shuiyu1123
0669d102ce
- Scene Demo: cube + sphere + ground plane with proper lighting - Audio: XAudio2 dynamic loading + WAV playback - Physics: expanded collision system - Renderer: improved pipeline with materials - UI: extended widget system - VisualScript: enhanced serializer + block interactions - ASM: optimized math/memory/profiling routines - Window: mouse input + scroll support
149 lines
4.1 KiB
NASM
149 lines
4.1 KiB
NASM
; ============================================================================
|
|
; profiling.asm - 性能分析工具 (x86-64 MASM)
|
|
; 使用 RDTSC/RDTSCP 进行高精度计时
|
|
; ============================================================================
|
|
|
|
include asm_macros.inc
|
|
|
|
.DATA
|
|
ALIGN 8
|
|
cpu_freq DQ 0 ; CPU 频率 (Hz)
|
|
cpu_freq_init DB 0 ; 是否已初始化
|
|
|
|
.CODE
|
|
|
|
; ============================================================================
|
|
; RDTSC 计时
|
|
; ============================================================================
|
|
|
|
; uint64_t prof_rdtsc(void) -> RAX
|
|
; 返回: TSC (Time Stamp Counter) 值
|
|
prof_rdtsc PROC
|
|
; 使用 RDTSCP 获取有序的 TSC
|
|
rdtscp
|
|
; EDX:EAX = TSC
|
|
shl rdx, 32
|
|
or rax, rdx
|
|
ret
|
|
prof_rdtsc ENDP
|
|
|
|
; uint64_t prof_get_freq(void) -> RAX
|
|
; 获取 CPU 频率 (Hz), 使用 QPC 校准 TSC
|
|
prof_get_freq PROC
|
|
; 检查是否已初始化
|
|
cmp byte ptr [cpu_freq_init], 0
|
|
jnz prof_freq_return
|
|
|
|
push rbx
|
|
push rdi
|
|
push rsi
|
|
sub rsp, 40
|
|
|
|
; 获取 QPC 频率
|
|
lea rcx, [rsp + 32]
|
|
call QueryPerformanceFrequency
|
|
|
|
; 获取起始 QPC 和 TSC
|
|
lea rcx, [rsp + 24]
|
|
call QueryPerformanceCounter
|
|
rdtscp
|
|
shl rdx, 32
|
|
or rax, rdx
|
|
mov rbx, rax ; rbx = 起始 TSC
|
|
|
|
; 等待约 10ms
|
|
mov rcx, 10 ; 10 毫秒
|
|
call Sleep
|
|
|
|
; 获取结束 QPC 和 TSC
|
|
lea rcx, [rsp + 16]
|
|
call QueryPerformanceCounter
|
|
rdtscp
|
|
shl rdx, 32
|
|
or rax, rdx
|
|
mov rdi, rax ; rdi = 结束 TSC
|
|
|
|
; qpc_diff = end_qpc - start_qpc
|
|
mov rax, [rsp + 16]
|
|
sub rax, [rsp + 24]
|
|
mov rsi, rax ; rsi = qpc_diff
|
|
|
|
; tsc_diff = end_tsc - start_tsc
|
|
mov rcx, rdi
|
|
sub rcx, rbx ; rcx = tsc_diff
|
|
|
|
; cpu_freq = tsc_diff * qpc_freq / qpc_diff
|
|
mov rax, rcx ; rax = tsc_diff
|
|
mov rdx, [rsp + 32] ; rdx = qpc_freq
|
|
mul rdx ; rdx:rax = tsc_diff * qpc_freq
|
|
div rsi ; rax = freq (Hz)
|
|
|
|
mov [cpu_freq], rax
|
|
mov byte ptr [cpu_freq_init], 1
|
|
|
|
add rsp, 40
|
|
pop rsi
|
|
pop rdi
|
|
pop rbx
|
|
|
|
prof_freq_return:
|
|
mov rax, [cpu_freq]
|
|
ret
|
|
prof_get_freq ENDP
|
|
|
|
; double prof_cycles_to_ms(uint64_t cycles) -> XMM0
|
|
; 将 TSC cycles 转换为毫秒: ms = cycles * 1000 / freq
|
|
prof_cycles_to_ms PROC
|
|
mov rdx, rcx ; 保存 cycles
|
|
call prof_get_freq ; rax = CPU freq (Hz)
|
|
|
|
cvtsi2sd xmm0, rdx ; xmm0 = cycles (double)
|
|
mov rcx, 1000
|
|
cvtsi2sd xmm1, rcx ; xmm1 = 1000.0
|
|
mulsd xmm0, xmm1 ; cycles * 1000
|
|
cvtsi2sd xmm1, rax ; xmm1 = freq
|
|
divsd xmm0, xmm1 ; ms
|
|
ret
|
|
prof_cycles_to_ms ENDP
|
|
|
|
; ============================================================================
|
|
; 内联性能宏 (在 C 代码中使用)
|
|
; ============================================================================
|
|
|
|
; 获取当前时间戳 (直接 RDTSC, 无函数调用开销)
|
|
; 在 C 中使用: __rdtsc() 内联函数
|
|
|
|
; 高精度计时器结构
|
|
; struct ProfTimer {
|
|
; uint64_t start;
|
|
; uint64_t end;
|
|
; double elapsed_ms;
|
|
; };
|
|
|
|
; void prof_timer_start(ProfTimer* timer)
|
|
prof_timer_start PROC
|
|
rdtscp
|
|
shl rdx, 32
|
|
or rax, rdx
|
|
mov [rcx], rax ; timer->start = tsc
|
|
ret
|
|
prof_timer_start ENDP
|
|
|
|
; void prof_timer_stop(ProfTimer* timer)
|
|
prof_timer_stop PROC
|
|
rdtscp
|
|
shl rdx, 32
|
|
or rax, rdx
|
|
mov [rcx + 8], rax ; timer->end = tsc
|
|
|
|
; 计算 elapsed_ms
|
|
mov rax, [rcx + 8]
|
|
sub rax, [rcx] ; cycles
|
|
; 转换为 ms
|
|
call prof_cycles_to_ms
|
|
movsd qword ptr [rcx + 16], xmm0 ; timer->elapsed_ms
|
|
ret
|
|
prof_timer_stop ENDP
|
|
|
|
END
|