发现keil 4.12版本的一个bug

只看该作者 · 2012-4-14 16:49

本帖最后由 yixuanyuxiao 于 2012-4-14 16:57 编辑

我最近在stm32上移植mp3的软解码程序，用的是helix的源码，调试过程中发现一个问题，在helix的源码里有一个内联函数在做优化等级3的优化时会出现未定义指令，觉得是keil4.12版本优化的一个bug，因为我家里的电脑上是4.23，一直是优化等级3编译，没出现过问题，到公司的电脑上编译就出问题了，公司电脑是4.12版本，本人菜鸟，分析不出具体原因，希望牛人分析下。现贴出代码
这是内联函数的c语言代码：

static __inline void imdct12 (int *x, int *out)
{

int a0, a1, a2;

int x0, x1, x2, x3, x4, x5;

x0 = *x;
x+=3;
x1 = *x;
x+=3;

x2 = *x;
x+=3;
x3 = *x;
x+=3;

x4 = *x;
x+=3;
x5 = *x;
x+=3;

x4 -= x5;

x3 -= x4;

x2 -= x3;

x3 -= x5;

x1 -= x2;

x0 -= x1;

x1 -= x3;

x0 >>= 1;

x1 >>= 1;

a0 = MULSHIFT32(c3_0, x2) << 1;

a1 = x0 + (x4 >> 1);

a2 = x0 - x4;

x0 = a1 + a0;

x2 = a2;

x4 = a1 - a0;

a0 = MULSHIFT32(c3_0, x3) << 1;

a1 = x1 + (x5 >> 1);

a2 = x1 - x5;

/* cos window odd samples, mul by 2, eat sign bit */

x1 = MULSHIFT32(c6[0], a1 + a0) << 2;

x3 = MULSHIFT32(c6[1], a2) << 2;

x5 = MULSHIFT32(c6[2], a1 - a0) << 2;

*out = x0 + x1;
out++;

*out = x2 + x3;
out++;

*out = x4 + x5;
out++;

*out = x4 - x5;
out++;

*out = x2 - x3;
out++;

*out = x0 - x1;
}

只看该作者 · 2012-4-14 16:50

这里是优化等级为0时的汇编代码：
471: static __inline void imdct12 (int *x, int *out)
0x08004C30 B027    ADD    sp,sp,#0x9C
0x08004C32 E8BD8FF0  POP    {r4-r11,pc}
0x08004C36 0000    MOVS    r0,r0
0x08004C38 EBA16ED9  SUB    lr,r1,r9,LSR #27
0x08004C3C BE8B    BKPT    0x8B
0x08004C3E 620D    STR    r5,[r1,#0x20]
0x08004C40 1A7E    SUBS    r6,r7,r1
0x08004C42 163A    ASRS    r2,r7,#24
0x08004C44 DD49    BLE    0x08004CDA
0x08004C46 5246    STRH    r6,[r0,r1]
0x08004C48 2E32    CMP    r6,#0x32
0x08004C4A 7E0E    LDRB    r6,[r1,#0x18]
472: {
473:       int a0, a1, a2;
474:       int x0, x1, x2, x3, x4, x5;
475:
0x08004C4C E92D4FFE  PUSH    {r1-r11,lr}
0x08004C50 4604    MOV    r4,r0
0x08004C52 460D    MOV    r5,r1
476:       x0 = *x;       x+=3; x1 = *x; x+=3;
0x08004C54 F8D48000  LDR    r8,[r4,#0x00]
0x08004C58 340C    ADDS    r4,r4,#0x0C
0x08004C5A 6826    LDR    r6,[r4,#0x00]
0x08004C5C 340C    ADDS    r4,r4,#0x0C
477:       x2 = *x;       x+=3; x3 = *x; x+=3;
0x08004C5E F8D4B000  LDR    r11,[r4,#0x00]
0x08004C62 340C    ADDS    r4,r4,#0x0C
0x08004C64 6827    LDR    r7,[r4,#0x00]
0x08004C66 340C    ADDS    r4,r4,#0x0C
478:       x4 = *x;       x+=3; x5 = *x; x+=3;
479:
0x08004C68 F8D49000  LDR    r9,[r4,#0x00]
0x08004C6C 340C    ADDS    r4,r4,#0x0C
0x08004C6E F8D4A000  LDR    r10,[r4,#0x00]
0x08004C72 340C    ADDS    r4,r4,#0x0C
480:       x4 -= x5;
0x08004C74 EBA9090A  SUB    r9,r9,r10
481:       x3 -= x4;
0x08004C78 EBA70709  SUB    r7,r7,r9
482:       x2 -= x3;
0x08004C7C EBAB0B07  SUB    r11,r11,r7
483:       x3 -= x5;
0x08004C80 EBA7070A  SUB    r7,r7,r10
484:       x1 -= x2;
0x08004C84 EBA6060B  SUB    r6,r6,r11
485:       x0 -= x1;
0x08004C88 EBA80806  SUB    r8,r8,r6
486:       x1 -= x3;
487:
0x08004C8C 1BF6    SUBS    r6,r6,r7
488:       x0 >>= 1;
0x08004C8E EA4F0868  ASR    r8,r8,#1
489:       x1 >>= 1;
490:
0x08004C92 1076    ASRS    r6,r6,#1
491:       a0 = MULSHIFT32(c3_0, x2) << 1;
0x08004C94 4659    MOV    r1,r11
0x08004C96 4828    LDR    r0,[pc,#160]  ; @0x08004D38
0x08004C98 F7FBFA9A  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004C9C 0040    LSLS    r0,r0,#1
0x08004C9E 9002    STR    r0,[sp,#0x08]
492:       a1 = x0 + (x4 >> 1);
0x08004CA0 EB080069  ADD    r0,r8,r9,ASR #1
0x08004CA4 9001    STR    r0,[sp,#0x04]
493:       a2 = x0 - x4;
0x08004CA6 EBA80009  SUB    r0,r8,r9
0x08004CAA 9000    STR    r0,[sp,#0x00]
494:       x0 = a1 + a0;
0x08004CAC E9DD0101  LDRD    r0,r1,[sp,#0x04]
0x08004CB0 EB000801  ADD    r8,r0,r1
495:       x2 = a2;
0x08004CB4 F8DDB000  LDR    r11,[sp,#0x00]
496:       x4 = a1 - a0;
497:
0x08004CB8 E9DD0101  LDRD    r0,r1,[sp,#0x04]
0x08004CBC EBA00901  SUB    r9,r0,r1
498:       a0 = MULSHIFT32(c3_0, x3) << 1;
0x08004CC0 4639    MOV    r1,r7
0x08004CC2 481D    LDR    r0,[pc,#116]  ; @0x08004D38
0x08004CC4 F7FBFA84  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004CC8 0040    LSLS    r0,r0,#1
0x08004CCA 9002    STR    r0,[sp,#0x08]
499:       a1 = x1 + (x5 >> 1);
0x08004CCC EB06006A  ADD    r0,r6,r10,ASR #1
0x08004CD0 9001    STR    r0,[sp,#0x04]
500:       a2 = x1 - x5;
501:
502:       /* cos window odd samples, mul by 2, eat sign bit */
0x08004CD2 EBA6000A  SUB    r0,r6,r10
0x08004CD6 9000    STR    r0,[sp,#0x00]
503:       x1 = MULSHIFT32(c6[0], a1 + a0) << 2;
0x08004CD8 E9DD2301  LDRD    r2,r3,[sp,#0x04]
0x08004CDC 18D1    ADDS    r1,r2,r3
0x08004CDE 4A17    LDR    r2,[pc,#92]  ; @0x08004D3C
0x08004CE0 6810    LDR    r0,[r2,#0x00]
0x08004CE2 F7FBFA75  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004CE6 0086    LSLS    r6,r0,#2
504:       x3 = MULSHIFT32(c6[1], a2) << 2;
0x08004CE8 4914    LDR    r1,[pc,#80]  ; @0x08004D3C
0x08004CEA 6848    LDR    r0,[r1,#0x04]
0x08004CEC 9900    LDR    r1,[sp,#0x00]
0x08004CEE F7FBFA6F  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004CF2 0087    LSLS    r7,r0,#2
505:       x5 = MULSHIFT32(c6[2], a1 - a0) << 2;
506:
0x08004CF4 E9DD2301  LDRD    r2,r3,[sp,#0x04]
0x08004CF8 1AD1    SUBS    r1,r2,r3
0x08004CFA 4A10    LDR    r2,[pc,#64]  ; @0x08004D3C
0x08004CFC 6890    LDR    r0,[r2,#0x08]
0x08004CFE F7FBFA67  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004D02 EA4F0A80  LSL    r10,r0,#2
507:       *out = x0 + x1; out++;
0x08004D06 EB080006  ADD    r0,r8,r6
0x08004D0A 6028    STR    r0,[r5,#0x00]
0x08004D0C 1D2D    ADDS    r5,r5,#4
508:       *out = x2 + x3; out++;
0x08004D0E EB0B0007  ADD    r0,r11,r7
0x08004D12 6028    STR    r0,[r5,#0x00]
0x08004D14 1D2D    ADDS    r5,r5,#4
509:       *out = x4 + x5; out++;
0x08004D16 EB09000A  ADD    r0,r9,r10
0x08004D1A 6028    STR    r0,[r5,#0x00]
0x08004D1C 1D2D    ADDS    r5,r5,#4
510:       *out = x4 - x5; out++;
0x08004D1E EBA9000A  SUB    r0,r9,r10
0x08004D22 6028    STR    r0,[r5,#0x00]
0x08004D24 1D2D    ADDS    r5,r5,#4
511:       *out = x2 - x3; out++;
0x08004D26 EBAB0007  SUB    r0,r11,r7
0x08004D2A 6028    STR    r0,[r5,#0x00]
0x08004D2C 1D2D    ADDS    r5,r5,#4
512:       *out = x0 - x1;
0x08004D2E EBA80006  SUB    r0,r8,r6
0x08004D32 6028    STR    r0,[r5,#0x00]
513: }
0x08004D34 E8BD8FFE  POP    {r1-r11,pc}
0x08004D38 EBA16ED9  SUB    lr,r1,r9,LSR #27
0x08004D3C 9590    STR    r5,[sp,#0x240]
0x08004D3E 0800    LSRS    r0,r0,#0

只看该作者 · 2012-4-14 16:57

局部变量太多，寄存器用光不会压栈直接使用，于是，就乱掉了。
新版已修正。
MDK是带升级服务的，公司的也可以升。

只看该作者 · 2012-4-14 16:58

这里是优化等级为3时的汇编代码：
471: static __inline void imdct12 (int *x, int *out)
0x080049F4 B011    ADD    sp,sp,#0x44
0x080049F6 E8BD8FF0  POP    {r4-r11,pc}
0x080049FA 0000    MOVS    r0,r0
0x080049FC EBA16ED9  SUB    lr,r1,r9,LSR #27
0x08004A00 BE8B    BKPT    0x8B
0x08004A02 620D    STR    r5,[r1,#0x20]
0x08004A04 1A7E    SUBS    r6,r7,r1
0x08004A06 163A    ASRS    r2,r7,#24
0x08004A08 DD49    BLE    0x08004A9E
0x08004A0A 5246    STRH    r6,[r0,r1]
0x08004A0C 2E32    CMP    r6,#0x32
0x08004A0E 7E0E    LDRB    r6,[r1,#0x18]
472: {
473:       int a0, a1, a2;
474:       int x0, x1, x2, x3, x4, x5;
475:
476:       x0 = *x;       x+=3; x1 = *x; x+=3;
477:       x2 = *x;       x+=3; x3 = *x; x+=3;
478:       x4 = *x;       x+=3; x5 = *x; x+=3;
479:
0x08004A10 E92D5FF0  PUSH    {r4-r12,lr}
480:       x4 -= x5;
0x08004A14 6B05    LDR    r5,[r0,#0x30]
0x08004A16 6BC4    LDR    r4,[r0,#0x3C]
481:       x3 -= x4;
0x08004A18 6A46    LDR    r6,[r0,#0x24]
480:       x4 -= x5;
0x08004A1A 1B2D    SUBS    r5,r5,r4
481:       x3 -= x4;
0x08004A1C 68C3    LDR    r3,[r0,#0x0C]
0x08004A1E 6802    LDR    r2,[r0,#0x00]
0x08004A20 6981    LDR    r1,[r0,#0x18]
0x08004A22 1B70    SUBS    r0,r6,r5
482:       x2 -= x3;
0x08004A24 1A09    SUBS    r1,r1,r0
483:       x3 -= x5;
0x08004A26 1B06    SUBS    r6,r0,r4
484:       x1 -= x2;
0x08004A28 1A58    SUBS    r0,r3,r1
485:       x0 -= x1;
0x08004A2A 1A12    SUBS    r2,r2,r0
486:       x1 -= x3;
487:
488:       x0 >>= 1;
489:       x1 >>= 1;
490:
0x08004A2C 1B80    SUBS    r0,r0,r6
491:       a0 = MULSHIFT32(c3_0, x2) << 1;
0x08004A2E F8DF807C  LDR.W r8,[pc,#124]  ; @0x08004AAE
489:       x1 >>= 1;
490:
491:       a0 = MULSHIFT32(c3_0, x2) << 1;
0x08004A32 EA4F0B60  ASR    r11,r0,#1
488:       x0 >>= 1;
489:       x1 >>= 1;
490:
491:       a0 = MULSHIFT32(c3_0, x2) << 1;
0x08004A36 1057    ASRS    r7,r2,#1
0x08004A38 4640    MOV    r0,r8
0x08004A3A F7FBFBC9  BL.W    xmp3_MULSHIFT32 (0x080001D0)
492:       a1 = x0 + (x4 >> 1);
493:       a2 = x0 - x4;
0x08004A3E EB070165  ADD    r1,r7,r5,ASR #1
491:       a0 = MULSHIFT32(c3_0, x2) << 1;
492:       a1 = x0 + (x4 >> 1);
493:       a2 = x0 - x4;
0x08004A42 0040    LSLS    r0,r0,#1
494:       x0 = a1 + a0;
495:       x2 = a2;
0x08004A44 EB010A00  ADD    r10,r1,r0
496:       x4 = a1 - a0;
497:
0x08004A48 EBA10900  SUB    r9,r1,r0
493:       a2 = x0 - x4;
494:       x0 = a1 + a0;
495:       x2 = a2;
496:       x4 = a1 - a0;
497:
0x08004A4C 1B7D    SUBS    r5,r7,r5
498:       a0 = MULSHIFT32(c3_0, x3) << 1;
0x08004A4E 4631    MOV    r1,r6
0x08004A50 4640    MOV    r0,r8
0x08004A52 F7FBFBBD  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004A56 0046    LSLS    r6,r0,#1
499:       a1 = x1 + (x5 >> 1);
0x08004A58 EB0B0864  ADD    r8,r11,r4,ASR #1
500:       a2 = x1 - x5;
501:
502:       /* cos window odd samples, mul by 2, eat sign bit */
0x08004A5C EBAB0704  SUB    r7,r11,r4
503:       x1 = MULSHIFT32(c6[0], a1 + a0) << 2;
0x08004A60 EB080106  ADD    r1,r8,r6
0x08004A64 4812    LDR    r0,[pc,#72]  ; @0x08004AB0
0x08004A66 F7FBFBB3  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004A6A 0084    LSLS    r4,r0,#2
504:       x3 = MULSHIFT32(c6[1], a2) << 2;
0x08004A6C 4639    MOV    r1,r7
0x08004A6E 4811    LDR    r0,[pc,#68]  ; @0x08004AB4
0x08004A70 F7FBFBAE  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004A74 0087    LSLS    r7,r0,#2
505:       x5 = MULSHIFT32(c6[2], a1 - a0) << 2;
506:
0x08004A76 EBA80106  SUB    r1,r8,r6
0x08004A7A 480F    LDR    r0,[pc,#60]  ; @0x08004AB8
0x08004A7C F7FBFBA8  BL.W    xmp3_MULSHIFT32 (0x080001D0)
0x08004A80 0081    LSLS    r1,r0,#2
507:       *out = x0 + x1; out++;
0x08004A82 980B    LDR    r0,[sp,#0x2C]
0x08004A84 EB0A0204  ADD    r2,r10,r4
0x08004A88 6002    STR    r2,[r0,#0x00]
508:       *out = x2 + x3; out++;
0x08004A8A 980B    LDR    r0,[sp,#0x2C]
0x08004A8C 19EA    ADDS    r2,r5,r7
509:       *out = x4 + x5; out++;
0x08004A8E 6042    STR    r2,[r0,#0x04]
0x08004A90 EB090201  ADD    r2,r9,r1
510:       *out = x4 - x5; out++;
0x08004A94 EBA90101  SUB    r1,r9,r1
0x08004A98 E9C02102  STRD    r2,r1,[r0,#0x08]
511:       *out = x2 - x3; out++;
0x08004A9C 1BE9    SUBS    r1,r5,r7
512:       *out = x0 - x1;
0x08004A9E 6101    STR    r1,[r0,#0x10]
0x08004AA0 EBAA0104  SUB    r1,r10,r4
0x08004AA4 6141    STR    r1,[r0,#0x14]
513: }
0x08004AA6 E8BD9FF0  POP    {r4-r12,pc}
0x08004AAA 0000    MOVS    r0,r0
0x08004AAC EBA16ED9  SUB    lr,r1,r9,LSR #27
0x08004AB0 751D    STRB    r5,[r3,#0x14]
0x08004AB2 7BA3    LDRB    r3,[r4,#0x0E]
0x08004AB4 799A    LDRB    r2,[r3,#0x06]
0x08004AB6 5A82    LDRH    r2,[r0,r2]
0x08004AB8 FB832120  DCD    0xFB832120 ; ? Undefined 应该是这个地方

只看该作者 · 2012-4-14 16:59

c语言程序再发一遍，楼主位的看起来好别扭，调不了。
这是内联函数的c语言代码：

static __inline void imdct12 (int *x, int *out)
{
int a0, a1, a2;
int x0, x1, x2, x3, x4, x5;

x0 = *x; x+=3; x1 = *x; x+=3;
x2 = *x; x+=3; x3 = *x; x+=3;
x4 = *x; x+=3; x5 = *x; x+=3;

x4 -= x5;
x3 -= x4;
x2 -= x3;
x3 -= x5;
x1 -= x2;
x0 -= x1;
x1 -= x3;

x0 >>= 1;
x1 >>= 1;

a0 = MULSHIFT32(c3_0, x2) << 1;
a1 = x0 + (x4 >> 1);
a2 = x0 - x4;
x0 = a1 + a0;
x2 = a2;
x4 = a1 - a0;

a0 = MULSHIFT32(c3_0, x3) << 1;
a1 = x1 + (x5 >> 1);
a2 = x1 - x5;

/* cos window odd samples, mul by 2, eat sign bit */
x1 = MULSHIFT32(c6[0], a1 + a0) << 2;
x3 = MULSHIFT32(c6[1], a2) << 2;
x5 = MULSHIFT32(c6[2], a1 - a0) << 2;

*out = x0 + x1; out++;
*out = x2 + x3; out++;
*out = x4 + x5; out++;
*out = x4 - x5; out++;
*out = x2 - x3; out++;
*out = x0 - x1;
}