Re: [Tinycc-devel] BUG: code generation problem in i386-gen.c:gcall_or_jmp()

[Christian Jullien]

Zhang,
Thank you for commits.

I don't really know if they make tcc better or not but I sure that:

- Windows 32/64 has no regression
- RPi has no regression
- Your commit messages and patch are clear very easy to understand.

To me (again I'm not a maintainer) they seem to be good contributions. To be 
perfect, they lack of new tests that show why tcc was wrong before your commits 
and why now it is doing right.
With those new tests, we can better tell you if your fixes work on all 
supported platforms. Feel free to add them. No one will protest to see new 
tests on a tcc.

Christian

-----Original Message-----
From: Tinycc-devel [mailto:address@hidden On Behalf Of ???
Sent: samedi 9 septembre 2017 14:56
To: address@hidden
Subject: Re: [Tinycc-devel] BUG: code generation problem in 
i386-gen.c:gcall_or_jmp()

OK, I will push my modifications to the mob branch.


在 2017年09月09日 18:26, Christian Jullien 写道:
> Hello Zhang
> 
> First I'm not an official maintainer of tcc but a regular user and observer 
> of its evolutions. I also check that it does not regress on the platforms I 
> use.
> 
> That said, your recent proposals seem strongly argumented with associated 
> tests. Since no one commented your proposal (that sad!).
> I suggest you to push them in mod so that we can see if we have regression or 
> not with them. Please make one commit at a time for a given fix so that we 
> can better insulate regression (if any).
> 
> I any cases, many thanks to participate to this project and to try to make it 
> better.
> 
> Christian
> 
> -----Original Message-----
> From: Tinycc-devel [mailto:address@hidden On Behalf Of ???
> Sent: samedi 9 septembre 2017 05:03
> To: address@hidden
> Subject: [Tinycc-devel] BUG: code generation problem in 
> i386-gen.c:gcall_or_jmp()
> 
> Hello,
> 
> I found 3 problems in i386-gen.c:gcall_or_jmp().
> 
> (1) There is a fix for extending return value to whole register, however, 
> this fix is in the constant case only, if a call is indirect, this fix will 
> not be applied and the result will be wrong.
> 
> (2) The fix is for 'call' and if 'is_jmp' == 1, it will generate redundant 
> code.
> 
> (3) Calling constant should use the indirect way. For example, if someone 
> write "((void (*)(void)) 0x12345678)()" in a DLL, the generated code will be 
> "CALL 0x12345678". However, because the imm in CALL instruction is PC 
> related, when the DLL is relocated, the call target will be wrong.
> 
> 
> 
> I provided a patch and two testcase.
> 
> 
> 
> 'test-movzx-movsx.c' is for problem (1) and (2).
> run it with:
>     tcc -run test-movzx-movsx.c
> 
> expected output is:
>     0000ABCE 0000ABCE
>     FFFFABCE FFFFABCE
>     000000CE 000000CE
>     FFFFFFCE FFFFFFCE
> 
>     0000ABCE 0000ABCE
>     FFFFABCE FFFFABCE
>     000000CE 000000CE
>     FFFFFFCE FFFFFFCE
> 
> 
> 
> 
> the other two files are for problem (3).
> run it with:
> 
> # set DLL base at 0x400000 to force a relocation tcc -shared 
> -Wl,-Ttext=0x00400000 test-call-constant-dll.c
> 
> tcc -L. -ltest-call-constant-dll -run test-call-constant-run.c
> 
> expected output is:
>     by pointer variable: FEEDBEEF
>     by pointer constant: FEEDBEEF
> 
> 
> 
> 
> Thanks
> Zhang Boyang
> 
> 
> _______________________________________________
> Tinycc-devel mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/tinycc-devel
> 

-- 
张博洋 - 复旦大学2014级计算机科学与技术
我的手机: 18600020982
我的个人网站: http://www.zbyzbyzby.com

_______________________________________________
Tinycc-devel mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/tinycc-devel