mirror of
https://github.com/rkujawa/rk65c02.git
synced 2026-03-11 03:16:11 +00:00
813bc612f5
- test_emulation_bench_loop.s: tight 256x256 loop (ALU, load/store, branches), STP - bench_emulation: standalone program times N runs each (interp/JIT), reports speedup - test/Makefile: bench_emulation target, no ATF - test_emulation: emul_bench_loop / emul_bench_loop_jit for correctness Made-with: Cursor
51 lines
773 B
ArmAsm
51 lines
773 B
ArmAsm
; Compute-intensive loop for interpreter vs JIT benchmark.
|
|
; Runs a large number of JIT-native instructions (ALU, load/store, branches)
|
|
; then stops with STP. Same result regardless of interp/JIT.
|
|
;
|
|
.org 0xC000
|
|
|
|
.set COUNTER_LO, 0x10
|
|
.set COUNTER_HI, 0x11
|
|
.set TMP, 0x12
|
|
.set OUTER, 0x13
|
|
|
|
start:
|
|
cld
|
|
lda #0
|
|
sta COUNTER_LO
|
|
sta COUNTER_HI
|
|
sta TMP
|
|
lda #0xFF
|
|
sta OUTER
|
|
|
|
outer_loop:
|
|
ldx #0xFF
|
|
inner_loop:
|
|
; Mix of JIT-native ops (~24 instructions per inner iteration)
|
|
clc
|
|
lda TMP
|
|
adc #1
|
|
and #0xFF
|
|
sta TMP
|
|
lda COUNTER_LO
|
|
adc #1
|
|
sta COUNTER_LO
|
|
bcc no_carry
|
|
inc COUNTER_HI
|
|
no_carry:
|
|
lda COUNTER_LO
|
|
ora COUNTER_HI
|
|
eor TMP
|
|
and #0x7F
|
|
sta TMP
|
|
lda COUNTER_LO
|
|
adc COUNTER_HI
|
|
and #0xFF
|
|
sta TMP
|
|
dex
|
|
bne inner_loop
|
|
dec OUTER
|
|
bpl outer_loop
|
|
|
|
stp
|