Repository navigation
Releases
Releases · parallelno/v6clang
v6clang 2026.10.06
Full Changelog: v2026.07.26...v2026.10.06
v6c 2026.07.26
Full Changelog: v2026.07.10...v2026.07.26
v6c 2026.07.10-1
Full Changelog: v2026.06.08...v2026.07.10-1
v6c 2026.06.08
Full Changelog: v2026.05.31...v2026.06.08
v6c 2026.05.31-2
Release Notes
New Optimization Passes
Type Narrowing & Induction Variable Optimization
- O52 — Index Induction Variable Rewriting: rewrites i16 loop indices to i8 when safe, eliminating 16-bit arithmetic overhead in tight loops
- O85 — TypeNarrowing: narrows i16 up-counters to i8 when all arithmetic users permit it; includes correctness guards for inner-loop PHI users and regression fixes
Code-Shape Peephole Passes
- O81 — SELECT_CC i8 Materialization Through Accumulator: lowers i8 conditional selects via accumulator instead of full register-pair paths
- O82 — MOV Chain Collapse and Dead High-Byte Elimination: collapses redundant MOV chains and removes dead high-byte moves
- O83 — POP/PUSH Pair Elimination for Dead Register Pairs: removes POP/PUSH round-trips when the register pair is dead at that point
- O84 — INX/DCX-Through-Spill Round-Trip Fold: folds INX/DCX increments that are spilled and immediately reloaded back into a single operation
- O86 — Constant-Amount i16 Shift Specialization: emits optimal fixed-shift sequences for all constant i16 shift amounts
- O87 — Negate Specialization: recognizes
-x,-1*x, andx*(-1)patterns and emits a directNEGsequence instead of a general multiply
Inter-Procedural Analysis (O10 / O61)
hasNoCallbackEvidencenow acceptsnorecursecallees directly, reducing conservative spill decisions in non-recursive call chains- Made
hasNoCallbackEvidencetransitive across visible callees for improved stack allocation across call trees
Peephole Improvements
- i16 multiply: optimized multiplication sequence with peephole to remove redundant
XCHGinstructions - MOV round-trip elimination: new peephole pass removes
MOV X, Y ; ... ; MOV Y, Xredundant copy chains; also eliminatesMOV rp, copy ; MOV copy, rpregister-pair round-trips - V6C_ADD16: emits a shorter expansion when the
DEregister pair is dead after the operation
Code Generator Fixes
- Dead reload elimination in V6C spill forwarding:
V6C_SPILL16/V6C_RELOADpairs with no live use of the reload are now removed - i8 SRL: constant-amount logical right shifts now use the existing
emitRotChainhelper, consistent with the left-shift path - Truncating shift return codegen: optimized the code sequence emitted when a shift result is immediately truncated before a return
- Truncating ashr expansion: improved arithmetic-shift-right expansion for the truncating-return case
- XRA zero rewrite: fixed an incorrect rewrite of
XRA Awhen flag-consuming instructions follow - CMP_SIGN instruction: added signed zero-test for i16 comparisons using a new
CMP_SIGNpseudo, enabling cheaper== 0/!= 0checks on 16-bit values - PrintAsmOperand: added support for global and external symbols in inline assembly operand printing
- Fixed prolog code generation bug affecting function entry setup
Runtime Library (compiler-rt)
Header Library
- Header consolidation: unified V6C-specific headers under a single include tree; added
v6c_rt_macros.h,v6c_macros.h, and new hardware wrappers - Interrupt handling refactor: rewrote interrupt enable/disable wrappers and hardware instruction stubs for correctness and consistency
- Replaced
annotate("v6c-rt-helper")with the real[[clang::v6c_rt]]attribute on all runtime helper functions BIT_MASKchanged from mutable toconstwith corrected inline-asm constraints
String & Memory Builtins
strlenandstrcmp: rewritten in optimized assembly, reducing cycle count in the common casememcpy/memset: signatures corrected to return the destination pointer (void *dst) per the C standard
CRT0 / Startup
- Added
DI(Disable Interrupts) beforeHLTincrt0for a clean halt - Improved
crt0.obuild process; fixed loop-counter bug in startup init sequence v6c_outfunction renamed from__v6c_outfor consistency with the rest of the ABI
Graphics Library
draw_line(assembly version): hand-written 8080 assembly implementation of Bresenham line drawingdraw_line2: improved slope calculation and endpoint handling for steeper linesdraw_scale_x_4_3: new function for 4:3 pixel-aspect-ratio scaling, used bydraw_circleto correct circle proportions on Vector 06c displaysdraw_pixel:randand draw functions refactored to usedraw_pixelas the common pixel-write primitive- Added a line-rendering benchmark sample
Samples
- New
samples/directory with three curated programs:01_hello/— minimal compile-and-run example02_bsort/— benchmark-style C sort03_demo/— larger drawing/demo program including line and circle rendering, with build helpers
- Removed unused legacy assembly stubs (demo and no-LPI variants)
Build System
- Quick-build option added to
build.ps1;build_release.ps1consolidated into the main build script - Build documentation streamlined: system requirements section simplified, commands updated to PowerShell style
- z88dk integration: benchmark runner now gracefully skips z88dk targets when the toolchain is not installed, with a clear error message
.lit_test_times.txtadded to.gitignoreand untracked from all subdirectories
Documentation
- Overriding the Initial Stack Pointer: new section in the programming guide
- README updated: added Samples section, expanded benchmark suite description to five programs (
bsort,sieve,fib_crc,fannkuch,lfsr16), clarified quick-start workflow for packaged release vs. source build TODO.mdrefactored; addedO63_split_spill_pseudo_flags.mddesign document for the planned flags-aware spill splitting pass- Added rejected-optimization design documents for future reference
Full Changelog: v2026.05.18-2...v2026.05.31-2
v6c 2026.05.18-2
Full Changelog: v2026.05.18-1...v2026.05.18-2
v6c 2026.05.18-1
Full Changelog: v2026.05.15...v2026.05.18-1
v6c 2026.05.15
Full Changelog: v2026.05.14...v2026.05.15
v6c 2026.05.10-5
Full Changelog: v2026.04.29-1...v2026.05.10-5
v6c 2026.04.29-1
Full Changelog: v2026.04.29...v2026.04.29-1