[ARM] Add MVE mul patterns

[dmgreen]

Summary

This simply adds integer and floating point VMUL patterns for MVE.

[samparker]

LGTM

llvm/lib/Target/ARM/ARMInstrMVE.td

class MVE_VMULt1<string suffix, bits<2> size, list<dag> pattern=[]>
   let Inst{4} = 0b1;
   let Inst{0} = 0b0;
 }
 
 def MVE_VMULt1i8  : MVE_VMULt1<"i8", 0b00>;
 def MVE_VMULt1i16 : MVE_VMULt1<"i16", 0b01>;
 def MVE_VMULt1i32 : MVE_VMULt1<"i32", 0b10>;
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(v16i8 (mul (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
+            (v16i8 (MVE_VMULt1i8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
+  def : Pat<(v8i16 (mul (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
+            (v8i16 (MVE_VMULt1i16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
+  def : Pat<(v4i32 (mul (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
+            (v4i32 (MVE_VMULt1i32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
+}
+
 class MVE_VQxDMULH<string iname, string suffix, bits<2> size, bit rounding,
                   list<dag> pattern=[]>
   : MVE_int<iname, suffix, size, pattern> {
 
   let Inst{28} = rounding;
   let Inst{25-23} = 0b110;
   let Inst{16} = 0b0;
   let Inst{12-8} = 0b01011;
class MVE_VMUL_fp<string suffix, bit size, list<dag> pattern=[]>
   let Inst{12-8} = 0b01101;
   let Inst{7} = Qn{3};
   let Inst{4} = 0b1;
 }
 
 def MVE_VMULf32 : MVE_VMUL_fp<"f32", 0b0>;
 def MVE_VMULf16 : MVE_VMUL_fp<"f16", 0b1>;
 
+let Predicates = [HasMVEFloat] in {
+  def : Pat<(v4f32 (fmul (v4f32 MQPR:$val1), (v4f32 MQPR:$val2))),
+            (v4f32 (MVE_VMULf32 (v4f32 MQPR:$val1), (v4f32 MQPR:$val2)))>;
+  def : Pat<(v8f16 (fmul (v8f16 MQPR:$val1), (v8f16 MQPR:$val2))),
+            (v8f16 (MVE_VMULf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2)))>;
+}
+
 class MVE_VCMLA<string suffix, bit size, list<dag> pattern=[]>
   : MVEFloatArithNeon<"vcmla", suffix, size, (outs MQPR:$Qd),
                          (ins MQPR:$Qd_src, MQPR:$Qn, MQPR:$Qm, complexrotateop:$rot),
                          "$Qd, $Qn, $Qm, $rot", vpred_n, "$Qd = $Qd_src", pattern> {
   bits<4> Qd;
   bits<4> Qn;
   bits<2> rot;
 

llvm/test/CodeGen/Thumb2/mve-simple-arith.ll

 ; CHECK-MVEFP-LABEL: sub_float16_t:
 ; CHECK-MVEFP:       @ %bb.0: @ %entry
 ; CHECK-MVEFP-NEXT:    vsub.f16 q0, q1, q0
 ; CHECK-MVEFP-NEXT:    bx lr
 entry:
   %0 = fsub nnan ninf nsz <8 x half> %src2, %src1
   ret <8 x half> %0
 }
+
+define arm_aapcs_vfpcc <16 x i8> @mul_int8_t(<16 x i8> %src1, <16 x i8> %src2) {
+; CHECK-LABEL: mul_int8_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.i8 q0, q0, q1
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = mul <16 x i8> %src1, %src2
+  ret <16 x i8> %0
+}
+
+define arm_aapcs_vfpcc <8 x i16> @mul_int16_t(<8 x i16> %src1, <8 x i16> %src2) {
+; CHECK-LABEL: mul_int16_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = mul <8 x i16> %src1, %src2
+  ret <8 x i16> %0
+}
+
+define arm_aapcs_vfpcc <4 x i32> @mul_int32_t(<4 x i32> %src1, <4 x i32> %src2) {
+; CHECK-LABEL: mul_int32_t:
+; CHECK:       @ %bb.0: @ %entry
+; CHECK-NEXT:    vmul.i32 q0, q0, q1
+; CHECK-NEXT:    bx lr
+entry:
+  %0 = mul nsw <4 x i32> %src1, %src2
+  ret <4 x i32> %0
+}
+
+define arm_aapcs_vfpcc <8 x half> @mul_float16_t(<8 x half> %src1, <8 x half> %src2) {
+; CHECK-MVE-LABEL: mul_float16_t:
+; CHECK-MVE:       @ %bb.0: @ %entry
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[0]
+; CHECK-MVE-NEXT:    vmov.u16 r1, q1[0]
+; CHECK-MVE-NEXT:    vmov s10, r1
+; CHECK-MVE-NEXT:    vmov.u16 r1, q0[1]
+; CHECK-MVE-NEXT:    vmov s8, r0
+; CHECK-MVE-NEXT:    movs r2, #0
+; CHECK-MVE-NEXT:    vmul.f16 s8, s10, s8
+; CHECK-MVE-NEXT:    vmov r0, s8
+; CHECK-MVE-NEXT:    vmov s8, r1
+; CHECK-MVE-NEXT:    vmov.u16 r1, q1[1]
+; CHECK-MVE-NEXT:    vmov s10, r1
+; CHECK-MVE-NEXT:    vmul.f16 s8, s10, s8
+; CHECK-MVE-NEXT:    vmov r1, s8
+; CHECK-MVE-NEXT:    vdup.16 q2, r2
+; CHECK-MVE-NEXT:    vmov.16 q2[0], r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[2]
+; CHECK-MVE-NEXT:    vmov s12, r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q1[2]
+; CHECK-MVE-NEXT:    vmov s14, r0
+; CHECK-MVE-NEXT:    vmov.16 q2[1], r1
+; CHECK-MVE-NEXT:    vmul.f16 s12, s14, s12
+; CHECK-MVE-NEXT:    vmov r0, s12
+; CHECK-MVE-NEXT:    vmov.16 q2[2], r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[3]
+; CHECK-MVE-NEXT:    vmov s12, r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q1[3]
+; CHECK-MVE-NEXT:    vmov s14, r0
+; CHECK-MVE-NEXT:    vmul.f16 s12, s14, s12
+; CHECK-MVE-NEXT:    vmov r0, s12
+; CHECK-MVE-NEXT:    vmov.16 q2[3], r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[4]
+; CHECK-MVE-NEXT:    vmov s12, r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q1[4]
+; CHECK-MVE-NEXT:    vmov s14, r0
+; CHECK-MVE-NEXT:    vmul.f16 s12, s14, s12
+; CHECK-MVE-NEXT:    vmov r0, s12
+; CHECK-MVE-NEXT:    vmov.16 q2[4], r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[5]
+; CHECK-MVE-NEXT:    vmov s12, r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q1[5]
+; CHECK-MVE-NEXT:    vmov s14, r0
+; CHECK-MVE-NEXT:    vmul.f16 s12, s14, s12
+; CHECK-MVE-NEXT:    vmov r0, s12
+; CHECK-MVE-NEXT:    vmov.16 q2[5], r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[6]
+; CHECK-MVE-NEXT:    vmov s12, r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q1[6]
+; CHECK-MVE-NEXT:    vmov s14, r0
+; CHECK-MVE-NEXT:    vmul.f16 s12, s14, s12
+; CHECK-MVE-NEXT:    vmov r0, s12
+; CHECK-MVE-NEXT:    vmov.16 q2[6], r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q0[7]
+; CHECK-MVE-NEXT:    vmov s0, r0
+; CHECK-MVE-NEXT:    vmov.u16 r0, q1[7]
+; CHECK-MVE-NEXT:    vmov s2, r0
+; CHECK-MVE-NEXT:    vmul.f16 s0, s2, s0
+; CHECK-MVE-NEXT:    vmov r0, s0
+; CHECK-MVE-NEXT:    vmov.16 q2[7], r0
+; CHECK-MVE-NEXT:    vmov q0, q2
+; CHECK-MVE-NEXT:    bx lr
+;
+; CHECK-MVEFP-LABEL: mul_float16_t:
+; CHECK-MVEFP:       @ %bb.0: @ %entry
+; CHECK-MVEFP-NEXT:    vmul.f16 q0, q1, q0
+; CHECK-MVEFP-NEXT:    bx lr
+entry:
+  %0 = fmul nnan ninf nsz <8 x half> %src2, %src1
+  ret <8 x half> %0
+}
+
+define arm_aapcs_vfpcc <4 x float> @mul_float32_t(<4 x float> %src1, <4 x float> %src2) {
+; CHECK-MVE-LABEL: mul_float32_t:
+; CHECK-MVE:       @ %bb.0: @ %entry
+; CHECK-MVE-NEXT:    vmul.f32 s8, s4, s0
+; CHECK-MVE-NEXT:    movs r0, #0
+; CHECK-MVE-NEXT:    vmul.f32 s10, s5, s1
+; CHECK-MVE-NEXT:    vmul.f32 s12, s6, s2
+; CHECK-MVE-NEXT:    vmul.f32 s4, s7, s3
+; CHECK-MVE-NEXT:    vdup.32 q0, r0
+; CHECK-MVE-NEXT:    vmov.f32 s0, s8
+; CHECK-MVE-NEXT:    vmov.f32 s1, s10
+; CHECK-MVE-NEXT:    vmov.f32 s2, s12
+; CHECK-MVE-NEXT:    vmov.f32 s3, s4
+; CHECK-MVE-NEXT:    bx lr
+;
+; CHECK-MVEFP-LABEL: mul_float32_t:
+; CHECK-MVEFP:       @ %bb.0: @ %entry
+; CHECK-MVEFP-NEXT:    vmul.f32 q0, q1, q0
+; CHECK-MVEFP-NEXT:    bx lr
+entry:
+  %0 = fmul nnan ninf nsz <4 x float> %src2, %src1
+  ret <4 x float> %0
+}