MSL_C: build the library optimized (Clanker Warning) by joeyballentine · Pull Request #770 · bfbbdecomp/bfbb

joeyballentine · 2026-08-03T04:21:17Z

Summary

MSL_C.PPCEABI.H is configured with -opt level=0, peephole, schedule, nospace and -inline off, but the original library was built optimized. At level 0 CodeWarrior spills user locals to callee-saved registers and emits a stack frame for functions the target objects implement as leaves, so almost nothing in the library could match.

Changing the library flags to -opt level=4, peephole, schedule, nospace and -inline auto,deferred matches 54 additional functions with no source changes:

	before	after
msl functions matched	138	192
msl code matched	31.38%	49.29%
DOL functions matched	6394	6448

abort_exit, arith, mem, mem_funcs, rand, signal, wchar_io, FILE_POS and math_ppc go to fully matching functions; string goes from 0/10 to 8/10 and printf from 0/14 to 7/14.

How the flags were chosen

I swept the plausible variants across every MSL_C unit and picked by total exact-matching functions rather than by any single file:

flags	exact functions
`-opt level=0 -inline off` (current)	27/84
`-opt level=4 -inline off`	22/84
`-opt level=4 -inline auto` / `-inline on`	34/84
`-opt level=4 -inline auto,deferred`	39/84
`-opt level=4,...,space`	22/84
`-opt nopeephole`	5/84
`-opt noschedule`	4/84

Measured end to end with the project's own progress metric, deferred is worth 41.92% -> 49.29% of msl code on top of level=4 alone. It is only +3 functions but roughly 8KB more matched code, landing large bodies in printf and ansi_fp — which is why it is included rather than dropped.

-sym on and -str pool were left as they are; neither changed any output in the sweep.

Why critical_regions.gamecube.c is reordered

deferred inlining makes CodeWarrior emit function bodies in reverse source order. The project already models this — tools/project.py sets reverse_fn_order from exactly this flag value.

critical_regions.gamecube.c is a Matching object, so it is linked into the DOL and its emission order has to stay correct. Its three stub functions are reordered so the object still emits __end_critical_region, __begin_critical_region, __kill_critical_regions, matching the target. Verified with readelf against build/GQPE78/obj/. No other linked object changed order.

signal.c now matches byte-for-byte and is marked Matching.

Verification

ninja is clean and the DOL sha1 check passes, so every Matching object still produces identical linked output.

I checked the other libraries for the same class of misconfiguration while I was here; cflags_bfbb and cflags_dolphin are already correct and no variant improved them.

MSL_C.PPCEABI.H was configured with `-opt level=0, ... -inline off`, but the original library was clearly built optimized. At level 0 CodeWarrior spills user locals to callee-saved registers and emits a stack frame for functions that the target objects implement as leaves. Switching to `-opt level=4, ... -inline auto,deferred` matches 54 additional functions with no source changes, taking the msl category from 31.4% to 49.3% of code matched (138 -> 192 functions). `deferred` inlining makes CodeWarrior emit function bodies in reverse source order, which the project already models as `reverse_fn_order` in tools/project.py. critical_regions.gamecube.c is therefore reordered to keep the emission order it contributes to the DOL correct; it is linked in, so the build sha1 check covers this. signal.c now matches byte-for-byte and is marked Matching.

github-actions · 2026-08-03T04:23:09Z

`main/MSL_C/PPC_EABI/abort_exit`

	Section	From	To	Bytes
✅	`.text`	80.39%	100.00%	+77

	Function	From	To	Bytes
✅	`exit`	82.83%	100.00%	+41
✅	`abort`	76.64%	100.00%	+36

`main/MSL_C/MSL_Common/alloc`

	Section	From	To	Bytes
📈	`.text`	6.84%	19.87%	+739

	Function	From	To	Bytes
✅	`__pool_free`	10.91%	100.00%	+78
✅	`deallocate_from_fixed_pools`	67.67%	100.00%	+111
✅	`deallocate_from_var_pools`	16.61%	100.00%	+550

`main/MSL_C/MSL_Common/ansi_files`

	Section	From	To	Bytes
📈	`.text`	32.04%	35.71%	+28

	Function	From	To	Bytes
✅	`__flush_all`	90.54%	100.00%	+10
✅	`__close_all`	89.14%	100.00%	+18

`main/MSL_C/MSL_Common_Embedded/ansi_fp`

	Section	From	To	Bytes
✅	`.data`	59.55%	100.00%	+145
✅	`.rodata`	83.26%	100.00%	+37
📈	`.text`	17.61%	99.33%	+11266

	Function	From	To	Bytes
📈	`__dec2num`	18.40%	97.81%	+3354
✅	`__num2dec`	36.19%	100.00%	+268
✅	`__num2dec_internal`	21.28%	100.00%	+746
✅	`__minus_dec`	33.58%	100.00%	+847
✅	`__equals_dec`	45.45%	100.00%	+146
✅	`__two_exp`	8.20%	100.00%	+5526
✅	`__timesdec`	40.13%	100.00%	+378

`main/MSL_C/MSL_Common/arith`

	Section	From	To	Bytes
✅	`.text`	55.00%	100.00%	+7

	Function	From	To	Bytes
✅	`abs`	55.00%	100.00%	+7

`main/MSL_C/MSL_Common/buffer_io`

	Section	From	To	Bytes
📈	`.text`	19.98%	30.54%	+85

	Function	From	To	Bytes
✅	`__flush_buffer`	56.22%	100.00%	+85

`main/MSL_C/MSL_Common/ctype`

	Section	From	To	Bytes
📈	`.text`	11.00%	60.00%	+29

	Function	From	To	Bytes
✅	`tolower`	18.33%	100.00%	+29

`main/MSL_C/MSL_Common/direct_io`

	Section	From	To	Bytes
📈	`.text`	39.63%	48.61%	+168

	Function	From	To	Bytes
✅	`__fwrite`	84.73%	100.00%	+119
✅	`fwrite`	60.16%	100.00%	+49

`main/MSL_C/MSL_Common/file_io`

	Section	From	To	Bytes
📈	`.text`	24.19%	43.75%	+338

	Function	From	To	Bytes
✅	`fflush`	89.19%	100.00%	+33
✅	`fclose`	31.46%	100.00%	+304

`main/MSL_C/MSL_Common/FILE_POS`

	Section	From	To	Bytes
✅	`.text`	63.47%	100.00%	+309

	Function	From	To	Bytes
✅	`fseek`	64.48%	100.00%	+38
✅	`_fseek`	70.44%	100.00%	+169
✅	`ftell`	39.10%	100.00%	+102

`main/MSL_C/MSL_Common/mbstring`

	Section	From	To	Bytes
📈	`.text`	49.58%	64.18%	+117

	Function	From	To	Bytes
✅	`wcstombs`	49.67%	100.00%	+140
️⚠️	`mbtowc`	16.46%	0.00%	--47
✅	`is_utf8_complete`	89.90%	100.00%	+23

`main/MSL_C/MSL_Common/mem`

	Section	From	To	Bytes
✅	`.text`	29.91%	100.00%	+257

	Function	From	To	Bytes
✅	`memcmp`	11.05%	100.00%	+67
✅	`__memrchr`	0.00%	100.00%	+44
✅	`memchr`	0.00%	100.00%	+44
✅	`memmove`	49.84%	100.00%	+102

`main/MSL_C/MSL_Common/mem_funcs`

	Section	From	To	Bytes
✅	`.text`	78.71%	100.00%	+153

	Function	From	To	Bytes
✅	`__copy_longs_rev_unaligned`	82.44%	100.00%	+30
✅	`__copy_longs_unaligned`	76.25%	100.00%	+45
✅	`__copy_longs_rev_aligned`	79.26%	100.00%	+34
✅	`__copy_longs_aligned`	77.32%	100.00%	+42

`main/MSL_C/MSL_Common/printf`

	Section	From	To	Bytes
📈	`.data`	50.00%	73.74%	+132
✅	`.rodata`	97.30%	100.00%	+1
📈	`.text`	76.30%	92.38%	+1368

	Function	From	To	Bytes
✅	`sprintf`	52.84%	100.00%	+105
✅	`vsprintf`	27.17%	100.00%	+87
✅	`vprintf`	66.63%	100.00%	+50
✅	`fprintf`	88.95%	100.00%	+24
✅	`__StringWrite`	36.15%	100.00%	+68
✅	`__FileWrite`	67.41%	100.00%	+28
✅	`__pformatter`	85.62%	100.00%	+274
📈	`float2str`	72.16%	89.11%	+300
✅	`round_decimal`	76.92%	100.00%	+69
📈	`double2hex`	73.50%	97.04%	+193
📈	`longlong2str`	84.05%	86.80%	+20
️⚠️	`long2str`	82.23%	80.62%	--8
📈	`parse_format`	82.15%	94.09%	+153

`main/MSL_C/MSL_Common/rand`

	Section	From	To	Bytes
✅	`.text`	85.00%	100.00%	+4

	Function	From	To	Bytes
✅	`rand`	85.00%	100.00%	+4

`main/MSL_C/MSL_Common/scanf`

	Section	From	To	Bytes
📈	`.text`	73.71%	92.96%	+802

	Function	From	To	Bytes
✅	`sscanf`	57.43%	100.00%	+83
✅	`__StringRead`	62.75%	100.00%	+53
📈	`__sformatter`	81.64%	93.81%	+301
📈	`parse_format`	62.72%	89.64%	+363

`main/MSL_C/MSL_Common/signal`

	Section	From	To	Bytes
✅	`.text`	70.00%	100.00%	+57

	Function	From	To	Bytes
✅	`raise`	70.00%	100.00%	+57

`main/MSL_C/MSL_Common/string`

	Section	From	To	Bytes
📈	`.text`	32.46%	68.73%	+445

	Function	From	To	Bytes
✅	`strstr`	67.50%	100.00%	+33
✅	`strchr`	0.00%	100.00%	+48
✅	`strncmp`	27.81%	100.00%	+46
✅	`strcmp`	59.07%	100.00%	+121
✅	`strcat`	35.00%	100.00%	+28
✅	`strncpy`	27.06%	100.00%	+49
✅	`strcpy`	51.04%	100.00%	+90
✅	`strlen`	0.00%	100.00%	+28

`main/MSL_C/MSL_Common/strtold`

	Section	From	To	Bytes
📈	`.text`	30.97%	83.55%	+2233

	Function	From	To	Bytes
📈	`__strtold`	32.03%	86.39%	+2233

`main/MSL_C/MSL_Common/strtoul`

	Section	From	To	Bytes
📈	`.data`	50.00%	52.94%	+4
📈	`.text`	68.24%	83.29%	+318

	Function	From	To	Bytes
📈	`__strtoull`	73.17%	92.95%	+204
📈	`__strtoul`	76.79%	89.58%	+113

`main/MSL_C/MSL_Common_Embedded/uart_console_io_gcn`

	Section	From	To	Bytes
📈	`.text`	69.61%	92.94%	+47

	Function	From	To	Bytes
📈	`__write_console`	68.37%	92.65%	+47

`main/MSL_C/MSL_Common/wchar_io`

	Section	From	To	Bytes
✅	`.text`	56.88%	100.00%	+58

	Function	From	To	Bytes
✅	`fwide`	56.88%	100.00%	+58

joeyballentine marked this pull request as draft August 3, 2026 04:21

joeyballentine changed the title ~~MSL_C: build the library optimized~~ MSL_C: build the library optimized (Clanker Warning) Aug 3, 2026

joeyballentine closed this Aug 3, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

MSL_C: build the library optimized (Clanker Warning) - #770

MSL_C: build the library optimized (Clanker Warning)#770
joeyballentine wants to merge 1 commit into
bfbbdecomp:mainfrom
joeyballentine:msl-optimization-flags

joeyballentine commented Aug 3, 2026

Uh oh!

github-actions Bot commented Aug 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

joeyballentine commented Aug 3, 2026

Summary

How the flags were chosen

Why critical_regions.gamecube.c is reordered

Verification

Uh oh!

github-actions Bot commented Aug 3, 2026

main/MSL_C/PPC_EABI/abort_exit

main/MSL_C/MSL_Common/alloc

main/MSL_C/MSL_Common/ansi_files

main/MSL_C/MSL_Common_Embedded/ansi_fp

main/MSL_C/MSL_Common/arith

main/MSL_C/MSL_Common/buffer_io

main/MSL_C/MSL_Common/ctype

main/MSL_C/MSL_Common/direct_io

main/MSL_C/MSL_Common/file_io

main/MSL_C/MSL_Common/FILE_POS

main/MSL_C/MSL_Common/mbstring

main/MSL_C/MSL_Common/mem

main/MSL_C/MSL_Common/mem_funcs

main/MSL_C/MSL_Common/printf

main/MSL_C/MSL_Common/rand

main/MSL_C/MSL_Common/scanf

main/MSL_C/MSL_Common/signal

main/MSL_C/MSL_Common/string

main/MSL_C/MSL_Common/strtold

main/MSL_C/MSL_Common/strtoul

main/MSL_C/MSL_Common_Embedded/uart_console_io_gcn

main/MSL_C/MSL_Common/wchar_io

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

`main/MSL_C/PPC_EABI/abort_exit`

`main/MSL_C/MSL_Common/alloc`

`main/MSL_C/MSL_Common/ansi_files`

`main/MSL_C/MSL_Common_Embedded/ansi_fp`

`main/MSL_C/MSL_Common/arith`

`main/MSL_C/MSL_Common/buffer_io`

`main/MSL_C/MSL_Common/ctype`

`main/MSL_C/MSL_Common/direct_io`

`main/MSL_C/MSL_Common/file_io`

`main/MSL_C/MSL_Common/FILE_POS`

`main/MSL_C/MSL_Common/mbstring`

`main/MSL_C/MSL_Common/mem`

`main/MSL_C/MSL_Common/mem_funcs`

`main/MSL_C/MSL_Common/printf`

`main/MSL_C/MSL_Common/rand`

`main/MSL_C/MSL_Common/scanf`

`main/MSL_C/MSL_Common/signal`

`main/MSL_C/MSL_Common/string`

`main/MSL_C/MSL_Common/strtold`

`main/MSL_C/MSL_Common/strtoul`

`main/MSL_C/MSL_Common_Embedded/uart_console_io_gcn`

`main/MSL_C/MSL_Common/wchar_io`