefile.inc
+++ b/tensorflow/lite/micro/tools/make/targets/cortex_m_generic_makefile.inc
@@ -1,4 +1,4 @@
-# Copyright 2022 The TensorFlow Authors. All Rights Reserved.
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -204,3 +204,5 @@
     $(TENSORFLOW_ROOT)tensorflow/lite/micro/examples/micro_speech/Makefile.inc
   MICRO_LITE_EXAMPLE_TESTS := $(filter-out $(EXCLUDED_EXAMPLE_TESTS), $(MICRO_LITE_EXAMPLE_TESTS))
 endif
+
+include $(MAKEFILE_DIR)/ext_libs/eyalroz_printf.inc
diff --git a/tensorflow/lite/micro/tools/project_generation/Makefile b/tensorflow/lite/micro/tools/project_generation/Makefile
index 092ba7e..7497f07 100644
--- a/tensorflow/lite/micro/tools/project_generation/Makefile
+++ b/tensorflow/lite/micro/tools/project_generation/Makefile
@@ -1,4 +1,4 @@
-# Copyright 2022 The TensorFlow Authors. All Rights Reserved.
+# Copyright 2023 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -92,6 +92,13 @@
 TFLM_CC_SRCS := $(shell find $(TENSORFLOW_ROOT)tensorflow -name "*.cc" -o -name "*.c")
 OBJS := $(addprefix $(OBJDIR)/, $(patsubst %.c,%.o,$(patsubst %.cc,%.o,$(TFLM_CC_SRCS))))
 
+# if the third party printf library is present, add the include paths
+TFLM_PRINTF_PATH := $(shell find third_party -name eyalroz_printf)
+ifneq ($(TFLM_PRINTF_PATH),)
+  INCLUDES += \
+    -I./third_party
+endif
+
 $(OBJDIR)/%.o: %.cc
 	@mkdir -p $(dir $@)
 	$(CXX) $(CXXFLAGS) $(INCLUDES) -c $< -o $@