; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=riscv32 -mattr=+m,+v %s -o - | FileCheck %s --check-prefixes=CHECK,RV32
; RUN: llc -mtriple=riscv64 -mattr=+m,+v %s -o - | FileCheck %s --check-prefixes=CHECK,RV64

; Companion to musttail-indirect-args.ll for scalable vector args.
; <vscale x 32 x i32> is too large to fit in v8-v23, so it is split into
; multiple parts and the callee receives a pointer to the parts via
; CCValAssign::Indirect. The musttail computed-arg path must scale per-part
; offsets by VSCALE -- mirrors the non-musttail spill path validated by
; calling-conv.ll's caller_scalable_vector_split_indirect.

declare <vscale x 32 x i32> @callee_musttail_scalable(<vscale x 32 x i32> %x, <vscale x 32 x i32> %y)

; Forwarded: both args are caller formals passed straight through. The
; incoming indirect pointers are forwarded zero-copy; no per-part offset
; computation happens in the caller, so VSCALE does not enter here.
define <vscale x 32 x i32> @caller_musttail_scalable_forwarded(<vscale x 32 x i32> %x, <vscale x 32 x i32> %y) nounwind {
; CHECK-LABEL: caller_musttail_scalable_forwarded:
; CHECK:       # %bb.0:
; CHECK-NEXT:    tail callee_musttail_scalable
  %r = musttail call <vscale x 32 x i32> @callee_musttail_scalable(<vscale x 32 x i32> %x, <vscale x 32 x i32> %y)
  ret <vscale x 32 x i32> %r
}

; Computed: both args are arithmetic on the formals, so dyn_cast<Argument>
; fails and we go through the store-into-incoming-pointer path. The
; second part's store must use VSCALE * PartOffset relative to
; IncomingPtr, otherwise the part lands at a fixed byte offset that
; differs from what the callee expects (vlenb * 8 above).
define <vscale x 32 x i32> @caller_musttail_scalable_computed(<vscale x 32 x i32> %x, <vscale x 32 x i32> %y) nounwind {
; CHECK-LABEL: caller_musttail_scalable_computed:
; CHECK:       # %bb.0:
; CHECK-NEXT:    csrr a1, vlenb
; CHECK-NEXT:    slli a1, a1, 3
; CHECK-NEXT:    vl8re32.v v24, (a0)
; CHECK-NEXT:    add a1, a0, a1
; CHECK-NEXT:    vl8re32.v v0, (a1)
; CHECK-NEXT:    vsetvli a2, zero, e32, m8, ta, ma
; CHECK-NEXT:    vadd.vv v8, v8, v24
; CHECK-NEXT:    vs8r.v v8, (a0)
; CHECK-NEXT:    vadd.vv v16, v16, v0
; CHECK-NEXT:    vs8r.v v16, (a1)
; CHECK-NEXT:    tail callee_musttail_scalable
  %s = add <vscale x 32 x i32> %x, %y
  %r = musttail call <vscale x 32 x i32> @callee_musttail_scalable(<vscale x 32 x i32> %s, <vscale x 32 x i32> %s)
  ret <vscale x 32 x i32> %r
}

; Mixed: first arg forwarded, second arg computed. Exercises both branches
; of the formal-arg-vs-computed path on the same call.
define <vscale x 32 x i32> @caller_musttail_scalable_mixed(<vscale x 32 x i32> %x, <vscale x 32 x i32> %y) nounwind {
; CHECK-LABEL: caller_musttail_scalable_mixed:
; CHECK:       # %bb.0:
; CHECK-NEXT:    csrr a1, vlenb
; CHECK-NEXT:    slli a1, a1, 3
; CHECK-NEXT:    vl8re32.v v24, (a0)
; CHECK-NEXT:    add a1, a0, a1
; CHECK-NEXT:    vl8re32.v v0, (a1)
; CHECK-NEXT:    vsetvli a2, zero, e32, m8, ta, ma
; CHECK-NEXT:    vadd.vv v24, v8, v24
; CHECK-NEXT:    vs8r.v v24, (a0)
; CHECK-NEXT:    vadd.vv v24, v16, v0
; CHECK-NEXT:    vs8r.v v24, (a1)
; CHECK-NEXT:    tail callee_musttail_scalable
  %s = add <vscale x 32 x i32> %x, %y
  %r = musttail call <vscale x 32 x i32> @callee_musttail_scalable(<vscale x 32 x i32> %x, <vscale x 32 x i32> %s)
  ret <vscale x 32 x i32> %r
}
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; RV32: {{.*}}
; RV64: {{.*}}