mirror of
https://github.com/python/cpython.git
synced 2025-08-22 17:55:18 +00:00
gh-106529: Split FOR_ITER_{LIST,TUPLE} into uops (#106696)
Also rename `_ITER_EXHAUSTED_XXX` to `_IS_ITER_EXHAUSTED_XXX` to make it clear this is a test.
This commit is contained in:
parent
128a6c1d88
commit
025995fead
6 changed files with 385 additions and 118 deletions
|
@ -378,6 +378,7 @@ translate_bytecode_to_trace(
|
|||
_Py_CODEUNIT *initial_instr = instr;
|
||||
int trace_length = 0;
|
||||
int max_length = buffer_size;
|
||||
int reserved = 0;
|
||||
|
||||
#ifdef Py_DEBUG
|
||||
char *uop_debug = Py_GETENV("PYTHONUOPSDEBUG");
|
||||
|
@ -385,6 +386,9 @@ translate_bytecode_to_trace(
|
|||
if (uop_debug != NULL && *uop_debug >= '0') {
|
||||
lltrace = *uop_debug - '0'; // TODO: Parse an int and all that
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef Py_DEBUG
|
||||
#define DPRINTF(level, ...) \
|
||||
if (lltrace >= (level)) { fprintf(stderr, __VA_ARGS__); }
|
||||
#else
|
||||
|
@ -397,6 +401,8 @@ translate_bytecode_to_trace(
|
|||
uop_name(OPCODE), \
|
||||
(uint64_t)(OPERAND)); \
|
||||
assert(trace_length < max_length); \
|
||||
assert(reserved > 0); \
|
||||
reserved--; \
|
||||
trace[trace_length].opcode = (OPCODE); \
|
||||
trace[trace_length].operand = (OPERAND); \
|
||||
trace_length++;
|
||||
|
@ -409,9 +415,23 @@ translate_bytecode_to_trace(
|
|||
(INDEX), \
|
||||
uop_name(OPCODE), \
|
||||
(uint64_t)(OPERAND)); \
|
||||
assert(reserved > 0); \
|
||||
reserved--; \
|
||||
trace[(INDEX)].opcode = (OPCODE); \
|
||||
trace[(INDEX)].operand = (OPERAND);
|
||||
|
||||
// Reserve space for n uops
|
||||
#define RESERVE_RAW(n, opname) \
|
||||
if (trace_length + (n) > max_length) { \
|
||||
DPRINTF(2, "No room for %s (need %d, got %d)\n", \
|
||||
(opname), (n), max_length - trace_length); \
|
||||
goto done; \
|
||||
} \
|
||||
reserved = (n); // Keep ADD_TO_TRACE / ADD_TO_STUB honest
|
||||
|
||||
// Reserve space for main+stub uops, plus 2 for SAVE_IP and EXIT_TRACE
|
||||
#define RESERVE(main, stub) RESERVE_RAW((main) + (stub) + 2, uop_name(opcode))
|
||||
|
||||
DPRINTF(4,
|
||||
"Optimizing %s (%s:%d) at byte offset %ld\n",
|
||||
PyUnicode_AsUTF8(code->co_qualname),
|
||||
|
@ -420,16 +440,20 @@ translate_bytecode_to_trace(
|
|||
2 * INSTR_IP(initial_instr, code));
|
||||
|
||||
for (;;) {
|
||||
RESERVE_RAW(2, "epilogue"); // Always need space for SAVE_IP and EXIT_TRACE
|
||||
ADD_TO_TRACE(SAVE_IP, INSTR_IP(instr, code));
|
||||
|
||||
int opcode = instr->op.code;
|
||||
int oparg = instr->op.arg;
|
||||
int extras = 0;
|
||||
|
||||
while (opcode == EXTENDED_ARG) {
|
||||
instr++;
|
||||
extras += 1;
|
||||
opcode = instr->op.code;
|
||||
oparg = (oparg << 8) | instr->op.arg;
|
||||
}
|
||||
|
||||
if (opcode == ENTER_EXECUTOR) {
|
||||
_PyExecutorObject *executor =
|
||||
(_PyExecutorObject *)code->co_executors->executors[oparg&255];
|
||||
|
@ -437,17 +461,14 @@ translate_bytecode_to_trace(
|
|||
DPRINTF(2, " * ENTER_EXECUTOR -> %s\n", _PyOpcode_OpName[opcode]);
|
||||
oparg = (oparg & 0xffffff00) | executor->vm_data.oparg;
|
||||
}
|
||||
|
||||
switch (opcode) {
|
||||
|
||||
case POP_JUMP_IF_FALSE:
|
||||
case POP_JUMP_IF_TRUE:
|
||||
{
|
||||
// Assume jump unlikely (TODO: handle jump likely case)
|
||||
// Reserve 5 entries (1 here, 2 stub, plus SAVE_IP + EXIT_TRACE)
|
||||
if (trace_length + 5 > max_length) {
|
||||
DPRINTF(1, "Ran out of space for POP_JUMP_IF_FALSE\n");
|
||||
goto done;
|
||||
}
|
||||
RESERVE(1, 2);
|
||||
_Py_CODEUNIT *target_instr =
|
||||
instr + 1 + _PyOpcode_Caches[_PyOpcode_Deopt[opcode]] + oparg;
|
||||
max_length -= 2; // Really the start of the stubs
|
||||
|
@ -461,9 +482,8 @@ translate_bytecode_to_trace(
|
|||
|
||||
case JUMP_BACKWARD:
|
||||
{
|
||||
if (instr + 2 - oparg == initial_instr
|
||||
&& trace_length + 3 <= max_length)
|
||||
{
|
||||
if (instr + 2 - oparg == initial_instr) {
|
||||
RESERVE(1, 0);
|
||||
ADD_TO_TRACE(JUMP_TO_TOP, 0);
|
||||
}
|
||||
else {
|
||||
|
@ -474,26 +494,45 @@ translate_bytecode_to_trace(
|
|||
|
||||
case JUMP_FORWARD:
|
||||
{
|
||||
RESERVE(0, 0);
|
||||
// This will emit two SAVE_IP instructions; leave it to the optimizer
|
||||
instr += oparg;
|
||||
break;
|
||||
}
|
||||
|
||||
case FOR_ITER_LIST:
|
||||
case FOR_ITER_TUPLE:
|
||||
case FOR_ITER_RANGE:
|
||||
{
|
||||
// Assume jump unlikely (can a for-loop exit be likely?)
|
||||
// Reserve 9 entries (4 here, 3 stub, plus SAVE_IP + EXIT_TRACE)
|
||||
if (trace_length + 9 > max_length) {
|
||||
DPRINTF(1, "Ran out of space for FOR_ITER_RANGE\n");
|
||||
goto done;
|
||||
RESERVE(4, 3);
|
||||
int check_op, exhausted_op, next_op;
|
||||
switch (opcode) {
|
||||
case FOR_ITER_LIST:
|
||||
check_op = _ITER_CHECK_LIST;
|
||||
exhausted_op = _IS_ITER_EXHAUSTED_LIST;
|
||||
next_op = _ITER_NEXT_LIST;
|
||||
break;
|
||||
case FOR_ITER_TUPLE:
|
||||
check_op = _ITER_CHECK_TUPLE;
|
||||
exhausted_op = _IS_ITER_EXHAUSTED_TUPLE;
|
||||
next_op = _ITER_NEXT_TUPLE;
|
||||
break;
|
||||
case FOR_ITER_RANGE:
|
||||
check_op = _ITER_CHECK_RANGE;
|
||||
exhausted_op = _IS_ITER_EXHAUSTED_RANGE;
|
||||
next_op = _ITER_NEXT_RANGE;
|
||||
break;
|
||||
default:
|
||||
Py_UNREACHABLE();
|
||||
}
|
||||
// Assume jump unlikely (can a for-loop exit be likely?)
|
||||
_Py_CODEUNIT *target_instr = // +1 at the end skips over END_FOR
|
||||
instr + 1 + _PyOpcode_Caches[_PyOpcode_Deopt[opcode]] + oparg + 1;
|
||||
max_length -= 3; // Really the start of the stubs
|
||||
ADD_TO_TRACE(_ITER_CHECK_RANGE, 0);
|
||||
ADD_TO_TRACE(_ITER_EXHAUSTED_RANGE, 0);
|
||||
ADD_TO_TRACE(check_op, 0);
|
||||
ADD_TO_TRACE(exhausted_op, 0);
|
||||
ADD_TO_TRACE(_POP_JUMP_IF_TRUE, max_length);
|
||||
ADD_TO_TRACE(_ITER_NEXT_RANGE, 0);
|
||||
ADD_TO_TRACE(next_op, 0);
|
||||
|
||||
ADD_TO_STUB(max_length + 0, POP_TOP, 0);
|
||||
ADD_TO_STUB(max_length + 1, SAVE_IP, INSTR_IP(target_instr, code));
|
||||
|
@ -507,10 +546,7 @@ translate_bytecode_to_trace(
|
|||
if (expansion->nuops > 0) {
|
||||
// Reserve space for nuops (+ SAVE_IP + EXIT_TRACE)
|
||||
int nuops = expansion->nuops;
|
||||
if (trace_length + nuops + 2 > max_length) {
|
||||
DPRINTF(1, "Ran out of space for %s\n", uop_name(opcode));
|
||||
goto done;
|
||||
}
|
||||
RESERVE(nuops, 0);
|
||||
for (int i = 0; i < nuops; i++) {
|
||||
uint64_t operand;
|
||||
int offset = expansion->uops[i].offset;
|
||||
|
@ -556,12 +592,14 @@ translate_bytecode_to_trace(
|
|||
}
|
||||
DPRINTF(2, "Unsupported opcode %s\n", uop_name(opcode));
|
||||
goto done; // Break out of loop
|
||||
}
|
||||
}
|
||||
} // End default
|
||||
|
||||
} // End switch (opcode)
|
||||
|
||||
instr++;
|
||||
// Add cache size for opcode
|
||||
instr += _PyOpcode_Caches[_PyOpcode_Deopt[opcode]];
|
||||
}
|
||||
} // End for (;;)
|
||||
|
||||
done:
|
||||
// Skip short traces like SAVE_IP, LOAD_FAST, SAVE_IP, EXIT_TRACE
|
||||
|
@ -610,6 +648,9 @@ done:
|
|||
}
|
||||
return 0;
|
||||
|
||||
#undef RESERVE
|
||||
#undef RESERVE_RAW
|
||||
#undef INSTR_IP
|
||||
#undef ADD_TO_TRACE
|
||||
#undef DPRINTF
|
||||
}
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue