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

[oliverlars]

Summary

Added support for VMUL to use an r register, this reduces pressure on the q registers.

[dmgreen]

Nice. Can you add tests for the %src and %sp being the other way around in the mul too? I think the same patterns should catch them, but it's worth having the tests.

[oliverlars]

added tests for swapping of %sp and %src

[dmgreen]

Thanks. LGTM

[oliverlars]

rG9bf27928e17a

llvm/lib/Target/ARM/ARMInstrMVE.td

class MVE_VMUL_qr_int<string iname, string suffix,
   let Inst{8} = 0b0;
   let Inst{5} = 0b1;
 }
 
 def MVE_VMUL_qr_i8  : MVE_VMUL_qr_int<"vmul", "i8",  0b00>;
 def MVE_VMUL_qr_i16 : MVE_VMUL_qr_int<"vmul", "i16", 0b01>;
 def MVE_VMUL_qr_i32 : MVE_VMUL_qr_int<"vmul", "i32", 0b10>;
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(v16i8 (mul (v16i8 MQPR:$val1), (v16i8 (ARMvdup GPR:$val2)))),
+            (v16i8 (MVE_VMUL_qr_i8 (v16i8 MQPR:$val1), (i32 GPR:$val2)))>;
+  def : Pat<(v8i16 (mul (v8i16 MQPR:$val1), (v8i16 (ARMvdup GPR:$val2)))),
+            (v8i16 (MVE_VMUL_qr_i16 (v8i16 MQPR:$val1), (i32 GPR:$val2)))>;
+  def : Pat<(v4i32 (mul (v4i32 MQPR:$val1), (v4i32 (ARMvdup GPR:$val2)))),
+            (v4i32 (MVE_VMUL_qr_i32 (v4i32 MQPR:$val1), (i32 GPR:$val2)))>;
+}
+
 class MVE_VxxMUL_qr<string iname, string suffix,
                     bit bit_28, bits<2> bits_21_20, list<dag> pattern=[]>
   : MVE_qDest_rSrc<iname, suffix, pattern> {
 
   let Inst{28} = bit_28;
   let Inst{21-20} = bits_21_20;
   let Inst{16} = 0b1;
   let Inst{12} = 0b0;

llvm/test/CodeGen/Thumb2/mve-vmulqr.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> @vmulqr_v4i32(<4 x i32> %src, i32 %src2, <4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: vmulqr_v4i32:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.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 = mul <4 x i32> %src, %sp
+  ret <4 x i32> %c
+}
+
+define arm_aapcs_vfpcc <8 x i16> @vmulqr_v8i16(<8 x i16> %src, i16 %src2, <8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: vmulqr_v8i16:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.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 = mul <8 x i16> %src, %sp
+  ret <8 x i16> %c
+}
+
+define arm_aapcs_vfpcc <16 x i8> @vmulqr_v16i8(<16 x i8> %src, i8 %src2, <16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: vmulqr_v16i8:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.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 = mul <16 x i8> %src, %sp
+  ret <16 x i8> %c
+}
+
+define arm_aapcs_vfpcc <4 x i32> @vmulqr_v4i32_2(<4 x i32> %src, i32 %src2, <4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: vmulqr_v4i32_2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.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 = mul <4 x i32> %sp, %src
+  ret <4 x i32> %c
+}
+
+define arm_aapcs_vfpcc <8 x i16> @vmulqr_v8i16_2(<8 x i16> %src, i16 %src2, <8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: vmulqr_v8i16_2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.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 = mul <8 x i16> %sp, %src
+  ret <8 x i16> %c
+}
+
+define arm_aapcs_vfpcc <16 x i8> @vmulqr_v16i8_2(<16 x i8> %src, i8 %src2, <16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: vmulqr_v16i8_2:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.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 = mul <16 x i8> %sp, %src
+  ret <16 x i8> %c
+}