Reputation: 41
Refer to: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0553a/CHDHGFEF.html
Writes the most significant signed 32 bits of the result in Rd.
SMMUL is meant to multiply to vairable together but I do not understand how it calculates the "most significant signed 32 bits of the result"
Thanks in advance
Upvotes: 1
Views: 2262
Reputation: 28087
When you multiply 2 32-bit value you get a 64-bit value and you have to save the result in two registers since a register is 32-bit. However may be you are not interested in lowest 32-bit and only highest 32-bit.
SMMUL
instruction provides you that. First calculates 64-bit result then it may simply discard/truncate lower 32-bits or it can round them into higher 32-bits (SMMULR
).
SMULL Rdlo, Rdhi, src1, src2
produces both halves in two output registers. (docs). Most Cortex-M microcontrollers either have both smmul
and smull
, or neither depending on whether they have a 64-bit multiplier. Cortex-M3 has smull
(Thumb-2 group) but not smmul
(DSP group).
If you only need the high half, SMMUL
avoids overwriting a second output register, and has the rounding version.
$ cat smmul.c
int smull(int a, int b, int c, int d) {
asm volatile("smmul r0, r2, r3");
// only works in debug builds. Use Extended Asm for real uses
}
int main() {
int a, b, c;
a = 0x00000001;
b = 0x00000001;
c = smull(0, 0, a, b);
printf("%x(%d) %x(%d) %x(%d)\n", a, a, b, b, c, c);
a = 0x40000000;
b = 0x00000004;
c = smull(0, 0, a, b);
printf("%x(%d) %x(%d) %x(%d)\n", a, a, b, b, c, c);
return 0;
}
$ smmul
1(1) 1(1) 0(0)
40000000(1073741824) 4(4) 1(1)
A less hacky version of the inline asm that actually tells the compiler what's going on:
// untested, otherwise I'd edit the original test case.
int smull(int a, int b, int c, int d)
{
// b is unused, kept only to match the original
int retval;
asm("smmul %0, %1, %2"
: "=r"(retval)
: "r"(c), "r"(d)
: // no clobbers
);
return retval;
}
Upvotes: 2