8000 gh-107557: Remove unnecessary SAVE_IP instructions by gvanrossum · Pull Request #108583 · python/cpython · GitHub
[go: up one dir, main page]
Skip to content

gh-107557: Remove unnecessary SAVE_IP instructions #108583

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 12 commits into from
Aug 29, 2023
Merged

gh-107557: Remove unnecessary SAVE_IP instructions #108583

Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
f289749
Add HAS_ERROR_FLAG and HAS_DEOPT_FLAG
gvanrossum Aug 26, 2023
290f955
Write metadata for all ops and all instrs
gvanrossum Aug 28, 2023
d9258d8
Squeeze out unneeded SAVE_IP from traces
gvanrossum Aug 28, 2023
51eaa2f
Factor squeeze_stubs() out of translate_bytecode_to_trace()
gvanrossum Aug 28, 2023
4c5c38d
Use squeeze_stubs() instead of inserting NOPs into gap created by rem…
gvanrossum Aug 28, 2023
869ae8d
When there's no gap we can set trace_length = buffer_size
gvanrossum Aug 28, 2023
bf9fefd
Always remove NOPs, even if no new ones were added
gvanrossum Aug 28, 2023
8d31dc0
Change meaning of last_instr to be base-0
gvanrossum Aug 28, 2023
afb7cc1
Tweaks to debug output
gvanrossum Aug 28, 2023
67b9e6b
Rename squeeze_stubs() to move_stubs()
gvanrossum Aug 29, 2023
ad5c50f
Shorten comment
gvanrossum Aug 29, 2023
ebce668
Initialize last_instr to 0, not -1
gvanrossum Aug 29, 2023
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
Prev Previous commit
Next Next commit
Factor squeeze_stubs() out of translate_bytecode_to_trace()
  • Loading branch information
@gvanrossum
gvanrossum committed Aug 28, 2023
commit 51eaa2fb54c5ebfdd696fab9510d4ceca06945b4
62 changes: 40 additions & 22 deletions Python/optimizer.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -372,6 +372,32 @@ static PyTypeObject UOpExecutor_Type = {
.tp_as_sequence = &uop_as_sequence,
};

static int
squeeze_stubs(
Copy link
Member

Choose a reason for hiding this comment

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

Suggested change
squeeze_stubs(
shift_stubs(

(the name implied to me that something happens to the size of the stubs).

Copy link
Member Author

Choose a reason for hiding this comment

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

How about move_stubs() (echoing that it uses memmove()).

_PyUOpInstruction *trace,
int trace_length,
int stubs_start,
int stubs_end
)
{
memmove(trace + trace_length,
trace + stubs_start,
(stubs_end - stubs_start) * sizeof(_PyUOpInstruction));
// Patch up the jump targets
for (int i = 0; i < trace_length; i++) {
if (trace[i].opcode == _POP_JUMP_IF_FALSE ||
trace[i].opcode == _POP_JUMP_IF_TRUE)
{
int target = trace[i].oparg;
if (target >= stubs_start) {
target += trace_length - stubs_start;
trace[i].oparg = target;
}
}
}
return trace_length + stubs_end - stubs_start;
}

#define TRACE_STACK_SIZE 5

static int
Expand Down Expand Up @@ -736,30 +762,22 @@ translate_bytecode_to_trace(
code->co_firstlineno,
2 * INSTR_IP(initial_instr, code),
trace_length);
if (max_length < buffer_size && trace_length < max_length) {
// Move the stubs back to be immediately after the main trace
// (which ends at trace_length)
DPRINTF(2,
"Moving %d stub uops back by %d\n",
buffer_size - max_length,
max_length - trace_length);
memmove(trace + trace_length,
trace + max_length,
(buffer_size - max_length) * sizeof(_PyUOpInstruction));
// Patch up the jump targets
for (int i = 0; i < trace_length; i++) {
if (trace[i].opcode == _POP_JUMP_IF_FALSE ||
trace[i].opcode == _POP_JUMP_IF_TRUE)
{
int target = trace[i].oparg;
if (target >= max_length) {
target += trace_length - max_length;
trace[i].oparg = target;
}
}
if (max_length < buffer_size) {
// There are stubs
if (trace_length < max_length) {
// There's a gap before the stubs
// Move the stubs back to be immediately after the main trace
// (which ends at trace_length)
DPRINTF(2,
"Moving %d stub uops back by %d\n",
buffer_size - max_length,
max_length - trace_length);
trace_length = squeeze_stubs(trace, trace_length, max_length, buffer_size);
}
else {
trace_length += buffer_size - max_length;
}
}
trace_length += buffer_size - max_length;
return trace_length;
}
else {
Expand Down
0