[llvm-dev] Need help figuring out a isNopCopy() assert

[Justin Hibbits via llvm-dev]

I'm trying to fix a bug in the PowerPC SPE backend that prevents a
bunch of FreeBSD ports from building, including gtk20.  The attached
file, generated from the following C source, triggers the "Def ==
PreviousDef" assertion in isNopCopy():

typedef float a;
typedef struct {
  a b, c;
} complex;
d(complex *e, complex *h) {
  double f = h->c, g = h->b;
  i(g);
  e->c = g * j(f);
}

This was a reduction from texlive's c_cos.c.

The applicable assertion failure is:

Assertion failed: (Def == PreviousDef), function isNopCopy, file
/home/chmeee/llvm_git/llvm/llvm/lib/CodeGen/MachineCopyPropagation.cpp,
line 338. PLEASE submit a bug report to https://bugs.llvm.org/ and
include the crash backtrace. Stack dump:
 0.      Program arguments: /ralga/scratch/chmeee/builds/llvm/bin/llc
 -mtriple powerpcspe reduced.ll 
1.      Running pass 'Function Pass Manager' on module 'reduced.ll'. 
2.      Running pass 'Machine Copy Propagation Pass' on function '@d'


I dumped the debug during this, and found the following interesting
bits:

0B      bb.0.entry:
          liveins: $r4
16B       %1:gprc_and_gprc_nor0 = COPY $r4
32B       %2:gprc = SPELWZ 0, undef %3:gprc_and_gprc_nor0 :: (load 4
from `float* undef`) 40B       %5:gprc = SPELWZ 0, killed
%1:gprc_and_gprc_nor0 :: (load 4 from %ir.b1) 48B       %4:sperc = COPY
killed %2:gprc 80B       %6:sperc = COPY killed %5:gprc

...
MCP: Copy is a deletion candidate:   renamable $s29 = COPY killed
renamable $r4
MCP: Copy is used - not dead:   renamable $s29 = COPY
killed renamable $r4 
MCP: Copy is used - not dead:   renamable $s29 = COPY killed renamable
$r4

$s29 is a SPE register (64-bits wide), while $r4 is a GPR (32-bits
wide), and given what I've found googling this assert, the assert
indicates two different register classes, which is obvious from this.
The question is *where* and *why* do I have this?

At first I thought to look through PPCFastISel.cpp, since there's some
code in there that does fast selection with assumptions of the
traditional PowerPC FPU.  However, that didn't pan out.  Now I'm
scratching my head, and looking for help on where to try next, as every
place I've found "TargetOpcode::COPY" and "PPC::COPY" appear to be
covered already, correctly.

Thanks,
Justin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: reduced.ll
Type: application/octet-stream
Size: 535 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200415/acf28f3d/attachment-0001.obj>