Generate fallback for aarch64

Author: corona10

Python/jit.c

@@ -590,21 +590,20 @@ combine_symbol_mask(const symbol_mask src, symbol_mask dest)
 // TODO: With dynasm, the per-architecture #ifdef branches below could be
 // replaced by a single portable emission sequence.
 
-// Max oparg for manual codegen. Conservative (AArch64 imm12 limit).
-// TODO: could be set per-architecture (x86 disp32 has a much higher limit).
-#define _LOAD_FAST_BORROW_MAX_OPARG 4085
-
 // Decode a _LOAD_FAST_BORROW* opcode into register variant and oparg.
 // Returns 1 if the opcode is a _LOAD_FAST_BORROW variant, 0 otherwise.
-// Falls back to stencil pipeline for oparg values too large for manual codegen.
+// On AArch64, falls back to stencil for oparg > 4085 (imm12 limit).
+// https://developer.arm.com/documentation/ddi0602/2024-06/Base-Instructions/LDR--immediate---Load-register--immediate--?lang=en
 static int
 _decode_load_fast_borrow(uint16_t opcode, uint16_t insn_oparg,
                          int *reg_variant, int *oparg)
 {
     if (opcode >= _LOAD_FAST_BORROW_r01 && opcode <= _LOAD_FAST_BORROW_r23) {
-        if (insn_oparg > _LOAD_FAST_BORROW_MAX_OPARG) {
+#if defined(__aarch64__) || defined(_M_ARM64)
+        if (insn_oparg > 4085) {
             return 0;
         }
+#endif
         *reg_variant = opcode - _LOAD_FAST_BORROW_r01;
         *oparg = insn_oparg;
         return 1;

Tools/jit/_targets.py

@@ -217,8 +217,12 @@ async def _build_stencils(self) -> dict[str, _stencils.StencilGroup]:
                 for case, opname in cases_and_opnames:
                     # _LOAD_FAST_BORROW uses manual codegen in jit.c,
                     # so skip stencil generation for its register variants.
+                    # AArch64 keeps stencils as fallback for huge opargs
+                    # (imm12 limit, oparg > 4085).
+                    # https://developer.arm.com/documentation/ddi0602/2024-06/Base-Instructions/LDR--immediate---Load-register--immediate--?lang=en
                     if opname.startswith("_LOAD_FAST_BORROW_r"):
-                        continue
+                        if not self.triple.startswith("aarch64"):
+                            continue
                     # Write out a copy of the template with *only* this case
                     # inserted. This is about twice as fast as #include'ing all
                     # of executor_cases.c.h each time we compile (since the C