This is the mail archive of the
mailing list for the GCC project.
Re: arm inline assembly 4.8 vs. 4.9 [solved]
- From: Gene Smith <gds at chartertn dot net>
- To: gcc-help at gcc dot gnu dot org
- Date: Wed, 11 Mar 2015 01:26:40 -0400
- Subject: Re: arm inline assembly 4.8 vs. 4.9 [solved]
- Authentication-results: sourceware.org; auth=none
- References: <mdnrsn$3lg$1 at ger dot gmane dot org>
On 3/10/2015 6:40 PM, Gene Smith wrote:
The following inline assembly works as I want with gcc 4.8.3 and
produces the result shown:
"ldr r0, =%[addrOfIoPorts]\n"
: /* no output */
: [addrOfIoPorts] "g" (&ioPorts) /* input */
: "r0"); /* clobber */
801b9e2: 4822 ldr r0, [pc, #136]
But when compiled with 4.9.3 the linker says "undefined reference to
`r3'. This is because the produced assembly becomes such that r3 looks
like a constant:
ldr r0, =r3
I have been unable to find a modification to the original code that
works with 4.9.3. Any ideas or suggestions?
A helpful reply via emailed suggested the use of other constraints:
"Using "g" allows "Any register, memory or immediate integer operand."
Such being the case, have you tried replacing it with each of "m", "r"
and "i"? I would expect "m" to work for you."
I think I had tried all of these. But tried again.
"m" causes "error: memory input 0 is not directly addressable"
"r" is same as "g": undefined reference to `r3'
"i" worked! Obviously I didn't try this. I should have looked closer in
the latest gcc manual under "constraints" -- "i": An immediate integer
operand (one with constant value) is allowed. This includes
symbolic constants whose values will be known only at assembly time or
later. (So maybe it was a bug that 4.8.3 worked with "g" in this context?)
No build errors and now see these again in obj dump under 4.9.3:
801b4c8: 486c ldr r0, [pc, #432] ; (801b67c ...