Bug 105304
| Summary: | Fix the ARM traditional DFG JIT related bugs | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Csaba Osztrogonác <ossy> |
| Component: | New Bugs | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED DUPLICATE | ||
| Severity: | Normal | CC: | gaborb, galpeter, hausmann, loki, ossy, rgabor, zherczeg |
| Priority: | P2 | ||
| Version: | 420+ | ||
| Hardware: | All | ||
| OS: | All | ||
| Bug Depends on: | |||
| Bug Blocks: | 108645 | ||
Csaba Osztrogonác
Now there are some failing tests because of a/some ARM traditional DFG bug:
- fast/canvas/canvas-putImageData.html
- fast/dom/gc-dom-tree-lifetime.html
- fast/dom/simultaneouslyRegsiteredTimerFireOrder.html
- fast/events/scoped/editing-commands.html
- fast/harness/perftests/runs-per-second-iterations.html
- fast/js/array-bad-time.html
- fast/js/dfg-array-pop-value-clearing.html
- fast/js/dfg-array-push-bad-time.html
- fast/js/dfg-arrayify-when-late-prevent-extensions.html
- fast/js/dfg-get-by-val-clobber.html
- html5lib/run-test10.html
- html5lib/run-test11.html
- html5lib/run-test19.html
- html5lib/run-test20.html
- html5lib/run-test21.html
- html5lib/run-test26.html
- html5lib/run-test9.html
- html5lib/runner.html
- html5lib/webkit-resumer.html
- inspector/debugger/script-formatter.html
- inspector/elements/undo-dom-edits-2.html
- inspector/utilities.html
These tests pass on ARM Thumb2 build and on ARM traditional with disabled DFG
build, but fail with ARM traditional with enabled DFG build (default setting).
See http://build.webkit.sed.hu/results/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/r137983%20%287524%29/results.html for details.
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Csaba Osztrogonác
( Additionally fast/events/tabindex-focus-blur-all.html fails on all ARM platform: ARM-Thumb2, ARM-traditional with/without DFG and fast/js/kde/Array.html fails on ARM-traditional with/without DFG, but passes on ARM-Thumb2 )
I'm going to file separated bug reports to fix these failures too.
Csaba Osztrogonác
Gábor, Zoltán, could you check it?
Csaba Osztrogonác
Some new tests introduced in http://trac.webkit.org/changeset/140276 fails on ARM traditional. (maybe on thumb2 too, but I haven't checked) And I haven't checked if these fails are related to DFG JIT or not. But will do soon.
Csaba Osztrogonác
There are 20 failing tests on ARM traditional because of
enabled DFG JIT, here is the updated list: (tested on r143570)
fast/dom/gc-dom-tree-lifetime.html
fast/events/scoped/editing-commands.html
fast/js/regress/boolean-test.html
fast/js/regress/cmpeq-obj-to-obj-other.html
fast/js/regress/function-test.html
fast/js/regress/number-test.html
fast/js/regress/object-test.html
fast/js/regress/string-test.html
fast/js/regress/undefined-test.html
html5lib/run-test10.html
html5lib/run-test11.html
html5lib/run-test19.html
html5lib/run-test20.html
html5lib/run-test21.html
html5lib/run-test26.html
html5lib/run-test9.html
html5lib/runner.html
html5lib/webkit-resumer.html
http/tests/misc/webtiming-one-redirect.php
inspector/elements/undo-dom-edits-2.html
Csaba Osztrogonác
There are 33 failing tests on ARM traditional related to buggy DFG JIT:
------------------------------------------------------------------------
fast/dom/gc-dom-tree-lifetime.html
fast/events/scoped/editing-commands.html
fast/js/regress/boolean-test.html
fast/js/regress/cmpeq-obj-to-obj-other.html
fast/js/regress/function-test.html
fast/js/regress/number-test.html
fast/js/regress/object-test.html
fast/js/regress/string-test.html
fast/js/regress/undefined-test.html
html5lib/generated/run-domjs-unsafe-data.html
html5lib/generated/run-domjs-unsafe-write.html
html5lib/generated/run-plain-text-unsafe-data.html
html5lib/generated/run-plain-text-unsafe-write.html
html5lib/generated/run-tests10-data.html
html5lib/generated/run-tests10-write.html
html5lib/generated/run-tests11-data.html
html5lib/generated/run-tests11-write.html
html5lib/generated/run-tests19-data.html
html5lib/generated/run-tests19-write.html
html5lib/generated/run-tests20-data.html
html5lib/generated/run-tests20-write.html
html5lib/generated/run-tests21-data.html
html5lib/generated/run-tests21-write.html
html5lib/generated/run-tests26-data.html
html5lib/generated/run-tests26-write.html
html5lib/generated/run-tests9-data.html
html5lib/generated/run-tests9-write.html
html5lib/generated/run-webkit01-data.html
html5lib/generated/run-webkit01-write.html
html5lib/webkit-resumer.html
http/tests/misc/webtiming-one-redirect.php
http/tests/xmlhttprequest/timeout/xmlhttprequest-timeout-overridesexpires.html
http/tests/xmlhttprequest/timeout/xmlhttprequest-timeout-worker-overridesexpires.html
r146872 with disabled DFG JIT: http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/8149
r146862 with enabled DFG JIT: http://build.webkit.sed.hu/builders/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/builds/8148
Csaba Osztrogonác
Most of these bugs were fixed by https://trac.webkit.org/changeset/150748 ( in a different bug report - https://bugs.webkit.org/show_bug.cgi?id=116306 )
But there are still 3 failing tests related to ARM traditional and DFG JIT:
- fast/dom/gc-dom-tree-lifetime.html
- fast/js/integer-division-neg2tothe32-by-neg1.html
- fast/js/regress/negative-zero-modulo.html
See the diffs for details:
- http://build.webkit.sed.hu/results/ARMv7%20Linux%20Qt5%20Release%20(Test)/r150785%20(8638)/fast/dom/gc-dom-tree-lifetime-pretty-diff.html
- http://build.webkit.sed.hu/results/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/r150785%20%288638%29/fast/js/integer-division-neg2tothe32-by-neg1-pretty-diff.html
- http://build.webkit.sed.hu/results/ARMv7%20Linux%20Qt5%20Release%20%28Test%29/r150785%20%288638%29/fast/js/regress/negative-zero-modulo-pretty-diff.html
Csaba Osztrogonác
There are only two failing tests:
- fast/js/integer-division-neg2tothe32-by-neg1.html
- fast/js/regress/negative-zero-modulo.html
and there is a new bug report with proposed fix,
so closing it with resolved/duplicated
*** This bug has been marked as a duplicate of bug 117542 ***