Feat/e2 conformance#7
Merged
Merged
Conversation
Clear three residual correctness bugs that the coverage matrix marked "done" but were genuinely wrong: - 32-bit adc.w/sbc.w shifted-register form was entirely missing (no case 10/11 in t32_dataproc_reg) -> IllegalInstruction. The "implemented" status only covered the immediate form. - 16-bit adcs/sbcs (op 5/6) computed the right result but never wrote C/V (only update_nz) -> flags silently stale. - SDIV INT_MIN/-1 is C signed overflow (UB) -> saturate to INT_MIN, now guarded (the old a/b was UB). SDIV/0 stays 0 for both signs -- the QEMU differential oracle (scripts/qemu_cortex_m3_oracle.sh) caught and rejected an initial "negative -> INT_MIN" misreading of the spec. Add carry-in-aware flag helpers (set_adc_flags/set_sbc_flags) using a 64-bit sum so C/V are correct even at the 0xFFFFFFFF+carry edge where the prior "fold Cin into operand" trick wrapped and mis-reported C. Used uniformly across the 16-reg / 32-imm / 32-shifted-reg forms. 11 objdump-verified conformance tests (result value + N/Z/C/V via MRS APSR). ctest 312/312 green, no E2E/CLI regression. All Tier-1 semantics cross-checked vs qemu-system-arm mps2-an385. See document/notes/017.
…t (E2 T2) Tier 2 of the E2 conformance milestone (same branch as T1): - scripts/qemu_cortex_m3_oracle.sh: assembles a tiny Thumb program (SDIV/UDIV edge cases + ADC/SBC shifted-reg with flags), runs it under qemu-system-arm -M mps2-an385 via a gdbstub, and diffs the result/xPSR buffers. Toolchain confirmed present (qemu-system-arm 11.0.1 + arm-none-eabi-gdb). All Tier-1 semantics match QEMU line-for-line; the oracle caught the SDIV/0 spec error corrected in T1. - test/firmware/armcc/build_corpus_opt.ps1: PowerShell build of the CubeF1 corpus at AC6 -O0/-O2/-Oz (run on Windows NTFS; WSL can't create Keil .__i). Applies the REGENERATE AC5->AC6 patches plus an opt-level injection into MiscControls, emitting <example>.ac6-<opt>.axf. Note AC6 has no -Os; min-size is -Oz. The -O0/-O2/-Oz firmware regression TESTS land once the user builds the .axf corpus with the script above; see document/notes/018.
…(E2 T2) D:\mf\STM32CubeF1 was cleared; rebuild it and correct the build script against the real vendored CubeF1 structure: - scripts/prepare_corpus.sh: rebuild the minimal CubeF1 subtree on /mnt/d/mf (== D:\mf) from third_party/STM32CubeF1, preserving the 6-level include depth the .uvprojx files expect. Trims CMSIS DSP/Lib/docs/NN (~111M) -> 22M total. Verified. - build_corpus_opt.ps1: the project file is Project.uvprojx (not STM32F103RB_Nucleo.uvprojx); OutputName=STM32F103RB_Nucleo. Fix two bugs in the first cut: wrong project name, and output clobber across opt levels (shared STM32F103RB_Nucleo\ dir). Now copies the whole MDK-ARM dir per opt (MDK-ARM-<opt>) for full output isolation, and injects -O<opt> via XML into only the C compiler (Cads) MiscControls rather than regex-hitting the assembler. See document/notes/018.
Three bugs surfaced running build_corpus_opt.ps1 on Windows:
- InsertAfter failed: PowerShell's XML ETS stringifies leaf elements,
so $t.ToolsetName became "ARM-ADS" (string), not a node. Use
SelectSingleNode to get real XmlNodes; also read/write via .InnerText.
- Opt flag was "-OO0": levels are "O0"/"O2"/"Oz" and the code prepended
another "-O". Flag is now "-" + level ("-O0"/"-O2"/"-Oz").
- $LASTEXITCODE is not populated by Start-Process; use -PassThru and
.ExitCode to read UV4's real exit code.
Also: suggest -NoProfile (skip $PROFILE / oh-my-posh noise). Verified
the .uvprojx has no default xmlns, so bare-name XPath resolves.
…urce needs existing target)
Add 9 opt-level firmware variants (GPIO/TIM/UART x O0/O2/Oz) built via build_corpus_opt.ps1, and wire BootsClean gates for each. Per example .text grows O0 > O2 > Oz, confirming the optimization levels are distinct; all 9 boot clean in micro-forge. ctest 321/321 green. This closes the Tier-2 opt-level regression gate: compiler codegen variants (e.g. adc.w/sdiv emitted at -O2 but not -O0) are now exercised beyond the single-level corpus. See document/notes/018.
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
1 participant
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.
No description provided.