flac dev - Jan 2016 - warning: inlining failed in call to 'FLAC__bitwriter_write_raw_uint32.constprop':

Building with MinGW-w64 GCC 5.3.0 via Makefile.lite, I get the
following warnings:

bitwriter.c: In function 'FLAC__bitwriter_write_utf8_uint64':
bitwriter.c:324:19: warning: inlining failed in call to
'FLAC__bitwriter_write_raw_uint32.constprop': --param
large-function-growth limit reached [-Winline]
 inline FLAC__bool FLAC__bitwriter_write_raw_uint32(FLAC__BitWriter
*bw, FLAC__uint32 val, unsigned bits)
                   ^
bitwriter.c:826:9: warning: called from here [-Winline]
   ok &= FLAC__bitwriter_write_raw_uint32(bw, 0x80 |
(FLAC__uint32)((val>>6)&0x3F), 8);
         ^
bitwriter.c:324:19: warning: inlining failed in call to
'FLAC__bitwriter_write_raw_uint32.constprop': --param
large-function-growth limit reached [-Winline]
 inline FLAC__bool FLAC__bitwriter_write_raw_uint32(FLAC__BitWriter
*bw, FLAC__uint32 val, unsigned bits)
                   ^
bitwriter.c:823:9: warning: called from here [-Winline]
   ok &= FLAC__bitwriter_write_raw_uint32(bw, 0xF8 |
(FLAC__uint32)(val>>24), 8);
         ^
bitwriter.c:324:19: warning: inlining failed in call to
'FLAC__bitwriter_write_raw_uint32.constprop': --param
large-function-growth limit reached [-Winline]
 inline FLAC__bool FLAC__bitwriter_write_raw_uint32(FLAC__BitWriter
*bw, FLAC__uint32 val, unsigned bits)
                   ^
bitwriter.c:826:9: warning: called from here [-Winline]
   ok &= FLAC__bitwriter_write_raw_uint32(bw, 0x80 |
(FLAC__uint32)((val>>6)&0x3F), 8);
         ^

Would it be better to remove the inline keyword from the declaration
of FLAC__bitwriter_write_raw_uint32 (and by extension, remove inline
from FLAC__bitwriter_write_zeroes and add it to
FLAC__bitwriter_write_unary_unsigned)? I suspect that the cache misses
caused by code bloat from so much inlining would outweigh the
comparable function call overhead, but performance testing would be
necessary to confirm this.

Best,
Evan

Evan Ramos wrote:
> Would it be better to remove the inline keyword from the declaration
> of FLAC__bitwriter_write_raw_uint32 (and by extension, remove inline
> from FLAC__bitwriter_write_zeroes and add it to
> FLAC__bitwriter_write_unary_unsigned)? I suspect that the cache misses
> caused by code bloat from so much inlining would outweigh the
> comparable function call overhead, but performance testing would be
> necessary to confirm this.
I suspect that the main reason for inline attribute of
FLAC__bitwriter_write_raw_uint32() is FLAC__bitwriter_write_raw_int32()
function. Both of them are used in e.g. libFLAC/stream_encoder_framing.c,
FLAC__subframe_add_(constant|fixed|lpc|verbatim) functions.