[libc-commits] [libc] [llvm] [libc] Add fixed point support to printf (PR #82707)
via libc-commits
libc-commits at lists.llvm.org
Thu Feb 22 15:44:50 PST 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-libc
Author: Michael Jones (michaelrj-google)
<details>
<summary>Changes</summary>
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.
---
Patch is 34.97 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/82707.diff
12 Files Affected:
- (modified) libc/docs/dev/printf_behavior.rst (+7)
- (modified) libc/src/stdio/printf_core/CMakeLists.txt (+1)
- (modified) libc/src/stdio/printf_core/converter.cpp (+8)
- (modified) libc/src/stdio/printf_core/converter_atlas.h (+5)
- (modified) libc/src/stdio/printf_core/converter_utils.h (+3)
- (modified) libc/src/stdio/printf_core/core_structs.h (+19-4)
- (added) libc/src/stdio/printf_core/fixed_converter.h (+386)
- (modified) libc/src/stdio/printf_core/float_dec_converter.h (-3)
- (modified) libc/src/stdio/printf_core/parser.h (+65)
- (modified) libc/src/stdio/printf_core/printf_config.h (+7)
- (modified) libc/test/src/stdio/sprintf_test.cpp (+160)
- (modified) utils/bazel/llvm-project-overlay/libc/BUILD.bazel (+9-7)
``````````diff
diff --git a/libc/docs/dev/printf_behavior.rst b/libc/docs/dev/printf_behavior.rst
index bc60aa43ee2b6b..7786b27a9fdcd5 100644
--- a/libc/docs/dev/printf_behavior.rst
+++ b/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.
diff --git a/libc/src/stdio/printf_core/CMakeLists.txt b/libc/src/stdio/printf_core/CMakeLists.txt
index 8da274395526df..cd75060db37579 100644
--- a/libc/src/stdio/printf_core/CMakeLists.txt
+++ b/libc/src/stdio/printf_core/CMakeLists.txt
@@ -76,6 +76,7 @@ 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
diff --git a/libc/src/stdio/printf_core/converter.cpp b/libc/src/stdio/printf_core/converter.cpp
index 52412aef3c5c15..613d693c3cfcb3 100644
--- a/libc/src/stdio/printf_core/converter.cpp
+++ b/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);
diff --git a/libc/src/stdio/printf_core/converter_atlas.h b/libc/src/stdio/printf_core/converter_atlas.h
index 6471f3f2955b75..2189ed11a551ea 100644
--- a/libc/src/stdio/printf_core/converter_atlas.h
+++ b/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
diff --git a/libc/src/stdio/printf_core/converter_utils.h b/libc/src/stdio/printf_core/converter_utils.h
index 54f0a870d0ac4a..948fe816e9b76d 100644
--- a/libc/src/stdio/printf_core/converter_utils.h
+++ b/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
diff --git a/libc/src/stdio/printf_core/core_structs.h b/libc/src/stdio/printf_core/core_structs.h
index 7634d45568ab84..681f85dd5a285d 100644
--- a/libc/src/stdio/printf_core/core_structs.h
+++ b/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};
}
}
diff --git a/libc/src/stdio/printf_core/fixed_converter.h b/libc/src/stdio/printf_core/fixed_converter.h
new file mode 100644
index 00000000000000..5d521f2957a560
--- /dev/null
+++ b/libc/src/stdio/printf_core/fixed_converter.h
@@ -0,0 +1,386 @@
+//===-- Fixed Point Converter for printf ------------------------*- C++ -*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIBC_SRC_STDIO_PRINTF_CORE_FIXED_CONVERTER_H
+#define LLVM_LIBC_SRC_STDIO_PRINTF_CORE_FIXED_CONVERTER_H
+
+#include "include/llvm-libc-macros/stdfix-macros.h"
+#include "src/__support/CPP/string_view.h"
+#include "src/__support/fixed_point/fx_bits.h"
+#include "src/__support/fixed_point/fx_rep.h"
+#include "src/__support/integer_to_string.h"
+#include "src/__support/libc_assert.h"
+#include "src/stdio/printf_core/converter_utils.h"
+#include "src/stdio/printf_core/core_structs.h"
+#include "src/stdio/printf_core/writer.h"
+
+#include <inttypes.h>
+#include <stddef.h>
+
+namespace LIBC_NAMESPACE {
+namespace printf_core {
+
+// This is just for assertions. It will be compiled out for release builds.
+LIBC_INLINE constexpr uint32_t const_ten_exp(uint32_t exponent) {
+ uint32_t result = 1;
+ LIBC_ASSERT(exponent < 11);
+ for (size_t i = 0; i < exponent; ++i)
+ result *= 10;
+
+ return result;
+}
+
+LIBC_INLINE int convert_fixed(Writer *writer, const FormatSection &to_conv) {
+ using SAStorageType = fixed_point::FXRep<short accum>::StorageType;
+ using AStorageType = fixed_point::FXRep<accum>::StorageType;
+ using LAStorageType = fixed_point::FXRep<long accum>::StorageType;
+ using SFStorageType = fixed_point::FXRep<short fract>::StorageType;
+ using FStorageType = fixed_point::FXRep<fract>::StorageType;
+ using LFStorageType = fixed_point::FXRep<long fract>::StorageType;
+
+ // Long accum should be the largest type, so we can store all the smaller
+ // numbers in things sized for it.
+ using LARep = fixed_point::FXRep<unsigned long accum>;
+ using StorageType = LARep::StorageType;
+
+ // All of the letters will be defined relative to variable a, which will be
+ // the appropriate case based on the name of the conversion. This converts any
+ // conversion name into the letter 'a' with the appropriate case.
+ const char a = (to_conv.conv_name & 32) | 'A';
+ FormatFlags flags = to_conv.flags;
+
+ bool is_negative;
+ int exponent;
+ StorageType integral;
+ StorageType fractional;
+
+ // TODO: See about simplifying this mess of a 3D matrix if statement.
+ // r = fract
+ // k = accum
+ // lowercase = signed
+ // uppercase = unsigned
+ // h = short
+ // l = long
+ // any other length modifier has no effect
+ if (to_conv.length_modifier == LengthModifier::h) {
+ // short types
+ if (to_conv.conv_name == 'r') {
+ auto fixed_bits = fixed_point::FXBits<short fract>(
+ static_cast<SFStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'R') {
+ auto fixed_bits = fixed_point::FXBits<unsigned short fract>(
+ static_cast<SFStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'k') {
+ auto fixed_bits = fixed_point::FXBits<short accum>(
+ static_cast<SAStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'K') {
+ auto fixed_bits = fixed_point::FXBits<unsigned short accum>(
+ static_cast<SAStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ } else if (to_conv.length_modifier == LengthModifier::l) {
+ // long types
+ if (to_conv.conv_name == 'r') {
+ auto fixed_bits = fixed_point::FXBits<long fract>(
+ static_cast<LFStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'R') {
+ auto fixed_bits = fixed_point::FXBits<unsigned long fract>(
+ static_cast<LFStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'k') {
+ auto fixed_bits = fixed_point::FXBits<long accum>(
+ static_cast<LAStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'K') {
+ auto fixed_bits = fixed_point::FXBits<unsigned long accum>(
+ static_cast<LAStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ } else {
+ // unspecified types
+ if (to_conv.conv_name == 'r') {
+ auto fixed_bits = fixed_point::FXBits<fract>(
+ static_cast<FStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'R') {
+ auto fixed_bits = fixed_point::FXBits<unsigned fract>(
+ static_cast<FStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'k') {
+ auto fixed_bits = fixed_point::FXBits<accum>(
+ static_cast<AStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ if (to_conv.conv_name == 'K') {
+ auto fixed_bits = fixed_point::FXBits<unsigned accum>(
+ static_cast<AStorageType>(to_conv.conv_val_raw));
+ integral = fixed_bits.get_integral();
+ fractional = fixed_bits.get_fraction();
+ exponent = fixed_bits.get_exponent();
+ is_negative = fixed_bits.get_sign();
+ }
+ }
+
+ LIBC_ASSERT(static_cast<size_t>(exponent) <=
+ (sizeof(StorageType) - sizeof(uint32_t)) * CHAR_BIT &&
+ "StorageType must be large enough to hold the fractional "
+ "component multiplied by a 32 bit number.");
+
+ char sign_char = 0;
+
+ // Check if the conv name is uppercase
+ if (a == 'A') {
+ // These flags are only for signed conversions, so this removes them if the
+ // conversion is unsigned.
+ flags = FormatFlags(flags &
+ ~(FormatFlags::FORCE_SIGN | FormatFlags::SPACE_PREFIX));
+ }
+
+ if (is_negative)
+ sign_char = '-';
+ else if ((flags & FormatFlags::FORCE_SIGN) == FormatFlags::FORCE_SIGN)
+ sign_char = '+'; // FORCE_SIGN has precedence over SPACE_PREFIX
+ else if ((flags & FormatFlags::SPACE_PREFIX) == FormatFlags::SPACE_PREFIX)
+ sign_char = ' ';
+
+ // If to_conv doesn't specify a precision, the precision defaults to 6.
+ const size_t precision = to_conv.precision < 0 ? 6 : to_conv.precision;
+ bool has_decimal_point =
+ (precision > 0) || ((flags & FormatFlags::ALTERNATE_FORM) != 0);
+
+ // The number of non-zero digits below the decimal point for a negative power
+ // of 2 in base 10 is equal to the magnitude of the power of 2.
+
+ // A quick proof:
+ // Let p be any positive integer.
+ // Let e = 2^(-p)
+ // Let t be a positive integer such that e * 10^t is an integer.
+ // By definition: The smallest allowed value of t must be equal to the number
+ // of non-zero digits below the decimal point in e.
+ // If we evaluate e * 10^t we get the following:
+ // e * 10^t = 2^(-p) * 10*t = 2^(-p) * 2^t * 5^t = 5^t * 2^(t-p)
+ // For 5^t * 2^(t-p) to be an integer, both exponents must be non-negative,
+ // since 5 and 2 are coprime.
+ // The smallest value of t such that t-p is non-negative is p.
+ // Therefor, the number of non-zero digits below the decimal point for a given
+ // negative power of 2 "p" is equal to the value of p.
+
+ constexpr size_t MAX_FRACTION_DIGITS = LARep::FRACTION_LEN;
+
+ char fraction_digits[MAX_FRACTION_DIGITS];
+
+ size_t valid_fraction_digits = 0;
+
+ // TODO: Factor this part out
+ while (fractional > 0) {
+ uint32_t cur_digits = 0;
+ // 10^9 is used since it's the largest power of 10 that fits in a uint32_t
+ constexpr uint32_t TEN_EXP_NINE = 1000000000;
+ constexpr size_t DIGITS_PER_BLOCK = 9;
+
+ // Multiply by 10^9, then grab the digits above the decimal point, then
+ // clear those digits in fractional.
+ fractional = fractional * TEN_EXP_NINE;
+ cur_digits = static_cast<uint32_t>(fractional >> exponent);
+ fractional = fractional % (StorageType(1) << exponent);
+
+ // we add TEN_EXP_NINE to force leading zeroes to show up, then we skip the
+ // first digit in the loop.
+ const IntegerToString<uint32_t> cur_fractional_digits(cur_digits +
+ TEN_EXP_NINE);
+ for (size_t i = 0;
+ i < DIGITS_PER_BLOCK && valid_fraction_digits < MAX_FRACTION_DIGITS;
+ ++i, ++valid_fraction_digits)
+ fraction_digits[valid_fraction_digits] =
+ cur_fractional_digits.view()[i + 1];
+
+ if (valid_fraction_digits >= MAX_FRACTION_DIGITS) {
+ LIBC_ASSERT(fractional == 0 && "If the fraction digit buffer is full, "
+ "there should be no remaining digits.");
+ /*
+ A visual explanation of what this assert is checking:
+
+ 32 digits (max for 32 bit fract)
+ +------------------------------++--+--- must be zero
+ | || |
+ 123456789012345678901234567890120000
+ | || || || |
+ +-------++-------++-------++-------+
+ 9 digit blocks
+ */
+ LIBC_ASSERT(cur_digits % const_ten_exp(
+ DIGITS_PER_BLOCK -
+ (MAX_FRACTION_DIGITS % DIGITS_PER_BLOCK)) ==
+ 0 &&
+ "Digits after the MAX_FRACTION_DIGITS should all be zero.");
+ valid_fraction_digits = MAX_FRACTION_DIGITS;
+ }
+ }
+
+ if (precision < valid_fraction_digits) {
+ // Handle rounding. Just do round to nearest, tie to even since it's
+ // unspecified.
+ RoundDirection round;
+ char first_digit_after = fraction_digits[precision];
+ if (first_digit_after > '5') {
+ round = RoundDirection::Up;
+ } else if (first_digit_after < '5') {
+ round = RoundDirection::Down;
+ } else {
+ // first_digit_after == '5'
+ // need to check the remaining digits, but default to even.
+ round = RoundDirection::Even;
+ for (size_t cur_digit_index = precision + 1;
+ cur_digit_index + 1 < valid_fraction_digits; ++cur_digit_index) {
+ if (fraction_digits[cur_digit_index] != '0') {
+ round = RoundDirection::Up;
+ break;
+ }
+ }
+ }
+
+ // If we need to actually perform rounding, do so.
+ if (round == RoundDirection::Up || round == RoundDirection::Even) {
+ bool keep_rounding = true;
+ int digit_to_round = static_cast<int>(precision) - 1;
+ for (; digit_to_round >= 0 && keep_rounding; --digit_to_round) {
+ keep_rounding = false;
+ char cur_digit = fraction_digits[digit_to_round];
+ // if the digit should not be rounded up
+ if (round == RoundDirection::Even && ((cur_digit - '0') % 2) == 0) {
+ // break out of the loop
+ break;
+ }
+ fraction_digits[digit_to_round] += 1;
+
+ // if the digit was a 9, instead replace with a 0.
+ if (cur_digit == '9') {
+ fraction_digits[digit_to_round] = '0';
+ keep_rounding = true;
+ }
+ }
+
+ // if every digit below the decimal point was rounded up but we need to
+ // keep rounding
+ if (keep_rounding &&
+ (round == RoundDirection::Up ||
+ (round == RoundDirection::Even && ((integral % 2) == 1)))) {
+ // add one to the integral portion to round it up.
+ ++integral;
+ }
+ }
+
+ valid_fraction_digits = precision;
+ }
+
+ const IntegerToString<StorageType> integral_str(integral);
+
+ // these are signed to prevent underflow due to negative values. The
+ // eventual values will always be non-negative.
+ size_t trailing_zeroes = 0;
+ int padding;
+
+ // If the precision is greater than the actual result, pad with 0s
+ if (precision > valid_fraction_digits)
+ trailing_zeroes = precision - (valid_fraction_digits);
+
+ constexpr cpp::string_view DECIMAL_POINT(".");
+
+ padding = static_cast<int>(to_conv.min_width - (sign_char > 0 ? 1 : 0) -
+ integral_str.size() -
+ static_cast<int>(has_decimal_point) -
+ valid_fraction_digits - trailing_zeroes);
+ if (padding < 0)
+ padding = 0;
+
+ if ((flags & FormatFlags::LEFT_JUSTIFIED) == FormatFlags::LEFT_JUSTIFIED) {
+ // The pattern is (sign), integral, (.), (fraction), (zeroes), (spaces)
+ if (sign_char > 0)
+ RET_IF_RESULT_NEGATIVE(writer->write(sign_char));
+ RET_IF_RESULT_NEGATIVE(writer->write(integral_str.view()));
+ if (has_decimal_point)
+ RET_IF_RESULT_NEGATIVE(writer->write(DECIMAL_POINT));
+ if (valid_fraction_digits > 0)
+ RET_IF_RESULT_NEGATIVE(
+ writer->write({fraction_digits, valid_fraction_digits}));
+ if (trailing_zeroes > 0)
+ RET_IF_RESULT_NEGATIVE(writer->write('0', trailing_zeroes));
+ if (padding > 0)
+ RET_IF_RESULT_NEGATIVE(...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/82707
More information about the libc-commits
mailing list