[ARM] Fix arm32be softfp mode miscompilation for neon sdiv

[hstk30-hw]

Related issue: #97782

[llvmbot]

@llvm/pr-subscribers-backend-arm

Author: None (hstk30-hw)

Changes

Related issue: #97782

---

Full diff: https://github.com/llvm/llvm-project/pull/97883.diff

2 Files Affected:

• (modified) llvm/lib/Target/ARM/ARMISelLowering.cpp (+1-1)
• (added) llvm/test/CodeGen/ARM/sdiv_shl.ll (+106)

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index a3fd72fe1351c..cc36e8b33472b 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -14798,7 +14798,7 @@ static SDValue PerformORCombine(SDNode *N,
                                              N0->getOperand(1),
                                              N0->getOperand(0),
                                              N1->getOperand(0));
-                return DAG.getNode(ISD::BITCAST, dl, VT, Result);
+                return DAG.getNode(ARMISD::VECTOR_REG_CAST, dl, VT, Result);
             }
         }
     }
diff --git a/llvm/test/CodeGen/ARM/sdiv_shl.ll b/llvm/test/CodeGen/ARM/sdiv_shl.ll
new file mode 100644
index 0000000000000..01615ce2c46af
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/sdiv_shl.ll
@@ -0,0 +1,106 @@
+; RUN: llc -mtriple armv7-linux -mattr=+neon %s -o - | FileCheck %s --check-prefix=LE
+; RUN: llc -mtriple armebv7-linux -mattr=+neon %s -o - | FileCheck %s --check-prefix=BE
+
+; The key is the last vrev64 should be vrev64.16 instead of vrev64.32
+
+define void @sdiv_shl(ptr %x, ptr %y) nounwind {
+; LE-LABEL: sdiv_shl:
+; LE:       @ %bb.0: @ %entry
+; LE-NEXT:    adr r2, .LCPI0_0
+; LE-NEXT:    vld1.64 {d18, d19}, [r1]
+; LE-NEXT:    adr r1, .LCPI0_1
+; LE-NEXT:    vld1.64 {d16, d17}, [r2:128]
+; LE-NEXT:    vshr.s16 q10, q9, #15
+; LE-NEXT:    vneg.s16 q8, q8
+; LE-NEXT:    vld1.64 {d22, d23}, [r1:128]
+; LE-NEXT:    adr r1, .LCPI0_2
+; LE-NEXT:    vshl.u16 q8, q10, q8
+; LE-NEXT:    vneg.s16 q10, q11
+; LE-NEXT:    vadd.i16 q8, q9, q8
+; LE-NEXT:    vshl.s16 q8, q8, q10
+; LE-NEXT:    vld1.64 {d20, d21}, [r1:128]
+; LE-NEXT:    vbit q8, q9, q10
+; LE-NEXT:    vst1.64 {d16, d17}, [r0]
+; LE:         .LCPI0_0:
+; LE-NEXT:    .short 16 @ 0x10
+; LE-NEXT:    .short 14 @ 0xe
+; LE-NEXT:    .short 15 @ 0xf
+; LE-NEXT:    .short 13 @ 0xd
+; LE-NEXT:    .short 12 @ 0xc
+; LE-NEXT:    .short 10 @ 0xa
+; LE-NEXT:    .short 11 @ 0xb
+; LE-NEXT:    .short 9 @ 0x9
+; LE-NEXT:  .LCPI0_1:
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 2 @ 0x2
+; LE-NEXT:    .short 1 @ 0x1
+; LE-NEXT:    .short 3 @ 0x3
+; LE-NEXT:    .short 4 @ 0x4
+; LE-NEXT:    .short 6 @ 0x6
+; LE-NEXT:    .short 5 @ 0x5
+; LE-NEXT:    .short 7 @ 0x7
+; LE-NEXT:  .LCPI0_2:
+; LE-NEXT:    .short 65535 @ 0xffff
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 0 @ 0x0
+; LE-NEXT:    .short 0 @ 0x0
+;
+; BE-LABEL: sdiv_shl:
+; BE:       @ %bb.0: @ %entry
+; BE-NEXT:    adr r2, .LCPI0_0
+; BE-NEXT:    vld1.64 {d18, d19}, [r1]
+; BE-NEXT:    adr r1, .LCPI0_1
+; BE-NEXT:    vld1.64 {d16, d17}, [r2:128]
+; BE-NEXT:    vrev64.16 q8, q8
+; BE-NEXT:    vrev64.16 q9, q9
+; BE-NEXT:    vneg.s16 q8, q8
+; BE-NEXT:    vld1.64 {d20, d21}, [r1:128]
+; BE-NEXT:    adr r1, .LCPI0_2
+; BE-NEXT:    vshr.s16 q11, q9, #15
+; BE-NEXT:    vrev64.16 q10, q10
+; BE-NEXT:    vshl.u16 q8, q11, q8
+; BE-NEXT:    vld1.64 {d22, d23}, [r1:128]
+; BE-NEXT:    vneg.s16 q10, q10
+; BE-NEXT:    vrev64.16 q11, q11
+; BE-NEXT:    vadd.i16 q8, q9, q8
+; BE-NEXT:    vshl.s16 q8, q8, q10
+; BE-NEXT:    vbit q8, q9, q11
+; BE-NEXT:    vrev64.16 q8, q8
+; BE-NEXT:    vst1.64 {d16, d17}, [r0]
+; BE:         .LCPI0_0:
+; BE-NEXT:    .short 16 @ 0x10
+; BE-NEXT:    .short 14 @ 0xe
+; BE-NEXT:    .short 15 @ 0xf
+; BE-NEXT:    .short 13 @ 0xd
+; BE-NEXT:    .short 12 @ 0xc
+; BE-NEXT:    .short 10 @ 0xa
+; BE-NEXT:    .short 11 @ 0xb
+; BE-NEXT:    .short 9 @ 0x9
+; BE-NEXT:  .LCPI0_1:
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 2 @ 0x2
+; BE-NEXT:    .short 1 @ 0x1
+; BE-NEXT:    .short 3 @ 0x3
+; BE-NEXT:    .short 4 @ 0x4
+; BE-NEXT:    .short 6 @ 0x6
+; BE-NEXT:    .short 5 @ 0x5
+; BE-NEXT:    .short 7 @ 0x7
+; BE-NEXT:  .LCPI0_2:
+; BE-NEXT:    .short 65535 @ 0xffff
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 0 @ 0x0
+; BE-NEXT:    .short 0 @ 0x0
+entry:
+  %0 = load <8 x i16>, ptr %y, align 8
+  %div = sdiv <8 x i16> %0, <i16 1, i16 4, i16 2, i16 8, i16 16, i16 64, i16 32, i16 128>
+  store <8 x i16> %div, ptr %x, align 8
+  ret void
+}

[davemgreen]

Thanks. From what I can tell this LGTM