[ARM] Add patterns for bitreverse intrinsic on MVE

[oliverlars]

Summary

BITREVERSE can use the VBRSR which will reverse and right shift. Shifting right by 0 will just reverse the bits.

[oliverlars]

Added -verify-machineinstrs and a test for 2i64

[oliverlars]

added test for 2i64 and fixed vbrsr not taking a register operand

[dmgreen]

LGTM. Thanks

llvm/test/CodeGen/Thumb2/mve-bitreverse.ll

10

This one is nice too, that it works this out.

[oliverlars]

rGe9510a6cadb1

llvm/lib/Target/ARM/ARMISelLowering.cpp

for (auto VT : IntTypes) {
     setOperationAction(ISD::SMIN, VT, Legal);
     setOperationAction(ISD::SMAX, VT, Legal);
     setOperationAction(ISD::UMIN, VT, Legal);
     setOperationAction(ISD::UMAX, VT, Legal);
     setOperationAction(ISD::ABS, VT, Legal);
     setOperationAction(ISD::SETCC, VT, Custom);
     setOperationAction(ISD::CTLZ, VT, Legal);
     setOperationAction(ISD::CTTZ, VT, Custom);
+    setOperationAction(ISD::BITREVERSE, VT, Legal);
 
     // No native support for these.
     setOperationAction(ISD::UDIV, VT, Expand);
     setOperationAction(ISD::SDIV, VT, Expand);
     setOperationAction(ISD::UREM, VT, Expand);
     setOperationAction(ISD::SREM, VT, Expand);
     setOperationAction(ISD::CTPOP, VT, Expand);
 

llvm/lib/Target/ARM/ARMInstrMVE.td

class MVE_VBRSR<string iname, string suffix, bits<2> size, list<dag> pattern=[]>
   let Inst{8} = 0b0;
   let Inst{5} = 0b1;
 }
 
 def MVE_VBRSR8  : MVE_VBRSR<"vbrsr", "8", 0b00>;
 def MVE_VBRSR16 : MVE_VBRSR<"vbrsr", "16", 0b01>;
 def MVE_VBRSR32 : MVE_VBRSR<"vbrsr", "32", 0b10>;
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(v16i8 ( bitreverse (v16i8 MQPR:$val1))),
+            (v16i8 ( MVE_VBRSR8 (v16i8 MQPR:$val1), 0 ))>;
+
+  def : Pat<(v4i32 ( bitreverse (v4i32 MQPR:$val1))),
+            (v4i32 ( MVE_VBRSR32 (v4i32 MQPR:$val1), 0 ))>;
+
+  def : Pat<(v8i16 ( bitreverse (v8i16 MQPR:$val1))),
+            (v8i16 ( MVE_VBRSR16 (v8i16 MQPR:$val1), 0 ))>;
+}
+
 class MVE_VMUL_qr_int<string iname, string suffix,
                       bits<2> size, list<dag> pattern=[]>
   : MVE_qDest_rSrc<iname, suffix, pattern> {
 
   let Inst{28} = 0b0;
   let Inst{21-20} = size;
   let Inst{16} = 0b1;
   let Inst{12} = 0b1;

llvm/test/CodeGen/Thumb2/mve-bitreverse.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -verify-machineinstrs -mattr=+mve %s -o - | FileCheck %s
+
+define arm_aapcs_vfpcc <2 x i64> @brv_2i64_t(<2 x i64> %src){
+; CHECK-LABEL: brv_2i64_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vrev64.8 q1, q0
+; CHECK-NEXT:    vbrsr.8 q0, q1, #0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = call <2 x i64> @llvm.bitreverse.v2i64(<2 x i64> %src)
+  ret <2 x i64> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @brv_4i32_t(<4 x i32> %src){
+; CHECK-LABEL: brv_4i32_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vbrsr.32 q0, q0, #0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = call <4 x i32> @llvm.bitreverse.v4i32(<4 x i32> %src)
+  ret <4 x i32> %0
+}
+
+define arm_aapcs_vfpcc <8 x i16> @brv_8i16_t(<8 x i16> %src){
+; CHECK-LABEL: brv_8i16_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vbrsr.16 q0, q0, #0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = call <8 x i16> @llvm.bitreverse.v8i16(<8 x i16> %src)
+  ret <8 x i16> %0
+}
+
+define arm_aapcs_vfpcc <16 x i8> @brv_16i8_t(<16 x i8> %src){
+; CHECK-LABEL: brv_16i8_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vbrsr.8 q0, q0, #0
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = call <16 x i8> @llvm.bitreverse.v16i8(<16 x i8> %src)
+  ret <16 x i8> %0
+}
+
+declare <2 x i64> @llvm.bitreverse.v2i64(<2 x i64>)
+declare <4 x i32> @llvm.bitreverse.v4i32(<4 x i32>)
+declare <8 x i16> @llvm.bitreverse.v8i16(<8 x i16>)
+declare <16 x i8> @llvm.bitreverse.v16i8(<16 x i8>)
+