Skip to content

Commit

Permalink
ENH: Update Double-convert third-party
Browse files Browse the repository at this point in the history 
Fix typos in comments
CMakeLists.txt: Export all symbols (InsightSoftwareConsortium#136)
test/cctest/CMakeLists.txt: Added /bigobj for MSVC tests (InsightSoftwareConsortium#135)
Add DOUBLE_CONVERSION_HAS_ATTRIBUTE to fix warnings on MSVC and enable for GCC. (InsightSoftwareConsortium#131)
Fix broken MSVC builds. (InsightSoftwareConsortium#130)
Add support for quiet and signaling NaNs to the ieee header. (InsightSoftwareConsortium#128)
Move buffer and buffer_pos down (InsightSoftwareConsortium#125)
    * Move buffer and buffer_pos down
    Simplifies code by removing two asserts
    Optimize code with -ftrivial-auto-var-init=pattern by avoiding initialization when buffer is not used
    * Disable -ftrivial-auto-var-init=pattern for a large buffer

Fix strtod.cc undefined constants (InsightSoftwareConsortium#123)
    When DOUBLE_CONVERSION_CORRECT_DOUBLE_OPERATIONS is not defined
    there is a build error when -Wall -Werror enabled because
    kMaxExactDoubleIntegerDecimalDigits, exact_powers_of_ten and
    kExactPowersOfTenSize are used only in else branch of this define
    (when it's defined).
    Fixes google/double-conversion#122

Add full license to test .cc files missing it. (InsightSoftwareConsortium#121)
    Some files in cctest had a 1-line copyright that doesn't list the
    license. This copies the project license to those files, preserving the
    date.

Add wasm32 as supported platform (InsightSoftwareConsortium#120)
    Summary:
    Emscripten is already included, adding wasm32 the same way for when
    build with "plain" clang wasm32 (without emscripten)

Pseiderer/add nios2 and xtensa 001 (InsightSoftwareConsortium#119)
    * double-conversion: enable for nios2
    Nios2 supports double conversion, tested using qemu:

Add support for microblaze.
Add support for e2k architecture. (InsightSoftwareConsortium#118)
Add min exponent width option in double-to-string conversion (InsightSoftwareConsortium#116)

Remove reference to `diy-fp.cc`
    That file was removed in a previous commit.

More Bignum fiddling. (InsightSoftwareConsortium#108)
    * Char has size 1.
    * More const.
    * Allow inlining.
    * Compact Bignum sizes.
    * Consistent naming.

Remove redundant parenthesis.
    Reported by seanm (github).

Optimise Bignum layout. (InsightSoftwareConsortium#107)
    * Use memset to clear bignum.
    * Improve data locality.
    * Reduce size of bignum.

Split Strtod() (InsightSoftwareConsortium#106)
    Add `StrtodTrimmed` method, exposing a later stage of the conversion pipeline.
    Some tools can do the first stage outside of the double-conversion library and would prefer not to pay the cost of doing it again.

Split double-conversion. (InsightSoftwareConsortium#104)
    Separates the two main classes into separate c and h files.
    Fix naming. (InsightSoftwareConsortium#103)
    Fix naming of `case_insensibility` to `case_insensitivity`.

Consistent macro prefix. (InsightSoftwareConsortium#101)

Use standard min/max. (InsightSoftwareConsortium#102)

Fix some issues with invalid hex-float literals.
    When converting `0x` the converter would assert (or access out of
        boundary).
    With `0x1.p1234556666FFFFF` the converter would overflow and not yield
    the correct exponent.

Usefulcat master (InsightSoftwareConsortium#98)
    * minor bug fix: use free() instead of delete[] to free memory allocated by strdup()
    * fix for uninitialized variable problem found by valgrind
    * disable assertions for 'optimize' build
    * removed -g option that was inadvertently added to CCFLAGS
    * ignore generated files

    Fix warning for g++ 4.9.3.

CMake: install to correct lib dir (InsightSoftwareConsortium#93)
    64-bit libraries should be installed in /usr/lib64, not in /usr/lib/
    Make the destination lib dir configurable.

Add big endian ARM support (InsightSoftwareConsortium#92)
    This fixes compilation on such platforms.

Switch to relative includes.
Fix 16-bit separators.

msvc: check if _MSC_VER is defined (InsightSoftwareConsortium#88)

Allow for compilation in emscripten (InsightSoftwareConsortium#86)
    Support emscripten

Add test cases.
    Fixes InsightSoftwareConsortium#62

Add support of ARC architecture (InsightSoftwareConsortium#82)
    More info about ARC architecture is here: [1] & [2].
    We need ARC supported here for many things like:
     - ICU (see [3])
     - Qt5 etc

Fix hex literal bug.
    Large hex literals would lose their minus sign.

Support separator characters.
Add support for hexadecimal float literals.
Fix bug where hex numbers would lose the minus sign.

Add comments for achitecture check.
    This should make it easier to add new architectures.

Add support for aarch64_be, or1k and microblazebe.
Add support for Windows on ARM and ARM64 (InsightSoftwareConsortium#76)

Use `static_assert` with newer compilers.
Add Native Client as support architecture.
Avoid undefined cast to make ASAN happy.
Avoid undefined cast to make ASAN happy.
Add `exports_files`
Processed length should include no trailing junk (InsightSoftwareConsortium#63)

Clarify output charset in DoubleToAscii documentation (InsightSoftwareConsortium#61)
    * Clarify output charset in DoubleToAscii documentation
    * Fixing typo in charset docs.

Fix warning for code that will never be executed (InsightSoftwareConsortium#59)

Rename macros.
    Renamed DISALLOW_COPY_AND_ASSIGN and DISALLOW_IMPLICIT_CONSTRUCTORS to
    DC_DISALLOW_COPY_AND_ASSIGN and DC_DISALLOW_IMPLICIT_CONSTRUCTORS.

    This makes it easier to use this library in projects that have macros
    with the same name (as is common in Google).

Some refactorings: remove unused static, replace deprecated headers, init member in constructor
    REF: replace deprecated headers
    REF: meaningless static definition in anonymous namespace
    REF: init member in constructor

Fix typo in variable name.

Suppress issue in clang analyzer.
CMake fixes.
    Remove unused CMake file.

Implement ALLOW_CASE_INSENSIBILITY mode for StringToDoubleConverter c…

Remove unnecessary INSTALL_INTERFACE expression.

Use template for CMake installation.

Fix mistake for build interface include dir.

Improve CMake changes.
    Update CMake package generation.

Implement ALLOW_CASE_INSENSIBILITY mode for StringToDoubleConverter class

Add assert and test.

Avoid negative shift.
    When filling in fractional digits in a fixed representation we
    might use all existing digits. When this happens, we can not look
    at the next digit to see if we should round up.
    Before this fix, we tried to shift by a negative amount to see if the
    (non-existing) next bit was set to 1 (requiring the number to be rounded up).
  • Loading branch information
@hjmjohnson
hjmjohnson committed Dec 26, 2020
1 parent e2f4724 commit 204bfc9
Show file tree
Hide file tree
Showing 20 changed files with 2,089 additions and 1,661 deletions.
24 changes: 14 additions & 10 deletions Modules/ThirdParty/DoubleConversion/src/double-conversion/CMakeLists.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,28 +22,32 @@ configure_file(double-conversion-configure.h.in double-conversion-configure.h)
include_directories(${CMAKE_CURRENT_BINARY_DIR})

set(headers
bignum-dtoa.h
bignum.h
cached-powers.h
diy-fp.h
double-conversion.h
double-to-string.h
fast-dtoa.h
fixed-dtoa.h
ieee.h
string-to-double.h
strtod.h
utils.h
${CMAKE_CURRENT_BINARY_DIR}/double-conversion-configure.h
)
)


add_library(itkdouble-conversion
bignum.cc
bignum-dtoa.cc
cached-powers.cc
diy-fp.cc
double-conversion.cc
fast-dtoa.cc
fixed-dtoa.cc
strtod.cc
${headers}
bignum.cc
bignum-dtoa.cc
cached-powers.cc
double-to-string.cc
fast-dtoa.cc
fixed-dtoa.cc
string-to-double.cc
strtod.cc
${headers}
)

if(ITK_LIBRARY_PROPERTIES)
Expand Down
4 changes: 2 additions & 2 deletions Modules/ThirdParty/DoubleConversion/src/double-conversion/SConscript
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,10 +3,10 @@ double_conversion_sources = [
'bignum.cc',
'bignum-dtoa.cc',
'cached-powers.cc',
'diy-fp.cc',
'double-conversion.cc',
'double-to-string.cc',
'fast-dtoa.cc',
'fixed-dtoa.cc',
'string-to-double.cc',
'strtod.cc'
]
Return('double_conversion_sources')
46 changes: 23 additions & 23 deletions Modules/ThirdParty/DoubleConversion/src/double-conversion/bignum-dtoa.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#include <math.h>
#include <cmath>

#include "bignum-dtoa.h"

Expand All @@ -35,7 +35,7 @@
namespace double_conversion {

static int NormalizedExponent(uint64_t significand, int exponent) {
ASSERT(significand != 0);
DOUBLE_CONVERSION_ASSERT(significand != 0);
while ((significand & Double::kHiddenBit) == 0) {
significand = significand << 1;
exponent = exponent - 1;
Expand Down Expand Up @@ -76,26 +76,26 @@ static void GenerateShortestDigits(Bignum* numerator, Bignum* denominator,
// Generates 'requested_digits' after the decimal point.
static void BignumToFixed(int requested_digits, int* decimal_point,
Bignum* numerator, Bignum* denominator,
Vector<char>(buffer), int* length);
Vector<char> buffer, int* length);
// Generates 'count' digits of numerator/denominator.
// Once 'count' digits have been produced rounds the result depending on the
// remainder (remainders of exactly .5 round upwards). Might update the
// decimal_point when rounding up (for example for 0.9999).
static void GenerateCountedDigits(int count, int* decimal_point,
Bignum* numerator, Bignum* denominator,
Vector<char>(buffer), int* length);
Vector<char> buffer, int* length);


void BignumDtoa(double v, BignumDtoaMode mode, int requested_digits,
Vector<char> buffer, int* length, int* decimal_point) {
ASSERT(v > 0);
ASSERT(!Double(v).IsSpecial());
DOUBLE_CONVERSION_ASSERT(v > 0);
DOUBLE_CONVERSION_ASSERT(!Double(v).IsSpecial());
uint64_t significand;
int exponent;
bool lower_boundary_is_closer;
if (mode == BIGNUM_DTOA_SHORTEST_SINGLE) {
float f = static_cast<float>(v);
ASSERT(f == v);
DOUBLE_CONVERSION_ASSERT(f == v);
significand = Single(f).Significand();
exponent = Single(f).Exponent();
lower_boundary_is_closer = Single(f).LowerBoundaryIsCloser();
Expand Down Expand Up @@ -134,7 +134,7 @@ void BignumDtoa(double v, BignumDtoaMode mode, int requested_digits,
// 4e-324. In this case the denominator needs fewer than 324*4 binary digits.
// The maximum double is 1.7976931348623157e308 which needs fewer than
// 308*4 binary digits.
ASSERT(Bignum::kMaxSignificantBits >= 324*4);
DOUBLE_CONVERSION_ASSERT(Bignum::kMaxSignificantBits >= 324*4);
InitialScaledStartValues(significand, exponent, lower_boundary_is_closer,
estimated_power, need_boundary_deltas,
&numerator, &denominator,
Expand Down Expand Up @@ -163,7 +163,7 @@ void BignumDtoa(double v, BignumDtoaMode mode, int requested_digits,
buffer, length);
break;
default:
UNREACHABLE();
DOUBLE_CONVERSION_UNREACHABLE();
}
buffer[*length] = '\0';
}
Expand Down Expand Up @@ -195,7 +195,7 @@ static void GenerateShortestDigits(Bignum* numerator, Bignum* denominator,
for (;;) {
uint16_t digit;
digit = numerator->DivideModuloIntBignum(*denominator);
ASSERT(digit <= 9); // digit is a uint16_t and therefore always positive.
DOUBLE_CONVERSION_ASSERT(digit <= 9); // digit is a uint16_t and therefore always positive.
// digit = numerator / denominator (integer division).
// numerator = numerator % denominator.
buffer[(*length)++] = static_cast<char>(digit + '0');
Expand Down Expand Up @@ -241,7 +241,7 @@ static void GenerateShortestDigits(Bignum* numerator, Bignum* denominator,
// loop would have stopped earlier.
// We still have an assert here in case the preconditions were not
// satisfied.
ASSERT(buffer[(*length) - 1] != '9');
DOUBLE_CONVERSION_ASSERT(buffer[(*length) - 1] != '9');
buffer[(*length) - 1]++;
} else {
// Halfway case.
Expand All @@ -252,7 +252,7 @@ static void GenerateShortestDigits(Bignum* numerator, Bignum* denominator,
if ((buffer[(*length) - 1] - '0') % 2 == 0) {
// Round down => Do nothing.
} else {
ASSERT(buffer[(*length) - 1] != '9');
DOUBLE_CONVERSION_ASSERT(buffer[(*length) - 1] != '9');
buffer[(*length) - 1]++;
}
}
Expand All @@ -264,9 +264,9 @@ static void GenerateShortestDigits(Bignum* numerator, Bignum* denominator,
// Round up.
// Note again that the last digit could not be '9' since this would have
// stopped the loop earlier.
// We still have an ASSERT here, in case the preconditions were not
// We still have an DOUBLE_CONVERSION_ASSERT here, in case the preconditions were not
// satisfied.
ASSERT(buffer[(*length) -1] != '9');
DOUBLE_CONVERSION_ASSERT(buffer[(*length) -1] != '9');
buffer[(*length) - 1]++;
return;
}
Expand All @@ -283,11 +283,11 @@ static void GenerateShortestDigits(Bignum* numerator, Bignum* denominator,
static void GenerateCountedDigits(int count, int* decimal_point,
Bignum* numerator, Bignum* denominator,
Vector<char> buffer, int* length) {
ASSERT(count >= 0);
DOUBLE_CONVERSION_ASSERT(count >= 0);
for (int i = 0; i < count - 1; ++i) {
uint16_t digit;
digit = numerator->DivideModuloIntBignum(*denominator);
ASSERT(digit <= 9); // digit is a uint16_t and therefore always positive.
DOUBLE_CONVERSION_ASSERT(digit <= 9); // digit is a uint16_t and therefore always positive.
// digit = numerator / denominator (integer division).
// numerator = numerator % denominator.
buffer[i] = static_cast<char>(digit + '0');
Expand All @@ -300,7 +300,7 @@ static void GenerateCountedDigits(int count, int* decimal_point,
if (Bignum::PlusCompare(*numerator, *numerator, *denominator) >= 0) {
digit++;
}
ASSERT(digit <= 10);
DOUBLE_CONVERSION_ASSERT(digit <= 10);
buffer[count - 1] = static_cast<char>(digit + '0');
// Correct bad digits (in case we had a sequence of '9's). Propagate the
// carry until we hat a non-'9' or til we reach the first digit.
Expand All @@ -325,7 +325,7 @@ static void GenerateCountedDigits(int count, int* decimal_point,
// Input verifies: 1 <= (numerator + delta) / denominator < 10.
static void BignumToFixed(int requested_digits, int* decimal_point,
Bignum* numerator, Bignum* denominator,
Vector<char>(buffer), int* length) {
Vector<char> buffer, int* length) {
// Note that we have to look at more than just the requested_digits, since
// a number could be rounded up. Example: v=0.5 with requested_digits=0.
// Even though the power of v equals 0 we can't just stop here.
Expand All @@ -341,7 +341,7 @@ static void BignumToFixed(int requested_digits, int* decimal_point,
} else if (-(*decimal_point) == requested_digits) {
// We only need to verify if the number rounds down or up.
// Ex: 0.04 and 0.06 with requested_digits == 1.
ASSERT(*decimal_point == -requested_digits);
DOUBLE_CONVERSION_ASSERT(*decimal_point == -requested_digits);
// Initially the fraction lies in range (1, 10]. Multiply the denominator
// by 10 so that we can compare more easily.
denominator->Times10();
Expand Down Expand Up @@ -370,7 +370,7 @@ static void BignumToFixed(int requested_digits, int* decimal_point,
// Returns an estimation of k such that 10^(k-1) <= v < 10^k where
// v = f * 2^exponent and 2^52 <= f < 2^53.
// v is hence a normalized double with the given exponent. The output is an
// approximation for the exponent of the decimal approimation .digits * 10^k.
// approximation for the exponent of the decimal approximation .digits * 10^k.
//
// The result might undershoot by 1 in which case 10^k <= v < 10^k+1.
// Note: this property holds for v's upper boundary m+ too.
Expand Down Expand Up @@ -420,7 +420,7 @@ static void InitialScaledStartValuesPositiveExponent(
Bignum* numerator, Bignum* denominator,
Bignum* delta_minus, Bignum* delta_plus) {
// A positive exponent implies a positive power.
ASSERT(estimated_power >= 0);
DOUBLE_CONVERSION_ASSERT(estimated_power >= 0);
// Since the estimated_power is positive we simply multiply the denominator
// by 10^estimated_power.

Expand Down Expand Up @@ -506,7 +506,7 @@ static void InitialScaledStartValuesNegativeExponentNegativePower(
// numerator = v * 10^-estimated_power * 2 * 2^-exponent.
// Remember: numerator has been abused as power_ten. So no need to assign it
// to itself.
ASSERT(numerator == power_ten);
DOUBLE_CONVERSION_ASSERT(numerator == power_ten);
numerator->MultiplyByUInt64(significand);

// denominator = 2 * 2^-exponent with exponent < 0.
Expand Down Expand Up @@ -548,7 +548,7 @@ static void InitialScaledStartValuesNegativeExponentNegativePower(
//
// Let ep == estimated_power, then the returned values will satisfy:
// v / 10^ep = numerator / denominator.
// v's boundarys m- and m+:
// v's boundaries m- and m+:
// m- / 10^ep == v / 10^ep - delta_minus / denominator
// m+ / 10^ep == v / 10^ep + delta_plus / denominator
// Or in other words:
Expand Down
Loading

0 comments on commit 204bfc9

Please sign in to comment.