daisytuner / docc / 19852283663
67%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 47
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 24.986% (+0.08%) from 24.903%
19852283663

push

github

web-flow 
Docc tmp support (#357)

* New DOCC_TMP support and output folders.

 * DOCC_TMP ENV indicates the base output folder where to put our temps.
 * if DOCC_TMP does not exist or is empty, use /tmp/$USER/DOCC instead
 * support -docc-work-dir=path on CMD to control the specific dir into which to output (subfolders with module hashes as usual). Renamed from previous -extract-dir
 * if no specific -docc-work-dir is set, use the [base output]/[outputFile]-[PID]. Makes it easier to find the outputs of a specific build. Similar to how compiler drivers use main-[hash].o
 + driver converts all -mllvm args with space into -mllvm= so they are one string
 + driver maps all plain -docc- args into BOTH compiler and linker args. Legacy args (with -mllvm or -Wl,-mllvm) are never touched

* Docc temp cleanup

 + DOCC will delete the temp folder it created after successful completion (in docc-ld) if not given -docc-save-temps or in CI mode
 * explicit -docc-work-dir for driver is an implicit -docc-save-temps
 * made sdfg file path in JSON index file relative to the index file (sdfg-registry internal only)

* Fixing tests with DOCC_TMP

 ~ Fixed: handle absolute -o paths correctly for DOCC_TMP
 ~ fixed: if 2nd linking for TT, never delete tmps, as it contains kernel files
 * updated HPCCG with newest version from our fork
 * pytest TestRunner puts DOCC tmp outputs into its explicitly created tmpdir (with -docc-work-dir, not DOCC_TMP)
 ~ fixed rtl_tests to also use explicit docc-work-dir for better debugging and prevent auto-cleanup of result files its testing

* Updating sdfglib-auto with fixes for tt-unique-names merge & updating daisy-mantevo workflow for updated HPCCG sources & enabled TT variant

* Disabling hpccg tt daisy workflow again. Aster has no TT.

* disabled --docc-einsum for HPCCG cuda to make it work reliably again (memory corruption is not in einsum, but in handling the CUDA kernel that results from it)

26 of 63 new or added lines in 4 files covered. (41.27%)

86 existing lines in 3 files now uncovered.

1806 of 7228 relevant lines covered (24.99%)

8.5 hits per line

New Missed Lines in Diff

Lines Coverage ∆ File
1
0.0
 0.0% plugin_llvm/src/passes/function_to_sdfg_pass.cpp
6
0.0
 0.0% plugin_llvm/src/analysis/sdfg_registry.cpp
30
42.31
 plugin_llvm/src/docc_paths.cpp

Uncovered Existing Lines

Lines Coverage ∆ File
2
87.5
 87.5% plugin_llvm/src/cmd_args.cpp
33
0.0
 0.0% plugin_llvm/src/passes/function_to_sdfg_pass.cpp
51
0.0
 0.0% plugin_llvm/src/analysis/sdfg_registry.cpp
Jobs
ID Job ID Ran Files Coverage
1 19852283663.1 47
24.99
 GitHub Action Run
Source Files on build 19852283663