Bug 101712 – [Qt] Fix the LLINT build from ARMv7 platform

• Component: New Bugs
• Status: RESOLVED FIXED

[Csaba Osztrogonác]

The Qt build is broken on ARMv7 (Thumb2) platform:
/home/oszi/WebKit/Source/JavaScriptCore/llint/LowLevelInterpreter.cpp:541:27: fatal error: LLIntAssembly.h: No such file or directory

The reason is simple. LLINT is enabled by Platform.h:
/* On some of the platforms where we have a JIT, we want to also have the
   low-level interpreter. */
#if !defined(ENABLE_LLINT) \
    && ENABLE(JIT) \
    && (OS(DARWIN) || OS(LINUX)) \
    && (PLATFORM(MAC) || PLATFORM(IOS) || PLATFORM(GTK) || (PLATFORM(QT) && OS(LINUX))) \
    && (CPU(X86) || CPU(X86_64) || CPU(ARM_THUMB2))
#define ENABLE_LLINT 1
#endif

But generating of LLIntAssembly.h is disabled all ARM platforms, not only ARM_TRADITIONAL.

Fix is coming soon.

[Csaba Osztrogonác]

LLINT is disabled on MIPS, but generating isn't disabled. But it doesn't
cause any problem, because the genarator script automatically skip it:

ruby /data/buildbot/mips-1/qt-linux-mipsel-mips32r2-release/build/Source/JavaScriptCore/offlineasm/asm.rb /data/buildbot/mips-1/qt-linux-mipsel-mips32r2-release/build/Source/JavaScriptCore/llint/LowLevelInterpreter.asm LLIntOffsetsExtractor generated/LLIntAssembly.h
offlineasm: Parsing /data/buildbot/mips-1/qt-linux-mipsel-mips32r2-release/build/Source/JavaScriptCore/llint/LowLevelInterpreter.asm and LLIntOffsetsExtractor and creating assembly file generated/LLIntAssembly.h.
offlineasm: No magic values found. Skipping assembly file generation.

Here is the output on ARM_TRADITIONAL after enabling generation:

ruby /home/oszi/WebKit/Source/JavaScriptCore/offlineasm/asm.rb /home/oszi/WebKit/Source/JavaScriptCore/llint/LowLevelInterpreter.asm LLIntOffsetsExtractor generated/LLIntAssembly.h
offlineasm: Parsing /home/oszi/WebKit/Source/JavaScriptCore/llint/LowLevelInterpreter.asm and LLIntOffsetsExtractor and creating assembly file generated/LLIntAssembly.h.

[Csaba Osztrogonác]

Created attachment 173208 [details]
Patch

[Csaba Osztrogonác]

Of course we could implement a complex condition for generating similar to Platform.h . But it needs configure tests, etc. Additionally generating is
needed LLINT_C_LOOP (interpreter) platforms too, not only JIT&&LLINT platforms.

I think enabling generation on more platform than necessary is safe,
it won't cause any problem.

[Csaba Osztrogonác]

Comment on attachment 173208 [details]
Patch

Clearing flags on attachment: 173208

Committed r134025: <http://trac.webkit.org/changeset/134025>

[Csaba Osztrogonác]

All reviewed patches have been landed.  Closing bug.

[Simon Hausmann]

Latest build error on ARM now: 

11:32:23 /work/build/qt/qtwebkit/Source/JavaScriptCore/release/libJavaScriptCore.a(JITPropertyAccess.o): In function `JSC::JIT::privateCompilePutByVal(JSC::ByValInfo*, JSC::ReturnAddressPtr, JSC::JITArrayMode)':
11:32:23 JITPropertyAccess.cpp:(.text+0x20fa): undefined reference to `JSC::AbstractMacroAssembler<JSC::ARMv7Assembler>::JumpList JSC::JIT::emitGenericContiguousPutByVal<(unsigned char)20>(JSC::Instruction*, JSC::AbstractMacroAssembler<JSC::ARMv7Assembler>::PatchableJump&)'
11:32:23 JITPropertyAccess.cpp:(.text+0x2134): undefined reference to `JSC::AbstractMacroAssembler<JSC::ARMv7Assembler>::JumpList JSC::JIT::emitGenericContiguousPutByVal<(unsigned char)22>(JSC::Instruction*, JSC::AbstractMacroAssembler<JSC::ARMv7Assembler>::PatchableJump&)'

[Csaba Osztrogonác]

Maybe http://trac.webkit.org/changeset/133953 is the culprit, let me check it.

[Csaba Osztrogonác]

(In reply to comment #7)
> Maybe http://trac.webkit.org/changeset/133953 is the culprit, let me check it.
It is innocent, but r133985 is the culprit - https://bugs.webkit.org/show_bug.cgi?id=101740