[ARM] Fix VMOVSR pattern requirements

The required target features for VMOVSR instruction (integer -> MVE
FP register copy) are HasFPRegs and UseVMOVSR (see
https://reviews.llvm.org/D60694).

The instruction selection pattern must use the same set of
requirements (i.e., HasFPRegs instead of HasVFP2). The discrepancy
can cause instruction selection failure when compiling code for
execute-only position-independence mode.

Reviewed By: dmgreen

Differential Revision: https://reviews.llvm.org/D147523

diff --git a/llvm/lib/Target/ARM/ARMInstrVFP.td b/llvm/lib/Target/ARM/ARMInstrVFP.td
index c1fecf3..5d940cc 100644 (file)
--- a/llvm/lib/Target/ARM/ARMInstrVFP.td
+++ b/llvm/lib/Target/ARM/ARMInstrVFP.td
@@ -1201,7 +1201,7 @@ def VMOVSR : AVConv4I<0b11100000, 0b1010,
   let D = VFPNeonDomain;
 }
 } // isMoveReg
-def : Pat<(arm_vmovsr GPR:$Rt), (VMOVSR GPR:$Rt)>, Requires<[HasVFP2, UseVMOVSR]>;
+def : Pat<(arm_vmovsr GPR:$Rt), (VMOVSR GPR:$Rt)>, Requires<[HasFPRegs, UseVMOVSR]>;
 
 let hasSideEffects = 0 in {
 def VMOVRRD  : AVConv3I<0b11000101, 0b1011,

diff --git a/llvm/test/CodeGen/Thumb2/float-ops-vmovsr-nofp-xo.ll b/llvm/test/CodeGen/Thumb2/float-ops-vmovsr-nofp-xo.ll
new file mode 100644 (file)
index 0000000..a71cd44
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/float-ops-vmovsr-nofp-xo.ll
@@ -0,0 +1,30 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc < %s -mtriple=thumbv8.1m.main-none-eabihf -mattr=+mve,+execute-only | FileCheck %s
+
+declare void @ext_fn(float)
+
+define void @test1() {
+; CHECK-LABEL: test1:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    .save {r7, lr}
+; CHECK-NEXT:    push {r7, lr}
+; CHECK-NEXT:    movs r0, #0
+; CHECK-NEXT:    vmov s0, r0
+; CHECK-NEXT:    bl ext_fn
+; CHECK-NEXT:    pop {r7, pc}
+entry:
+  call void @ext_fn(float 0.000000e+00)
+  ret void
+}
+
+define float @test2() {
+; CHECK-LABEL: test2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    movw r0, #63911
+; CHECK-NEXT:    movt r0, #16673
+; CHECK-NEXT:    vmov s0, r0
+; CHECK-NEXT:    bx lr
+entry:
+  ret float 0x40243F34E0000000
+}
+