diff --git a/NEWS.md b/NEWS.md
index f9fb0835afffc9..f7f25a5b6a54b5 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -6,7 +6,10 @@ New language features
 ---------------------
 
 * `Module(:name, false, false)` can be used to create a `module` that does not import `Core`. ([#40110])
-* `@inline` and `@noinline` annotations may now be used in function bodies. ([#41312])
+* `@inline` and `@noinline` annotations can be used within a function body to give an extra
+  hint about the inlining cost to the compiler. ([#41312])
+* `@inline` and `@noinline` annotations can now be applied to a function callsite or block
+  to enforce the involved function calls to be (or not to be) inlined. ([#41312])
 * The default behavior of observing `@inbounds` declarations is now an option via `auto` in `--check-bounds=yes|no|auto` ([#41551])
 
 Language changes
@@ -39,7 +42,7 @@ New library features
 
 * `@test_throws "some message" triggers_error()` can now be used to check whether the displayed error text
   contains "some message" regardless of the specific exception type.
-  Regular expressions, lists of strings, and matching functions are also supported. ([#41888)
+  Regular expressions, lists of strings, and matching functions are also supported. ([#41888])
 
 Standard library changes
 ------------------------
diff --git a/base/compiler/abstractinterpretation.jl b/base/compiler/abstractinterpretation.jl
index 4c3f7d46fd98b7..ede05572edc3a8 100644
--- a/base/compiler/abstractinterpretation.jl
+++ b/base/compiler/abstractinterpretation.jl
@@ -592,7 +592,7 @@ function maybe_get_const_prop_profitable(interp::AbstractInterpreter, result::Me
         return nothing
     end
     mi = mi::MethodInstance
-    if !force && !const_prop_methodinstance_heuristic(interp, method, mi)
+    if !force && !const_prop_methodinstance_heuristic(interp, match, mi, sv)
         add_remark!(interp, sv, "[constprop] Disabled by method instance heuristic")
         return nothing
     end
@@ -696,7 +696,8 @@ end
 # This is a heuristic to avoid trying to const prop through complicated functions
 # where we would spend a lot of time, but are probably unlikely to get an improved
 # result anyway.
-function const_prop_methodinstance_heuristic(interp::AbstractInterpreter, method::Method, mi::MethodInstance)
+function const_prop_methodinstance_heuristic(interp::AbstractInterpreter, match::MethodMatch, mi::MethodInstance, sv::InferenceState)
+    method = match.method
     if method.is_for_opaque_closure
         # Not inlining an opaque closure can be very expensive, so be generous
         # with the const-prop-ability. It is quite possible that we can't infer
@@ -714,7 +715,8 @@ function const_prop_methodinstance_heuristic(interp::AbstractInterpreter, method
     if isdefined(code, :inferred) && !cache_inlineable
         cache_inf = code.inferred
         if !(cache_inf === nothing)
-            cache_inlineable = inlining_policy(interp)(cache_inf) !== nothing
+            src = inlining_policy(interp, cache_inf, get_curr_ssaflag(sv))
+            cache_inlineable = src !== nothing
         end
     end
     if !cache_inlineable
@@ -1908,7 +1910,9 @@ function typeinf_local(interp::AbstractInterpreter, frame::InferenceState)
                     if isa(fname, SlotNumber)
                         changes = StateUpdate(fname, VarState(Any, false), changes, false)
                     end
-                elseif hd === :inbounds || hd === :meta || hd === :loopinfo || hd === :code_coverage_effect
+                elseif hd === :code_coverage_effect ||
+                       (hd !== :boundscheck && # :boundscheck can be narrowed to Bool
+                        hd !== nothing && is_meta_expr_head(hd))
                     # these do not generate code
                 else
                     t = abstract_eval_statement(interp, stmt, changes, frame)
diff --git a/base/compiler/inferencestate.jl b/base/compiler/inferencestate.jl
index d76a4ebd18f6e0..216c397af31e44 100644
--- a/base/compiler/inferencestate.jl
+++ b/base/compiler/inferencestate.jl
@@ -296,3 +296,5 @@ function print_callstack(sv::InferenceState)
         sv = sv.parent
     end
 end
+
+get_curr_ssaflag(sv::InferenceState) = sv.src.ssaflags[sv.currpc]
diff --git a/base/compiler/optimize.jl b/base/compiler/optimize.jl
index 1898aa8b757782..09c778d57d695f 100644
--- a/base/compiler/optimize.jl
+++ b/base/compiler/optimize.jl
@@ -21,23 +21,28 @@ function push!(et::EdgeTracker, ci::CodeInstance)
     push!(et, ci.def)
 end
 
-struct InliningState{S <: Union{EdgeTracker, Nothing}, T, P}
+struct InliningState{S <: Union{EdgeTracker, Nothing}, T, I<:AbstractInterpreter}
     params::OptimizationParams
     et::S
     mi_cache::T
-    policy::P
+    interp::I
 end
 
-function default_inlining_policy(@nospecialize(src))
+function inlining_policy(interp::AbstractInterpreter, @nospecialize(src), stmt_flag::UInt8)
     if isa(src, CodeInfo) || isa(src, Vector{UInt8})
         src_inferred = ccall(:jl_ir_flag_inferred, Bool, (Any,), src)
-        src_inlineable = ccall(:jl_ir_flag_inlineable, Bool, (Any,), src)
+        src_inlineable = is_stmt_inline(stmt_flag) || ccall(:jl_ir_flag_inlineable, Bool, (Any,), src)
         return src_inferred && src_inlineable ? src : nothing
+    elseif isa(src, OptimizationState) && isdefined(src, :ir)
+        return (is_stmt_inline(stmt_flag) || src.src.inlineable) ? src.ir : nothing
+    else
+        # maybe we want to make inference keep the source in a local cache if a statement is going to inlined
+        # and re-optimize it here with disabling further inlining to avoid infinite optimization loop
+        # (we can even naively try to re-infer it entirely)
+        # but it seems like that "single-level-inlining" is more trouble and complex than it's worth
+        # see https://github.com/JuliaLang/julia/pull/41328/commits/0fc0f71a42b8c9d04b0dafabf3f1f17703abf2e7
+        return nothing
     end
-    if isa(src, OptimizationState) && isdefined(src, :ir)
-        return src.src.inlineable ? src.ir : nothing
-    end
-    return nothing
 end
 
 include("compiler/ssair/driver.jl")
@@ -57,7 +62,7 @@ mutable struct OptimizationState
         inlining = InliningState(params,
             EdgeTracker(s_edges, frame.valid_worlds),
             WorldView(code_cache(interp), frame.world),
-            inlining_policy(interp))
+            interp)
         return new(frame.linfo,
                    frame.src, nothing, frame.stmt_info, frame.mod,
                    frame.sptypes, frame.slottypes, false,
@@ -86,7 +91,7 @@ mutable struct OptimizationState
         inlining = InliningState(params,
             nothing,
             WorldView(code_cache(interp), get_world_counter()),
-            inlining_policy(interp))
+            interp)
         return new(linfo,
                    src, nothing, stmt_info, mod,
                    sptypes_from_meth_instance(linfo), slottypes, false,
@@ -125,9 +130,15 @@ const SLOT_ASSIGNEDONCE = 16 # slot is assigned to only once
 const SLOT_USEDUNDEF    = 32 # slot has uses that might raise UndefVarError
 # const SLOT_CALLED      = 64
 
-# This statement was marked as @inbounds by the user. If replaced by inlining,
-# any contained boundschecks may be removed
-const IR_FLAG_INBOUNDS       = 0x01
+# NOTE make sure to sync the flag definitions below with julia.h and `jl_code_info_set_ir` in method.c
+
+# This statement is marked as @inbounds by user.
+# Ff replaced by inlining, any contained boundschecks may be removed.
+const IR_FLAG_INBOUNDS       = 0x01 << 0
+# This statement is marked as @inline by user
+const IR_FLAG_INLINE         = 0x01 << 1
+# This statement is marked as @noinline by user
+const IR_FLAG_NOINLINE       = 0x01 << 2
 # This statement may be removed if its result is unused. In particular it must
 # thus be both pure and effect free.
 const IR_FLAG_EFFECT_FREE    = 0x01 << 4
@@ -173,6 +184,9 @@ function isinlineable(m::Method, me::OptimizationState, params::OptimizationPara
     return inlineable
 end
 
+is_stmt_inline(stmt_flag::UInt8)   = stmt_flag & IR_FLAG_INLINE   != 0
+is_stmt_noinline(stmt_flag::UInt8) = stmt_flag & IR_FLAG_NOINLINE != 0
+
 # These affect control flow within the function (so may not be removed
 # if there is no usage within the function), but don't affect the purity
 # of the function as a whole.
@@ -358,42 +372,22 @@ function convert_to_ircode(ci::CodeInfo, code::Vector{Any}, coverage::Bool, sv::
     end
     renumber_ir_elements!(code, changemap, labelmap)
 
-    inbounds_depth = 0 # Number of stacked inbounds
     meta = Any[]
-    flags = fill(0x00, length(code))
     for i = 1:length(code)
-        stmt = code[i]
-        if isexpr(stmt, :inbounds)
-            arg1 = stmt.args[1]
-            if arg1 === true # push
-                inbounds_depth += 1
-            elseif arg1 === false # clear
-                inbounds_depth = 0
-            elseif inbounds_depth > 0 # pop
-                inbounds_depth -= 1
-            end
-            stmt = nothing
-        else
-            stmt = normalize(stmt, meta)
-        end
-        code[i] = stmt
-        if !(stmt === nothing)
-            if inbounds_depth > 0
-                flags[i] |= IR_FLAG_INBOUNDS
-            end
-        end
+        code[i] = remove_meta!(code[i], meta)
     end
-    strip_trailing_junk!(ci, code, stmtinfo, flags)
+    strip_trailing_junk!(ci, code, stmtinfo)
     cfg = compute_basic_blocks(code)
     types = Any[]
-    stmts = InstructionStream(code, types, stmtinfo, ci.codelocs, flags)
+    stmts = InstructionStream(code, types, stmtinfo, ci.codelocs, ci.ssaflags)
     ir = IRCode(stmts, cfg, collect(LineInfoNode, ci.linetable::Union{Vector{LineInfoNode},Vector{Any}}), sv.slottypes, meta, sv.sptypes)
     return ir
 end
 
-function normalize(@nospecialize(stmt), meta::Vector{Any})
+function remove_meta!(@nospecialize(stmt), meta::Vector{Any})
     if isa(stmt, Expr)
-        if stmt.head === :meta
+        head = stmt.head
+        if head === :meta
             args = stmt.args
             if length(args) > 0
                 push!(meta, stmt)
diff --git a/base/compiler/ssair/inlining.jl b/base/compiler/ssair/inlining.jl
index 7e7baff1367415..c76e74e50c6d08 100644
--- a/base/compiler/ssair/inlining.jl
+++ b/base/compiler/ssair/inlining.jl
@@ -604,6 +604,7 @@ function rewrite_apply_exprargs!(ir::IRCode, todo::Vector{Pair{Int, Any}}, idx::
         argexprs::Vector{Any}, atypes::Vector{Any}, arginfos::Vector{Any},
         arg_start::Int, istate::InliningState)
 
+    flag = ir.stmts[idx][:flag]
     new_argexprs = Any[argexprs[arg_start]]
     new_atypes = Any[atypes[arg_start]]
     # loop over original arguments and flatten any known iterators
@@ -659,8 +660,8 @@ function rewrite_apply_exprargs!(ir::IRCode, todo::Vector{Pair{Int, Any}}, idx::
                 info = call.info
                 handled = false
                 if isa(info, ConstCallInfo)
-                    if maybe_handle_const_call!(ir, state1.id, new_stmt, info, new_sig,
-                        call.rt, istate, false, todo)
+                    if !is_stmt_noinline(flag) && maybe_handle_const_call!(
+                        ir, state1.id, new_stmt, info, new_sig,call.rt, istate, flag, false, todo)
                         handled = true
                     else
                         info = info.call
@@ -671,7 +672,7 @@ function rewrite_apply_exprargs!(ir::IRCode, todo::Vector{Pair{Int, Any}}, idx::
                         MethodMatchInfo[info] : info.matches
                     # See if we can inline this call to `iterate`
                     analyze_single_call!(ir, todo, state1.id, new_stmt,
-                        new_sig, call.rt, info, istate)
+                        new_sig, call.rt, info, istate, flag)
                 end
                 if i != length(thisarginfo.each)
                     valT = getfield_tfunc(call.rt, Const(1))
@@ -719,7 +720,8 @@ function compileable_specialization(et::Union{EdgeTracker, Nothing}, (; linfo)::
     return mi
 end
 
-function resolve_todo(todo::InliningTodo, state::InliningState)
+function resolve_todo(todo::InliningTodo, state::InliningState, flag::UInt8)
+    mi = todo.mi
     (; match) = todo.spec::DelayedInliningSpec
 
     #XXX: update_valid_age!(min_valid[1], max_valid[1], sv)
@@ -735,7 +737,7 @@ function resolve_todo(todo::InliningTodo, state::InliningState)
             isconst, src = false, inferred_src
         end
     else
-        linfo = get(state.mi_cache, todo.mi, nothing)
+        linfo = get(state.mi_cache, mi, nothing)
         if linfo isa CodeInstance
             if invoke_api(linfo) == 2
                 # in this case function can be inlined to a constant
@@ -751,13 +753,11 @@ function resolve_todo(todo::InliningTodo, state::InliningState)
     et = state.et
 
     if isconst && et !== nothing
-        push!(et, todo.mi)
+        push!(et, mi)
         return ConstantCase(src)
     end
 
-    if src !== nothing
-        src = state.policy(src)
-    end
+    src = inlining_policy(state.interp, src, flag)
 
     if src === nothing
         return compileable_specialization(et, match)
@@ -767,13 +767,13 @@ function resolve_todo(todo::InliningTodo, state::InliningState)
         src = copy(src)
     end
 
-    et !== nothing && push!(et, todo.mi)
-    return InliningTodo(todo.mi, src)
+    et !== nothing && push!(et, mi)
+    return InliningTodo(mi, src)
 end
 
-function resolve_todo(todo::UnionSplit, state::InliningState)
+function resolve_todo(todo::UnionSplit, state::InliningState, flag::UInt8)
     UnionSplit(todo.fully_covered, todo.atype,
-        Pair{Any,Any}[sig=>resolve_todo(item, state) for (sig, item) in todo.cases])
+        Pair{Any,Any}[sig=>resolve_todo(item, state, flag) for (sig, item) in todo.cases])
 end
 
 function validate_sparams(sparams::SimpleVector)
@@ -784,7 +784,7 @@ function validate_sparams(sparams::SimpleVector)
 end
 
 function analyze_method!(match::MethodMatch, atypes::Vector{Any},
-                         state::InliningState, @nospecialize(stmttyp))
+                         state::InliningState, @nospecialize(stmttyp), flag::UInt8)
     method = match.method
     methsig = method.sig
 
@@ -804,7 +804,7 @@ function analyze_method!(match::MethodMatch, atypes::Vector{Any},
 
     et = state.et
 
-    if !state.params.inlining
+    if !state.params.inlining || is_stmt_noinline(flag)
         return compileable_specialization(et, match)
     end
 
@@ -818,7 +818,7 @@ function analyze_method!(match::MethodMatch, atypes::Vector{Any},
     # If we don't have caches here, delay resolving this MethodInstance
     # until the batch inlining step (or an external post-processing pass)
     state.mi_cache === nothing && return todo
-    return resolve_todo(todo, state)
+    return resolve_todo(todo, state, flag)
 end
 
 function InliningTodo(mi::MethodInstance, ir::IRCode)
@@ -1043,7 +1043,7 @@ is_builtin(s::Signature) =
     s.ft ⊑ Builtin
 
 function inline_invoke!(ir::IRCode, idx::Int, sig::Signature, (; match, result)::InvokeCallInfo,
-        state::InliningState, todo::Vector{Pair{Int, Any}})
+        state::InliningState, todo::Vector{Pair{Int, Any}}, flag::UInt8)
     stmt = ir.stmts[idx][:inst]
     calltype = ir.stmts[idx][:type]
 
@@ -1057,17 +1057,17 @@ function inline_invoke!(ir::IRCode, idx::Int, sig::Signature, (; match, result):
     atypes = atypes[4:end]
     pushfirst!(atypes, atype0)
 
-    if isa(result, InferenceResult)
+    if isa(result, InferenceResult) && !is_stmt_noinline(flag)
         (; mi) = item = InliningTodo(result, atypes, calltype)
         validate_sparams(mi.sparam_vals) || return nothing
         if argtypes_to_type(atypes) <: mi.def.sig
-            state.mi_cache !== nothing && (item = resolve_todo(item, state))
+            state.mi_cache !== nothing && (item = resolve_todo(item, state, flag))
             handle_single_case!(ir, stmt, idx, item, true, todo)
             return nothing
         end
     end
 
-    result = analyze_method!(match, atypes, state, calltype)
+    result = analyze_method!(match, atypes, state, calltype, flag)
     handle_single_case!(ir, stmt, idx, result, true, todo)
     return nothing
 end
@@ -1162,7 +1162,7 @@ end
 
 function analyze_single_call!(ir::IRCode, todo::Vector{Pair{Int, Any}}, idx::Int, @nospecialize(stmt),
         sig::Signature, @nospecialize(calltype), infos::Vector{MethodMatchInfo},
-        state::InliningState)
+        state::InliningState, flag::UInt8)
     cases = Pair{Any, Any}[]
     signature_union = Union{}
     only_method = nothing  # keep track of whether there is one matching method
@@ -1196,7 +1196,7 @@ function analyze_single_call!(ir::IRCode, todo::Vector{Pair{Int, Any}}, idx::Int
                 fully_covered = false
                 continue
             end
-            case = analyze_method!(match, sig.atypes, state, calltype)
+            case = analyze_method!(match, sig.atypes, state, calltype, flag)
             if case === nothing
                 fully_covered = false
                 continue
@@ -1223,7 +1223,7 @@ function analyze_single_call!(ir::IRCode, todo::Vector{Pair{Int, Any}}, idx::Int
             match = meth[1]
         end
         fully_covered = true
-        case = analyze_method!(match, sig.atypes, state, calltype)
+        case = analyze_method!(match, sig.atypes, state, calltype, flag)
         case === nothing && return
         push!(cases, Pair{Any,Any}(match.spec_types, case))
     end
@@ -1245,7 +1245,7 @@ end
 
 function maybe_handle_const_call!(ir::IRCode, idx::Int, stmt::Expr,
         info::ConstCallInfo, sig::Signature, @nospecialize(calltype),
-        state::InliningState,
+        state::InliningState, flag::UInt8,
         isinvoke::Bool, todo::Vector{Pair{Int, Any}})
     # when multiple matches are found, bail out and later inliner will union-split this signature
     # TODO effectively use multiple constant analysis results here
@@ -1257,7 +1257,7 @@ function maybe_handle_const_call!(ir::IRCode, idx::Int, stmt::Expr,
     validate_sparams(mi.sparam_vals) || return true
     mthd_sig = mi.def.sig
     mistypes = mi.specTypes
-    state.mi_cache !== nothing && (item = resolve_todo(item, state))
+    state.mi_cache !== nothing && (item = resolve_todo(item, state, flag))
     if sig.atype <: mthd_sig
         handle_single_case!(ir, stmt, idx, item, isinvoke, todo)
         return true
@@ -1295,6 +1295,8 @@ function assemble_inline_todo!(ir::IRCode, state::InliningState)
             info = info.info
         end
 
+        flag = ir.stmts[idx][:flag]
+
         # Inference determined this couldn't be analyzed. Don't question it.
         if info === false
             continue
@@ -1304,7 +1306,8 @@ function assemble_inline_todo!(ir::IRCode, state::InliningState)
         # it'll have performed a specialized analysis for just this case. Use its
         # result.
         if isa(info, ConstCallInfo)
-            if maybe_handle_const_call!(ir, idx, stmt, info, sig, calltype, state, sig.f === Core.invoke, todo)
+            if !is_stmt_noinline(flag) && maybe_handle_const_call!(
+                ir, idx, stmt, info, sig, calltype, state, flag, sig.f === Core.invoke, todo)
                 continue
             else
                 info = info.call
@@ -1312,7 +1315,7 @@ function assemble_inline_todo!(ir::IRCode, state::InliningState)
         end
 
         if isa(info, OpaqueClosureCallInfo)
-            result = analyze_method!(info.match, sig.atypes, state, calltype)
+            result = analyze_method!(info.match, sig.atypes, state, calltype, flag)
             handle_single_case!(ir, stmt, idx, result, false, todo)
             continue
         end
@@ -1320,7 +1323,7 @@ function assemble_inline_todo!(ir::IRCode, state::InliningState)
         # Handle invoke
         if sig.f === Core.invoke
             if isa(info, InvokeCallInfo)
-                inline_invoke!(ir, idx, sig, info, state, todo)
+                inline_invoke!(ir, idx, sig, info, state, todo, flag)
             end
             continue
         end
@@ -1334,7 +1337,7 @@ function assemble_inline_todo!(ir::IRCode, state::InliningState)
             continue
         end
 
-        analyze_single_call!(ir, todo, idx, stmt, sig, calltype, infos, state)
+        analyze_single_call!(ir, todo, idx, stmt, sig, calltype, infos, state, flag)
     end
     todo
 end
@@ -1399,7 +1402,8 @@ end
 function late_inline_special_case!(ir::IRCode, sig::Signature, idx::Int, stmt::Expr, params::OptimizationParams)
     f, ft, atypes = sig.f, sig.ft, sig.atypes
     typ = ir.stmts[idx][:type]
-    if params.inlining && length(atypes) == 3 && istopfunction(f, :!==)
+    isinlining = params.inlining
+    if isinlining && length(atypes) == 3 && istopfunction(f, :!==)
         # special-case inliner for !== that precedes _methods_by_ftype union splitting
         # and that works, even though inference generally avoids inferring the `!==` Method
         if isa(typ, Const)
@@ -1411,7 +1415,7 @@ function late_inline_special_case!(ir::IRCode, sig::Signature, idx::Int, stmt::E
         not_call = Expr(:call, GlobalRef(Core.Intrinsics, :not_int), cmp_call_ssa)
         ir[SSAValue(idx)] = not_call
         return true
-    elseif params.inlining && length(atypes) == 3 && istopfunction(f, :(>:))
+    elseif isinlining && length(atypes) == 3 && istopfunction(f, :(>:))
         # special-case inliner for issupertype
         # that works, even though inference generally avoids inferring the `>:` Method
         if isa(typ, Const) && _builtin_nothrow(<:, Any[atypes[3], atypes[2]], typ)
@@ -1421,7 +1425,7 @@ function late_inline_special_case!(ir::IRCode, sig::Signature, idx::Int, stmt::E
         subtype_call = Expr(:call, GlobalRef(Core, :(<:)), stmt.args[3], stmt.args[2])
         ir[SSAValue(idx)] = subtype_call
         return true
-    elseif params.inlining && f === TypeVar && 2 <= length(atypes) <= 4 && (atypes[2] ⊑ Symbol)
+    elseif isinlining && f === TypeVar && 2 <= length(atypes) <= 4 && (atypes[2] ⊑ Symbol)
         ir[SSAValue(idx)] = Expr(:call, GlobalRef(Core, :_typevar), stmt.args[2],
             length(stmt.args) < 4 ? Bottom : stmt.args[3],
             length(stmt.args) == 2 ? Any : stmt.args[end])
diff --git a/base/compiler/ssair/slot2ssa.jl b/base/compiler/ssair/slot2ssa.jl
index 777e7e2bb0af7f..2e3d1da1c168fa 100644
--- a/base/compiler/ssair/slot2ssa.jl
+++ b/base/compiler/ssair/slot2ssa.jl
@@ -183,7 +183,7 @@ function rename_uses!(ir::IRCode, ci::CodeInfo, idx::Int, @nospecialize(stmt), r
     return fixemup!(stmt->true, stmt->renames[slot_id(stmt)], ir, ci, idx, stmt)
 end
 
-function strip_trailing_junk!(ci::CodeInfo, code::Vector{Any}, info::Vector{Any}, flags::Vector{UInt8})
+function strip_trailing_junk!(ci::CodeInfo, code::Vector{Any}, info::Vector{Any})
     # Remove `nothing`s at the end, we don't handle them well
     # (we expect the last instruction to be a terminator)
     ssavaluetypes = ci.ssavaluetypes::Vector{Any}
@@ -193,7 +193,7 @@ function strip_trailing_junk!(ci::CodeInfo, code::Vector{Any}, info::Vector{Any}
             resize!(ssavaluetypes, i)
             resize!(ci.codelocs, i)
             resize!(info, i)
-            resize!(flags, i)
+            resize!(ci.ssaflags, i)
             break
         end
     end
@@ -205,7 +205,7 @@ function strip_trailing_junk!(ci::CodeInfo, code::Vector{Any}, info::Vector{Any}
         push!(ssavaluetypes, Union{})
         push!(ci.codelocs, 0)
         push!(info, nothing)
-        push!(flags, 0x00)
+        push!(ci.ssaflags, 0x00)
     end
     nothing
 end
diff --git a/base/compiler/typeinfer.jl b/base/compiler/typeinfer.jl
index a9ab2d231c13c5..5d3325a2a5e3c6 100644
--- a/base/compiler/typeinfer.jl
+++ b/base/compiler/typeinfer.jl
@@ -343,7 +343,7 @@ function maybe_compress_codeinfo(interp::AbstractInterpreter, linfo::MethodInsta
             nslots = length(ci.slotflags)
             resize!(ci.slottypes::Vector{Any}, nslots)
             resize!(ci.slotnames, nslots)
-            return ccall(:jl_compress_ir, Any, (Any, Any), def, ci)
+            return ccall(:jl_compress_ir, Vector{UInt8}, (Any, Any), def, ci)
         else
             return ci
         end
diff --git a/base/compiler/types.jl b/base/compiler/types.jl
index 677450f288c707..5f8f6563124584 100644
--- a/base/compiler/types.jl
+++ b/base/compiler/types.jl
@@ -217,7 +217,6 @@ may_discard_trees(::AbstractInterpreter) = true
 verbose_stmt_info(::AbstractInterpreter) = false
 
 method_table(interp::AbstractInterpreter) = InternalMethodTable(get_world_counter(interp))
-inlining_policy(::AbstractInterpreter) = default_inlining_policy
 
 """
 By default `AbstractInterpreter` implements the following inference bail out logic:
diff --git a/base/compiler/utilities.jl b/base/compiler/utilities.jl
index ed09d5316473a4..8dfe1f65f0d539 100644
--- a/base/compiler/utilities.jl
+++ b/base/compiler/utilities.jl
@@ -59,7 +59,7 @@ end
 
 # Meta expression head, these generally can't be deleted even when they are
 # in a dead branch but can be ignored when analyzing uses/liveness.
-is_meta_expr_head(head::Symbol) = (head === :inbounds || head === :boundscheck || head === :meta || head === :loopinfo)
+is_meta_expr_head(head::Symbol) = head === :boundscheck || head === :meta || head === :loopinfo
 
 sym_isless(a::Symbol, b::Symbol) = ccall(:strcmp, Int32, (Ptr{UInt8}, Ptr{UInt8}), a, b) < 0
 
@@ -196,7 +196,7 @@ function specialize_method(method::Method, @nospecialize(atypes), sparams::Simpl
     if preexisting
         # check cached specializations
         # for an existing result stored there
-        return ccall(:jl_specializations_lookup, Any, (Any, Any), method, atypes)
+        return ccall(:jl_specializations_lookup, Any, (Any, Any), method, atypes)::Union{Nothing,MethodInstance}
     end
     return ccall(:jl_specializations_get_linfo, Ref{MethodInstance}, (Any, Any, Any), method, atypes, sparams)
 end
diff --git a/base/compiler/validation.jl b/base/compiler/validation.jl
index c152dfb9fa6a54..6e05c96cd79369 100644
--- a/base/compiler/validation.jl
+++ b/base/compiler/validation.jl
@@ -16,6 +16,8 @@ const VALID_EXPR_HEADS = IdDict{Symbol,UnitRange{Int}}(
     :leave => 1:1,
     :pop_exception => 1:1,
     :inbounds => 1:1,
+    :inline => 1:1,
+    :noinline => 1:1,
     :boundscheck => 0:0,
     :copyast => 1:1,
     :meta => 0:typemax(Int),
@@ -141,7 +143,7 @@ function validate_code!(errors::Vector{>:InvalidCodeError}, c::CodeInfo, is_top_
                 head === :const || head === :enter || head === :leave || head === :pop_exception ||
                 head === :method || head === :global || head === :static_parameter ||
                 head === :new || head === :splatnew || head === :thunk || head === :loopinfo ||
-                head === :throw_undef_if_not || head === :code_coverage_effect
+                head === :throw_undef_if_not || head === :code_coverage_effect || head === :inline || head === :noinline
                 validate_val!(x)
             else
                 # TODO: nothing is actually in statement position anymore
diff --git a/base/expr.jl b/base/expr.jl
index f583a58f78dc5c..1af1e9486068ea 100644
--- a/base/expr.jl
+++ b/base/expr.jl
@@ -209,9 +209,53 @@ end
 
 !!! compat "Julia 1.8"
     The usage within a function body requires at least Julia 1.8.
+
+---
+    @inline block
+
+Give a hint to the compiler that calls within `block` are worth inlining.
+
+```julia
+# The compiler will try to inline `f`
+@inline f(...)
+
+# The compiler will try to inline `f`, `g` and `+`
+@inline f(...) + g(...)
+```
+
+!!! note
+    A callsite annotation always has the precedence over the annotation applied to the
+    definition of the called function:
+    ```julia
+    @noinline function explicit_noinline(args...)
+        # body
+    end
+
+    let
+        @inline explicit_noinline(args...) # will be inlined
+    end
+    ```
+
+!!! note
+    When there are nested callsite annotations, the innermost annotation has the precedence:
+    ```julia
+    @noinline let a0, b0 = ...
+        a = @inline f(a0)  # the compiler will try to inline this call
+        b = f(b0)          # the compiler will NOT try to inline this call
+        return a, b
+    end
+    ```
+
+!!! warning
+    Although a callsite annotation will try to force inlining in regardless of the cost model,
+    there are still chances it can't succeed in it. Especially, recursive calls can not be
+    inlined even if they are annotated as `@inline`d.
+
+!!! compat "Julia 1.8"
+    The callsite annotation requires at least Julia 1.8.
 """
-macro inline(ex)
-    esc(isa(ex, Expr) ? pushmeta!(ex, :inline) : ex)
+macro inline(x)
+    return annotate_meta_def_or_block(x, :inline)
 end
 
 """
@@ -244,11 +288,51 @@ end
 !!! compat "Julia 1.8"
     The usage within a function body requires at least Julia 1.8.
 
+---
+    @noinline block
+
+Give a hint to the compiler that it should not inline the calls within `block`.
+
+```julia
+# The compiler will try to not inline `f`
+@noinline f(...)
+
+# The compiler will try to not inline `f`, `g` and `+`
+@noinline f(...) + g(...)
+```
+
+!!! note
+    A callsite annotation always has the precedence over the annotation applied to the
+    definition of the called function:
+    ```julia
+    @inline function explicit_inline(args...)
+        # body
+    end
+
+    let
+        @noinline explicit_inline(args...) # will not be inlined
+    end
+    ```
+
+!!! note
+    When there are nested callsite annotations, the innermost annotation has the precedence:
+    ```julia
+    @inline let a0, b0 = ...
+        a = @noinline f(a0)  # the compiler will NOT try to inline this call
+        b = f(b0)            # the compiler will try to inline this call
+        return a, b
+    end
+    ```
+
+!!! compat "Julia 1.8"
+    The callsite annotation requires at least Julia 1.8.
+
+---
 !!! note
     If the function is trivial (for example returning a constant) it might get inlined anyway.
 """
-macro noinline(ex)
-    esc(isa(ex, Expr) ? pushmeta!(ex, :noinline) : ex)
+macro noinline(x)
+    return annotate_meta_def_or_block(x, :noinline)
 end
 
 """
@@ -301,6 +385,15 @@ end
 
 ## some macro utilities ##
 
+unwrap_macrocalls(@nospecialize(x)) = x
+function unwrap_macrocalls(ex::Expr)
+    inner = ex
+    while inner.head === :macrocall
+        inner = inner.args[end]::Expr
+    end
+    return inner
+end
+
 function pushmeta!(ex::Expr, sym::Symbol, args::Any...)
     if isempty(args)
         tag = sym
@@ -308,10 +401,7 @@ function pushmeta!(ex::Expr, sym::Symbol, args::Any...)
         tag = Expr(sym, args...)::Expr
     end
 
-    inner = ex
-    while inner.head === :macrocall
-        inner = inner.args[end]::Expr
-    end
+    inner = unwrap_macrocalls(ex)
 
     idx, exargs = findmeta(inner)
     if idx != 0
@@ -361,8 +451,23 @@ function findmetaarg(metaargs, sym)
     return 0
 end
 
-function is_short_function_def(ex)
-    ex.head === :(=) || return false
+function annotate_meta_def_or_block(@nospecialize(ex), meta::Symbol)
+    inner = unwrap_macrocalls(ex)
+    if is_function_def(inner)
+        # annotation on a definition
+        return esc(pushmeta!(ex, meta))
+    else
+        # annotation on a block
+        return Expr(:block,
+                    Expr(meta, true),
+                    Expr(:local, Expr(:(=), :val, esc(ex))),
+                    Expr(meta, false),
+                    :val)
+    end
+end
+
+function is_short_function_def(@nospecialize(ex))
+    isexpr(ex, :(=)) || return false
     while length(ex.args) >= 1 && isa(ex.args[1], Expr)
         (ex.args[1].head === :call) && return true
         (ex.args[1].head === :where || ex.args[1].head === :(::)) || return false
@@ -370,9 +475,11 @@ function is_short_function_def(ex)
     end
     return false
 end
+is_function_def(@nospecialize(ex)) =
+    return isexpr(ex, :function) || is_short_function_def(ex) || isexpr(ex, :->)
 
 function findmeta(ex::Expr)
-    if ex.head === :function || is_short_function_def(ex) || ex.head === :->
+    if is_function_def(ex)
         body = ex.args[2]::Expr
         body.head === :block || error(body, " is not a block expression")
         return findmeta_block(ex.args)
diff --git a/base/meta.jl b/base/meta.jl
index b483630a92f8fd..649ffe9d1a19c3 100644
--- a/base/meta.jl
+++ b/base/meta.jl
@@ -440,7 +440,7 @@ function _partially_inline!(@nospecialize(x), slot_replacements::Vector{Any},
                 @assert isa(arg, Union{GlobalRef, Symbol})
                 return x
             end
-        elseif !is_meta_expr_head(head)
+        elseif !Core.Compiler.is_meta_expr_head(head)
             partially_inline!(x.args, slot_replacements, type_signature, static_param_values,
                               slot_offset, statement_offset, boundscheck)
         end
@@ -450,6 +450,4 @@ end
 
 _instantiate_type_in_env(x, spsig, spvals) = ccall(:jl_instantiate_type_in_env, Any, (Any, Any, Ptr{Any}), x, spsig, spvals)
 
-is_meta_expr_head(head::Symbol) = (head === :inbounds || head === :boundscheck || head === :meta || head === :loopinfo)
-
 end # module
diff --git a/src/ast.scm b/src/ast.scm
index bc8d847279fc91..e5148a507a4fdb 100644
--- a/src/ast.scm
+++ b/src/ast.scm
@@ -289,7 +289,7 @@
 ;; predicates and accessors
 
 (define (quoted? e)
-  (memq (car e) '(quote top core globalref outerref line break inert meta inbounds loopinfo)))
+  (memq (car e) '(quote top core globalref outerref line break inert meta inbounds inline noinline loopinfo)))
 (define (quotify e) `',e)
 (define (unquote e)
   (if (and (pair? e) (memq (car e) '(quote inert)))
diff --git a/src/codegen.cpp b/src/codegen.cpp
index 65702955691e2a..e0b86c51bf7d15 100644
--- a/src/codegen.cpp
+++ b/src/codegen.cpp
@@ -4434,7 +4434,7 @@ static void emit_stmtpos(jl_codectx_t &ctx, jl_value_t *expr, int ssaval_result)
     jl_value_t **args = (jl_value_t**)jl_array_data(ex->args);
     jl_sym_t *head = ex->head;
     if (head == meta_sym || head == inbounds_sym || head == coverageeffect_sym
-            || head == aliasscope_sym || head == popaliasscope_sym) {
+            || head == aliasscope_sym || head == popaliasscope_sym || head == inline_sym || head == noinline_sym) {
         // some expression types are metadata and can be ignored
         // in statement position
         return;
@@ -4869,7 +4869,7 @@ static jl_cgval_t emit_expr(jl_codectx_t &ctx, jl_value_t *expr, ssize_t ssaval)
     }
     else if (head == leave_sym || head == coverageeffect_sym
             || head == pop_exception_sym || head == enter_sym || head == inbounds_sym
-            || head == aliasscope_sym || head == popaliasscope_sym) {
+            || head == aliasscope_sym || head == popaliasscope_sym || head == inline_sym || head == noinline_sym) {
         jl_errorf("Expr(:%s) in value position", jl_symbol_name(head));
     }
     else if (head == boundscheck_sym) {
diff --git a/src/interpreter.c b/src/interpreter.c
index 4686efe054edf3..f999542d68c4f3 100644
--- a/src/interpreter.c
+++ b/src/interpreter.c
@@ -312,7 +312,7 @@ static jl_value_t *eval_value(jl_value_t *e, interpreter_state *s)
         return jl_true;
     }
     else if (head == meta_sym || head == coverageeffect_sym || head == inbounds_sym || head == loopinfo_sym ||
-             head == aliasscope_sym || head == popaliasscope_sym) {
+             head == aliasscope_sym || head == popaliasscope_sym || head == inline_sym || head == noinline_sym) {
         return jl_nothing;
     }
     else if (head == gc_preserve_begin_sym || head == gc_preserve_end_sym) {
diff --git a/src/julia-syntax.scm b/src/julia-syntax.scm
index f00ea0c9ba6d9d..428b0513b7e526 100644
--- a/src/julia-syntax.scm
+++ b/src/julia-syntax.scm
@@ -3498,7 +3498,7 @@ f(x) = yt(x)
          thunk with-static-parameters toplevel-only
          global globalref outerref const-if-global thismodule
          const atomic null true false ssavalue isdefined toplevel module lambda
-         error gc_preserve_begin gc_preserve_end import using export)))
+         error gc_preserve_begin gc_preserve_end import using export inline noinline)))
 
 (define (local-in? s lam)
   (or (assq s (car  (lam:vinfo lam)))
@@ -4592,7 +4592,7 @@ f(x) = yt(x)
                (cons (car e) args)))
 
             ;; metadata expressions
-            ((line meta inbounds loopinfo gc_preserve_end aliasscope popaliasscope)
+            ((line meta inbounds loopinfo gc_preserve_end aliasscope popaliasscope inline noinline)
              (let ((have-ret? (and (pair? code) (pair? (car code)) (eq? (caar code) 'return))))
                (cond ((eq? (car e) 'line)
                       (set! current-loc e)
@@ -4737,7 +4737,7 @@ f(x) = yt(x)
           (begin (set! linetable (cons (make-lineinfo name file line) linetable))
                  (set! current-loc 1)))
       (if (or reachable
-              (and (pair? e) (memq (car e) '(meta inbounds gc_preserve_begin gc_preserve_end aliasscope popaliasscope))))
+              (and (pair? e) (memq (car e) '(meta inbounds gc_preserve_begin gc_preserve_end aliasscope popaliasscope inline noinline))))
           (begin (set! code (cons e code))
                  (set! i (+ i 1))
                  (set! locs (cons current-loc locs)))))
diff --git a/src/julia.h b/src/julia.h
index e6edc71b42dfa1..e53b33bef674da 100644
--- a/src/julia.h
+++ b/src/julia.h
@@ -255,7 +255,8 @@ typedef struct _jl_code_info_t {
     jl_value_t *ssavaluetypes; // types of ssa values (or count of them)
     jl_array_t *ssaflags; // flags associated with each statement:
         // 0 = inbounds
-        // 1,2 = <reserved> inlinehint,always-inline,noinline
+        // 1 = inline
+        // 2 = noinline
         // 3 = <reserved> strict-ieee (strictfp)
         // 4 = effect-free (may be deleted if unused)
         // 5-6 = <unused>
diff --git a/src/macroexpand.scm b/src/macroexpand.scm
index 5e55c7bbb29c16..f17f4d3510dc60 100644
--- a/src/macroexpand.scm
+++ b/src/macroexpand.scm
@@ -352,7 +352,7 @@
                                    ,(resolve-expansion-vars-with-new-env (caddr arg) env m parent-scope inarg))))
                              (else
                               `(global ,(resolve-expansion-vars-with-new-env arg env m parent-scope inarg))))))
-           ((using import export meta line inbounds boundscheck loopinfo) (map unescape e))
+           ((using import export meta line inbounds boundscheck loopinfo inline noinline) (map unescape e))
            ((macrocall) e) ; invalid syntax anyways, so just act like it's quoted.
            ((symboliclabel) e)
            ((symbolicgoto) e)
diff --git a/src/method.c b/src/method.c
index 9c255f786b74da..c17566ac6f9361 100644
--- a/src/method.c
+++ b/src/method.c
@@ -84,7 +84,8 @@ static jl_value_t *resolve_globals(jl_value_t *expr, jl_module_t *module, jl_sve
             e->head == quote_sym || e->head == inert_sym ||
             e->head == meta_sym || e->head == inbounds_sym ||
             e->head == boundscheck_sym || e->head == loopinfo_sym ||
-            e->head == aliasscope_sym || e->head == popaliasscope_sym) {
+            e->head == aliasscope_sym || e->head == popaliasscope_sym ||
+            e->head == inline_sym || e->head == noinline_sym) {
             // ignore these
         }
         else {
@@ -253,6 +254,11 @@ JL_DLLEXPORT void jl_resolve_globals_in_ir(jl_array_t *stmts, jl_module_t *m, jl
     }
 }
 
+jl_value_t *expr_arg1(jl_value_t *expr) {
+    jl_array_t *args = ((jl_expr_t*)expr)->args;
+    return jl_array_ptr_ref(args, 0);
+}
+
 // copy a :lambda Expr into its CodeInfo representation,
 // including popping of known meta nodes
 static void jl_code_info_set_ir(jl_code_info_t *li, jl_expr_t *ir)
@@ -274,8 +280,17 @@ static void jl_code_info_set_ir(jl_code_info_t *li, jl_expr_t *ir)
     jl_gc_wb(li, li->code);
     size_t n = jl_array_len(body);
     jl_value_t **bd = (jl_value_t**)jl_array_ptr_data((jl_array_t*)li->code);
+    li->ssaflags = jl_alloc_array_1d(jl_array_uint8_type, n);
+    jl_gc_wb(li, li->ssaflags);
+    int inbounds_depth = 0; // number of stacked inbounds
+    // isempty(inline_flags): no user annotation
+    // last(inline_flags) == 1: inline region
+    // last(inline_flags) == 0: noinline region
+    arraylist_t *inline_flags = arraylist_new((arraylist_t*)malloc_s(sizeof(arraylist_t)), 0);
     for (j = 0; j < n; j++) {
         jl_value_t *st = bd[j];
+        int is_flag_stmt = 0;
+        // check :meta expression
         if (jl_is_expr(st) && ((jl_expr_t*)st)->head == meta_sym) {
             size_t k, ins = 0, na = jl_expr_nargs(st);
             jl_array_t *meta = ((jl_expr_t*)st)->args;
@@ -297,10 +312,60 @@ static void jl_code_info_set_ir(jl_code_info_t *li, jl_expr_t *ir)
             else
                 jl_array_del_end(meta, na - ins);
         }
+        // check other flag expressions
+        else if (jl_is_expr(st) && ((jl_expr_t*)st)->head == inbounds_sym) {
+            is_flag_stmt = 1;
+            jl_value_t *arg1 = expr_arg1(st);
+            if (arg1 == (jl_value_t*)jl_true)       // push
+                inbounds_depth += 1;
+            else if (arg1 == (jl_value_t*)jl_false) // clear
+                inbounds_depth = 0;
+            else if (inbounds_depth > 0)            // pop
+                inbounds_depth -= 1;
+            bd[j] = jl_nothing;
+        }
+        else if (jl_is_expr(st) && ((jl_expr_t*)st)->head == inline_sym) {
+            is_flag_stmt = 1;
+            jl_value_t *arg1 = expr_arg1(st);
+            if (arg1 == (jl_value_t*)jl_true) // enter inline region
+                arraylist_push(inline_flags, (void*)1);
+            else {                            // exit inline region
+                assert(arg1 == (jl_value_t*)jl_false);
+                arraylist_pop(inline_flags);
+            }
+            bd[j] = jl_nothing;
+        }
+        else if (jl_is_expr(st) && ((jl_expr_t*)st)->head == noinline_sym) {
+            is_flag_stmt = 1;
+            jl_value_t *arg1 = expr_arg1(st);
+            if (arg1 == (jl_value_t*)jl_true) // enter noinline region
+                arraylist_push(inline_flags, (void*)0);
+            else {                             // exit noinline region
+                assert(arg1 == (jl_value_t*)jl_false);
+                arraylist_pop(inline_flags);
+            }
+            bd[j] = jl_nothing;
+        }
         else if (jl_is_expr(st) && ((jl_expr_t*)st)->head == return_sym) {
             jl_array_ptr_set(body, j, jl_new_struct(jl_returnnode_type, jl_exprarg(st, 0)));
         }
+
+        if (is_flag_stmt)
+            jl_array_uint8_set(li->ssaflags, j, 0);
+        else {
+            uint8_t flag = 0;
+            if (inbounds_depth > 0)
+                flag |= 1 << 0;
+            if (inline_flags->len > 0) {
+                void* inline_flag = inline_flags->items[inline_flags->len - 1];
+                flag |= 1 << (inline_flag ? 1 : 2);
+            }
+            jl_array_uint8_set(li->ssaflags, j, flag);
+        }
     }
+    assert(inline_flags->len == 0); // malformed otherwise
+    arraylist_free(inline_flags);
+    free(inline_flags);
     jl_array_t *vinfo = (jl_array_t*)jl_exprarg(ir, 1);
     jl_array_t *vis = (jl_array_t*)jl_array_ptr_ref(vinfo, 0);
     size_t nslots = jl_array_len(vis);
@@ -313,7 +378,6 @@ static void jl_code_info_set_ir(jl_code_info_t *li, jl_expr_t *ir)
     jl_gc_wb(li, li->slotflags);
     li->ssavaluetypes = jl_box_long(nssavalue);
     jl_gc_wb(li, li->ssavaluetypes);
-    li->ssaflags = jl_alloc_array_1d(jl_array_uint8_type, 0);
 
     // Flags that need to be copied to slotflags
     const uint8_t vinfo_mask = 8 | 16 | 32 | 64;
diff --git a/test/compiler/inline.jl b/test/compiler/inline.jl
index 9358dce3325f75..a04e29e608e6e7 100644
--- a/test/compiler/inline.jl
+++ b/test/compiler/inline.jl
@@ -389,11 +389,11 @@ function isinvoke(@nospecialize(x), pred)
     end
     return false
 end
-code_typed1(args...; kwargs...) = (first∘first)(code_typed(args...; kwargs...))::Core.CodeInfo
+code_typed1(args...; kwargs...) = (first(only(code_typed(args...; kwargs...)))::Core.CodeInfo).code
 
 @testset "@inline/@noinline annotation before definition" begin
-    m = Module()
-    @eval m begin
+    M = Module()
+    @eval M begin
         @inline function _def_inline(x)
             # this call won't be resolved and thus will prevent inlining to happen if we don't
             # annotate `@inline` at the top of this function body
@@ -414,32 +414,32 @@ code_typed1(args...; kwargs...) = (first∘first)(code_typed(args...; kwargs...)
         def_noinline_noconflict(x) = _def_noinline_noconflict(x)
     end
 
-    let ci = code_typed1(m.def_inline, (Int,))
-        @test all(ci.code) do x
+    let code = code_typed1(M.def_inline, (Int,))
+        @test all(code) do x
             !isinvoke(x, :_def_inline)
         end
     end
-    let ci = code_typed1(m.def_noinline, (Int,))
-        @test any(ci.code) do x
+    let code = code_typed1(M.def_noinline, (Int,))
+        @test any(code) do x
             isinvoke(x, :_def_noinline)
         end
     end
     # test that they don't conflict with other "before-definition" macros
-    let ci = code_typed1(m.def_inline_noconflict, (Int,))
-        @test all(ci.code) do x
+    let code = code_typed1(M.def_inline_noconflict, (Int,))
+        @test all(code) do x
             !isinvoke(x, :_def_inline_noconflict)
         end
     end
-    let ci = code_typed1(m.def_noinline_noconflict, (Int,))
-        @test any(ci.code) do x
+    let code = code_typed1(M.def_noinline_noconflict, (Int,))
+        @test any(code) do x
             isinvoke(x, :_def_noinline_noconflict)
         end
     end
 end
 
 @testset "@inline/@noinline annotation within a function body" begin
-    m = Module()
-    @eval m begin
+    M = Module()
+    @eval M begin
         function _body_inline(x)
             @inline
             # this call won't be resolved and thus will prevent inlining to happen if we don't
@@ -471,36 +471,126 @@ end
         end
     end
 
-    let ci = code_typed1(m.body_inline, (Int,))
-        @test all(ci.code) do x
+    let code = code_typed1(M.body_inline, (Int,))
+        @test all(code) do x
             !isinvoke(x, :_body_inline)
         end
     end
-    let ci = code_typed1(m.body_noinline, (Int,))
-        @test any(ci.code) do x
+    let code = code_typed1(M.body_noinline, (Int,))
+        @test any(code) do x
             isinvoke(x, :_body_noinline)
         end
     end
     # test annotations for `do` blocks
-    let ci = code_typed1(m.do_inline, (Int,))
+    let code = code_typed1(M.do_inline, (Int,))
         # what we test here is that both `simple_caller` and the anonymous function that the
         # `do` block creates should inlined away, and as a result there is only the unresolved call
-        @test all(ci.code) do x
+        @test all(code) do x
             !isinvoke(x, :simple_caller) &&
             !isinvoke(x, mi->startswith(string(mi.def.name), '#'))
         end
     end
-    let ci = code_typed1(m.do_noinline, (Int,))
+    let code = code_typed1(M.do_noinline, (Int,))
         # the anonymous function that the `do` block created shouldn't be inlined here
-        @test any(ci.code) do x
+        @test any(code) do x
             isinvoke(x, mi->startswith(string(mi.def.name), '#'))
         end
     end
 end
 
+@testset "callsite @inline/@noinline annotations" begin
+    M = Module()
+    @eval M begin
+        # this global variable prevents inference to fold everything as constant, and/or the optimizer to inline the call accessing to this
+        g = 0
+
+        @noinline noinlined_explicit(x) = x
+        force_inline_explicit(x)        = @inline noinlined_explicit(x)
+        force_inline_block_explicit(x)  = @inline noinlined_explicit(x) + noinlined_explicit(x)
+        noinlined_implicit(x)          = g
+        force_inline_implicit(x)       = @inline noinlined_implicit(x)
+        force_inline_block_implicit(x) = @inline noinlined_implicit(x) + noinlined_implicit(x)
+
+        @inline inlined_explicit(x)      = x
+        force_noinline_explicit(x)       = @noinline inlined_explicit(x)
+        force_noinline_block_explicit(x) = @noinline inlined_explicit(x) + inlined_explicit(x)
+        inlined_implicit(x)              = x
+        force_noinline_implicit(x)       = @noinline inlined_implicit(x)
+        force_noinline_block_implicit(x) = @noinline inlined_implicit(x) + inlined_implicit(x)
+
+        # test callsite annotations for constant-prop'ed calls
+
+        @noinline Base.@aggressive_constprop noinlined_constprop_explicit(a) = a+g
+        force_inline_constprop_explicit()                                    = @inline noinlined_constprop_explicit(0)
+        Base.@aggressive_constprop noinlined_constprop_implicit(a) = a+g
+        force_inline_constprop_implicit()                          = @inline noinlined_constprop_implicit(0)
+
+        @inline Base.@aggressive_constprop inlined_constprop_explicit(a) = a+g
+        force_noinline_constprop_explicit()                              = @noinline inlined_constprop_explicit(0)
+        @inline Base.@aggressive_constprop inlined_constprop_implicit(a) = a+g
+        force_noinline_constprop_implicit()                              = @noinline inlined_constprop_implicit(0)
+
+        @noinline notinlined(a) = a
+        function nested(a0, b0)
+            @noinline begin
+                a = @inline notinlined(a0) # this call should be inlined
+                b = notinlined(b0) # this call should NOT be inlined
+                return a, b
+            end
+        end
+    end
+
+    let code = code_typed1(M.force_inline_explicit, (Int,))
+        @test all(x->!isinvoke(x, :noinlined_explicit), code)
+    end
+    let code = code_typed1(M.force_inline_block_explicit, (Int,))
+        @test all(code) do x
+            !isinvoke(x, :noinlined_explicit) &&
+            !isinvoke(x, :(+))
+        end
+    end
+    let code = code_typed1(M.force_inline_implicit, (Int,))
+        @test all(x->!isinvoke(x, :noinlined_implicit), code)
+    end
+    let code = code_typed1(M.force_inline_block_implicit, (Int,))
+        @test all(x->!isinvoke(x, :noinlined_explicit), code)
+    end
+
+    let code = code_typed1(M.force_noinline_explicit, (Int,))
+        @test any(x->isinvoke(x, :inlined_explicit), code)
+    end
+    let code = code_typed1(M.force_noinline_block_explicit, (Int,))
+        @test count(x->isinvoke(x, :inlined_explicit), code) == 2
+    end
+    let code = code_typed1(M.force_noinline_implicit, (Int,))
+        @test any(x->isinvoke(x, :inlined_implicit), code)
+    end
+    let code = code_typed1(M.force_noinline_block_implicit, (Int,))
+        @test count(x->isinvoke(x, :inlined_implicit), code) == 2
+    end
+
+    let code = code_typed1(M.force_inline_constprop_explicit)
+        @test all(x->!isinvoke(x, :noinlined_constprop_explicit), code)
+    end
+    let code = code_typed1(M.force_inline_constprop_implicit)
+        @test all(x->!isinvoke(x, :noinlined_constprop_implicit), code)
+    end
+
+    let code = code_typed1(M.force_noinline_constprop_explicit)
+        @test any(x->isinvoke(x, :inlined_constprop_explicit), code)
+    end
+    let code = code_typed1(M.force_noinline_constprop_implicit)
+        @test any(x->isinvoke(x, :inlined_constprop_implicit), code)
+    end
+
+    let code = code_typed1(M.nested, (Int,Int))
+        @test count(x->isinvoke(x, :notinlined), code) == 1
+    end
+end
+
 # force constant-prop' for `setproperty!`
-let m = Module()
-    ci = @eval m begin
+# https://github.com/JuliaLang/julia/pull/41882
+let code = @eval Module() begin
         # if we don't force constant-prop', `T = fieldtype(Foo, ::Symbol)` will be union-split to
         # `Union{Type{Any},Type{Int}` and it will make `convert(T, nothing)` too costly
         # and it leads to inlining failure
@@ -518,7 +608,7 @@ let m = Module()
         $code_typed1(setter, (Vector{Foo},))
     end
 
-    @test !any(x->isinvoke(x, :setproperty!), ci.code)
+    @test !any(x->isinvoke(x, :setproperty!), code)
 end
 
 # Issue #41299 - inlining deletes error check in :>