[ARM] Make v2f62 scalar_to_vector legal

[dmgreen]

Summary

Because we mark all operations as expand for v2f64, scalar_to_vector would end up lowering through a stack store/reload. But it is pretty simple to implement, only inserting a D reg into an undef vector. This helps clear up some inefficient codegen from soft calling conventions.

[simon_tatham]

You've got a typo in the subject line :-) v2f64, not v2f62.

[dmgreen]

Ah, yep. Thanks

llvm/lib/Target/ARM/ARMISelLowering.cpp

void ARMTargetLowering::addMVEVectorTypes(bool HasMVEFP) {
   const MVT LongTypes[] = { MVT::v2i64, MVT::v2f64 };
   for (auto VT : LongTypes) {
     addRegisterClass(VT, &ARM::MQPRRegClass);
     setAllExpand(VT);
     setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
     setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
   }
+  setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v2f64, Legal);
+
   // We can do bitwise operations on v2i64 vectors
   setOperationAction(ISD::AND, MVT::v2i64, Legal);
   setOperationAction(ISD::OR, MVT::v2i64, Legal);
   setOperationAction(ISD::XOR, MVT::v2i64, Legal);
 
   // It is legal to extload from v4i8 to v4i16 or v4i32.
   addAllExtLoads(MVT::v8i16, MVT::v8i8, Legal);
   addAllExtLoads(MVT::v4i32, MVT::v4i16, Legal);

llvm/lib/Target/ARM/ARMInstrMVE.td

def : Pat<(insertelt (v8f16 MQPR:$src1), (f16 HPR:$src2), imm:$lane),
             (MVE_VMOV_to_lane_16 MQPR:$src1, (COPY_TO_REGCLASS (f16 HPR:$src2), rGPR), imm:$lane)>;
   def : Pat<(extractelt (v8f16 MQPR:$src), imm_even:$lane),
             (EXTRACT_SUBREG MQPR:$src, (SSubReg_f16_reg imm_even:$lane))>;
   def : Pat<(extractelt (v8f16 MQPR:$src), imm_odd:$lane),
             (COPY_TO_REGCLASS
               (VMOVH (EXTRACT_SUBREG MQPR:$src, (SSubReg_f16_reg imm_odd:$lane))),
               HPR)>;
 
+  def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
+            (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
   def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
             (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
   def : Pat<(v4f32 (scalar_to_vector GPR:$src)),
             (MVE_VMOV_to_lane_32 (v4f32 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;
   def : Pat<(v8f16 (scalar_to_vector (f16 HPR:$src))),
             (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), (f16 HPR:$src), ssub_0)>;
   def : Pat<(v8f16 (scalar_to_vector GPR:$src)),
             (MVE_VMOV_to_lane_16 (v8f16 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;

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

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -verify-machineinstrs %s -o - | FileCheck %s
 
 define half @extret_f16_sf(<8 x half> %a, <8 x half> %b, half* nocapture %p) {
 ; CHECK-LABEL: extret_f16_sf:
 ; CHECK:       @ %bb.0:
-; CHECK-NEXT:    .save {r4, r6, r7, lr}
-; CHECK-NEXT:    push {r4, r6, r7, lr}
-; CHECK-NEXT:    .setfp r7, sp, #8
-; CHECK-NEXT:    add r7, sp, #8
-; CHECK-NEXT:    .pad #16
-; CHECK-NEXT:    sub sp, #16
-; CHECK-NEXT:    mov r4, sp
-; CHECK-NEXT:    bfc r4, #0, #4
-; CHECK-NEXT:    mov sp, r4
-; CHECK-NEXT:    strd r0, r1, [sp]
-; CHECK-NEXT:    add.w r0, r7, #8
-; CHECK-NEXT:    mov r1, sp
-; CHECK-NEXT:    vldrw.u32 q0, [r0]
-; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    sub.w r4, r7, #8
-; CHECK-NEXT:    ldr r0, [r7, #24]
-; CHECK-NEXT:    vadd.f16 q0, q1, q0
+; CHECK-NEXT:    vmov d0, r0, r1
+; CHECK-NEXT:    mov r0, sp
+; CHECK-NEXT:    vldrw.u32 q1, [r0]
+; CHECK-NEXT:    ldr r0, [sp, #16]
+; CHECK-NEXT:    vadd.f16 q0, q0, q1
 ; CHECK-NEXT:    vmovx.f16 s0, s0
 ; CHECK-NEXT:    vstr.16 s0, [r0]
 ; CHECK-NEXT:    vmov r0, s0
-; CHECK-NEXT:    mov sp, r4
-; CHECK-NEXT:    pop {r4, r6, r7, pc}
+; CHECK-NEXT:    bx lr
   %c = fadd <8 x half> %a, %b
   %e = extractelement <8 x half> %c, i32 1
   store half %e, half* %p, align 2
   ret half %e
 }
 
 define arm_aapcs_vfpcc half @extret_f16_hf(<8 x half> %a, <8 x half> %b, half* nocapture %p) {
 ; CHECK-LABEL: extret_f16_hf:
 ; CHECK:       @ %bb.0:
 ; CHECK-NEXT:    vadd.f16 q0, q0, q1
 ; CHECK-NEXT:    vmovx.f16 s0, s0
 ; CHECK-NEXT:    vstr.16 s0, [r0]
 ; CHECK-NEXT:    bx lr
   %c = fadd <8 x half> %a, %b
   %e = extractelement <8 x half> %c, i32 1
   store half %e, half* %p, align 2
   ret half %e
 }
 
 define float @extret_f32_sf(<4 x float> %a, <4 x float> %b, float* nocapture %p) {
 ; CHECK-LABEL: extret_f32_sf:
 ; CHECK:       @ %bb.0:
-; CHECK-NEXT:    .save {r4, r6, r7, lr}
-; CHECK-NEXT:    push {r4, r6, r7, lr}
-; CHECK-NEXT:    .setfp r7, sp, #8
-; CHECK-NEXT:    add r7, sp, #8
-; CHECK-NEXT:    .pad #16
-; CHECK-NEXT:    sub sp, #16
-; CHECK-NEXT:    mov r4, sp
-; CHECK-NEXT:    bfc r4, #0, #4
-; CHECK-NEXT:    mov sp, r4
-; CHECK-NEXT:    strd r0, r1, [sp]
-; CHECK-NEXT:    add.w r0, r7, #8
-; CHECK-NEXT:    mov r1, sp
-; CHECK-NEXT:    vldrw.u32 q0, [r0]
-; CHECK-NEXT:    vldrw.u32 q1, [r1]
-; CHECK-NEXT:    ldr r1, [r7, #24]
-; CHECK-NEXT:    sub.w r4, r7, #8
-; CHECK-NEXT:    vadd.f32 q0, q1, q0
+; CHECK-NEXT:    vmov d0, r0, r1
+; CHECK-NEXT:    mov r0, sp
+; CHECK-NEXT:    vldrw.u32 q1, [r0]
+; CHECK-NEXT:    ldr r1, [sp, #16]
+; CHECK-NEXT:    vadd.f32 q0, q0, q1
 ; CHECK-NEXT:    vmov r0, s1
 ; CHECK-NEXT:    str r0, [r1]
-; CHECK-NEXT:    mov sp, r4
-; CHECK-NEXT:    pop {r4, r6, r7, pc}
+; CHECK-NEXT:    bx lr
   %c = fadd <4 x float> %a, %b
   %e = extractelement <4 x float> %c, i32 1
   store float %e, float* %p, align 2
   ret float %e
 }
 
 define arm_aapcs_vfpcc float @extret_f32_hf(<4 x float> %a, <4 x float> %b, float* nocapture %p) {
 ; CHECK-LABEL: extret_f32_hf:

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

 entry:
   %res = insertelement <8 x half> undef, half %a, i32 0
   ret <8 x half> %res
 }
 
 define arm_aapcs_vfpcc <2 x double> @insert_f64(double %a) {
 ; CHECK-LABEL: insert_f64:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    .save {r4, r6, r7, lr}
-; CHECK-NEXT:    push {r4, r6, r7, lr}
-; CHECK-NEXT:    .setfp r7, sp, #8
-; CHECK-NEXT:    add r7, sp, #8
-; CHECK-NEXT:    .pad #16
-; CHECK-NEXT:    sub sp, #16
-; CHECK-NEXT:    mov r4, sp
-; CHECK-NEXT:    bfc r4, #0, #4
-; CHECK-NEXT:    mov sp, r4
-; CHECK-NEXT:    sub.w r4, r7, #8
-; CHECK-NEXT:    vstr d0, [sp]
-; CHECK-NEXT:    mov r0, sp
-; CHECK-NEXT:    vldrw.u32 q0, [r0]
-; CHECK-NEXT:    mov sp, r4
-; CHECK-NEXT:    pop {r4, r6, r7, pc}
+; CHECK-NEXT:    @ kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT:    bx lr
 entry:
   %res = insertelement <2 x double> undef, double %a, i32 0
   ret <2 x double> %res
 }
 
 define arm_aapcs_vfpcc i64 @scalar_to_vector_i32(<8 x i16> %v) {
 ; CHECK-LABEL: scalar_to_vector_i32:
 ; CHECK:       @ %bb.0: @ %entry