Skip to content

zend_compile: Optimize sprintf() into a rope #14546

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

Merged
merged 7 commits into from
Jun 13, 2024
Merged

zend_compile: Optimize sprintf() into a rope #14546

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
3d0f581
zend_compile: Add `zend_compile_rope_finalize()`
TimWolla Jun 11, 2024
72f9b24
zend_compile: Use clearer parameter names for `zend_compile_rope_fina…
TimWolla Jun 11, 2024
ccc1ac8
zend_compile: Fix `zend_compile_rope_finalize()` for ropes containing…
TimWolla Jun 11, 2024
9fc7a73
zend_compile: Optimize `sprintf()` into a rope
TimWolla May 31, 2024
cd788a2
zend_compile: Eagerly handle empty format strings in `sprintf()` opti…
TimWolla Jun 12, 2024
490f200
zend_compile: Add additional explanatory comments to zend_compile_fun…
TimWolla Jun 13, 2024
f05206d
Add zero-argument test to sprintf_rope_optimization_001.phpt
TimWolla Jun 13, 2024
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
268 changes: 221 additions & 47 deletions Zend/zend_compile.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4712,6 +4712,171 @@ static void zend_compile_ns_call(znode *result, znode *name_node, zend_ast *args
}
/* }}} */

static zend_op *zend_compile_rope_add(znode *result, uint32_t num, znode *elem_node);
Copy link
Contributor

Choose a reason for hiding this comment

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

The benchmark shows decreased performance:

image

Copy link
Member Author

Choose a reason for hiding this comment

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

Please explain how this change would be able to affect bench.php.

Please also explain how the addition of two comments to the source code significantly changed the results compared to these: https://github.com/php/php-src/actions/runs/9484406283

Copy link
Contributor

Choose a reason for hiding this comment

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

maybe the performance degradation is comming from #14054 and the base branch used is detected wrongly?

static zend_op *zend_compile_rope_add_ex(zend_op *opline, znode *result, uint32_t num, znode *elem_node);
static void zend_compile_rope_finalize(znode *result, uint32_t j, zend_op *init_opline, zend_op *opline);

static zend_result zend_compile_func_sprintf(znode *result, zend_ast_list *args) /* {{{ */
{
/* Bail out if we do not have a format string. */
if (args->children < 1) {
return FAILURE;
}

zend_eval_const_expr(&args->child[0]);
/* Bail out if the format string is not constant. */
if (args->child[0]->kind != ZEND_AST_ZVAL) {
return FAILURE;
}

zval *format_string = zend_ast_get_zval(args->child[0]);
if (Z_TYPE_P(format_string) != IS_STRING) {
return FAILURE;
}
if (Z_STRLEN_P(format_string) >= 256) {
return FAILURE;
}

char *p;
char *end;
uint32_t string_placeholder_count;

string_placeholder_count = 0;
p = Z_STRVAL_P(format_string);
end = p + Z_STRLEN_P(format_string);

for (;;) {
p = memchr(p, '%', end - p);
if (!p) {
break;
}

char *q = p + 1;
if (q == end) {
return FAILURE;
}

switch (*q) {
case 's':
string_placeholder_count++;
break;
case '%':
break;
default:
return FAILURE;
}

p = q;
p++;
}

/* Bail out if the number of placeholders does not match the number of values. */
if (string_placeholder_count != (args->children - 1)) {
return FAILURE;
}

/* Handle empty format strings. */
if (Z_STRLEN_P(format_string) == 0) {
result->op_type = IS_CONST;
ZVAL_EMPTY_STRING(&result->u.constant);

return SUCCESS;
}

znode *elements = NULL;

if (string_placeholder_count > 0) {
elements = safe_emalloc(sizeof(*elements), string_placeholder_count, 0);
}

/* Compile the value expressions first for error handling that is consistent
* with a function call: Values that fail to convert to a string may emit errors.
*/
for (uint32_t i = 0; i < string_placeholder_count; i++) {
zend_compile_expr(elements + i, args->child[1 + i]);
if (elements[i].op_type == IS_CONST) {
if (Z_TYPE(elements[i].u.constant) != IS_ARRAY) {
convert_to_string(&elements[i].u.constant);
}
Comment on lines +4798 to +4800
Copy link
Member

@Girgias Girgias Jun 13, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

You could potentially check if it is an object if it implements the Stringable interface? It would miss oddities like GMP not implementing the interface, although it can be cast to a string (which we really should fix...) but should work except if the __toString() method throws an error.

EDIT: I realized the issue is that one would need to possibly trigger autoloading to check the class....

Copy link
Member

Choose a reason for hiding this comment

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

These are compile time const expressions, they cannot currently contain objects (not even enums).

}
}

uint32_t rope_elements = 0;
uint32_t rope_init_lineno = -1;
zend_op *opline = NULL;

string_placeholder_count = 0;
p = Z_STRVAL_P(format_string);
end = p + Z_STRLEN_P(format_string);
char *offset = p;
for (;;) {
p = memchr(p, '%', end - p);
if (!p) {
break;
}

char *q = p + 1;
ZEND_ASSERT(q < end);
ZEND_ASSERT(*q == 's' || *q == '%');

if (*q == '%') {
/* Optimization to not create a dedicated rope element for the literal '%':
* Include the first '%' within the "constant" part instead of dropping the
* full placeholder.
*/
p++;
}

if (p != offset) {
znode const_node;
const_node.op_type = IS_CONST;
ZVAL_STRINGL(&const_node.u.constant, offset, p - offset);
if (rope_elements == 0) {
rope_init_lineno = get_next_op_number();
}
opline = zend_compile_rope_add(result, rope_elements++, &const_node);
Copy link
Member

Choose a reason for hiding this comment

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

Nit: For the case of %%, the merging of constant strings would actually be beneficial within zend_compile_rope_add(). Otherwise we're compiling sprintf('%%foo') this to '%' . 'foo', essentially. Although the optimizer probably takes care of this.

Copy link
Member Author

Choose a reason for hiding this comment

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

I expect %% to be fairly rare within format strings. The previous remark about “clarity” also applies here: Merging of consecutive constants should ideally happen in a generic fashion, instead of requiring every user to reimplement it.

Copy link
Member

Choose a reason for hiding this comment

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

Right. zend_compile_encaps_list() essentially does the same thing, which is what I was referring to. Hence, moving it to zend_compile_rope_add() would make it usable for both. That said, I wasn't expecting you to do that, but just noted it as a possible follow-up, and/or a potential // TODO to be added.

}

if (*q == 's') {
/* Perform the cast of constant arrays when actually evaluating corresponding placeholder
* for correct error reporting.
*/
if (elements[string_placeholder_count].op_type == IS_CONST) {
if (Z_TYPE(elements[string_placeholder_count].u.constant) == IS_ARRAY) {
zend_emit_op_tmp(&elements[string_placeholder_count], ZEND_CAST, &elements[string_placeholder_count], NULL)->extended_value = IS_STRING;
}
}
if (rope_elements == 0) {
rope_init_lineno = get_next_op_number();
}
opline = zend_compile_rope_add(result, rope_elements++, &elements[string_placeholder_count]);

string_placeholder_count++;
}

p = q;
p++;
offset = p;
}
if (end != offset) {
/* Add the constant part after the last placeholder. */
znode const_node;
const_node.op_type = IS_CONST;
ZVAL_STRINGL(&const_node.u.constant, offset, end - offset);
if (rope_elements == 0) {
rope_init_lineno = get_next_op_number();
}
opline = zend_compile_rope_add(result, rope_elements++, &const_node);
}
ZEND_ASSERT(opline != NULL);

zend_op *init_opline = CG(active_op_array)->opcodes + rope_init_lineno;
zend_compile_rope_finalize(result, rope_elements, init_opline, opline);
efree(elements);

return SUCCESS;
}

static zend_result zend_try_compile_special_func_ex(znode *result, zend_string *lcname, zend_ast_list *args, zend_function *fbc, uint32_t type) /* {{{ */
{
if (zend_string_equals_literal(lcname, "strlen")) {
Expand Down Expand Up @@ -4778,6 +4943,8 @@ static zend_result zend_try_compile_special_func_ex(znode *result, zend_string *
return zend_compile_func_array_slice(result, args);
} else if (zend_string_equals_literal(lcname, "array_key_exists")) {
return zend_compile_func_array_key_exists(result, args);
} else if (zend_string_equals_literal(lcname, "sprintf")) {
return zend_compile_func_sprintf(result, args);
} else {
return FAILURE;
}
Expand Down Expand Up @@ -10188,6 +10355,59 @@ static zend_op *zend_compile_rope_add(znode *result, uint32_t num, znode *elem_n
}
/* }}} */

static void zend_compile_rope_finalize(znode *result, uint32_t rope_elements, zend_op *init_opline, zend_op *opline)
{
if (rope_elements == 1) {
if (opline->op2_type == IS_CONST) {
GET_NODE(result, opline->op2);
ZVAL_UNDEF(CT_CONSTANT(opline->op2));
SET_UNUSED(opline->op2);
MAKE_NOP(opline);
} else {
opline->opcode = ZEND_CAST;
opline->extended_value = IS_STRING;
opline->op1_type = opline->op2_type;
opline->op1 = opline->op2;
SET_UNUSED(opline->op2);
zend_make_tmp_result(result, opline);
}
} else if (rope_elements == 2) {
opline->opcode = ZEND_FAST_CONCAT;
opline->extended_value = 0;
opline->op1_type = init_opline->op2_type;
opline->op1 = init_opline->op2;
zend_make_tmp_result(result, opline);
MAKE_NOP(init_opline);
} else {
uint32_t var;

init_opline->extended_value = rope_elements;
opline->opcode = ZEND_ROPE_END;
zend_make_tmp_result(result, opline);
var = opline->op1.var = get_temporary_variable();

/* Allocates the necessary number of zval slots to keep the rope */
uint32_t i = ((rope_elements * sizeof(zend_string*)) + (sizeof(zval) - 1)) / sizeof(zval);
while (i > 1) {
get_temporary_variable();
i--;
}

/* Update all the previous opcodes to use the same variable */
while (opline != init_opline) {
opline--;
if (opline->opcode == ZEND_ROPE_ADD &&
opline->result.var == (uint32_t)-1) {
opline->op1.var = var;
opline->result.var = var;
} else if (opline->opcode == ZEND_ROPE_INIT &&
opline->result.var == (uint32_t)-1) {
opline->result.var = var;
}
}
}
}

static void zend_compile_encaps_list(znode *result, zend_ast *ast) /* {{{ */
{
uint32_t i, j;
Expand Down Expand Up @@ -10263,53 +10483,7 @@ static void zend_compile_encaps_list(znode *result, zend_ast *ast) /* {{{ */
opline = zend_compile_rope_add_ex(opline, result, j++, &last_const_node);
}
init_opline = CG(active_op_array)->opcodes + rope_init_lineno;
if (j == 1) {
if (opline->op2_type == IS_CONST) {
GET_NODE(result, opline->op2);
MAKE_NOP(opline);
} else {
opline->opcode = ZEND_CAST;
opline->extended_value = IS_STRING;
opline->op1_type = opline->op2_type;
opline->op1 = opline->op2;
SET_UNUSED(opline->op2);
zend_make_tmp_result(result, opline);
}
} else if (j == 2) {
opline->opcode = ZEND_FAST_CONCAT;
opline->extended_value = 0;
opline->op1_type = init_opline->op2_type;
opline->op1 = init_opline->op2;
zend_make_tmp_result(result, opline);
MAKE_NOP(init_opline);
} else {
uint32_t var;

init_opline->extended_value = j;
opline->opcode = ZEND_ROPE_END;
zend_make_tmp_result(result, opline);
var = opline->op1.var = get_temporary_variable();

/* Allocates the necessary number of zval slots to keep the rope */
i = ((j * sizeof(zend_string*)) + (sizeof(zval) - 1)) / sizeof(zval);
while (i > 1) {
get_temporary_variable();
i--;
}

/* Update all the previous opcodes to use the same variable */
while (opline != init_opline) {
opline--;
if (opline->opcode == ZEND_ROPE_ADD &&
opline->result.var == (uint32_t)-1) {
opline->op1.var = var;
opline->result.var = var;
} else if (opline->opcode == ZEND_ROPE_INIT &&
opline->result.var == (uint32_t)-1) {
opline->result.var = var;
}
}
}
zend_compile_rope_finalize(result, j, init_opline, opline);
}
/* }}} */

Expand Down
Loading
Loading