[libc] Add fixed point support to printf (llvm#82707)

This patch adds the r, R, k, and K conversion specifiers to printf, with
accompanying tests. They are guarded behind the
LIBC_COPT_PRINTF_DISABLE_FIXED_POINT flag as well as automatic fixed
point support detection.

Author: michaelrj-google

libc/config/config.json

@@ -15,6 +15,10 @@
     "LIBC_CONF_PRINTF_FLOAT_TO_STR_USE_MEGA_LONG_DOUBLE_TABLE": {
       "value": false,
       "doc": "Use large table for better printf long double performance."
+    },
+    "LIBC_CONF_PRINTF_DISABLE_FIXED_POINT": {
+      "value": false,
+      "doc": "Disable printing fixed point values in printf and friends."
     }
   },
   "string": {

libc/docs/configure.rst

@@ -26,6 +26,7 @@ overrides in ``config/<platform>/config.json`` and ``config/<platform>/<arch>/co
 to learn about the defaults for your platform and target.
 
 * **"printf" options**
+    - ``LIBC_CONF_PRINTF_DISABLE_FIXED_POINT``: Disable printing fixed point values in printf and friends.
     - ``LIBC_CONF_PRINTF_DISABLE_FLOAT``: Disable printing floating point values in printf and friends.
     - ``LIBC_CONF_PRINTF_DISABLE_INDEX_MODE``: Disable index mode in the printf format string.
     - ``LIBC_CONF_PRINTF_DISABLE_WRITE_INT``: Disable handling of %n in printf format string.

libc/docs/dev/printf_behavior.rst

@@ -62,6 +62,13 @@ When set, this flag disables support for floating point numbers and all their
 conversions (%a, %f, %e, %g); any floating point number conversion will be
 treated as invalid. This reduces code size.
 
+LIBC_COPT_PRINTF_DISABLE_FIXED_POINT
+------------------------------------
+When set, this flag disables support for fixed point numbers and all their
+conversions (%r, %k); any fixed point number conversion will be treated as
+invalid. This reduces code size. This has no effect if the current compiler does
+not support fixed point numbers.
+
 LIBC_COPT_PRINTF_NO_NULLPTR_CHECKS
 ----------------------------------
 When set, this flag disables the nullptr checks in %n and %s.
@@ -191,3 +198,8 @@ original conversion.
 The %p conversion will display a null pointer as if it was the string
 "(nullptr)" passed to a "%s" conversion, with all other options remaining the
 same as the original conversion.
+
+The %r, %R, %k, and %K fixed point number format specifiers are accepted as
+defined in ISO/IEC TR 18037 (the fixed point number extension). These are
+available when the compiler is detected as having support for fixed point
+numbers and the LIBC_COPT_PRINTF_DISABLE_FIXED_POINT flag is not set.

libc/fuzzing/stdio/CMakeLists.txt

@@ -15,3 +15,17 @@ add_libc_fuzzer(
     libc.src.stdio.snprintf
     libc.src.__support.FPUtil.fp_bits
 )
+
+if(LIBC_COMPILER_HAS_FIXED_POINT)
+  add_libc_fuzzer(
+    printf_fixed_conv_fuzz
+    NEED_MPFR
+    SRCS
+      printf_fixed_conv_fuzz.cpp
+    DEPENDS
+      libc.src.stdio.snprintf
+      libc.src.__support.fixed_point.fx_bits
+    COMPILE_OPTIONS
+      -ffixed-point # TODO: add -ffixed-point to fuzz tests automatically
+  )
+endif()

libc/fuzzing/stdio/printf_fixed_conv_fuzz.cpp

@@ -0,0 +1,133 @@
+//===-- printf_fixed_conv_fuzz.cpp ----------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+///
+/// Fuzzing test for llvm-libc printf %f/e/g/a implementations.
+///
+//===----------------------------------------------------------------------===//
+#include "src/stdio/snprintf.h"
+
+#include "include/llvm-libc-macros/stdfix-macros.h"
+#include "src/__support/fixed_point/fx_bits.h"
+#include "src/__support/fixed_point/fx_rep.h"
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include "utils/MPFRWrapper/mpfr_inc.h"
+
+constexpr int MAX_SIZE = 10000;
+
+inline bool simple_streq(char *first, char *second, int length) {
+  for (int i = 0; i < length; ++i)
+    if (first[i] != second[i])
+      return false;
+
+  return true;
+}
+
+inline int clamp(int num, int max) {
+  if (num > max)
+    return max;
+  if (num < -max)
+    return -max;
+  return num;
+}
+
+enum class TestResult {
+  Success,
+  BufferSizeFailed,
+  LengthsDiffer,
+  StringsNotEqual,
+};
+
+template <typename F>
+inline TestResult test_vals(const char *fmt, uint64_t num, int prec,
+                            int width) {
+  typename LIBC_NAMESPACE::fixed_point::FXRep<F>::StorageType raw_num = num;
+
+  auto raw_num_bits = LIBC_NAMESPACE::fixed_point::FXBits<F>(raw_num);
+
+  // This needs to be a float with enough bits of precision to hold the fixed
+  // point number.
+  static_assert(sizeof(long double) > sizeof(long accum));
+
+  // build a long double that is equivalent to the fixed point number.
+  long double ld_num =
+      static_cast<long double>(raw_num_bits.get_integral()) +
+      (static_cast<long double>(raw_num_bits.get_fraction()) /
+       static_cast<long double>(1ll << raw_num_bits.get_exponent()));
+
+  if (raw_num_bits.get_sign())
+    ld_num = -ld_num;
+
+  // Call snprintf on a nullptr to get the buffer size.
+  int buffer_size = LIBC_NAMESPACE::snprintf(nullptr, 0, fmt, width, prec, num);
+
+  if (buffer_size < 0)
+    return TestResult::BufferSizeFailed;
+
+  char *test_buff = new char[buffer_size + 1];
+  char *reference_buff = new char[buffer_size + 1];
+
+  int test_result = 0;
+  int reference_result = 0;
+
+  test_result = LIBC_NAMESPACE::snprintf(test_buff, buffer_size + 1, fmt, width,
+                                         prec, num);
+
+  // The fixed point format is defined to be %f equivalent.
+  reference_result = mpfr_snprintf(reference_buff, buffer_size + 1, "%*.*Lf",
+                                   width, prec, ld_num);
+
+  // All of these calls should return that they wrote the same amount.
+  if (test_result != reference_result || test_result != buffer_size)
+    return TestResult::LengthsDiffer;
+
+  if (!simple_streq(test_buff, reference_buff, buffer_size))
+    return TestResult::StringsNotEqual;
+
+  delete[] test_buff;
+  delete[] reference_buff;
+  return TestResult::Success;
+}
+
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+  // const uint8_t raw_data[] = {0x8d,0x43,0x40,0x0,0x0,0x0,};
+  // data = raw_data;
+  // size = sizeof(raw_data);
+  int prec = 0;
+  int width = 0;
+
+  LIBC_NAMESPACE::fixed_point::FXRep<long accum>::StorageType raw_num = 0;
+
+  // Copy as many bytes of data as will fit into num, prec, and with. Any extras
+  // are ignored.
+  for (size_t cur = 0; cur < size; ++cur) {
+    if (cur < sizeof(raw_num)) {
+      raw_num = (raw_num << 8) + data[cur];
+    } else if (cur < sizeof(raw_num) + sizeof(prec)) {
+      prec = (prec << 8) + data[cur];
+    } else if (cur < sizeof(raw_num) + sizeof(prec) + sizeof(width)) {
+      width = (width << 8) + data[cur];
+    }
+  }
+
+  width = clamp(width, MAX_SIZE);
+  prec = clamp(prec, MAX_SIZE);
+
+  TestResult result;
+  result = test_vals<long accum>("%*.*lk", raw_num, prec, width);
+  if (result != TestResult::Success)
+    __builtin_trap();
+
+  result = test_vals<unsigned long accum>("%*.*lK", raw_num, prec, width);
+  if (result != TestResult::Success)
+    __builtin_trap();
+
+  return 0;
+}

libc/src/stdio/printf_core/CMakeLists.txt

@@ -10,6 +10,9 @@ endif()
 if(LIBC_CONF_PRINTF_FLOAT_TO_STR_USE_MEGA_LONG_DOUBLE_TABLE)
   list(APPEND printf_config_copts "-DLIBC_COPT_FLOAT_TO_STR_USE_MEGA_LONG_DOUBLE_TABLE")
 endif()
+if(LIBC_CONF_PRINTF_DISABLE_FIXED_POINT)
+  list(APPEND printf_config_copts "-DLIBC_COPT_PRINTF_DISABLE_FIXED_POINT")
+endif()
 if(printf_config_copts)
   list(PREPEND printf_config_copts "COMPILE_OPTIONS")
 endif()
@@ -76,6 +79,7 @@ add_object_library(
     float_inf_nan_converter.h
     float_hex_converter.h
     float_dec_converter.h
+    fixed_converter.h #TODO: Check if this should be disabled when fixed unavail
   DEPENDS
     .writer
     .core_structs

libc/src/stdio/printf_core/converter.cpp

@@ -9,6 +9,7 @@
 #include "src/stdio/printf_core/converter.h"
 
 #include "src/stdio/printf_core/core_structs.h"
+#include "src/stdio/printf_core/printf_config.h"
 #include "src/stdio/printf_core/writer.h"
 
 // This option allows for replacing all of the conversion functions with custom
@@ -75,6 +76,13 @@ int convert(Writer *writer, const FormatSection &to_conv) {
   case 'G':
     return convert_float_dec_auto(writer, to_conv);
 #endif // LIBC_COPT_PRINTF_DISABLE_FLOAT
+#ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT
+  case 'r':
+  case 'R':
+  case 'k':
+  case 'K':
+    return convert_fixed(writer, to_conv);
+#endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT
 #ifndef LIBC_COPT_PRINTF_DISABLE_WRITE_INT
   case 'n':
     return convert_write_int(writer, to_conv);

libc/src/stdio/printf_core/converter_atlas.h

@@ -31,6 +31,11 @@
 #include "src/stdio/printf_core/float_hex_converter.h"
 #endif // LIBC_COPT_PRINTF_DISABLE_FLOAT
 
+#ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT
+// defines convert_fixed
+#include "src/stdio/printf_core/fixed_converter.h"
+#endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT
+
 #ifndef LIBC_COPT_PRINTF_DISABLE_WRITE_INT
 #include "src/stdio/printf_core/write_int_converter.h"
 #endif // LIBC_COPT_PRINTF_DISABLE_WRITE_INT

libc/src/stdio/printf_core/converter_utils.h

@@ -51,6 +51,9 @@ LIBC_INLINE uintmax_t apply_length_modifier(uintmax_t num, LengthModifier lm) {
       return result;                                                           \
   }
 
+// This is used to represent which direction the number should be rounded.
+enum class RoundDirection { Up, Down, Even };
+
 } // namespace printf_core
 } // namespace LIBC_NAMESPACE
 

libc/src/stdio/printf_core/core_structs.h

@@ -10,7 +10,9 @@
 #define LLVM_LIBC_SRC_STDIO_PRINTF_CORE_CORE_STRUCTS_H
 
 #include "src/__support/CPP/string_view.h"
+#include "src/__support/CPP/type_traits.h"
 #include "src/__support/FPUtil/FPBits.h"
+#include "src/stdio/printf_core/printf_config.h"
 
 #include <inttypes.h>
 #include <stddef.h>
@@ -77,7 +79,13 @@ struct FormatSection {
   }
 };
 
-enum PrimaryType : uint8_t { Unknown = 0, Float = 1, Pointer = 2, Integer = 3 };
+enum PrimaryType : uint8_t {
+  Unknown = 0,
+  Float = 1,
+  Pointer = 2,
+  Integer = 3,
+  FixedPoint = 4,
+};
 
 // TypeDesc stores the information about a type that is relevant to printf in
 // a relatively compact manner.
@@ -95,9 +103,16 @@ template <typename T> LIBC_INLINE constexpr TypeDesc type_desc_from_type() {
   } else {
     constexpr bool IS_POINTER = cpp::is_pointer_v<T>;
     constexpr bool IS_FLOAT = cpp::is_floating_point_v<T>;
-    return TypeDesc{sizeof(T), IS_POINTER ? PrimaryType::Pointer
-                               : IS_FLOAT ? PrimaryType::Float
-                                          : PrimaryType::Integer};
+#ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT
+    constexpr bool IS_FIXED_POINT = cpp::is_fixed_point_v<T>;
+#else
+    constexpr bool IS_FIXED_POINT = false;
+#endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT
+
+    return TypeDesc{sizeof(T), IS_POINTER       ? PrimaryType::Pointer
+                               : IS_FLOAT       ? PrimaryType::Float
+                               : IS_FIXED_POINT ? PrimaryType::FixedPoint
+                                                : PrimaryType::Integer};
   }
 }
 
@@ -109,6 +124,7 @@ constexpr int FILE_WRITE_ERROR = -1;
 constexpr int FILE_STATUS_ERROR = -2;
 constexpr int NULLPTR_WRITE_ERROR = -3;
 constexpr int INT_CONVERSION_ERROR = -4;
+constexpr int FIXED_POINT_CONVERSION_ERROR = -5;
 
 } // namespace printf_core
 } // namespace LIBC_NAMESPACE