My entry for https://hackaday.com/2016/11/21/step-up-to-the-1-kb-challenge/ includes a couple of huge functions which are not generated from any C code which I have written.
000004e4 <.divsi3_skip_div0_test>:
4e4: b410 push {r4}
4e6: 1c04 adds r4, r0, #0
4e8: 404c eors r4, r1
4ea: 46a4 mov ip, r4
4ec: 2301 movs r3, #1
4ee: 2200 movs r2, #0
4f0: 2900 cmp r1, #0
4f2: d500 bpl.n 4f6 <.divsi3_skip_div0_test+0x12>
4f4: 4249 negs r1, r1
4f6: 2800 cmp r0, #0
4f8: d500 bpl.n 4fc <.divsi3_skip_div0_test+0x18>
4fa: 4240 negs r0, r0
4fc: 4288 cmp r0, r1
4fe: d32c bcc.n 55a <.divsi3_skip_div0_test+0x76>
500: 2401 movs r4, #1
502: 0724 lsls r4, r4, #28
504: 42a1 cmp r1, r4
506: d204 bcs.n 512 <.divsi3_skip_div0_test+0x2e>
508: 4281 cmp r1, r0
50a: d202 bcs.n 512 <.divsi3_skip_div0_test+0x2e>
50c: 0109 lsls r1, r1, #4
50e: 011b lsls r3, r3, #4
510: e7f8 b.n 504 <.divsi3_skip_div0_test+0x20>
512: 00e4 lsls r4, r4, #3
514: 42a1 cmp r1, r4
516: d204 bcs.n 522 <.divsi3_skip_div0_test+0x3e>
518: 4281 cmp r1, r0
51a: d202 bcs.n 522 <.divsi3_skip_div0_test+0x3e>
51c: 0049 lsls r1, r1, #1
51e: 005b lsls r3, r3, #1
520: e7f8 b.n 514 <.divsi3_skip_div0_test+0x30>
522: 4288 cmp r0, r1
524: d301 bcc.n 52a <.divsi3_skip_div0_test+0x46>
526: 1a40 subs r0, r0, r1
528: 431a orrs r2, r3
52a: 084c lsrs r4, r1, #1
52c: 42a0 cmp r0, r4
52e: d302 bcc.n 536 <.divsi3_skip_div0_test+0x52>
530: 1b00 subs r0, r0, r4
532: 085c lsrs r4, r3, #1
534: 4322 orrs r2, r4
536: 088c lsrs r4, r1, #2
538: 42a0 cmp r0, r4
53a: d302 bcc.n 542 <.divsi3_skip_div0_test+0x5e>
53c: 1b00 subs r0, r0, r4
53e: 089c lsrs r4, r3, #2
540: 4322 orrs r2, r4
542: 08cc lsrs r4, r1, #3
544: 42a0 cmp r0, r4
546: d302 bcc.n 54e <.divsi3_skip_div0_test+0x6a>
548: 1b00 subs r0, r0, r4
54a: 08dc lsrs r4, r3, #3
54c: 4322 orrs r2, r4
54e: 2800 cmp r0, #0
550: d003 beq.n 55a <.divsi3_skip_div0_test+0x76>
552: 091b lsrs r3, r3, #4
554: d001 beq.n 55a <.divsi3_skip_div0_test+0x76>
556: 0909 lsrs r1, r1, #4
558: e7e3 b.n 522 <.divsi3_skip_div0_test+0x3e>
55a: 1c10 adds r0, r2, #0
55c: 4664 mov r4, ip
55e: 2c00 cmp r4, #0
560: d500 bpl.n 564 <.divsi3_skip_div0_test+0x80>
562: 4240 negs r0, r0
564: bc10 pop {r4}
566: 4770 bx lr
568: 2800 cmp r0, #0
56a: d006 beq.n 57a <.divsi3_skip_div0_test+0x96>
56c: db03 blt.n 576 <.divsi3_skip_div0_test+0x92>
56e: 2000 movs r0, #0
570: 43c0 mvns r0, r0
572: 0840 lsrs r0, r0, #1
574: e001 b.n 57a <.divsi3_skip_div0_test+0x96>
576: 2080 movs r0, #128 ; 0x80
578: 0600 lsls r0, r0, #24
57a: b407 push {r0, r1, r2}
57c: 4802 ldr r0, [pc, #8] ; (588 <.divsi3_skip_div0_test+0xa4>)
57e: a102 add r1, pc, #8 ; (adr r1, 588 <.divsi3_skip_div0_test+0xa4>)
580: 1840 adds r0, r0, r1
582: 9002 str r0, [sp, #8]
584: bd03 pop {r0, r1, pc}
586: 46c0 nop ; (mov r8, r8)
588: 00000019 .word 0x00000019
and:
000005a4 <memcpy>:
5a4: b5f0 push {r4, r5, r6, r7, lr}
5a6: 2a0f cmp r2, #15
5a8: d935 bls.n 616 <memcpy+0x72>
5aa: 1c03 adds r3, r0, #0
5ac: 430b orrs r3, r1
5ae: 079c lsls r4, r3, #30
5b0: d135 bne.n 61e <memcpy+0x7a>
5b2: 1c16 adds r6, r2, #0
5b4: 3e10 subs r6, #16
5b6: 0936 lsrs r6, r6, #4
5b8: 0135 lsls r5, r6, #4
5ba: 1945 adds r5, r0, r5
5bc: 3510 adds r5, #16
5be: 1c0c adds r4, r1, #0
5c0: 1c03 adds r3, r0, #0
5c2: 6827 ldr r7, [r4, #0]
5c4: 601f str r7, [r3, #0]
5c6: 6867 ldr r7, [r4, #4]
5c8: 605f str r7, [r3, #4]
5ca: 68a7 ldr r7, [r4, #8]
5cc: 609f str r7, [r3, #8]
5ce: 68e7 ldr r7, [r4, #12]
5d0: 3410 adds r4, #16
5d2: 60df str r7, [r3, #12]
5d4: 3310 adds r3, #16
5d6: 42ab cmp r3, r5
5d8: d1f3 bne.n 5c2 <memcpy+0x1e>
5da: 1c73 adds r3, r6, #1
5dc: 011b lsls r3, r3, #4
5de: 18c5 adds r5, r0, r3
5e0: 18c9 adds r1, r1, r3
5e2: 230f movs r3, #15
5e4: 4013 ands r3, r2
5e6: 2b03 cmp r3, #3
5e8: d91b bls.n 622 <memcpy+0x7e>
5ea: 1f1c subs r4, r3, #4
5ec: 08a4 lsrs r4, r4, #2
5ee: 3401 adds r4, #1
5f0: 00a4 lsls r4, r4, #2
5f2: 2300 movs r3, #0
5f4: 58ce ldr r6, [r1, r3]
5f6: 50ee str r6, [r5, r3]
5f8: 3304 adds r3, #4
5fa: 42a3 cmp r3, r4
5fc: d1fa bne.n 5f4 <memcpy+0x50>
5fe: 18ed adds r5, r5, r3
600: 18c9 adds r1, r1, r3
602: 2303 movs r3, #3
604: 401a ands r2, r3
606: d005 beq.n 614 <memcpy+0x70>
608: 2300 movs r3, #0
60a: 5ccc ldrb r4, [r1, r3]
60c: 54ec strb r4, [r5, r3]
60e: 3301 adds r3, #1
610: 4293 cmp r3, r2
612: d1fa bne.n 60a <memcpy+0x66>
614: bdf0 pop {r4, r5, r6, r7, pc}
616: 1c05 adds r5, r0, #0
618: 2a00 cmp r2, #0
61a: d1f5 bne.n 608 <memcpy+0x64>
61c: e7fa b.n 614 <memcpy+0x70>
61e: 1c05 adds r5, r0, #0
620: e7f2 b.n 608 <memcpy+0x64>
622: 1c1a adds r2, r3, #0
624: e7f8 b.n 618 <memcpy+0x74>
626: 46c0 nop ; (mov r8, r8)
I am guessing that I could code far smaller, but less time-efficient, replacements myself.
Is this likely?
Where will I find the source which I need to edit - I'm guessing that I should look for the source of a libc under gcc-arm-none-eabi/lib/gcc/arm-none-eabi/4.8.3/. I think that I've found the compiled symbols, but I can't find the source.
~/gcc-arm-none-eabi$ grep -R divsi3_skip_div0_test *
Binary file lib/gcc/arm-none-eabi/4.8.3/libgcc.a matches
Binary file lib/gcc/arm-none-eabi/4.8.3/thumb/libgcc.a matches
Binary file lib/gcc/arm-none-eabi/4.8.3/armv6-m/libgcc.a matches
Binary file lib/gcc/arm-none-eabi/4.8.3/fpu/libgcc.a matches
Binary file lib/gcc/arm-none-eabi/4.8.3/armv7-ar/thumb/libgcc.a matches
Binary file lib/gcc/arm-none-eabi/4.8.3/armv7-ar/thumb/softfp/libgcc.a matches
Binary file lib/gcc/arm-none-eabi/4.8.3/armv7-ar/thumb/fpu/libgcc.a matches
Alternatively, is there a way to tell gcc not to use memcpy when copying structures? (They are 10 bytes, so three thumb instructions should do the job.) I've tried adding -mno-memcpy and -Wa,mno-memcpy but neither are recognised.
Update:
I've solved the memcpy part of this question - adding a partial, but sufficient, memcpy function stops the other from being added.
size_t memcpy(uint8_t *restrict dst, uint8_t *restrict const src, size_t size) {
int i;
for (i = 0; i < size; i++) {
dst[i] = src[i];
}
return i;
}
It's much smaller, but less efficient and won't handle dst < src + size overlap.
000003ec <memcpy>:
3ec: b510 push {r4, lr}
3ee: 2300 movs r3, #0
3f0: 4293 cmp r3, r2
3f2: d003 beq.n 3fc <memcpy+0x10>
3f4: 5ccc ldrb r4, [r1, r3]
3f6: 54c4 strb r4, [r0, r3]
3f8: 3301 adds r3, #1
3fa: e7f9 b.n 3f0 <memcpy+0x4>
3fc: 1c18 adds r0, r3, #0
3fe: bd10 pop {r4, pc}
To clarify, I'm now only asking what I might do to replace the .divsi3_skip_div0_test code with a less efficient, but smaller, code.
It is not clear to me where the source of this code is, or how to edit its source. It looks to be more complicated to replace than memcpy as it does not look like a C function, as its name begins with a ..