[ARM] Add patterns for VSUB with q and r registers

[oliverlars]

Summary

Added patterns for VSUB to support q and r registers, which reduces pressure on q registers.

[oliverlars]

re-added run line and added tests for swapping %src and %sp

[dmgreen]

LGTM. Nice one.

llvm/lib/Target/ARM/ARMInstrMVE.td

3628

You could combine this into the same Predicates block (it doesn't matter a lot either way). Up to you.

[oliverlars]

rGa050307c0572

llvm/lib/Target/ARM/ARMInstrMVE.td

let Predicates = [HasMVEInt] in {
   def : Pat<(v16i8 (add (v16i8 MQPR:$val1), (v16i8 (ARMvdup GPR:$val2)))),
             (v16i8 (MVE_VADD_qr_i8 (v16i8 MQPR:$val1), (i32 GPR:$val2)))>;
   def : Pat<(v8i16 (add (v8i16 MQPR:$val1), (v8i16 (ARMvdup GPR:$val2)))),
             (v8i16 (MVE_VADD_qr_i16 (v8i16 MQPR:$val1), (i32 GPR:$val2)))>;
   def : Pat<(v4i32 (add (v4i32 MQPR:$val1), (v4i32 (ARMvdup GPR:$val2)))),
             (v4i32 (MVE_VADD_qr_i32 (v4i32 MQPR:$val1), (i32 GPR:$val2)))>;
 }
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(v16i8 (sub (v16i8 MQPR:$val1), (v16i8 (ARMvdup GPR:$val2)))),
+            (v16i8 (MVE_VSUB_qr_i8 (v16i8 MQPR:$val1), (i32 GPR:$val2)))>;
+  def : Pat<(v8i16 (sub (v8i16 MQPR:$val1), (v8i16 (ARMvdup GPR:$val2)))),
+            (v8i16 (MVE_VSUB_qr_i16 (v8i16 MQPR:$val1), (i32 GPR:$val2)))>;
+  def : Pat<(v4i32 (sub (v4i32 MQPR:$val1), (v4i32 (ARMvdup GPR:$val2)))),
+            (v4i32 (MVE_VSUB_qr_i32 (v4i32 MQPR:$val1), (i32 GPR:$val2)))>;
+}
+
 class MVE_VQDMULL_qr<string iname, string suffix, bit size,
                      bit T, list<dag> pattern=[]>
   : MVE_qDest_rSrc<iname, suffix, pattern> {
 
   let Inst{28} = size;
   let Inst{21-20} = 0b11;
   let Inst{16} = 0b0;
   let Inst{12} = T;

llvm/test/CodeGen/Thumb2/mve-vsubqr.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 <4 x i32> @vsubqr_v4i32(<4 x i32> %src, i32 %src2, <4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: vsubqr_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vsub.i32 q0, q0, r0
+; CHECK-NEXT:    bx lr
+entry:
+  %i = insertelement <4 x i32> undef, i32 %src2, i32 0
+  %sp = shufflevector <4 x i32> %i, <4 x i32> undef, <4 x i32> zeroinitializer
+  %c = sub <4 x i32> %src, %sp
+  ret <4 x i32> %c
+}
+
+define arm_aapcs_vfpcc <8 x i16> @vsubqr_v8i16(<8 x i16> %src, i16 %src2, <8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: vsubqr_v8i16:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vsub.i16 q0, q0, r0
+; CHECK-NEXT:    bx lr
+entry:
+  %i = insertelement <8 x i16> undef, i16 %src2, i32 0
+  %sp = shufflevector <8 x i16> %i, <8 x i16> undef, <8 x i32> zeroinitializer
+  %c = sub <8 x i16> %src, %sp
+  ret <8 x i16> %c
+}
+
+define arm_aapcs_vfpcc <16 x i8> @vsubqr_v16i8(<16 x i8> %src, i8 %src2, <16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: vsubqr_v16i8:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vsub.i8 q0, q0, r0
+; CHECK-NEXT:    bx lr
+entry:
+  %i = insertelement <16 x i8> undef, i8 %src2, i32 0
+  %sp = shufflevector <16 x i8> %i, <16 x i8> undef, <16 x i32> zeroinitializer
+  %c = sub <16 x i8> %src, %sp
+  ret <16 x i8> %c
+}
+
+define arm_aapcs_vfpcc <4 x i32> @vsubqr_v4i32_2(<4 x i32> %src, i32 %src2, <4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: vsubqr_v4i32_2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vdup.32 q1, r0
+; CHECK-NEXT:    vsub.i32 q0, q1, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %i = insertelement <4 x i32> undef, i32 %src2, i32 0
+  %sp = shufflevector <4 x i32> %i, <4 x i32> undef, <4 x i32> zeroinitializer
+  %c = sub <4 x i32> %sp, %src
+  ret <4 x i32> %c
+}
+
+define arm_aapcs_vfpcc <8 x i16> @vsubqr_v8i16_2(<8 x i16> %src, i16 %src2, <8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: vsubqr_v8i16_2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vdup.16 q1, r0
+; CHECK-NEXT:    vsub.i16 q0, q1, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %i = insertelement <8 x i16> undef, i16 %src2, i32 0
+  %sp = shufflevector <8 x i16> %i, <8 x i16> undef, <8 x i32> zeroinitializer
+  %c = sub <8 x i16> %sp, %src
+  ret <8 x i16> %c
+}
+
+define arm_aapcs_vfpcc <16 x i8> @vsubqr_v16i8_2(<16 x i8> %src, i8 %src2, <16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: vsubqr_v16i8_2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vdup.8 q1, r0
+; CHECK-NEXT:    vsub.i8 q0, q1, q0
+; CHECK-NEXT:    bx lr
+entry:
+  %i = insertelement <16 x i8> undef, i8 %src2, i32 0
+  %sp = shufflevector <16 x i8> %i, <16 x i8> undef, <16 x i32> zeroinitializer
+  %c = sub <16 x i8> %sp, %src
+  ret <16 x i8> %c
+}