[llvm] r176066 - [fast-isel] Make sure the FastLowerArguments function checks to make sure the

[David Blaikie]

On Tue, Feb 26, 2013 at 12:03 PM, Chad Rosier <mcrosier at apple.com> wrote:
>
> On Feb 26, 2013, at 11:51 AM, David Blaikie <dblaikie at gmail.com> wrote:
>
> On Tue, Feb 26, 2013 at 11:48 AM, Chad Rosier <mcrosier at apple.com> wrote:
>
>
> On Feb 25, 2013, at 6:33 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>
>
>
> On Mon, Feb 25, 2013 at 5:05 PM, Chad Rosier <mcrosier at apple.com> wrote:
>
>
> Author: mcrosier
> Date: Mon Feb 25 19:05:31 2013
> New Revision: 176066
>
> URL: http://llvm.org/viewvc/llvm-project?rev=176066&view=rev
> Log:
> [fast-isel] Make sure the FastLowerArguments function checks to make sure
> the
> arguments type is a simple type.
> rdar://13290455
>
>
>
> Test case?
>
>
> David,
> I'm not aware of an easy way to test this.
>
> ; RUN: not llc < %s -fast-isel -fast-isel-abort -fast-isel-abort-args
> -verify-machineinstrs -mtriple=x86_64-apple-darwin10
>
> ; Abort because we don't know how to handle non-simple type i31.
> define i31 @t1(i31 %a, i31 %b, i31 %c) {
> entry:
>  %add = add nsw i31 %b, %a
>  %add1 = add nsw i31 %add, %c
>  ret i31 %add1
> }
>
> The above test case will exercise the issue, but results in llc aborting and
> the 'not' on the run line doesn't have the correct effect.
>
>
> I'm by no means especially familiar with the backend, just looking at
> the change it seemed like something that I would expect to have an
> easily observable effect.
>
>
> I assume the -fast-isel-abort-args/-fast-isel-abort is what's meant to
> cause the abort if fast isel can't handle something?

That's usually totally fine by me - you can include "Requires:
Asserts" so we don't bother running the test on non-asserts build
where it wouldn't catch the regression anyway.

- David

>
>
> Correct.
>
> Do we produce the
> same code before/after your change, but the only issue is that
> previously we would do it with fast isel & now we don't?
>
>
> No, this was causing an assert when ArgVT.getSimpleVT() was called.
>
> What was the
> correctness problem (why make any change at all) then? If there was a
> correctness problem, can't we just test that in the actually assembly
> output rather than testing whether fast isel aborts?
>
>
> The issue was a compiler assertion.
>
> I could just remove the -fast-isel-abort-args flag to avoid causing the
> abort.  Would that suffice as a test?
> I could comment that this previously caused an assert.
>
>  Chad
>
> - David
>
>
> Chad
>
>
>
>
> Modified:
>    llvm/trunk/lib/Target/ARM/ARMFastISel.cpp
>    llvm/trunk/lib/Target/X86/X86FastISel.cpp
>
> Modified: llvm/trunk/lib/Target/ARM/ARMFastISel.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMFastISel.cpp?rev=176066&r1=176065&r2=176066&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Target/ARM/ARMFastISel.cpp (original)
> +++ llvm/trunk/lib/Target/ARM/ARMFastISel.cpp Mon Feb 25 19:05:31 2013
> @@ -2922,6 +2922,7 @@ bool ARMFastISel::FastLowerArguments() {
>       return false;
>
>     EVT ArgVT = TLI.getValueType(ArgTy);
> +    if (!ArgVT.isSimple()) return false;
>     switch (ArgVT.getSimpleVT().SimpleTy) {
>     case MVT::i8:
>     case MVT::i16:
>
> Modified: llvm/trunk/lib/Target/X86/X86FastISel.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86FastISel.cpp?rev=176066&r1=176065&r2=176066&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86FastISel.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86FastISel.cpp Mon Feb 25 19:05:31 2013
> @@ -1555,6 +1555,7 @@ bool X86FastISel::FastLowerArguments() {
>       return false;
>
>     EVT ArgVT = TLI.getValueType(ArgTy);
> +    if (!ArgVT.isSimple()) return false;
>     switch (ArgVT.getSimpleVT().SimpleTy) {
>     case MVT::i32:
>     case MVT::i64:
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>