Re: [avr-llvm-devel] Regression tests

[Borja F. <bor...@gm...>]

This is very weird. Do you get this problem with other code when you use
this particular number (3335977107) like with a regular call or with an
and/xor/or?

2012/6/24 Nicklas Bo Jensen <nbj...@gm...>

> I also get the wrong result compiling that snippet...
>
> I have tried to revert the last patches, those that not needed for syncing
> with the llvm repo, same result. I have also updated llvm. Same result.
>
> Please let me know if you have any ideas what it could be :)
>
> On Sat, Jun 23, 2012 at 8:15 PM, Borja Ferrer <bor...@gm...>wrote:
>
>> Thanks for the patch there!
>> Very interesting because I'm not getting that error in icall at all, I'm
>> getting the second code snippet you pasted.
>>
>> With this code:
>> long icall(long (*x)(long))
>> {
>>   return x(0xC6D6F893)+5;
>> }
>>
>> I get:
>>     movw    r19:r18, r25:r24
>>
>>     ldi    r22, 147
>>     ldi    r23, 248
>>     ldi    r24, 214
>>     ldi    r25, 198
>>     movw    r31:r30, r19:r18
>>     icall
>>     subi    r22, 251
>>     sbci    r23, 255
>>     sbci    r24, 255
>>     sbci    r25, 255
>>     ret
>>
>> Can you re check to see what's going on?
>>
>> The comments in zext.ll and sext.ll are there for reference, they cover
>> the cases implemented in the pseudo instruction expander, that way we're
>> able to know what test covers each case.
>> Yes i have comments for the memory tests, I'll write them in the next
>> email. We should continue with memory operations, there are many more
>> things to test there that i'll list next time. For now we have to see
>> what's going on with that failing test.
>>
>>
>>
>> 2012/6/23 Nicklas Bo Jensen <nbj...@gm...>
>>
>>> I've committed the small change in Targets.cpp as you described.
>>>
>>> We are breaking a test now. In call.s, icall is loading arguments
>>> incorrectly now.
>>>
>>> The instruction %1 = call i32 %foo(i32 3335977107)
>>>  gives:
>>> ldi r24, 147
>>> ldi r25, 248
>>>  ldi r22, 214
>>> ldi r23, 198
>>> but should give:
>>> ldi r22, 147
>>> ldi r23, 248
>>> ldi r24, 214
>>>  ldi r25, 198
>>>
>>> Also in your new sext.ll and zext.ll tests there are some comments above
>>> each function, should these be deleted?
>>>
>>> Last did you have any comments above the memory tests?
>>>
>>> Any preferences for what I should continue on?
>>>
>>> Thanks!
>>>
>>>
>>> On Tue, Jun 19, 2012 at 1:47 PM, Borja Ferrer <bor...@gm...>wrote:
>>>
>>>> Nicklas, in the meantime while the patch gets fixed you can manually
>>>> fix the error by adding that virtual BuiltinVaListKind
>>>> getBuiltinVaListKind() const function in the AVR class, in the above link
>>>> you have an example on how to do it.
>>>>
>>>>
>>>> 2012/6/19 Borja Ferrer <bor...@gm...>
>>>>
>>>>> John can you fix this issue, i have a different local copy of
>>>>> Targets.cpp that i don't want to change now with the patch in SVN. The fix
>>>>> is in this commit
>>>>>
>>>>>
>>>>> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Basic/Targets.cpp?r1=158334&r2=158592&sortby=date&diff_format=h
>>>>>
>>>>> so updating the patch file will do it.
>>>>>
>>>>>
>>>>> 2012/6/18 Borja Ferrer <bor...@gm...>
>>>>>
>>>>>> You've done it right, the problem is here:
>>>>>>
>>>>>> http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/Basic/Targets.cpp?r1=158334&r2=158592&sortby=date&diff_format=h
>>>>>>
>>>>>> it's caused by a recent API change that has broken out of tree
>>>>>> targets, i'll fix it soon.
>>>>>>
>>>>>>
>>>>>> 2012/6/18 Nicklas Bo Jensen <nbj...@gm...>
>>>>>>
>>>>>>> Yes of course.
>>>>>>>
>>>>>>> I'm having some problems compiling getting the following error:
>>>>>>>
>>>>>>> /avr-llvm/llvm/tools/clang/lib/Basic/Targets.cpp:4172:16: error:
>>>>>>> allocating an object of
>>>>>>>       abstract class type 'AVRTargetInfo'
>>>>>>>     return new AVRTargetInfo(T);
>>>>>>>                ^
>>>>>>> /avr-llvm/llvm/tools/clang/lib/Basic/../../include/clang/Basic/TargetInfo.h:406:29:
>>>>>>> note:
>>>>>>>       unimplemented pure virtual method 'getBuiltinVaListKind' in
>>>>>>> 'AVRTargetInfo'
>>>>>>>   virtual BuiltinVaListKind getBuiltinVaListKind() const = 0;
>>>>>>>                             ^
>>>>>>>  1 error generated.
>>>>>>>
>>>>>>> I think I have applied the patches from svn correctly, at least did
>>>>>>> it a few times. I'm not stuck, but any ideas would be nice.
>>>>>>>
>>>>>>> On Mon, Jun 18, 2012 at 1:46 PM, Borja Ferrer <bor...@gm...
>>>>>>> > wrote:
>>>>>>>
>>>>>>>> Ok great!
>>>>>>>>
>>>>>>>> One more thing about the icall test, it can be further reduced
>>>>>>>> because:
>>>>>>>>   %1 = alloca i32 (i32)*
>>>>>>>>   store i32 (i32)* %foo, i32 (i32)** %1
>>>>>>>>   %2 = load i32 (i32)** %1
>>>>>>>>
>>>>>>>> all that code is redundant, you're getting it because you've ran it
>>>>>>>> with -O0. If you think about it, you've allocated an i32 var in the stack
>>>>>>>> just to copy the value of foo but then it's not used at all. This code is
>>>>>>>> typical from unoptimized builds, so you can reduce it to something like:
>>>>>>>>   %3 = call i32 %foo(i32 3335977107)
>>>>>>>>   %4 = add nsw i32 %3, 5
>>>>>>>>
>>>>>>>> here you're using the foo incoming arg directly.
>>>>>>>> So as a recommendation, I would also run clang with -03 to get
>>>>>>>> optimized outputs that are usually clean of redundant allocas.
>>>>>>>>
>>>>>>>>
>>>>>>>> 2012/6/18 Nicklas Bo Jensen <nbj...@gm...>
>>>>>>>>
>>>>>>>>> I've committed the updated tests, excluding mem.ll awaiting your
>>>>>>>>> feedback on it (Besides not using immediate 0).
>>>>>>>>>
>>>>>>>>> On Mon, Jun 18, 2012 at 1:10 AM, Borja Ferrer <
>>>>>>>>> bor...@gm...> wrote:
>>>>>>>>>
>>>>>>>>>> return test:
>>>>>>>>>> 1) you can remove the @return8_reg() test since it's completely
>>>>>>>>>> commented out.
>>>>>>>>>> 2) in return64_arg2 you're always going to push r29:r28 as we
>>>>>>>>>> disccused in a previous email, so you can remove the SPLOW/HI vars, also SP
>>>>>>>>>> is going to be always the Y reg pair. Using Z here would be a bug.
>>>>>>>>>> 3) in return64_trunc same as in point 2.
>>>>>>>>>> 4) add a :TODO: test returning byval structs.
>>>>>>>>>>
>>>>>>>>>> call test:
>>>>>>>>>> 1) in the icall test, the code you're testing is a bit fragile.
>>>>>>>>>> Basically if you run it with -O3 it will get optimized away and you wont
>>>>>>>>>> get an icall. Instead, generate the icall using a function pointer from an
>>>>>>>>>> incoming argument.
>>>>>>>>>> 2) do the same as in 1 but passing an argument to the icall (any
>>>>>>>>>> data type will do). (remember i had to fix this last week, so testing it is
>>>>>>>>>> imporant in case it breaks again in the future)
>>>>>>>>>> 3) do the same as in 1 using the return value of the icall (any
>>>>>>>>>> data type will do).
>>>>>>>>>>
>>>>>>>>>> One general thing concerning all written tests up to now, can you
>>>>>>>>>> change all imm values (including the ones already commited), so that you
>>>>>>>>>> don't test things like "ldi reg, 0". So for a 16bit type instead of using a
>>>>>>>>>> 5 use a value were none of the registers are zero like 0xABCD instead of
>>>>>>>>>> 0x0005 where the hi part is 0. In the future we'll add an optimization that
>>>>>>>>>> relies on r0 having a 0, so many tests will break because of this since we
>>>>>>>>>> could get a mov reg, r0 instead of an ldi.
>>>>>>>>>>
>>>>>>>>>> I'll reply tomorrow with comments about the memory tests.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> 2012/6/17 Borja Ferrer <bor...@gm...>
>>>>>>>>>>
>>>>>>>>>>> Thanks, I'm reviewing them now
>>>>>>>>>>>
>>>>>>>>>>> El viernes, 15 de junio de 2012, Nicklas Bo Jensen escribió:
>>>>>>>>>>>
>>>>>>>>>>> Hi,
>>>>>>>>>>>>
>>>>>>>>>>>> Please review these tests.
>>>>>>>>>>>>
>>>>>>>>>>>> Any comments?
>>>>>>>>>>>>
>>>>>>>>>>>> On Wed, Jun 6, 2012 at 2:46 AM, Borja Ferrer <
>>>>>>>>>>>> bor...@gm...> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> Fixed in SVN, let me know any other issues. Thanks for
>>>>>>>>>>>> reporting it!
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> 2012/6/5 Borja Ferrer <bor...@gm...>
>>>>>>>>>>>>
>>>>>>>>>>>> Something has changed in llvm or maybe it never worked, can't
>>>>>>>>>>>> remember, but I'm unable to pass arguments through an icall, I'm
>>>>>>>>>>>> investigating for a fix. In the meantime you can disable that assertion
>>>>>>>>>>>> until I come with a proper fix.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> 2012/6/4 Borja Ferrer <bor...@gm...>
>>>>>>>>>>>>
>>>>>>>>>>>> I'll take a look to it shortly.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> 2012/6/3 Nicklas Bo Jensen <nbj...@gm...>
>>>>>>>>>>>>
>>>>>>>>>>>> Ok. Sure your explanations are really clear :) Thanks!
>>>>>>>>>>>>
>>>>>>>>>>>> I'm having some trouble testing icall, which i'm doing using a
>>>>>>>>>>>> function pointer. I've attached a small C example, the corresponding
>>>>>>>>>>>> cleaned up llvm code and the output of running llc on it. There is a failed
>>>>>>>>>>>> assertion.
>>>>>>>>>>>>
>>>>>>>>>>>> Any thoughts?
>>>>>>>>>>>>
>>>>>>>>>>>> On Fri, Jun 1, 2012 at 7:04 PM, Borja Ferrer <
>>>>>>>>>>>> bor...@gm...> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> Just checking for the sub instructions that allocate and
>>>>>>>>>>>> restore the stack should be enough, also check that the allocated size is
>>>>>>>>>>>> correct, the rest of stuff can be omitted. I would also check the whole
>>>>>>>>>>>> sequence including the interrupt instructions in a single test, no need to
>>>>>>>>>>>> do it everywhere. When we get to test allocas we'll have to test for
>>>>>>>>>>>> prologue and epilogue code, which is very similar to the one you're seeing
>>>>>>>>>>>> here.
>>>>>>>>>>>> About testing push and pops: no need to test that, it's part of
>>>>>>>>>>>> the prologue/epilogue of the function, so it's not related to calls, also,
>>>>>>>>>>>> that is handled by llvm so I wouldn't test that at all. Remember to test
>>>>>>>>>>>> the icall instruction, as far as I remember it needs to be tweaked because
>>>>>>>>>>>> for now it doesn't save the registers between calls like regular calls do.
>>>>>>>>>>>>
>>>>>>>>>>>> Ok, memory instructions, most of the instructions you see in
>>>>>>>>>>>> that file that end up with a "w" are pseudo instructions (excluding
>>>>>>>>>>>> sbiw/adiw, movw or other real ones that work with 16bit data) meaning that
>>>>>>>>>>>> they dont get into the final assembly since they don't exist in the
>>>>>>>>>>>> instruction set. It's a trick to make llvm think they're supported
>>>>>>>>>>>> instructions that way, movws and the real 16bit insts get generated,
>>>>>>>>>>>> however these pseudo ops get expanded into 8bit ops, that's why you wont
>>>>>>>>>>>> never see those in the final assembly, and if you ever see one then it's a
>>>>>>>>>>>> bug. So if you do an and with 16bit types, llvm selects the andw pseudo op,
>>>>>>>>>>>> and a later pass expands it into two real 8bit and insts that are supported
>>>>>>>>>>>> by the avr inst set, dont' know if this makes things clear.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> 2012/5/31 Nicklas Bo Jensen <nbj...@gm...>
>>>>>>>>>>>>
>>>>>>>>>>>> Checking for stack allocation and restore could be something
>>>>>>>>>>>> like:
>>>>>>>>>>>>
>>>>>>>>>>>> ; CHECK: in [[REG1:r[0-9]+]], 61
>>>>>>>>>>>> ; CHECK: in [[REG2:r[0-9]+]], 62
>>>>>>>>>>>> ; CHECK: sbiw [[REG2]]:[[REG1]], 4
>>>>>>>>>>>> ; CHECK: out 62, [[REG2]]
>>>>>>>>>>>> ; CHECK: out 61, [[REG1]]
>>>>>>>>>>>> ; ... other tests
>>>>>>>>>>>> ; CHECK: in [[REG1:r[0-9]+]], 61
>>>>>>>>>>>> ; CHECK: in [[REG2:r[0-9]+]], 62
>>>>>>>>>>>> ; CHECK: subi [[REG1]], 254
>>>>>>>>>>>> ; CHECK: sbci [[REG2]], 255
>>>>>>>>>>>> ; CHECK: out 62, [[REG2]]
>>>>>>>>>>>> ; CHECK: out 61, [[REG1]]
>>>>>>>>>>>>
>>>>>>>>>>>> Is it something like that you have in mind? Should disabling
>>>>>>>>>>>> and enabling interupts (cli+writeback of SREG) also be tested?
>>>>>>>>>>>>
>>>>>>>>>>>> Checkin
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>