[ARM] MVE VNeg instruction patterns

[dmgreen]

Summary

This selects vneg instructions, which can be especially useful with shifts.

[SjoerdMeijer]

LGTM

llvm/trunk/lib/Target/ARM/ARMInstrMVE.td

let Predicates = [HasMVEInt] in {
   def : Pat<(v4i32 (abs (v4i32 MQPR:$v))),
             (v4i32 (MVE_VABSs32 $v))>;
 }
 
 def MVE_VNEGs8  : MVE_VABSNEG_int<"vneg", "s8",  0b00, 0b1>;
 def MVE_VNEGs16 : MVE_VABSNEG_int<"vneg", "s16", 0b01, 0b1>;
 def MVE_VNEGs32 : MVE_VABSNEG_int<"vneg", "s32", 0b10, 0b1>;
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(v16i8 (vnegq (v16i8 MQPR:$v))),
+            (v16i8 (MVE_VNEGs8 $v))>;
+  def : Pat<(v8i16 (vnegq (v8i16 MQPR:$v))),
+            (v8i16 (MVE_VNEGs16 $v))>;
+  def : Pat<(v4i32 (vnegq (v4i32 MQPR:$v))),
+            (v4i32 (MVE_VNEGs32 $v))>;
+}
+
 class MVE_VQABSNEG<string iname, string suffix, bits<2> size,
                    bit negate, list<dag> pattern=[]>
   : MVEIntSingleSrc<iname, suffix, size, pattern> {
 
   let Inst{28} = 0b1;
   let Inst{25-23} = 0b111;
   let Inst{21-20} = 0b11;
   let Inst{17-16} = 0b00;

llvm/trunk/test/CodeGen/Thumb2/mve-neg.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s
+
+define arm_aapcs_vfpcc <16 x i8> @neg_v16i8(<16 x i8> %s1) {
+; CHECK-LABEL: neg_v16i8:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vneg.s8 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = sub nsw <16 x i8> zeroinitializer, %s1
+  ret <16 x i8> %0
+}
+
+define arm_aapcs_vfpcc <8 x i16> @neg_v8i16(<8 x i16> %s1) {
+; CHECK-LABEL: neg_v8i16:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vneg.s16 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = sub nsw <8 x i16> zeroinitializer, %s1
+  ret <8 x i16> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @neg_v4i32(<4 x i32> %s1) {
+; CHECK-LABEL: neg_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vneg.s32 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = sub nsw <4 x i32> zeroinitializer, %s1
+  ret <4 x i32> %0
+}