[ARM] Add sign and zero extend patterns for MVE

[dmgreen]

Summary

The vmovlb instructions can be uses to sign or zero extend registers from one size to another. This adds some patterns for them and relevant testing. The VBICIMM generation is also put behind a hasNEON check (as is already done for VORRIMM).

Code originally by David Sherwood.

[t.p.northover]

Looks reasonable to me.

llvm/lib/Target/ARM/ARMISelLowering.cpp

static SDValue PerformANDCombine(SDNode *N,
   SelectionDAG &DAG = DCI.DAG;
 
   if(!DAG.getTargetLoweringInfo().isTypeLegal(VT))
     return SDValue();
 
   APInt SplatBits, SplatUndef;
   unsigned SplatBitSize;
   bool HasAnyUndefs;
-  if (BVN &&
+  if (BVN && Subtarget->hasNEON() &&
       BVN->isConstantSplat(SplatBits, SplatUndef, SplatBitSize, HasAnyUndefs)) {
     if (SplatBitSize <= 64) {
       EVT VbicVT;
       SDValue Val = isNEONModifiedImm((~SplatBits).getZExtValue(),
                                       SplatUndef.getZExtValue(), SplatBitSize,
                                       DAG, dl, VbicVT, VT.is128BitVector(),
                                       OtherModImm);
       if (Val.getNode()) {

llvm/lib/Target/ARM/ARMInstrMVE.td

multiclass MVE_VMOVL_shift_half<string iname, string suffix, bits<2> sz, bit U,
   }
 }
 
 defm MVE_VMOVLs8 : MVE_VMOVL_shift_half<"vmovl", "s8", 0b01, 0b0>;
 defm MVE_VMOVLu8 : MVE_VMOVL_shift_half<"vmovl", "u8", 0b01, 0b1>;
 defm MVE_VMOVLs16 : MVE_VMOVL_shift_half<"vmovl", "s16", 0b10, 0b0>;
 defm MVE_VMOVLu16 : MVE_VMOVL_shift_half<"vmovl", "u16", 0b10, 0b1>;
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(sext_inreg (v4i32 MQPR:$src), v4i16),
+            (MVE_VMOVLs16bh MQPR:$src)>;
+  def : Pat<(sext_inreg (v8i16 MQPR:$src), v8i8),
+            (MVE_VMOVLs8bh MQPR:$src)>;
+  def : Pat<(sext_inreg (v4i32 MQPR:$src), v4i8),
+            (MVE_VMOVLs16bh (MVE_VMOVLs8bh MQPR:$src))>;
+
+  // zext_inreg 16 -> 32
+  def : Pat<(and (v4i32 MQPR:$src), (v4i32 (ARMvmovImm (i32 0xCFF)))),
+            (MVE_VMOVLu16bh MQPR:$src)>;
+  // zext_inreg 8 -> 16
+  def : Pat<(and (v8i16 MQPR:$src), (v8i16 (ARMvmovImm (i32 0x8FF)))),
+            (MVE_VMOVLu8bh MQPR:$src)>;
+}
+
+
 class MVE_VSHLL_imm<string iname, string suffix, bit U, bit th,
                     dag immops, list<dag> pattern=[]>
   : MVE_shift_imm<(outs MQPR:$Qd), !con((ins MQPR:$Qm), immops),
                   iname, suffix, "$Qd, $Qm, $imm", vpred_r, "", pattern> {
   let Inst{28} = U;
   let Inst{25-23} = 0b101;
   let Inst{21} = 0b1;
   let Inst{12} = th;

llvm/test/CodeGen/Thumb2/mve-sext.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 <8 x i16> @sext_v8i8_v8i16(<8 x i8> %src) {
+; CHECK-LABEL: sext_v8i8_v8i16:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmovlb.s8 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = sext <8 x i8> %src to <8 x i16>
+  ret <8 x i16> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @sext_v4i16_v4i32(<4 x i16> %src) {
+; CHECK-LABEL: sext_v4i16_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmovlb.s16 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = sext <4 x i16> %src to <4 x i32>
+  ret <4 x i32> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @sext_v4i8_v4i32(<4 x i8> %src) {
+; CHECK-LABEL: sext_v4i8_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmovlb.s8 q0, q0
+; CHECK-NEXT:    vmovlb.s16 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = sext <4 x i8> %src to <4 x i32>
+  ret <4 x i32> %0
+}
+
+
+define arm_aapcs_vfpcc <8 x i16> @zext_v8i8_v8i16(<8 x i8> %src) {
+; CHECK-LABEL: zext_v8i8_v8i16:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmovlb.u8 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = zext <8 x i8> %src to <8 x i16>
+  ret <8 x i16> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @zext_v4i16_v4i32(<4 x i16> %src) {
+; CHECK-LABEL: zext_v4i16_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmovlb.u16 q0, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = zext <4 x i16> %src to <4 x i32>
+  ret <4 x i32> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @zext_v4i8_v4i32(<4 x i8> %src) {
+; CHECK-LABEL: zext_v4i8_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmov.i32 q1, #0xff
+; CHECK-NEXT:    vand q0, q0, q1
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = zext <4 x i8> %src to <4 x i32>
+  ret <4 x i32> %0
+}
+
+
+define arm_aapcs_vfpcc <8 x i8> @trunc_v8i16_v8i8(<8 x i16> %src) {
+; CHECK-LABEL: trunc_v8i16_v8i8:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = trunc <8 x i16> %src to <8 x i8>
+  ret <8 x i8> %0
+}
+
+define arm_aapcs_vfpcc <4 x i16> @trunc_v4i32_v4i16(<4 x i32> %src) {
+; CHECK-LABEL: trunc_v4i32_v4i16:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = trunc <4 x i32> %src to <4 x i16>
+  ret <4 x i16> %0
+}
+
+define arm_aapcs_vfpcc <4 x i8> @trunc_v4i32_v4i8(<4 x i32> %src) {
+; CHECK-LABEL: trunc_v4i32_v4i8:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = trunc <4 x i32> %src to <4 x i8>
+  ret <4 x i8> %0
+}