; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -mattr=+sve,+sve2,+f8f32mm < %s -passes=msan -S | FileCheck %s
;
; Forked from llvm/test/CodeGen/AArch64/sve2-fmmla-f32mf8.ll
;
; Strictly handled:
; - llvm.aarch64.sve.fp8.fmmla.nxv4f32
;
; Heuristically handled: (none)

target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64--linux-android9001"

define dso_local <vscale x 4 x float> @fmmla_f32mf8(<vscale x 4 x float> %acc, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b, i64 noundef %fpmr) #0 sanitize_memory {
; CHECK-LABEL: define dso_local <vscale x 4 x float> @fmmla_f32mf8(
; CHECK-SAME: <vscale x 4 x float> [[ACC:%.*]], <vscale x 16 x i8> [[A:%.*]], <vscale x 16 x i8> [[B:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    call void @llvm.donothing()
; CHECK-NEXT:    [[OUT:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.fp8.fmmla.nxv4f32(<vscale x 4 x float> [[ACC]], <vscale x 16 x i8> [[A]], <vscale x 16 x i8> [[B]])
; CHECK-NEXT:    store <vscale x 4 x i32> zeroinitializer, ptr @__msan_retval_tls, align 8
; CHECK-NEXT:    ret <vscale x 4 x float> [[OUT]]
;
entry:
  %out = call <vscale x 4 x float> @llvm.aarch64.sve.fp8.fmmla.nxv4f32.nxv16i82(<vscale x 4 x float> %acc, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
  ret <vscale x 4 x float> %out
}