<html>
    <head>
      <base href="https://bugs.llvm.org/">
    </head>
    <body><table border="1" cellspacing="0" cellpadding="8">
        <tr>
          <th>Bug ID</th>
          <td><a class="bz_bug_link 
          bz_status_NEW "
   title="NEW - [GlobalISel] CodeModel::Large sequences are not emitted"
   href="https://bugs.llvm.org/show_bug.cgi?id=35958">35958</a>
          </td>
        </tr>

        <tr>
          <th>Summary</th>
          <td>[GlobalISel] CodeModel::Large sequences are not emitted
          </td>
        </tr>

        <tr>
          <th>Product</th>
          <td>new-bugs
          </td>
        </tr>

        <tr>
          <th>Version</th>
          <td>trunk
          </td>
        </tr>

        <tr>
          <th>Hardware</th>
          <td>PC
          </td>
        </tr>

        <tr>
          <th>OS</th>
          <td>Linux
          </td>
        </tr>

        <tr>
          <th>Status</th>
          <td>NEW
          </td>
        </tr>

        <tr>
          <th>Severity</th>
          <td>normal
          </td>
        </tr>

        <tr>
          <th>Priority</th>
          <td>P
          </td>
        </tr>

        <tr>
          <th>Component</th>
          <td>new bugs
          </td>
        </tr>

        <tr>
          <th>Assignee</th>
          <td>unassignedbugs@nondot.org
          </td>
        </tr>

        <tr>
          <th>Reporter</th>
          <td>roger.ferreribanez@arm.com
          </td>
        </tr>

        <tr>
          <th>CC</th>
          <td>llvm-bugs@lists.llvm.org, qcolombet@apple.com
          </td>
        </tr></table>
      <p>
        <div>
        <pre>Created <span class=""><a href="attachment.cgi?id=19680" name="attach_19680" title="C Testcase">attachment 19680</a> <a href="attachment.cgi?id=19680&action=edit" title="C Testcase">[details]</a></span>
C Testcase

Hi all,

the following testcase needs to be built with -mcmodel=large as it accesses
huge arrays.

-- t.c
#define BIG    (1024*1024*1024)

int foo1[BIG];
int foo2[BIG];

int main(void)
{
  return foo1[0] + foo2[0];
}
-- end of t.c

unfortunately GlobalISel seems to be unable to select the "large" sequence of
instructions.

$ aarch64-linux-gnu-clang  -DADJUST=0 -o t.exe t.c -mcmodel=large 
/tmp/t-0e2765.o: In function `main':
t.c:(.text+0x4): relocation truncated to fit: R_AARCH64_ADR_PREL_PG_HI21
against symbol `foo1' defined in COMMON section in /tmp/t-0e2765.o
clang-6.0: error: linker command failed with exit code 1 (use -v to see
invocation)

Looking at the generated GMir

%bb.1: derived from LLVM BB %entry
        %1:_(s32) = G_CONSTANT i32 0
        %4:_(p0) = G_GLOBAL_VALUE @foo1
        %3:_(p0) = COPY %4(p0)
        %7:_(p0) = G_GLOBAL_VALUE @foo2
        %6:_(p0) = COPY %7(p0)
        ...

looks like G_GLOBAL_VALUE is lowered by the InstructionSelector to the pseudo
MOVaddr:

Selecting: 
  %4:gpr(p0) = G_GLOBAL_VALUE @foo1
(... tons of gibberish ...)
Into:
  %4:gpr64(p0) = MOVaddr target-flags(aarch64-page) @foo1,
target-flags(aarch64-pageoff, aarch64-nc) @foo1

which is later expanded to

%x8 = ADRP target-flags(aarch64-page) @foo1
renamable %x8 = ADDXri %x8, target-flags(aarch64-pageoff, aarch64-nc) @foo1, 0

which ends being

adrp    x8, foo1
add     x8, x8, :lo12:foo1

which will cause the linker failure.

Disabling GlobalISel via -mllvm -global-isel=0 we see that the expected
sequence is generated

movz    x9, #:abs_g0_nc:foo1
movk    x9, #:abs_g1_nc:foo1
movk    x9, #:abs_g2_nc:foo1
movk    x9, #:abs_g3:foo1

SelectionDAG strategy is a bit different: it wraps the ISD::GlobalAddress
during legalization into a AArch64ISD::WrapperLarge and then uses a Pat<...> to
expand it.

Kind regards,
Roger</pre>
        </div>
      </p>


      <hr>
      <span>You are receiving this mail because:</span>

      <ul>
          <li>You are on the CC list for the bug.</li>
      </ul>
    </body>
</html>