Question

9.7.6 Benchmul.exe

Benchmul 是另一个来自 Plum-Hall 基准程序套装的程序。这个程序测试在一个循环中执行 1000 次相乘的整数乘法。反汇编的程序在图 9-35 被显示，反编译的 C 程序在图 9-36，最初的 C 程序图 9-37。这个程序有下列调用图：

main scanf printf

Benchmul 使用两个长整型变量经过程序许多次循环，和三个执行该运算的整型变量；这些变量其中一个实际上在该程序中没有被使用。从反汇编的代码可知，长整型变量位于栈上偏移-4 和-8 处，而整型变量在偏移-12、-10 和寄存器变量 si 上。最后的 C 代码和最初的 C 代码是相同的，而且这个程序达到 86.36%的指令缩减率，如图 9-38 所示。

	main	PROC NEAR
000	0002FA	55	PUSH	bp
001	0002FB	8BEC	MOV	bp,sp
002	0002FD	83EC0C	SUB	sp,0Ch
003	000300	56	PUSH	si
004	000301	8D46FC	LEA	ax,[bp-4]
005	000304	50	PUSH	ax
006	000305	B89401	MOV	ax,194h
007	000308	50	PUSH	ax
008	000309	E8D014	CALL	near ptr scanf
009	00030C	59	POP	cx
010	00030D	59	POP	cx
011	00030E	FF76FE	PUSH	word ptr [bp-2]
012	000311	FF76FC	PUSH	word ptr [bp-4]
013	000314	B89801	MOV	ax,198h
014	000317	50	PUSH	ax
015	000318	E8380C	CALL	near ptr printf
016	00031B	83C406	ADD	sp,6
017	00031E	8D46F4	LEA	ax,[bp-0Ch]
018	000321	50	PUSH	ax
019	000322	B8B201	MOV	ax,1B2h
020	000325	50	PUSH	ax
021	000326	E8B314	CALL	near ptr scanf
022	000329	59	POP	cx
023	00032A	59	POP	cx
024	00032B	8D46F6	LEA	ax,[bp-0Ah]
025	00032E	50	PUSH	ax
026	00032F	B8B501	MOV	ax,1B5h
027	000332	50	PUSH	ax
028	000333	E8A614	CALL	near ptr scanf
029	000336	59	POP	cx
030	000337	59	POP	cx
031	000338	C746FA0000	MOV	word ptr [bp-6],0
032	00033D	C746F80100	MOV	word ptr [bp-8],1
034	0003AA	8B56FA	L1: MOV	dx,[bp-6]
035	0003AD	8B46F8	MOV	ax,[bp-8]
036	0003B0	3B56FE	CMP	dx,[bp-2]
037	0003B3	7C8F	JL	L2
038	0003B5	7F05	JG	L3
039	0003B7	3B46FC	CMP	ax,[bp-4]
040	0003BA	7688	JBE	L2
041	0003BC	FF76F4	L3: PUSH	word ptr [bp-0Ch]
042	0003BF	B8B801	MOV	ax,1B8h
043	0003C2	50	PUSH	ax
044	0003C3	E88D0B	CALL	near ptr printf
045	0003C6	59	POP	cx
046	0003C7	59	POP	cx
047	0003C8	5E	POP	si
048	0003C9	8BE5	MOV	sp,bp
049	0003CB	5D	POP	bp
050	0003CC	C3	RET
051	000344	BE0100	L2: MOV	si,1
053	00039D	83FE28	L4: CMP	si,28h
054	0003A0	7EA7	JLE	L5
055	0003A2	8346F801	ADD	word ptr [bp-8],1
056	0003A6	8356FA00	ADC	word ptr [bp-6],0
057			JMP	L1	;Synthetic inst
058	000349	8B46F4	L5: MOV	ax,[bp-0Ch]
059	00034C	F766F4	MUL	word ptr [bp-0Ch]
060	00034F	F766F4	MUL	word ptr [bp-0Ch]
061	000352	F766F4	MUL	word ptr [bp-0Ch]
062	000355	F766F4	MUL	word ptr [bp-0Ch]
063	000358	F766F4	MUL	word ptr [bp-0Ch]
064	00035B	F766F4	MUL	word ptr [bp-0Ch]
065	00035E	F766F4	MUL	word ptr [bp-0Ch]
066	000361	F766F4	MUL	word ptr [bp-0Ch]
067	000364	F766F4	MUL	word ptr [bp-0Ch]
068	000367	F766F4	MUL	word ptr [bp-0Ch]
069	00036A	F766F4	MUL	word ptr [bp-0Ch]
070	00036D	F766F4	MUL	word ptr [bp-0Ch]
071	000370	F766F4	MUL	word ptr [bp-0Ch]
072	000373	F766F4	MUL	word ptr [bp-0Ch]
073	000376	F766F4	MUL	word ptr [bp-0Ch]
074	000379	F766F4	MUL	word ptr [bp-0Ch]
075	00037C	F766F4	MUL	word ptr [bp-0Ch]
076	00037F	F766F4	MUL	word ptr [bp-0Ch]
077	000382	F766F4	MUL	word ptr [bp-0Ch]
078	000385	F766F4	MUL	word ptr [bp-0Ch]
079	000388	F766F4	MUL	word ptr [bp-0Ch]
080	00038B	F766F4	MUL	word ptr [bp-0Ch]
081	00038E	F766F4	MUL	word ptr [bp-0Ch]
082	000391	F766F4	MUL	word ptr [bp-0Ch]
083	000394	BA0300	MOV	dx,3
084	000397	F7E2	MUL	dx
085	000399	8946F4	MOV	[bp-0Ch],ax
086	00039C	46	INC	si
087			JMP	L4	;Synthetic inst
	main	ENDP

图 9-35: Benchmul.a2

/* * Input file : benchmul.exe * File type : EXE */ #include "dcc.h" void main () /* Takes no parameters. * High-level language prologue code. */ { int loc1; int loc2; long loc3; long loc4; int loc5; scanf ("%ld", &loc4); printf ("executing %ld iterations\n", loc4); scanf ("%d", &loc1); scanf ("%d", &loc2); loc3 = 1; while ((loc3 <= loc4)) { loc5 = 1; while ((loc5 <= 40)) { loc1 = (((((((((((((((((((((((((loc1 * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * loc1) * 3); loc5 = (loc5 + 1); } loc3 = (loc3 + 1); } printf ("a=%d\n", loc1); }

图 9-36: Benchmul.b

/* benchmul - benchmark for int multiply * Thomas Plum, Plum Hall Inc, 609-927-3770 * If machine traps overflow, use an unsigned type * Let T be the execution time in milliseconds * Then average time per operator = T/major usec * (Because the inner loop has exactly 1000 operations) */ #define STOR_CL auto #define TYPE int #include <stdio.h> main (int ac, char *av[]) { STOR_CL TYPE a, b, c; long d, major; scanf ("%ld", &major); printf("executing %ld iterations\n", major); scanf ("%d", &a); scanf ("%d", &b); for (d = 1; d <= major; ++d) { /* inner loop executes 1000 selected operations */ for (c = 1; c <= 40; ++c) { a = 3 *a*a*a*a*a*a*a*a * a*a*a*a*a*a*a*a * a*a*a*a*a*a*a*a * a; /* 25 * */ } } printf("a=%d\n", a); }

图 9-37: Benchmul.c

子程序	低级	高级	% 缩减率
main	88	12	86.36
合计	88	12	86.36

图 9-38: Benchmul 统计