GeezerVison is showing github.com — GH-115802: JIT using the "medium" code model on `x86_64-unknown-linux-gnu` by brandtbucher · Pull Request #130097 · python/cpython
Conversation
brandtbucher
commented
Feb 13, 2025
•
edited by bedevere-app
Bot
Loading
Uh oh!
There was an error while loading. Please reload this page.
This is a perfect middle-ground between the "large" code model (which we used to use) and the "small" code model (which we currently use):
- Local data, like
OPARG, is encoded directly in the instruction stream (currently they're loaded indirectly).
- Extern data, like
&_PyEval_BinaryOps, is encoded directly in the instruction stream (currently they're loaded indirectly).
- Local jumps, like
_JIT_ERROR_TARGET, use 32-bit jumps (currently they use "relaxable" 64-bit indirect jumps).
- Extern jumps, like
_Py_Dealloc, use "relaxable" 64-bit indirect jumps (same as today).
This only works on one platform, but it's an important one. Looks to be 0.5%-1% faster on benchmarks, as well as a very slight (~0.15%) memory savings due to having to JIT less auxiliary data for storing addresses.
Here's the before-and-after of _LOAD_SMALL_INT:
void
emit__LOAD_SMALL_INT(
unsigned char *code, unsigned char *data, _PyExecutorObject *executor,
const _PyUOpInstruction *instruction, jit_state *state)
{
//
// _LOAD_SMALL_INT.o: file format elf64-x86-64
//
// Disassembly of section .text:
//
// 0000000000000000 <_JIT_ENTRY>:
// 0: 0f b7 05 00 00 00 00 movzwl (%rip), %eax # 0x7 <_JIT_ENTRY+0x7>
// 0000000000000003: R_X86_64_GOTPCREL _JIT_OPARG-0x4
// 7: c1 e0 05 shll $0x5, %eax
// a: 48 8b 0d 00 00 00 00 movq (%rip), %rcx # 0x11 <_JIT_ENTRY+0x11>
// 000000000000000d: R_X86_64_REX_GOTPCRELX _PyRuntime-0x4
// 11: 48 01 c8 addq %rcx, %rax
// 14: 48 05 f8 36 00 00 addq $0x36f8, %rax # imm = 0x36F8
// 1a: 49 89 45 00 movq %rax, (%r13)
// 1e: 49 83 c5 08 addq $0x8, %r13
// 22: ff 25 00 00 00 00 jmpq *(%rip) # 0x28 <_JIT_ENTRY+0x28>
// 0000000000000024: R_X86_64_GOTPCRELX _JIT_CONTINUE-0x4
const unsigned char code_body[34] = {
0x0f, 0xb7, 0x05, 0x00, 0x00, 0x00, 0x00, 0xc1,
0xe0, 0x05, 0x48, 0x8b, 0x0d, 0x00, 0x00, 0x00,
0x00, 0x48, 0x01, 0xc8, 0x48, 0x05, 0xf8, 0x36,
0x00, 0x00, 0x49, 0x89, 0x45, 0x00, 0x49, 0x83,
0xc5, 0x08,
};
// 0: OPARG
// 8: &_PyRuntime+0x0
patch_64(data + 0x0, instruction->oparg);
patch_64(data + 0x8, (uintptr_t)&_PyRuntime);
memcpy(code, code_body, sizeof(code_body));
patch_32r(code + 0x3, (uintptr_t)data + -0x4);
patch_x86_64_32rx(code + 0xd, (uintptr_t)data + 0x4);
}void
emit__LOAD_SMALL_INT(
unsigned char *code, unsigned char *data, _PyExecutorObject *executor,
const _PyUOpInstruction *instruction, jit_state *state)
{
//
// _LOAD_SMALL_INT.o: file format elf64-x86-64
//
// Disassembly of section .text:
//
// 0000000000000000 <_JIT_ENTRY>:
// 0: 48 b8 00 00 00 00 00 00 00 00 movabsq $0x0, %rax
// 0000000000000002: R_X86_64_64 _JIT_OPARG
// a: 0f b7 c0 movzwl %ax, %eax
// d: c1 e0 05 shll $0x5, %eax
// 10: 48 b9 00 00 00 00 00 00 00 00 movabsq $0x0, %rcx
// 0000000000000012: R_X86_64_64 _PyRuntime+0x36f8
// 1a: 48 01 c1 addq %rax, %rcx
// 1d: 49 89 4d 00 movq %rcx, (%r13)
// 21: 49 83 c5 08 addq $0x8, %r13
// 25: e9 00 00 00 00 jmp 0x2a <_JIT_ENTRY+0x2a>
// 0000000000000026: R_X86_64_PLT32 _JIT_CONTINUE-0x4
const unsigned char code_body[37] = {
0x48, 0xb8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x0f, 0xb7, 0xc0, 0xc1, 0xe0, 0x05,
0x48, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x48, 0x01, 0xc1, 0x49, 0x89, 0x4d,
0x00, 0x49, 0x83, 0xc5, 0x08,
};
memcpy(code, code_body, sizeof(code_body));
patch_64(code + 0x2, instruction->oparg);
patch_64(code + 0x12, (uintptr_t)&_PyRuntime + 0x36f8);
}
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Oh, this is neat! Less indirection. LGTM!
Sorry, something went wrong.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Milestone
No milestone
Development
Successfully merging this pull request may close these issues.
Uh oh!
There was an error while loading. Please reload this page.
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
|