Mark FMOV constant materialization as being as cheap as a move.

[resistor]

Summary

This prevents us from doing things like LICM'ing it out of a loop,
which is usually a net loss because we end up having to spill a
callee-saved FPR to accomodate it.

This does perturb instruction scheduling around this instruction,
so a number of tests had to be updated to account for it.

[t.p.northover]

Looks fine to me.

llvm/lib/Target/AArch64/AArch64InstrInfo.td

let usesCustomInserter = 1 in
      def CATCHRET : Pseudo<(outs), (ins am_brcond:$dst, am_brcond:$src), [(catchret bb:$dst, bb:$src)]>,
                     Sched<[]>;
 }
 
 //===----------------------------------------------------------------------===//
 // Floating point immediate move.
 //===----------------------------------------------------------------------===//
 
-let isReMaterializable = 1 in {
+let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
 defm FMOV : FPMoveImmediate<"fmov">;
 }
 
 //===----------------------------------------------------------------------===//
 // Advanced SIMD two vector instructions.
 //===----------------------------------------------------------------------===//
 
 defm UABDL   : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl",

llvm/test/CodeGen/AArch64/arm64-aapcs.ll

 }
 
 declare void @variadic(i32 %a, ...)
 
   ; Under AAPCS variadic functions have the same calling convention as
   ; others. The extra arguments should go in registers rather than on the stack.
 define void @test_variadic() {
   call void(i32, ...) @variadic(i32 0, i64 1, double 2.0)
-; CHECK: fmov d0, #2.0
 ; CHECK: mov w1, #1
+; CHECK: fmov d0, #2.0
 ; CHECK: bl variadic
   ret void
 }
 
 ; We weren't marking x7 as used after deciding that the i128 didn't fit into
 ; registers and putting the first half on the stack, so the *second* half went
 ; into x7. Yuck!
 define i128 @test_i128_shadow([7 x i64] %x0_x6, i128 %sp) {

llvm/test/CodeGen/AArch64/fmov-imm-licm.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
+
+; The purpose of this test is to check that an FMOV instruction that
+; only materializes an immediate is not MachineLICM'd out of a loop.
+; We check this in two ways: by looking for the FMOV inside the loop,
+; and also by checking that we're not spilling any FP callee-saved
+; registers.
+
+%struct.Node = type { %struct.Node*, i8* }
+
+define void @process_nodes(%struct.Node* %0) {
+; CHECK-LABEL: process_nodes:
+; CHECK-NOT:   stp {{d[0-9]+}}
+; CHECK-LABEL: .LBB0_2:
+; CHECK:       fmov s0, #1.00000000
+; CHECK:       bl do_it
+entry:
+  %1 = icmp eq %struct.Node* %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, %struct.Node* nonnull %2)
+  %3 = getelementptr inbounds %struct.Node, %struct.Node* %2, i64 0, i32 0
+  %4 = load %struct.Node*, %struct.Node** %3, align 8
+  %5 = icmp eq %struct.Node* %4, null
+  br i1 %5, label %exit, label %loop
+
+exit:
+  ret void
+}
+
+declare void @do_it(float, %struct.Node*)

llvm/test/CodeGen/AArch64/fp-cond-sel.ll

 ; CHECK-DAG: fmov s[[FLT0:[0-9]+]], wzr
 ; CHECK-DAG: fmov s[[FLT1:[0-9]+]], #1.0
 ; CHECK: fcsel {{s[0-9]+}}, s[[FLT0]], s[[FLT1]], hi
 
   %rhs64 = sext i32 %rhs32 to i64
   %tst2 = icmp sle i64 %lhs64, %rhs64
   %val2 = select i1 %tst2, double 1.0, double 0.0
   store double %val2, double* @vardouble
-; FLT0 is reused from above on ARM64.
-; CHECK: fmov d[[FLT1:[0-9]+]], #1.0
+; CHECK-DAG: fmov d[[FLT0:[0-9]+]], xzr
+; CHECK-DAG: fmov d[[FLT1:[0-9]+]], #1.0
 ; CHECK: fcsel {{d[0-9]+}}, d[[FLT1]], d[[FLT0]], le
 
   call void @use_float(float 0.0)
   call void @use_float(float 1.0)
 
   call void @use_double(double 0.0)
   call void @use_double(double 1.0)
 
   ret void
 ; CHECK: ret
 }

llvm/test/CodeGen/AArch64/func-calls.ll

; CHECK-LABEL: check_stack_args:
   call i32 @struct_on_stack(i8 0, i16 12, i32 42, i64 99, i128 1,
                             i32* @var32, %myStruct* byval @varstruct,
                             i32 999, double 1.0)
   ; Want to check that the final double is passed in registers and
   ; that varstruct is passed on the stack. Rather dependent on how a
   ; memcpy gets created, but the following works for now.
 
 ; CHECK-DAG: str {{q[0-9]+}}, [sp]
-; CHECK-DAG: fmov d[[FINAL_DOUBLE:[0-9]+]], #1.0
-; CHECK: mov v0.16b, v[[FINAL_DOUBLE]].16b
+; CHECK-DAG: fmov d0, #1.0
 
 ; CHECK-NONEON-DAG: str {{q[0-9]+}}, [sp]
-; CHECK-NONEON-DAG: fmov d[[FINAL_DOUBLE:[0-9]+]], #1.0
-; CHECK-NONEON: fmov d0, d[[FINAL_DOUBLE]]
+; CHECK-NONEON-DAG: fmov d0, #1.0
 
 ; CHECK: bl struct_on_stack
 ; CHECK-NOFP-NOT: fmov
 
   call void @stacked_fpu(float -1.0, double 1.0, float 4.0, float 2.0,
                          float -2.0, float -8.0, float 16.0, float 1.0,
                          float 64.0)
 

llvm/test/CodeGen/AArch64/pow.ll

; CHECK-NEXT:    b pow
   %r = call nsz ninf double @llvm.pow.f64(double %x, double 2.5e-01)
   ret double %r
 }
 
 define <4 x float> @pow_v4f32_one_fourth_not_enough_fmf(<4 x float> %x) nounwind {
 ; CHECK-LABEL: pow_v4f32_one_fourth_not_enough_fmf:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    sub sp, sp, #48 // =48
-; CHECK-NEXT:    str d8, [sp, #32] // 8-byte Folded Spill
-; CHECK-NEXT:    fmov s8, #0.25000000
 ; CHECK-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
 ; CHECK-NEXT:    mov s0, v0.s[1]
-; CHECK-NEXT:    mov v1.16b, v8.16b
-; CHECK-NEXT:    str x30, [sp, #40] // 8-byte Folded Spill
+; CHECK-NEXT:    fmov s1, #0.25000000
+; CHECK-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
 ; CHECK-NEXT:    bl powf
 ; CHECK-NEXT:    str d0, [sp] // 16-byte Folded Spill
+; CHECK-NEXT:    fmov s1, #0.25000000
 ; CHECK-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
-; CHECK-NEXT:    mov v1.16b, v8.16b
 ; CHECK-NEXT:    // kill: def $s0 killed $s0 killed $q0
 ; CHECK-NEXT:    bl powf
 ; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
 ; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
 ; CHECK-NEXT:    mov v0.s[1], v1.s[0]
 ; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
 ; CHECK-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
-; CHECK-NEXT:    mov v1.16b, v8.16b
+; CHECK-NEXT:    fmov s1, #0.25000000
 ; CHECK-NEXT:    mov s0, v0.s[2]
 ; CHECK-NEXT:    bl powf
 ; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
 ; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
 ; CHECK-NEXT:    mov v1.s[2], v0.s[0]
 ; CHECK-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
 ; CHECK-NEXT:    str q1, [sp] // 16-byte Folded Spill
-; CHECK-NEXT:    mov v1.16b, v8.16b
+; CHECK-NEXT:    fmov s1, #0.25000000
 ; CHECK-NEXT:    mov s0, v0.s[3]
 ; CHECK-NEXT:    bl powf
 ; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
-; CHECK-NEXT:    ldr x30, [sp, #40] // 8-byte Folded Reload
-; CHECK-NEXT:    ldr d8, [sp, #32] // 8-byte Folded Reload
+; CHECK-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
 ; CHECK-NEXT:    // kill: def $s0 killed $s0 def $q0
 ; CHECK-NEXT:    mov v1.s[3], v0.s[0]
 ; CHECK-NEXT:    mov v0.16b, v1.16b
 ; CHECK-NEXT:    add sp, sp, #48 // =48
 ; CHECK-NEXT:    ret
   %r = call afn nsz <4 x float> @llvm.pow.v4f32(<4 x float> %x, <4 x float> <float 2.5e-1, float 2.5e-1, float 2.5e-01, float 2.5e-01>)
   ret <4 x float> %r
 }
 
 define <2 x double> @pow_v2f64_one_fourth_not_enough_fmf(<2 x double> %x) nounwind {
 ; CHECK-LABEL: pow_v2f64_one_fourth_not_enough_fmf:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    sub sp, sp, #48 // =48
-; CHECK-NEXT:    str d8, [sp, #32] // 8-byte Folded Spill
-; CHECK-NEXT:    fmov d8, #0.25000000
 ; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
 ; CHECK-NEXT:    mov d0, v0.d[1]
-; CHECK-NEXT:    mov v1.16b, v8.16b
-; CHECK-NEXT:    str x30, [sp, #40] // 8-byte Folded Spill
+; CHECK-NEXT:    fmov d1, #0.25000000
+; CHECK-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
 ; CHECK-NEXT:    bl pow
 ; CHECK-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
+; CHECK-NEXT:    fmov d1, #0.25000000
 ; CHECK-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
-; CHECK-NEXT:    mov v1.16b, v8.16b
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
 ; CHECK-NEXT:    bl pow
 ; CHECK-NEXT:    ldr q1, [sp, #16] // 16-byte Folded Reload
-; CHECK-NEXT:    ldr x30, [sp, #40] // 8-byte Folded Reload
-; CHECK-NEXT:    ldr d8, [sp, #32] // 8-byte Folded Reload
+; CHECK-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
 ; CHECK-NEXT:    mov v0.d[1], v1.d[0]
 ; CHECK-NEXT:    add sp, sp, #48 // =48
 ; CHECK-NEXT:    ret
   %r = call nsz nnan reassoc <2 x double> @llvm.pow.v2f64(<2 x double> %x, <2 x double> <double 2.5e-1, double 2.5e-1>)
   ret <2 x double> %r
 }
 

llvm/test/CodeGen/AArch64/swifterror.ll

 ; variable number of arguments.
 declare void @llvm.va_start(i8*) nounwind
 define float @foo_vararg(%swift_error** swifterror %error_ptr_ref, ...) {
 ; CHECK-APPLE-LABEL: foo_vararg:
 ; CHECK-APPLE: mov w0, #16
 ; CHECK-APPLE: malloc
 
 ; First vararg
-; CHECK-APPLE-AARCH64: ldr {{w[0-9]+}}, [{{.*}}[[TMP:x[0-9]+]], #16]
 ; CHECK-APPLE-AARCH64: mov [[ID:w[0-9]+]], #1
+; CHECK-APPLE-AARCH64: ldr {{w[0-9]+}}, [{{.*}}[[TMP:x[0-9]+]], #16]
 ; CHECK-APPLE-AARCH64: add [[ARGS:x[0-9]+]], [[TMP]], #16
+; Third vararg
+; CHECK-APPLE-AARCH64: ldr {{w[0-9]+}}, [{{.*}}[[TMP]], #32]
 ; CHECK-APPLE-AARCH64: strb [[ID]], [x0, #8]
 ; Second vararg
 ; CHECK-APPLE-AARCH64: ldr {{w[0-9]+}}, [{{.*}}[[TMP]], #24]
-; Third vararg
-; CHECK-APPLE-AARCH64: ldr {{w[0-9]+}}, [{{.*}}[[TMP]], #32]
 
 ; CHECK-APPLE-ARM64_32: mov [[ID:w[0-9]+]], #1
 ; CHECK-APPLE-ARM64_32: add [[ARGS:x[0-9]+]], [[TMP:x[0-9]+]], #16
 ; CHECK-APPLE-ARM64_32: strb [[ID]], [x0, #8]
 
 
 entry:
   %call = call i8* @malloc(i64 16)

llvm/test/Transforms/LoopStrengthReduce/AArch64/small-constant.ll

 ;     if (x > threshold)
 ;       return x;
 ;   }
 ;   return -7;
 ; }
 define float @test1(float* nocapture readonly %arr, i64 %start, float %threshold) {
 ; CHECK-LABEL: test1:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fmov s2, #-7.00000000
 ; CHECK-NEXT:    cbz x1, .LBB0_4
 ; CHECK-NEXT:  // %bb.1: // %for.body.preheader
 ; CHECK-NEXT:    add x8, x0, #28 // =28
 ; CHECK-NEXT:  .LBB0_2: // %for.body
 ; CHECK-NEXT:    // =>This Inner Loop Header: Depth=1
 ; CHECK-NEXT:    ldr s1, [x8, x1, lsl #2]
 ; CHECK-NEXT:    fcmp s1, s0
 ; CHECK-NEXT:    b.gt .LBB0_5
 ; CHECK-NEXT:  // %bb.3: // %for.cond
 ; CHECK-NEXT:    // in Loop: Header=BB0_2 Depth=1
 ; CHECK-NEXT:    add x1, x1, #1 // =1
 ; CHECK-NEXT:    cbnz x1, .LBB0_2
 ; CHECK-NEXT:  .LBB0_4:
-; CHECK-NEXT:    mov v0.16b, v2.16b
+; CHECK-NEXT:    fmov s0, #-7.00000000
 ; CHECK-NEXT:    ret
 ; CHECK-NEXT:  .LBB0_5: // %cleanup2
 ; CHECK-NEXT:    mov v0.16b, v1.16b
 ; CHECK-NEXT:    ret
 entry:
   %cmp11 = icmp eq i64 %start, 0
   br i1 %cmp11, label %cleanup2, label %for.body
 
cleanup2:                                         ; preds = %for.cond, %for.body, %entry
   ret float %1
 }
 
 ; Same as test1, except i has another use:
 ;     if (x > threshold) ---> if (x > threshold + i)
 define float @test2(float* nocapture readonly %arr, i64 %start, float %threshold) {
 ; CHECK-LABEL: test2:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fmov s2, #-7.00000000
 ; CHECK-NEXT:    cbz x1, .LBB1_4
 ; CHECK-NEXT:  // %bb.1: // %for.body.preheader
 ; CHECK-NEXT:    add x8, x0, #28 // =28
 ; CHECK-NEXT:  .LBB1_2: // %for.body
 ; CHECK-NEXT:    // =>This Inner Loop Header: Depth=1
 ; CHECK-NEXT:    ldr s1, [x8, x1, lsl #2]
-; CHECK-NEXT:    scvtf s3, x1
-; CHECK-NEXT:    fadd s3, s3, s0
-; CHECK-NEXT:    fcmp s1, s3
+; CHECK-NEXT:    scvtf s2, x1
+; CHECK-NEXT:    fadd s2, s2, s0
+; CHECK-NEXT:    fcmp s1, s2
 ; CHECK-NEXT:    b.gt .LBB1_5
 ; CHECK-NEXT:  // %bb.3: // %for.cond
 ; CHECK-NEXT:    // in Loop: Header=BB1_2 Depth=1
 ; CHECK-NEXT:    add x1, x1, #1 // =1
 ; CHECK-NEXT:    cbnz x1, .LBB1_2
 ; CHECK-NEXT:  .LBB1_4:
-; CHECK-NEXT:    mov v0.16b, v2.16b
+; CHECK-NEXT:    fmov s0, #-7.00000000
 ; CHECK-NEXT:    ret
 ; CHECK-NEXT:  .LBB1_5: // %cleanup4
 ; CHECK-NEXT:    mov v0.16b, v1.16b
 ; CHECK-NEXT:    ret
 entry:
   %cmp14 = icmp eq i64 %start, 0
   br i1 %cmp14, label %cleanup4, label %for.body