Match and link baselib/synth - #3184
Merged
Merged
Conversation
Restore the four-word tail store as a constant-trip loop, advance the node cursor in bytes, and let LICM hoist the copy shift. 93.37% -> 95.53%.
Use k * 0x40 indexing in the per-entry fixup loop, fold nshift into the data-size expression, and advance the node cursor before the stream cursor. 95.53% -> 96.69%.
Report for GALE01 (ff035cf - d0766a6)📈 Matched code: 90.25% (+0.05%, +2020 bytes) ✅ 5 new matches
|
Replace the two data_pad placeholders with the assertion text pooled by two mwld-stripped static helpers, define the .sbss objects that were left extern, order .sdata around HSD_AudioMalloc's assert, and restore the unreferenced .data tail object. .data 77.9% -> byte-identical; .sbss/.bss/.sdata2 byte-identical.
Pass &ve directly instead of through a block-local pointer and reserve the eight bytes that sit between the voice array and ve, which puts the whole AXSetVoiceVe sequence and the volume envelope stores on the right slots.
The setup loop reuses the acquire loop's counter and indexes both the voice array and the per-voice AX blocks directly, so MWCC strength- reduces the two walkers in the preheader and the counter's zero also feeds the node's zeroed fields. Hoist the SFX entry struct to file scope and reach the per-voice blocks through it. 97.03% -> 98.91%.
HSD_Synth_80389334 stores its sfx_id argument at node offset 4, which the struct declared as a next pointer; HSD_Synth_8038B5AC clears the same field. Retype it and store through it. 98.91% -> 99.50%.
Splitting the multiply from the header add stops MWCC folding the round-up constant into the multiply temp, so the header block now matches the DOL instruction for instruction.
Keep the unaligned size in its own value so the add lands in a fresh register, and give the shift-up copy loop its own counter. Everything from the header read to the unrolled copy now matches except one volatile pair.
initialize.c writes HSD_Synth_804D6018, so it cannot be static here. Define it in synth.c's data and declare it in synth.h instead of repeating the extern at the use site. Linking synth now succeeds and the DOL differs only inside the two functions that are still short of 100%.
Splitting the round-up into a mask of its own keeps the aligned size out of the optimiser's temp pool, and declaring the copy counter ahead of the header read orders the three long-lived values the way the DOL has them. HSD_SynthSFXSampleLoadCallback: 32 differing instructions -> 16.
alloc_size plus OSRoundUp32B, as in HSD_SynthSFXHeaderLoadCallback, reproduces the header block exactly and drops the dead sum read. Folding src_idx into the copy loop's initializer leaves the two dead alias homes the frame has room for, and the unsigned source index keeps the four header loads off a strength-reduced walker. 16 -> 7 differing instructions.
The include-cleaner pass on master dropped the transitive path that used to supply OSHeapHandle to synth.h, so the merged tree failed to compile.
The two cursor updates are independent, but their order decides which pair of registers the allocator gives each one. Retail stores HSD_Synth_804D7734 first, which is what makes the node advance keep r0 across the mask; both cursor relocations in the loop tail now match.
The ITD flag arrives as an int, not a u8. Storing it into the byte-wide field costs a truncation that the peephole then deletes, but the scheduler has already spent an integer slot on it -- which is what keeps the argument copy from landing between the two folded stores. With that slot occupied the node claims its slot before its id, matching retail.
A logical right shift by two already discards the low bits, so the `& ~3` was a no-op.
The bucket index is masked in place and both cursors advance by a word count, which is what the register allocation needs: the in-place mask emits a record-form `andi.` that the peephole folds into the following `rlwinm`, and the extra shift in the node advance gives that chain the height to claim the integer slot the mask frees.
`ninja apply` recovers the compiler's own literal-pool and static names. It also supersedes a split-era guess at 0x804DE7F0, which the splitter had carved into two words: it is one 8-byte pooled `double` (`@313`, align 8). Nothing referenced the phantom second half.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
this TU is the 2nd and 3rd failing reference when trapping all non-100% TUs and
HSD_SynthSFXSampleLoadCallbackis the 3rd failing function when booting the game with all non-100% functions trapping.