Skip to content

ext/bcmath: optimized divmod() and mod() take 2 #18058

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 13 commits into
base: master
Choose a base branch
from
+236 −145
Open

ext/bcmath: optimized divmod() and mod() take 2 #18058

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
03955d6
Added `bc_divide_copy_numerator()`
SakiTakamachi Mar 14, 2025
aaf1f6a
At the end of the calculation, set the value of `numerator_vectors` t…
SakiTakamachi Mar 14, 2025
b2c1b3f
Changed `bc_divide` to `bc_divide_ex`, and changed three division fun…
SakiTakamachi Mar 14, 2025
c0000cd
Set the sign of `quot` efficiently
SakiTakamachi Mar 14, 2025
0865562
Use `quot` only if `use_quot == true`, in `bc_divide_ex`
SakiTakamachi Mar 14, 2025
5e4b175
Added `bc_convert_vector_to_char()` and `bc_convert_vector_to_char_wi…
SakiTakamachi Mar 14, 2025
13a98ae
Use `quot` only if `use_quot == true`, in other funcs
SakiTakamachi Mar 14, 2025
b949c5e
Roughly write code using `use_rem`
SakiTakamachi Mar 14, 2025
484d9df
Using `rem` in `bc_divide_by_one`
SakiTakamachi Mar 14, 2025
2d63079
Create a process to initialize `bc_num` of `rem`
SakiTakamachi Mar 14, 2025
c7916b5
Using `rem` in `bc_divide_by_pow_10`
SakiTakamachi Mar 14, 2025
a2ba77c
Using `rem` in `bc_do_div`
SakiTakamachi Mar 14, 2025
d3a5713
fixed raisemod
SakiTakamachi Mar 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 0 additions & 1 deletion ext/bcmath/config.m4
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,6 @@ if test "$PHP_BCMATH" != "no"; then
libbcmath/src/compare.c
libbcmath/src/convert.c
libbcmath/src/div.c
libbcmath/src/divmod.c
libbcmath/src/doaddsub.c
libbcmath/src/floor_or_ceil.c
libbcmath/src/long2num.c
Expand Down
2 changes: 1 addition & 1 deletion ext/bcmath/config.w32
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ ARG_ENABLE("bcmath", "bc style precision math functions", "yes");
if (PHP_BCMATH == "yes") {
EXTENSION("bcmath", "bcmath.c", null, "/DZEND_ENABLE_STATIC_TSRMLS_CACHE=1");
ADD_SOURCES("ext/bcmath/libbcmath/src", "add.c div.c init.c neg.c \
raisemod.c sub.c compare.c divmod.c int2num.c long2num.c \
raisemod.c sub.c compare.c int2num.c long2num.c \
num2long.c recmul.c sqrt.c zero.c doaddsub.c \
floor_or_ceil.c nearzero.c num2str.c raise.c rmzero.c str2num.c \
round.c convert.c", "bcmath");
Expand Down
11 changes: 6 additions & 5 deletions ext/bcmath/libbcmath/src/bcmath.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,11 +147,7 @@ bc_num bc_multiply(bc_num n1, bc_num n2, size_t scale);
*(result) = mul_ex; \
} while (0)

bool bc_divide(bc_num n1, bc_num n2, bc_num *quot, size_t scale);

bool bc_modulo(bc_num num1, bc_num num2, bc_num *resul, size_t scale);

bool bc_divmod(bc_num num1, bc_num num2, bc_num *quo, bc_num *rem, size_t scale);
bool bc_divide_ex(bc_num numerator, bc_num divisor, bc_num *quot, bc_num *rem, size_t scale, bool use_quot, bool use_rem);

bc_num bc_floor_or_ceil(bc_num num, bool is_floor);

Expand Down Expand Up @@ -188,4 +184,9 @@ bool bc_sqrt(bc_num *num, size_t scale);
#define bc_free_num(num) _bc_free_num_ex((num), 0)
#define bc_num2str(num) bc_num2str_ex((num), (num->n_scale))

/* div and mod */
#define bc_divide(n1, n2, quot, scale) bc_divide_ex((n1), (n2), (quot), NULL, (scale), true, false)
#define bc_modulo(n1, n2, rem, scale) bc_divide_ex((n1), (n2), NULL, (rem), (scale), false, true)
#define bc_divmod(n1, n2, quot, rem, scale) bc_divide_ex((n1), (n2), (quot), (rem), (scale), true, true)

#endif
26 changes: 26 additions & 0 deletions ext/bcmath/libbcmath/src/convert.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,4 +111,30 @@ static inline void bc_convert_vector_to_char(const BC_VECTOR *vector, char *nptr
}
}

static inline void bc_convert_vector_to_char_with_skip(const BC_VECTOR *vector, char *nptr, char *nend, size_t arr_size, size_t skip)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This function either needs a better name or a doc block explaining what it does on a high level. What is skip and why is it necessary? I don't know at this point.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@nielsdos

For example, if BC_VECTOR = [7890, 3456, 12], and want to construct a char array like 1234567, need to skip three digits 890 in 7890.

This kind of skipping is necessary when storing the rem (from division) into a char array, especially when the number of digits in BC_VECTOR doesn’t match the required scale for the rem.

↑ Would this explanation be clear enough to convey the intent?

{
/* bulk skip */
size_t array_skip = skip / BC_VECTOR_SIZE;
arr_size -= array_skip;
vector += array_skip;

/* skip */
skip %= BC_VECTOR_SIZE;
if (skip > 0) {
BC_VECTOR current_vector = *vector;
current_vector /= BC_POW_10_LUT[skip];
size_t write_size = MIN(nend - nptr + 1, BC_VECTOR_SIZE - skip);
for (size_t i = 0; i < write_size; i++) {
*nend-- = current_vector % BASE;
current_vector /= BASE;
}
vector++;
arr_size--;
}

if (arr_size > 0) {
bc_convert_vector_to_char(vector, nptr, nend, arr_size);
}
}

#endif
235 changes: 190 additions & 45 deletions ext/bcmath/libbcmath/src/div.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,7 @@ static inline void bc_fast_div(
}
/* last */
quot_vectors[0] = numerator_vectors[0] / divisor_vector;
numerator_vectors[0] -= divisor_vector * quot_vectors[0];
}

/*
Expand Down Expand Up @@ -248,12 +249,15 @@ static inline void bc_standard_div(
div_carry = numerator_vectors[numerator_top_index - i];
numerator_vectors[numerator_top_index - i] = 0;
}
numerator_vectors[numerator_top_index - quot_arr_size + 1] = div_carry;
}

static void bc_do_div(
const char *numerator, size_t numerator_size, size_t numerator_readable_size,
const char *divisor, size_t divisor_size,
bc_num *quot, size_t quot_size
bc_num *quot, size_t quot_size,
bc_num *rem, size_t rem_over_size, size_t rem_write_size,
bool use_quot, bool use_rem
) {
size_t numerator_arr_size = BC_ARR_SIZE_FROM_LEN(numerator_size);
size_t divisor_arr_size = BC_ARR_SIZE_FROM_LEN(divisor_size);
Expand Down Expand Up @@ -290,57 +294,134 @@ static void bc_do_div(
}

/* Convert to bc_num */
char *qptr = (*quot)->n_value;
char *qend = qptr + (*quot)->n_len + (*quot)->n_scale - 1;
if (use_quot) {
char *qptr = (*quot)->n_value;
char *qend = qptr + (*quot)->n_len + (*quot)->n_scale - 1;
bc_convert_vector_to_char(quot_vectors, qptr, qend, quot_real_arr_size);
}
if (use_rem) {
char *rptr = (*rem)->n_value;
char *rend = rptr + rem_write_size - 1;

size_t rem_arr_size = (rem_write_size + rem_over_size + BC_VECTOR_SIZE - 1) / BC_VECTOR_SIZE;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Didn't we have a macro for the rounding-up divide ?

if (UNEXPECTED(rem_arr_size > numerator_arr_size)) {
/* If numerator_arr_size is exceeded because the integer part is zero */
rem_arr_size = numerator_arr_size;
*rptr = 0;
}
BC_VECTOR *rem_vectors = numerator_vectors;

bc_convert_vector_to_char(quot_vectors, qptr, qend, quot_real_arr_size);
if (rem_over_size > 0) {
bc_convert_vector_to_char_with_skip(rem_vectors, rptr, rend, rem_arr_size, rem_over_size);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I still don't understand the skip

} else {
bc_convert_vector_to_char(rem_vectors, rptr, rend, rem_arr_size);
}
}

if (allocation_arr_size > BC_STACK_VECTOR_SIZE) {
efree(numerator_vectors);
}
}

static inline void bc_divide_by_one(bc_num numerator, bc_num *quot, size_t quot_scale)
static inline void bc_divide_copy_numerator(bc_num numerator, bc_num *num, size_t scale)
{
quot_scale = MIN(numerator->n_scale, quot_scale);
*quot = bc_new_num_nonzeroed(numerator->n_len, quot_scale);
char *qptr = (*quot)->n_value;
memcpy(qptr, numerator->n_value, numerator->n_len + quot_scale);
scale = MIN(numerator->n_scale, scale);
*num = bc_new_num_nonzeroed(numerator->n_len, scale);
memcpy((*num)->n_value, numerator->n_value, numerator->n_len + scale);
}

static inline void bc_divide_by_pow_10(
const char *numeratorptr, size_t numerator_readable_size, bc_num *quot, size_t quot_size, size_t quot_scale)
static inline void bc_divide_by_one(
bc_num numerator, bc_num divisor, bc_num *quot, bc_num *rem,
size_t quot_scale, size_t rem_scale, bool use_quot, bool use_rem)
{
char *qptr = (*quot)->n_value;
for (size_t i = quot_size; i <= quot_scale; i++) {
*qptr++ = 0;
if (use_quot) {
bc_divide_copy_numerator(numerator, quot, quot_scale);
(*quot)->n_sign = numerator->n_sign == divisor->n_sign ? PLUS : MINUS;
}
if (use_rem) {
/* When dividing by 1, the integer part of rem is always 0. */
rem_scale = MIN(numerator->n_scale, rem_scale);
if (rem_scale == 0) {
*rem = bc_copy_num(BCG(_zero_));
} else {
*rem = bc_new_num_nonzeroed(1, rem_scale); /* 1 is for 0 */
(*rem)->n_value[0] = 0;
/* copy fractional part */
memcpy((*rem)->n_value + 1, numerator->n_value + numerator->n_len, rem_scale);
if (bc_is_zero(*rem)) {
(*rem)->n_sign = PLUS;
(*rem)->n_scale = 0;
} else {
(*rem)->n_sign = numerator->n_sign;
}
}
}
}

size_t numerator_use_size = quot_size > numerator_readable_size ? numerator_readable_size : quot_size;
memcpy(qptr, numeratorptr, numerator_use_size);
qptr += numerator_use_size;

if (numerator_use_size < (*quot)->n_len) {
/* e.g. 12.3 / 0.01 <=> 1230 */
for (size_t i = numerator_use_size; i < (*quot)->n_len; i++) {
static inline void bc_divide_by_pow_10(
const char *numeratorptr, size_t numerator_readable_size, size_t numerator_leading_zeros,
bc_num *quot, size_t quot_size, size_t quot_scale, bool use_quot,
bc_num *rem, size_t rem_size, bool use_rem,
size_t numerator_rem_len_diff)
{
if (use_quot) {
char *qptr = (*quot)->n_value;
for (size_t i = quot_size; i <= quot_scale; i++) {
*qptr++ = 0;
}
(*quot)->n_scale = 0;
} else {
char *qend = (*quot)->n_value + (*quot)->n_len + (*quot)->n_scale;
(*quot)->n_scale -= qend - qptr;

size_t numerator_use_size = quot_size > numerator_readable_size ? numerator_readable_size : quot_size;
memcpy(qptr, numeratorptr, numerator_use_size);
qptr += numerator_use_size;

if (numerator_use_size < (*quot)->n_len) {
/* e.g. 12.3 / 0.01 <=> 1230 */
for (size_t i = numerator_use_size; i < (*quot)->n_len; i++) {
*qptr++ = 0;
}
(*quot)->n_scale = 0;
} else {
char *qend = (*quot)->n_value + (*quot)->n_len + (*quot)->n_scale;
(*quot)->n_scale -= qend - qptr;
}
}
if (use_rem) {
size_t rem_leading_zeros = numerator_leading_zeros + quot_size - numerator_rem_len_diff;
if (rem_size <= rem_leading_zeros) {
bc_free_num(rem);
*rem = bc_copy_num(BCG(_zero_));
return;
}
/* The values after this have already been copied, so just need to set them to 0. */
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copied by whom?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

About 10 lines above the line where this function is called, there’s the following line:

memcpy((*rem)->n_value + rem_write_size, numerator->n_value + rem_write_size + numerator_rem_len_diff, copy_size);

The intention here was that everything except the leading zeros has already been copied.

for (size_t i = 0; i < rem_leading_zeros; i++) {
(*rem)->n_value[i] = 0;
}
_bc_rm_leading_zeros(*rem);
if (bc_is_zero(*rem)) {
(*rem)->n_sign = PLUS;
(*rem)->n_scale = 0;
}
}
}

bool bc_divide(bc_num numerator, bc_num divisor, bc_num *quot, size_t scale)
bool bc_divide_ex(bc_num numerator, bc_num divisor, bc_num *quot, bc_num *rem, size_t scale, bool use_quot, bool use_rem)
{
/* divide by zero */
if (bc_is_zero(divisor)) {
return false;
}

bc_free_num(quot);
size_t quot_scale = scale;
size_t quot_scale = 0;
size_t rem_scale = 0;
if (use_quot) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Instead of using a separate argument for use_quot, use_rem. Why not check if quot/rem is a NULL pointer?

bc_free_num(quot);
}
if (use_rem) {
bc_free_num(rem);
rem_scale = scale;
} else {
quot_scale = scale;
}

/* If numerator is zero, the quotient is always zero. */
if (bc_is_zero(numerator)) {
Expand All @@ -349,8 +430,7 @@ bool bc_divide(bc_num numerator, bc_num divisor, bc_num *quot, size_t scale)

/* If divisor is 1 / -1, the quotient's n_value is equal to numerator's n_value. */
if (_bc_do_compare(divisor, BCG(_one_), divisor->n_scale, false) == BCMATH_EQUAL) {
bc_divide_by_one(numerator, quot, quot_scale);
(*quot)->n_sign = numerator->n_sign == divisor->n_sign ? PLUS : MINUS;
bc_divide_by_one(numerator, divisor, quot, rem, quot_scale, rem_scale, use_quot, use_rem);
return true;
}

Expand All @@ -372,10 +452,12 @@ bool bc_divide(bc_num numerator, bc_num divisor, bc_num *quot, size_t scale)
numerator_size -= numerator_leading_zeros;

/* check and remove divisor leading zeros */
size_t divisor_leading_zeros = 0;
while (*divisorptr == 0) {
divisorptr++;
divisor_size--;
divisor_leading_zeros++;
}
divisor_size -= divisor_leading_zeros;

if (divisor_size > numerator_size) {
goto quot_zero;
Expand All @@ -393,39 +475,102 @@ bool bc_divide(bc_num numerator, bc_num divisor, bc_num *quot, size_t scale)
numerator_size -= divisor_trailing_zeros;

size_t quot_size = numerator_size - divisor_size + 1; /* numerator_size >= divisor_size */
if (quot_size > quot_scale) {
*quot = bc_new_num_nonzeroed(quot_size - quot_scale, quot_scale);
} else {
*quot = bc_new_num_nonzeroed(1, quot_scale); /* 1 is for 0 */
if (use_quot) {
if (quot_size > quot_scale) {
*quot = bc_new_num_nonzeroed(quot_size - quot_scale, quot_scale);
} else {
*quot = bc_new_num_nonzeroed(1, quot_scale); /* 1 is for 0 */
}
(*quot)->n_sign = numerator->n_sign == divisor->n_sign ? PLUS : MINUS;
}

/**
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm missing the high level explanation here on how you compute the remainder length

* If the calculation uses more digits than the scale of rem, writing the vector directly to rem
* will exceed the size, so calculate the excess size in advance.
*/
size_t rem_over_size = 0;

/**
* Conversely, there are cases where the vector does not fill the rem size.
* In this case, the size to be written is calculated in advance to determine the start position for writing to rem.
*/
size_t rem_write_size = 0;

size_t rem_size = 0;
size_t numerator_rem_len_diff = 0;
if (use_rem) {
size_t divisor_int_size = divisor->n_len > divisor_leading_zeros ? divisor->n_len - divisor_leading_zeros : 0;
size_t divisor_frac_size = divisor->n_scale > divisor_trailing_zeros ? divisor->n_scale - divisor_trailing_zeros : 0;
rem_scale = MIN(MAX(numerator->n_scale, divisor_frac_size), rem_scale);

*rem = bc_new_num_nonzeroed(divisor_int_size > 0 ? divisor_int_size : 1, rem_scale); // 1 is for 0
(*rem)->n_sign = numerator->n_sign;

if (divisor_frac_size > rem_scale) {
rem_over_size = divisor_frac_size - rem_scale;
rem_write_size = (*rem)->n_len + rem_scale;
} else {
if (divisor_frac_size > 0) {
rem_write_size = (*rem)->n_len + divisor_frac_size;
} else {
/* e.g. 123 % 30 */
rem_write_size = (*rem)->n_len - (divisor_trailing_zeros - divisor->n_scale);
}
}

rem_size = (*rem)->n_len + (*rem)->n_scale;
if (rem_size > rem_write_size) {
size_t copy_size = rem_size - rem_write_size;
numerator_rem_len_diff = numerator->n_len - (*rem)->n_len;
memcpy((*rem)->n_value + rem_write_size, numerator->n_value + rem_write_size + numerator_rem_len_diff, copy_size);
}
}

/* Size that can be read from numeratorptr */
size_t numerator_readable_size = numerator->n_len + numerator->n_scale - numerator_leading_zeros;

/* If divisor is 1 here, return the result of adjusting the decimal point position of numerator. */
if (divisor_size == 1 && *divisorptr == 1) {
bc_divide_by_pow_10(numeratorptr, numerator_readable_size, quot, quot_size, quot_scale);
(*quot)->n_sign = numerator->n_sign == divisor->n_sign ? PLUS : MINUS;
bc_divide_by_pow_10(
numeratorptr, numerator_readable_size, numerator_leading_zeros,
quot, quot_size, quot_scale, use_quot,
rem, rem_size, use_rem, numerator_rem_len_diff
);
return true;
}

/* do divide */
bc_do_div(
numeratorptr, numerator_size, numerator_readable_size,
divisorptr, divisor_size,
quot, quot_size
quot, quot_size,
rem, rem_over_size, rem_write_size,
use_quot, use_rem
);

_bc_rm_leading_zeros(*quot);
if (bc_is_zero(*quot)) {
(*quot)->n_sign = PLUS;
(*quot)->n_scale = 0;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You no longer reset the scale in this case, but you only removed the leading zeros (on line 425) and not the trailing zeros. Can you explain please?

} else {
(*quot)->n_sign = numerator->n_sign == divisor->n_sign ? PLUS : MINUS;
if (use_quot) {
_bc_rm_leading_zeros(*quot);
if (bc_is_zero(*quot)) {
(*quot)->n_sign = PLUS;
(*quot)->n_scale = 0;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I guess my previous question is answered because this line returned, but you should instead amend the previous commit so that it wasn't deleted in the first place. That keeps the history cleaner.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@nielsdos

Sorry, I think a line was accidentally removed when I rebased and merged in master.

I had only reviewed the final version of the code, so I didn’t notice that one of the intermediate commits included an unintended deletion.

}
}
if (use_rem) {
_bc_rm_leading_zeros(*rem);
if (bc_is_zero(*rem)) {
(*rem)->n_sign = PLUS;
(*rem)->n_scale = 0;
}
}
return true;

quot_zero:
*quot = bc_copy_num(BCG(_zero_));
if (use_quot) {
*quot = bc_copy_num(BCG(_zero_));
}
if (use_rem) {
bc_divide_copy_numerator(numerator, rem, rem_scale);
(*rem)->n_sign = numerator->n_sign;
}
return true;
}
Loading