[RISCV] Add isAsCheapAsAMove to FLI instructions.

[craig.topper]

Summary

This can prevent unnecessarily hoisting out of loops.

Test case cribbed from AArch64.

I also intend to make them rematerializable.

[reames]

LGTM

llvm/lib/Target/RISCV/RISCVInstrInfoZfa.td

class FPUnaryOp_r_rtz<bits<7> funct7, bits<5> rs2val, DAGOperand rdty,
   let rs2 = rs2val;
 }
 
 //===----------------------------------------------------------------------===//
 // Instructions
 //===----------------------------------------------------------------------===//
 
 let Predicates = [HasStdExtZfa] in {
+let isAsCheapAsAMove = 1 in
 def FLI_S : FPUnaryOp_imm<0b1111000, 0b00001, 0b000, OPC_OP_FP, (outs FPR32:$rd),
             (ins loadfp32imm:$imm), "fli.s", "$rd, $imm">;
 
 def FMINM_S: FPALU_rr<0b0010100, 0b010, "fminm.s", FPR32, /*Commutable*/ 1>;
 def FMAXM_S: FPALU_rr<0b0010100, 0b011, "fmaxm.s", FPR32, /*Commutable*/ 1>;
 
 def FROUND_S : FPUnaryOp_r_frm<0b0100000, 0b00100, FPR32, FPR32, "fround.s">;
 def FROUNDNX_S : FPUnaryOp_r_frm<0b0100000, 0b00101, FPR32, FPR32, "froundnx.s">;
 
 def FLTQ_S : FPCmp_rr<0b1010000, 0b101, "fltq.s", FPR32>;
 def FLEQ_S : FPCmp_rr<0b1010000, 0b100, "fleq.s", FPR32>;
 } // Predicates = [HasStdExtZfa]
 
 let Predicates = [HasStdExtZfa, HasStdExtD] in {
+let isAsCheapAsAMove = 1 in
 def FLI_D : FPUnaryOp_imm<0b1111001, 0b00001, 0b000, OPC_OP_FP, (outs FPR64:$rd),
             (ins loadfp64imm:$imm), "fli.d", "$rd, $imm">;
 
 def FMINM_D: FPALU_rr<0b0010101, 0b010, "fminm.d", FPR64, /*Commutable*/ 1>;
 def FMAXM_D: FPALU_rr<0b0010101, 0b011, "fmaxm.d", FPR64, /*Commutable*/ 1>;
 
 def FROUND_D : FPUnaryOp_r_frm<0b0100001, 0b00100, FPR64, FPR64, "fround.d">;
 def FROUNDNX_D : FPUnaryOp_r_frm<0b0100001, 0b00101, FPR64, FPR64, "froundnx.d">;
def FMVP_D_X : FPBinaryOp_rr<0b1011001, 0b000, FPR64, GPR, "fmvp.d.x">,
                Sched<[WriteFMovI32ToF32, ReadFMovI32ToF32]>;
 let isCodeGenOnly = 1, mayRaiseFPException = 0 in
 def FMV_X_W_FPR64 : FPUnaryOp_r<0b1110000, 0b00000, 0b000, GPR, FPR64,
                                 "fmv.x.w">,
                     Sched<[WriteFMovF32ToI32, ReadFMovF32ToI32]>;
 } // Predicates = [HasStdExtZfa, HasStdExtD, IsRV32]
 
 let Predicates = [HasStdExtZfa, HasStdExtZfhOrZvfh] in
+let isAsCheapAsAMove = 1 in
 def FLI_H : FPUnaryOp_imm<0b1111010, 0b00001, 0b000, OPC_OP_FP, (outs FPR16:$rd),
             (ins loadfp16imm:$imm), "fli.h", "$rd, $imm">;
 
 let Predicates = [HasStdExtZfa, HasStdExtZfh] in {
 def FMINM_H: FPALU_rr<0b0010110, 0b010, "fminm.h", FPR16, /*Commutable*/ 1>;
 def FMAXM_H: FPALU_rr<0b0010110, 0b011, "fmaxm.h", FPR16, /*Commutable*/ 1>;
 
 def FROUND_H : FPUnaryOp_r_frm<0b0100010, 0b00100, FPR16, FPR16, "fround.h">;

llvm/test/CodeGen/RISCV/fli-licm.ll

 %struct.Node = type { ptr, i8* }
 
 define void @process_nodes(ptr %0) nounwind {
 ; RV32-LABEL: process_nodes:
 ; RV32:       # %bb.0: # %entry
 ; RV32-NEXT:    addi sp, sp, -16
 ; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
 ; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
-; RV32-NEXT:    fsw fs0, 4(sp) # 4-byte Folded Spill
 ; RV32-NEXT:    beqz a0, .LBB0_3
 ; RV32-NEXT:  # %bb.1: # %loop.preheader
 ; RV32-NEXT:    mv s0, a0
-; RV32-NEXT:    fli.s fs0, 1.0
 ; RV32-NEXT:  .LBB0_2: # %loop
 ; RV32-NEXT:    # =>This Inner Loop Header: Depth=1
-; RV32-NEXT:    fmv.s fa0, fs0
+; RV32-NEXT:    fli.s fa0, 1.0
 ; RV32-NEXT:    mv a0, s0
 ; RV32-NEXT:    call do_it@plt
 ; RV32-NEXT:    lw s0, 0(s0)
 ; RV32-NEXT:    bnez s0, .LBB0_2
 ; RV32-NEXT:  .LBB0_3: # %exit
 ; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
-; RV32-NEXT:    flw fs0, 4(sp) # 4-byte Folded Reload
 ; RV32-NEXT:    addi sp, sp, 16
 ; RV32-NEXT:    ret
 ;
 ; RV64-LABEL: process_nodes:
 ; RV64:       # %bb.0: # %entry
-; RV64-NEXT:    addi sp, sp, -32
-; RV64-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
-; RV64-NEXT:    sd s0, 16(sp) # 8-byte Folded Spill
-; RV64-NEXT:    fsw fs0, 12(sp) # 4-byte Folded Spill
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
 ; RV64-NEXT:    beqz a0, .LBB0_3
 ; RV64-NEXT:  # %bb.1: # %loop.preheader
 ; RV64-NEXT:    mv s0, a0
-; RV64-NEXT:    fli.s fs0, 1.0
 ; RV64-NEXT:  .LBB0_2: # %loop
 ; RV64-NEXT:    # =>This Inner Loop Header: Depth=1
-; RV64-NEXT:    fmv.s fa0, fs0
+; RV64-NEXT:    fli.s fa0, 1.0
 ; RV64-NEXT:    mv a0, s0
 ; RV64-NEXT:    call do_it@plt
 ; RV64-NEXT:    ld s0, 0(s0)
 ; RV64-NEXT:    bnez s0, .LBB0_2
 ; RV64-NEXT:  .LBB0_3: # %exit
-; RV64-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
-; RV64-NEXT:    ld s0, 16(sp) # 8-byte Folded Reload
-; RV64-NEXT:    flw fs0, 12(sp) # 4-byte Folded Reload
-; RV64-NEXT:    addi sp, sp, 32
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64-NEXT:    addi sp, sp, 16
 ; RV64-NEXT:    ret
 entry:
   %1 = icmp eq ptr %0, null
   br i1 %1, label %exit, label %loop
 
 loop:
   %2 = phi %struct.Node* [ %4, %loop ], [ %0, %entry ]
   tail call void @do_it(float 1.000000e+00, ptr nonnull %2)