; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 5
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S | FileCheck %s -check-prefix=NO-ZVFBFMIN
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S -prefer-predicate-over-epilogue=scalar-epilogue | FileCheck %s -check-prefix=NO-ZVFBFMIN
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v -S -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue | FileCheck %s -check-prefix=NO-ZVFBFMIN
; RUN: opt < %s -passes=loop-vectorize -mtriple riscv64 -mattr=+v,+zvfbfmin -S | FileCheck %s -check-prefix=ZVFBFMIN

define void @fadd(ptr noalias %a, ptr noalias %b, i64 %n) {
; NO-ZVFBFMIN-LABEL: define void @fadd(
; NO-ZVFBFMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; NO-ZVFBFMIN-NEXT:  [[ENTRY:.*]]:
; NO-ZVFBFMIN-NEXT:    br label %[[LOOP:.*]]
; NO-ZVFBFMIN:       [[LOOP]]:
; NO-ZVFBFMIN-NEXT:    [[I:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
; NO-ZVFBFMIN-NEXT:    [[A_GEP:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[I]]
; NO-ZVFBFMIN-NEXT:    [[B_GEP:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[I]]
; NO-ZVFBFMIN-NEXT:    [[X:%.*]] = load bfloat, ptr [[A_GEP]], align 2
; NO-ZVFBFMIN-NEXT:    [[Y:%.*]] = load bfloat, ptr [[B_GEP]], align 2
; NO-ZVFBFMIN-NEXT:    [[Z:%.*]] = fadd bfloat [[X]], [[Y]]
; NO-ZVFBFMIN-NEXT:    store bfloat [[Z]], ptr [[A_GEP]], align 2
; NO-ZVFBFMIN-NEXT:    [[I_NEXT]] = add i64 [[I]], 1
; NO-ZVFBFMIN-NEXT:    [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]
; NO-ZVFBFMIN-NEXT:    br i1 [[DONE]], label %[[EXIT:.*]], label %[[LOOP]]
; NO-ZVFBFMIN:       [[EXIT]]:
; NO-ZVFBFMIN-NEXT:    ret void
;
; ZVFBFMIN-LABEL: define void @fadd(
; ZVFBFMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; ZVFBFMIN-NEXT:  [[ENTRY:.*:]]
; ZVFBFMIN-NEXT:    br label %[[VECTOR_PH:.*]]
; ZVFBFMIN:       [[VECTOR_PH]]:
; ZVFBFMIN-NEXT:    br label %[[VECTOR_BODY:.*]]
; ZVFBFMIN:       [[VECTOR_BODY]]:
; ZVFBFMIN-NEXT:    [[TMP0:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; ZVFBFMIN-NEXT:    [[AVL:%.*]] = phi i64 [ [[N]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; ZVFBFMIN-NEXT:    [[TMP6:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 8, i1 true)
; ZVFBFMIN-NEXT:    [[TMP1:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[TMP0]]
; ZVFBFMIN-NEXT:    [[TMP2:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[TMP0]]
; ZVFBFMIN-NEXT:    [[WIDE_LOAD:%.*]] = call <vscale x 8 x bfloat> @llvm.vp.load.nxv8bf16.p0(ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])
; ZVFBFMIN-NEXT:    [[WIDE_LOAD1:%.*]] = call <vscale x 8 x bfloat> @llvm.vp.load.nxv8bf16.p0(ptr align 2 [[TMP2]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])
; ZVFBFMIN-NEXT:    [[TMP11:%.*]] = fadd <vscale x 8 x bfloat> [[WIDE_LOAD]], [[WIDE_LOAD1]]
; ZVFBFMIN-NEXT:    call void @llvm.vp.store.nxv8bf16.p0(<vscale x 8 x bfloat> [[TMP11]], ptr align 2 [[TMP1]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]])
; ZVFBFMIN-NEXT:    [[TMP13:%.*]] = zext i32 [[TMP6]] to i64
; ZVFBFMIN-NEXT:    [[INDEX_EVL_NEXT]] = add i64 [[TMP13]], [[TMP0]]
; ZVFBFMIN-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP13]]
; ZVFBFMIN-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
; ZVFBFMIN-NEXT:    br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; ZVFBFMIN:       [[MIDDLE_BLOCK]]:
; ZVFBFMIN-NEXT:    br label %[[EXIT:.*]]
; ZVFBFMIN:       [[SCALAR_PH:.*]]:
; ZVFBFMIN-NEXT:    br label %[[LOOP:.*]]
; ZVFBFMIN:       [[LOOP]]:
; ZVFBFMIN-NEXT:    [[I:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
; ZVFBFMIN-NEXT:    [[A_GEP:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[I]]
; ZVFBFMIN-NEXT:    [[B_GEP:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[I]]
; ZVFBFMIN-NEXT:    [[X:%.*]] = load bfloat, ptr [[A_GEP]], align 2
; ZVFBFMIN-NEXT:    [[Y:%.*]] = load bfloat, ptr [[B_GEP]], align 2
; ZVFBFMIN-NEXT:    [[Z:%.*]] = fadd bfloat [[X]], [[Y]]
; ZVFBFMIN-NEXT:    store bfloat [[Z]], ptr [[A_GEP]], align 2
; ZVFBFMIN-NEXT:    [[I_NEXT]] = add i64 [[I]], 1
; ZVFBFMIN-NEXT:    [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]
; ZVFBFMIN-NEXT:    br i1 [[DONE]], label %[[EXIT]], label %[[LOOP]]
; ZVFBFMIN:       [[EXIT]]:
; ZVFBFMIN-NEXT:    ret void
;
entry:
  br label %loop
loop:
  %i = phi i64 [0, %entry], [%i.next, %loop]
  %a.gep = getelementptr bfloat, ptr %a, i64 %i
  %b.gep = getelementptr bfloat, ptr %b, i64 %i
  %x = load bfloat, ptr %a.gep
  %y = load bfloat, ptr %b.gep
  %z = fadd bfloat %x, %y
  store bfloat %z, ptr %a.gep
  %i.next = add i64 %i, 1
  %done = icmp eq i64 %i.next, %n
  br i1 %done, label %exit, label %loop
exit:
  ret void
}

define void @vfwmaccbf16.vv(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %n) {
; NO-ZVFBFMIN-LABEL: define void @vfwmaccbf16.vv(
; NO-ZVFBFMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; NO-ZVFBFMIN-NEXT:  [[ENTRY:.*]]:
; NO-ZVFBFMIN-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 4
; NO-ZVFBFMIN-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; NO-ZVFBFMIN:       [[VECTOR_PH]]:
; NO-ZVFBFMIN-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 4
; NO-ZVFBFMIN-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
; NO-ZVFBFMIN-NEXT:    br label %[[VECTOR_BODY:.*]]
; NO-ZVFBFMIN:       [[VECTOR_BODY]]:
; NO-ZVFBFMIN-NEXT:    [[TMP0:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; NO-ZVFBFMIN-NEXT:    [[TMP1:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[TMP0]]
; NO-ZVFBFMIN-NEXT:    [[TMP2:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[TMP0]]
; NO-ZVFBFMIN-NEXT:    [[TMP3:%.*]] = getelementptr float, ptr [[C]], i64 [[TMP0]]
; NO-ZVFBFMIN-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x bfloat>, ptr [[TMP1]], align 2
; NO-ZVFBFMIN-NEXT:    [[WIDE_LOAD1:%.*]] = load <4 x bfloat>, ptr [[TMP2]], align 2
; NO-ZVFBFMIN-NEXT:    [[WIDE_LOAD2:%.*]] = load <4 x float>, ptr [[TMP3]], align 4
; NO-ZVFBFMIN-NEXT:    [[TMP7:%.*]] = fpext <4 x bfloat> [[WIDE_LOAD]] to <4 x float>
; NO-ZVFBFMIN-NEXT:    [[TMP8:%.*]] = fpext <4 x bfloat> [[WIDE_LOAD1]] to <4 x float>
; NO-ZVFBFMIN-NEXT:    [[TMP9:%.*]] = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> [[TMP7]], <4 x float> [[TMP8]], <4 x float> [[WIDE_LOAD2]])
; NO-ZVFBFMIN-NEXT:    store <4 x float> [[TMP9]], ptr [[TMP3]], align 4
; NO-ZVFBFMIN-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[TMP0]], 4
; NO-ZVFBFMIN-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; NO-ZVFBFMIN-NEXT:    br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; NO-ZVFBFMIN:       [[MIDDLE_BLOCK]]:
; NO-ZVFBFMIN-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; NO-ZVFBFMIN-NEXT:    br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; NO-ZVFBFMIN:       [[SCALAR_PH]]:
; NO-ZVFBFMIN-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
; NO-ZVFBFMIN-NEXT:    br label %[[LOOP:.*]]
; NO-ZVFBFMIN:       [[LOOP]]:
; NO-ZVFBFMIN-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
; NO-ZVFBFMIN-NEXT:    [[A_GEP:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[I]]
; NO-ZVFBFMIN-NEXT:    [[B_GEP:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[I]]
; NO-ZVFBFMIN-NEXT:    [[C_GEP:%.*]] = getelementptr float, ptr [[C]], i64 [[I]]
; NO-ZVFBFMIN-NEXT:    [[X:%.*]] = load bfloat, ptr [[A_GEP]], align 2
; NO-ZVFBFMIN-NEXT:    [[Y:%.*]] = load bfloat, ptr [[B_GEP]], align 2
; NO-ZVFBFMIN-NEXT:    [[Z:%.*]] = load float, ptr [[C_GEP]], align 4
; NO-ZVFBFMIN-NEXT:    [[X_EXT:%.*]] = fpext bfloat [[X]] to float
; NO-ZVFBFMIN-NEXT:    [[Y_EXT:%.*]] = fpext bfloat [[Y]] to float
; NO-ZVFBFMIN-NEXT:    [[FMULADD:%.*]] = call float @llvm.fmuladd.f32(float [[X_EXT]], float [[Y_EXT]], float [[Z]])
; NO-ZVFBFMIN-NEXT:    store float [[FMULADD]], ptr [[C_GEP]], align 4
; NO-ZVFBFMIN-NEXT:    [[I_NEXT]] = add i64 [[I]], 1
; NO-ZVFBFMIN-NEXT:    [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]
; NO-ZVFBFMIN-NEXT:    br i1 [[DONE]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; NO-ZVFBFMIN:       [[EXIT]]:
; NO-ZVFBFMIN-NEXT:    ret void
;
; ZVFBFMIN-LABEL: define void @vfwmaccbf16.vv(
; ZVFBFMIN-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; ZVFBFMIN-NEXT:  [[ENTRY:.*:]]
; ZVFBFMIN-NEXT:    br label %[[VECTOR_PH:.*]]
; ZVFBFMIN:       [[VECTOR_PH]]:
; ZVFBFMIN-NEXT:    br label %[[VECTOR_BODY:.*]]
; ZVFBFMIN:       [[VECTOR_BODY]]:
; ZVFBFMIN-NEXT:    [[TMP6:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; ZVFBFMIN-NEXT:    [[AVL:%.*]] = phi i64 [ [[N]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; ZVFBFMIN-NEXT:    [[TMP11:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 4, i1 true)
; ZVFBFMIN-NEXT:    [[TMP7:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[TMP6]]
; ZVFBFMIN-NEXT:    [[TMP8:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[TMP6]]
; ZVFBFMIN-NEXT:    [[TMP9:%.*]] = getelementptr float, ptr [[C]], i64 [[TMP6]]
; ZVFBFMIN-NEXT:    [[WIDE_LOAD:%.*]] = call <vscale x 4 x bfloat> @llvm.vp.load.nxv4bf16.p0(ptr align 2 [[TMP7]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP11]])
; ZVFBFMIN-NEXT:    [[WIDE_LOAD1:%.*]] = call <vscale x 4 x bfloat> @llvm.vp.load.nxv4bf16.p0(ptr align 2 [[TMP8]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP11]])
; ZVFBFMIN-NEXT:    [[WIDE_LOAD2:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[TMP9]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP11]])
; ZVFBFMIN-NEXT:    [[TMP13:%.*]] = fpext <vscale x 4 x bfloat> [[WIDE_LOAD]] to <vscale x 4 x float>
; ZVFBFMIN-NEXT:    [[TMP14:%.*]] = fpext <vscale x 4 x bfloat> [[WIDE_LOAD1]] to <vscale x 4 x float>
; ZVFBFMIN-NEXT:    [[TMP15:%.*]] = call <vscale x 4 x float> @llvm.fmuladd.nxv4f32(<vscale x 4 x float> [[TMP13]], <vscale x 4 x float> [[TMP14]], <vscale x 4 x float> [[WIDE_LOAD2]])
; ZVFBFMIN-NEXT:    call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[TMP15]], ptr align 4 [[TMP9]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP11]])
; ZVFBFMIN-NEXT:    [[TMP12:%.*]] = zext i32 [[TMP11]] to i64
; ZVFBFMIN-NEXT:    [[INDEX_EVL_NEXT]] = add i64 [[TMP12]], [[TMP6]]
; ZVFBFMIN-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP12]]
; ZVFBFMIN-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
; ZVFBFMIN-NEXT:    br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; ZVFBFMIN:       [[MIDDLE_BLOCK]]:
; ZVFBFMIN-NEXT:    br label %[[EXIT:.*]]
; ZVFBFMIN:       [[SCALAR_PH:.*]]:
; ZVFBFMIN-NEXT:    br label %[[LOOP:.*]]
; ZVFBFMIN:       [[LOOP]]:
; ZVFBFMIN-NEXT:    [[I:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
; ZVFBFMIN-NEXT:    [[A_GEP:%.*]] = getelementptr bfloat, ptr [[A]], i64 [[I]]
; ZVFBFMIN-NEXT:    [[B_GEP:%.*]] = getelementptr bfloat, ptr [[B]], i64 [[I]]
; ZVFBFMIN-NEXT:    [[C_GEP:%.*]] = getelementptr float, ptr [[C]], i64 [[I]]
; ZVFBFMIN-NEXT:    [[X:%.*]] = load bfloat, ptr [[A_GEP]], align 2
; ZVFBFMIN-NEXT:    [[Y:%.*]] = load bfloat, ptr [[B_GEP]], align 2
; ZVFBFMIN-NEXT:    [[Z:%.*]] = load float, ptr [[C_GEP]], align 4
; ZVFBFMIN-NEXT:    [[X_EXT:%.*]] = fpext bfloat [[X]] to float
; ZVFBFMIN-NEXT:    [[Y_EXT:%.*]] = fpext bfloat [[Y]] to float
; ZVFBFMIN-NEXT:    [[FMULADD:%.*]] = call float @llvm.fmuladd.f32(float [[X_EXT]], float [[Y_EXT]], float [[Z]])
; ZVFBFMIN-NEXT:    store float [[FMULADD]], ptr [[C_GEP]], align 4
; ZVFBFMIN-NEXT:    [[I_NEXT]] = add i64 [[I]], 1
; ZVFBFMIN-NEXT:    [[DONE:%.*]] = icmp eq i64 [[I_NEXT]], [[N]]
; ZVFBFMIN-NEXT:    br i1 [[DONE]], label %[[EXIT]], label %[[LOOP]]
; ZVFBFMIN:       [[EXIT]]:
; ZVFBFMIN-NEXT:    ret void
;
entry:
  br label %loop
loop:
  %i = phi i64 [0, %entry], [%i.next, %loop]
  %a.gep = getelementptr bfloat, ptr %a, i64 %i
  %b.gep = getelementptr bfloat, ptr %b, i64 %i
  %c.gep = getelementptr float, ptr %c, i64 %i
  %x = load bfloat, ptr %a.gep
  %y = load bfloat, ptr %b.gep
  %z = load float, ptr %c.gep
  %x.ext = fpext bfloat %x to float
  %y.ext = fpext bfloat %y to float
  %fmuladd = call float @llvm.fmuladd.f32(float %x.ext, float %y.ext, float %z)
  store float %fmuladd, ptr %c.gep
  %i.next = add i64 %i, 1
  %done = icmp eq i64 %i.next, %n
  br i1 %done, label %exit, label %loop
exit:
  ret void
}