[LLVMbugs] [Bug 15416] New: ARM Code GEN ICE on LoopVectorize'd code containing float cosine function

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Fri Mar 1 20:32:04 PST 2013


            Bug ID: 15416
           Summary: ARM Code GEN ICE on LoopVectorize'd code containing
                    float cosine function
           Product: tools
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: llc
          Assignee: unassignedbugs at nondot.org
          Reporter: aschwaighofer at apple.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Release+Asserts/bin/clang -O3 -mcpu=swift -fvectorize -arch armv7 -isysroot
testfcos.c -S

cat > testfcos.c

#include <math.h>

int test(double *dvec1, double *dvec2, int size) {
  int i;
  for( i = 0; i < 128; i++ ) {
    dvec1[i] = 1.0000001 * cosf((float)(size - i));
    dvec2[i] = 1.0  + 0.0000000001 * sinf((float)i);

fatal error: error in backend: Cannot select: 0x7ff59a11db10: v2f32 = fcos
0x7ff59a11dd10 [ORD=15] [ID=47]
  0x7ff59a11dd10: v2f32 = sint_to_fp 0x7ff59a11c210 [ORD=14] [ID=44]
    0x7ff59a11c210: v2i32 = sub 0x7ff59a11d510, 0x7ff59a11c810 [ORD=13] [ID=42]
      0x7ff59a11d510: v2i32,ch = CopyFromReg 0x7ff599c4b598, 0x7ff59a11de10
[ORD=13] [ID=24]
        0x7ff59a11de10: v2i32 = Register %vreg0 [ORD=13] [ID=6]
      0x7ff59a11c810: v2i32 = add 0x7ff59a124310, 0x7ff59a122a10 [ORD=12]
        0x7ff59a124310: v2i32 = ARMISD::VDUP 0x7ff59a11da10 [ID=30]
          0x7ff59a11da10: i32,ch = CopyFromReg 0x7ff599c4b598, 0x7ff59a11df10
[ORD=10] [ID=23]
            0x7ff59a11df10: i32 = Register %vreg3 [ORD=10] [ID=3]
        0x7ff59a122a10: v2i32 = bitcast 0x7ff59a11bf10 [ID=39]
          0x7ff59a11bf10: f64,ch = load 0x7ff599c4b598, 0x7ff59a120d10,
0x7ff59a11bc10<LD8[ConstantPool]> [ID=34]
            0x7ff59a120d10: i32 = ARMISD::Wrapper 0x7ff59a11d710 [ID=27]
              0x7ff59a11d710: i32 = TargetConstantPool<<2 x i32> <i32 0, i32
1>> 0 [ID=20]
            0x7ff59a11bc10: i32 = undef [ORD=18] [ID=7]

Is the TargetLowering/TargetTransformInfo code going wrong?

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/20130302/7deadfcd/attachment.html>

More information about the llvm-bugs mailing list