I'm writing a mixed C and assembly programs. I will show you the detail of program below.
I use the C Standard Library function printf("%a",value). But I got fairly incorrect output.
This is part of the assembly code. Written in file mys.s:
...
FLDS S1, =1078530011
FLDS S2, =1066192077
FMULS S3, S1, S2
FMRS R4, S3
FMSR S1, R4
FLDS S2, =1066192077
FMULS S3, S1, S2
FMRS R5, S3
FMSR S0, R5
BL putfloat
MOV R13, R7
...
It implements something like:
putfloat(1.1*1.1*3.1415927410125732)
the implementation of 'putfloat()' show as followed:
void putfloat(float a) {
printf("%a", a);
}
Written in file myc.c
the ouput should be:
0x1.e691e6p+1
But I actually got:
0x1.00000420a3ae7p+513
I use gcc tooltrain to compile,assemble and link my project.
gcc version is 10.2.1 20210110 (Raspbian 10.2.1-6+rpi1)
The command to build my project is like:
gcc -march=armv7-a -mfloat-abi=hard myc.c mys.s -g -o run
I suppose it is enough to focus only on the function call part of my program, unless I ignore some other elements.
I use gdb to debug my program, setting a breakpoint on 'BL putfloat' .
Since I tell the procedure call standard to be 'hard', the parameter should be passed through s0, a register in floating point co-processor of ARMv7-a architecture.
(gdb) c
Continuing.
Breakpoint 5, mainentry () at ./output/95_float_new.s:292
292 BL putfloat
(gdb) print $s0
$2 = 3.80132747
The value in s0 is fairly correct.
If putfloat() print the value formatly, the output show on terminal should be
0x1.e691e6p+1
But I got the definately wrong result.
(gdb) n 0x1.00000420a3ae7p+513 293 MOV R13, R7
I would appreciate it if someone point out where I went wrong.
What else affects the result of a function in C library besides the argument list?
Or I'm not following arm procedure call standard?
This is the disassembly of putfloat() and some information about the registers.
Breakpoint 2, mainentry () at ./output/95_float_new.s:292
292 BL putfloat
(gdb) s
putfloat (a=3.80132747) at ./sysy/sylib.c:37
37 printf("%a\n", a);
(gdb) disassemble
Dump of assembler code for function putfloat:
0x00010aa4 <+0>: push {r7, lr}
0x00010aa6 <+2>: sub sp, #8
0x00010aa8 <+4>: add r7, sp, #0
0x00010aaa <+6>: vstr s0, [r7, #4]
0x00010aae <+10>: vldr s15, [r7, #4]
0x00010ab2 <+14>: vcvt.f64.f32 d7, s15
0x00010ab6 <+18>: vmov r2, r3, d7
0x00010aba <+22>: movw r0, #4576 ; 0x11e0
0x00010abe <+26>: movt r0, #1
=> 0x00010ac2 <+30>: blx 0x1040c <printf@plt>
0x00010ac6 <+34>: nop
0x00010ac8 <+36>: adds r7, #8
0x00010aca <+38>: mov sp, r7
0x00010acc <+40>: pop {r7, pc}
End of assembler dump.
(gdb) print (char*)0x111e0
$1 = 0x111e0 "%a\n"
(gdb) info r
r0 0x111e0 70112
r1 0x420a3ae7 1107966695
r2 0x60000000 1610612736
r3 0x400e691e 1074686238
r4 0x405d2b0b 1079847691
r5 0x407348f3 1081297139
r6 0x10478 66680
r7 0xbefffb0c 3204446988
r8 0x0 0
r9 0x0 0
r10 0xb6fff000 3070226432
r11 0x0 0
r12 0xbefffb48 3204447048
sp 0xbefffb0c 0xbefffb0c
lr 0x108cc 67788
pc 0x10ac2 0x10ac2 <putfloat+30>
cpsr 0x60000030 1610612784
fpscr 0x80000010 -2147483632
This is the complete code of mys.s
.section .data
global1:
.zero 4
global2:
.zero 4
global3:
.zero 4
global4:
.zero 4
global5:
.zero 4
global6:
.zero 4
global7:
.zero 4
global8:
.zero 4
global9:
.zero 4
global10:
.zero 4
global11:
.zero 4
global12:
.zero 4
global13:
.zero 4
global14:
.zero 4
global15:
.zero 4
.section .bss
.section .text
.global main
.extern putint
.extern getint
.extern getch
.extern putch
.arch armv7-a
.arm
error:
PUSH {R7, R14}
MOV R7, R13
errorentry:
LDR R0, =101
BL putch
MOV R13, R7
LDR R0, =114
BL putch
MOV R13, R7
LDR R0, =114
BL putch
MOV R13, R7
LDR R0, =111
BL putch
MOV R13, R7
LDR R0, =114
BL putch
MOV R13, R7
LDR R0, =10
BL putch
MOV R13, R7
B errorendlabel
errorendlabel:
POP {R7, R15}
POP {R7, R15}
.pool
ok:
PUSH {R7, R14}
MOV R7, R13
okentry:
LDR R0, =111
BL putch
MOV R13, R7
LDR R0, =107
BL putch
MOV R13, R7
LDR R0, =10
BL putch
MOV R13, R7
B okendlabel
okendlabel:
POP {R7, R15}
POP {R7, R15}
.pool
assert:
PUSH {R7, R14}
PUSH {R4, R5}
MOV R7, R13
MOV R4, R0
assertentry:
LDR R0, =0
CMP R4, R0
LDR R5, =0
MOVEQ R5, #1
LDR R0, =0
CMP R5, R0
BEQ assertlabel1
B assertlabel2
assertlabel2:
BL error
MOV R13, R7
B assertlabel3
assertlabel3:
B assertendlabel
assertendlabel:
POP {R4, R5}
POP {R7, R15}
assertlabel1:
B assertlabel3
POP {R4, R5}
POP {R7, R15}
.pool
float_abs:
PUSH {R7, R14}
PUSH {R4, R5}
SUB R13, R13, #4
MOV R7, R13
MOV R4, R0
float_absentry:
FMSR S1, R4
FLDS S2, =0
FCMPS S1, S2
FMSTAT
LDR R5, =0
MOVLT R5, #1
LDR R0, =0
CMP R5, R0
BEQ float_abslabel4
B float_abslabel5
float_abslabel5:
FLDS S1, =0
FMSR S2, R4
FSUBS S3, S1, S2
FMRS R5, S3
MOV R0, R5
ADD R13, R13, #4
POP {R4, R5}
POP {R7, R15}
float_absendlabel:
ADD R13, R13, #4
POP {R4, R5}
POP {R7, R15}
float_abslabel4:
MOV R0, R4
ADD R13, R13, #4
POP {R4, R5}
POP {R7, R15}
ADD R13, R13, #4
POP {R4, R5}
POP {R7, R15}
.pool
float_eq:
PUSH {R7, R14}
PUSH {R4, R5, R6}
SUB R13, R13, #12
MOV R7, R13
MOV R4, R0
MOV R5, R1
float_eqentry:
FMSR S1, R4
FMSR S2, R5
FSUBS S3, S1, S2
FMRS R6, S3
MOV R0, R6
BL float_abs
MOV R13, R7
MOV R4, R0
FMSR S1, R4
FLDS S2, =897988541
FCMPS S1, S2
FMSTAT
LDR R5, =0
MOVLT R5, #1
LDR R0, =0
CMP R5, R0
BEQ float_eqlabel7
B float_eqlabel8
float_eqlabel8:
FLDS S1, =1065353216
vcvt.s32.f32 S2, S1
FMRS R0, S2
ADD R13, R13, #12
POP {R4, R5, R6}
POP {R7, R15}
float_eqendlabel:
ADD R13, R13, #12
POP {R4, R5, R6}
POP {R7, R15}
float_eqlabel7:
LDR R0, =0
ADD R13, R13, #12
POP {R4, R5, R6}
POP {R7, R15}
ADD R13, R13, #12
POP {R4, R5, R6}
POP {R7, R15}
.pool
main:
PUSH {R7, R14}
PUSH {R4, R5}
SUB R13, R13, #12
MOV R7, R13
mainentry:
LDR R1, =1107966695
LDR R0, =1107966695
BL float_eq
MOV R13, R7
MOV R4, R0
MOV R0, R4
BL assert
MOV R13, R7
FLDS S1, =1078530011
FLDS S2, =1066192077
FMULS S3, S1, S2
FMRS R4, S3
FMSR S1, R4
FLDS S2, =1066192077
FMULS S3, S1, S2
FMRS R5, S3
FMSR S0, R5
BL putfloat
MOV R13, R7
LDR R0, =32
BL putch
MOV R13, R7
LDR R0, =10
BL putch
MOV R13, R7
LDR R0, =0
ADD R13, R13, #12
POP {R4, R5}
POP {R7, R15}
mainendlabel:
ADD R13, R13, #12
POP {R4, R5}
POP {R7, R15}
ADD R13, R13, #12
POP {R4, R5}
POP {R7, R15}
.pool
You need to align the stack pointer to 8 bytes at any "public interface", which includes any call to a C function. Your SUB R13, R13, #12 messes this up because 12 is not a multiple of 8.
It's common for library functions to misbehave when called with a misaligned stack, though in other cases the symptom is usually just a crash, e.g. glibc scanf Segmentation faults when called from a function that doesn't align RSP on x86-64. It's interesting that here, the result is not a crash but instead incorrect output. One could probably find out why by disassembling the code of printf. If I do so I'll post back.