[llvm-bugs] [Bug 30342] New: [ppc] LLVM generates an extra loop

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=30342

            Bug ID: 30342
           Summary: [ppc] LLVM generates an extra loop
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: PowerPC
          Assignee: unassignedbugs at nondot.org
          Reporter: carrot at google.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Compile the following code with options

--target=powerpc64le-grtev4-linux-gnu -m64 -O2 -mvsx -mcpu=power8

typedef unsigned short T;

int foo(T* array_a, T* array_b, int size_a, int size_b, T* buf) {
    int i_a = 0, i_b = 0, idx = 0;
    while (i_a < size_a && i_b < size_b) {
      if (array_a[i_a] == array_b[i_b]) {
        buf[idx++] = array_a[i_a];
        i_a++;
        i_b++;
      } else if (array_a[i_a] > array_b[i_b]) {
        i_b++;
      } else {
        i_a++;
      }
    }
    return idx;
}

Trunk llvm generates:

  9 # BB#0:                                 # %entry
 10         li 12, 0
 11         cmpwi    5, 0
 12         std 28, -32(1)                  # 8-byte Folded Spill
 13         std 29, -24(1)                  # 8-byte Folded Spill
 14         std 30, -16(1)                  # 8-byte Folded Spill
 15         cmpwi 1, 6, 0
 16         crand 20, 1, 5
 17         bc 4, 20, .LBB0_10
 18 # BB#1:                                 # %while.body.lr.ph.lr.ph.preheader
 19         extsw 8, 6
 20         extsw 9, 5
 21         addi 4, 4, -2
 22         li 11, 0
 23         li 10, 0
 24         li 30, 0
 25 .LBB0_2:                                # %while.body.lr.ph.lr.ph
 26                                         # =>This Loop Header: Depth=1
 27                                         #     Child Loop BB0_3 Depth 2
 28                                         #       Child Loop BB0_4 Depth 3
 29         extsw 12, 30
 30 .LBB0_3:                                # %while.body.lr.ph
 31                                         #   Parent Loop BB0_2 Depth=1
 32                                         # =>  This Loop Header: Depth=2
 33                                         #       Child Loop BB0_4 Depth 3
 34         extsw 11, 11
 35         sldi 0, 12, 1
 36         addi 30, 11, 1
 37         lhzx 0, 3, 0
 38         sldi 29, 11, 1
 39         cmpd     30, 8
 40         isel 30, 30, 8, 1
 41         sub      28, 30, 11
 42         add 30, 4, 29
 43         mtctr 28
 44         .p2align        5
 45 .LBB0_4:                                # %while.body
 46                                         #   Parent Loop BB0_2 Depth=1
 47                                         #     Parent Loop BB0_3 Depth=2
 48                                         # =>    This Inner Loop Header:
Depth=3
 49         lhzu 29, 2(30)
 50         cmplw    0, 29
 51         beq      0, .LBB0_8
 52 # BB#5:                                 # %if.else
 53                                         #   in Loop: Header=BB0_4 Depth=3
 54         ble 0, .LBB0_7
 55 # BB#6:                                 # %if.then21
 56                                         #   in Loop: Header=BB0_4 Depth=3
 57         addi 11, 11, 1
 58         bdnz .LBB0_4
 59         b .LBB0_9
 60         .p2align        4
 61 .LBB0_7:                                # %if.else23
 62                                         #   in Loop: Header=BB0_3 Depth=2
 63         addi 12, 12, 1
 64         cmpw 1, 11, 6
 65         cmpd     12, 9
 66         crand 20, 0, 4
 67         bc 12, 20, .LBB0_3
 68         b .LBB0_9
 69         .p2align        4
 70 .LBB0_8:                                # %if.then
 71                                         #   in Loop: Header=BB0_2 Depth=1
 72         addi 30, 12, 1
 73         addi 11, 11, 1
 74         addi 12, 10, 1
 75         sldi 10, 10, 1
 76         sthx 0, 7, 10
 77         mr 10, 12
 78         cmpw     30, 5
 79         cmpw 1, 11, 6
 80         crand 20, 0, 4
 81         bc 12, 20, .LBB0_2
 82         b .LBB0_10
 83 .LBB0_9:
 84         mr 12, 10
 85 .LBB0_10:                               # %while.end
 86         extsw 3, 12
 87         ld 30, -16(1)                   # 8-byte Folded Reload
 88         ld 29, -24(1)                   # 8-byte Folded Reload
 89         ld 28, -32(1)                   # 8-byte Folded Reload
 90         blr

Although there is only loop in the source code, llvm generates an extra loop
start from .LBB0_4. And the loop preparation work in .LBB0_3 is very slow
because of isel and mtctr.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160909/54839fd2/attachment.html>