a0d699a8e6
This helps to disambiguate accesses in the caller and the callee after LLVM inlining in some apps. I did not see any performance changes, but this is one step towards enabling other optimizations in the apps that I am looking at. The definition of llvm.noalias says: ``` ... indicates that memory locations accessed via pointer values based on the argument or return value are not also accessed, during the execution of the function, via pointer values not based on the argument or return value. This guarantee only holds for memory locations that are modified, by any means, during the execution of the function. ``` I believe this exactly matches Fortran rules for the dummy arguments that are modified during their subprogram execution. I also set llvm.noalias and llvm.nocapture on the !fir.box<> arguments, because the corresponding descriptors cannot be captured and cannot alias anything (not based on them) during the execution of the subprogram.
71 lines
4.7 KiB
Plaintext
71 lines
4.7 KiB
Plaintext
// RUN: fir-opt --function-attr="set-nocapture=true" %s | FileCheck %s
|
|
|
|
// If a function has a body and is not bind(c), and if the dummy argument doesn't have the target,
|
|
// asynchronous, volatile, or pointer attribute, then add llvm.nocapture to the dummy argument.
|
|
|
|
func.func @_QParg_nocapture(%arg0: !fir.ref<i32> {fir.bindc_name = "tar", fir.target}, %arg1: !fir.ref<i32> {fir.asynchronous, fir.bindc_name = "asynch"}, %arg2: !fir.ref<i32> {fir.bindc_name = "vol", fir.volatile}, %arg3: !fir.ref<!fir.box<!fir.ptr<i32>>> {fir.bindc_name = "ptr"}, %arg4: !fir.ref<i32> {fir.bindc_name = "nocap"}) {
|
|
%0 = fir.dummy_scope : !fir.dscope
|
|
%1 = fir.declare %arg0 dummy_scope %0 {fortran_attrs = #fir.var_attrs<target>, uniq_name = "_QFarg_nocaptureEtar"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
|
|
%2 = fir.declare %arg1 dummy_scope %0 {fortran_attrs = #fir.var_attrs<asynchronous>, uniq_name = "_QFarg_nocaptureEasynch"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
|
|
%3 = fir.declare %arg2 dummy_scope %0 {uniq_name = "_QFarg_nocaptureEvol"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
|
|
%4 = fir.declare %arg3 dummy_scope %0 {fortran_attrs = #fir.var_attrs<pointer>, uniq_name = "_QFarg_nocaptureEptr"} : (!fir.ref<!fir.box<!fir.ptr<i32>>>, !fir.dscope) -> !fir.ref<!fir.box<!fir.ptr<i32>>>
|
|
%5 = fir.declare %arg4 dummy_scope %0 {uniq_name = "_QFarg_nocaptureEnocap"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
|
|
return
|
|
}
|
|
// CHECK-LABEL: func.func @_QParg_nocapture(
|
|
// CHECK-SAME: %[[ARG0:.*]]: !fir.ref<i32> {fir.bindc_name = "tar", fir.target},
|
|
// CHECK-SAME: %[[ARG1:.*]]: !fir.ref<i32> {fir.asynchronous, fir.bindc_name = "asynch"},
|
|
// CHECK-SAME: %[[ARG2:.*]]: !fir.ref<i32> {fir.bindc_name = "vol", fir.volatile},
|
|
// CHECK-SAME: %[[ARG3:.*]]: !fir.ref<!fir.box<!fir.ptr<i32>>> {fir.bindc_name = "ptr", llvm.nocapture},
|
|
// CHECK-SAME: %[[ARG4:.*]]: !fir.ref<i32> {fir.bindc_name = "nocap", llvm.nocapture}) {
|
|
// CHECK: return
|
|
// CHECK-NEXT: }
|
|
|
|
func.func @arg_nocapture_bindc(%arg0: !fir.ref<i32> {fir.bindc_name = "tar", fir.target}, %arg1: !fir.ref<i32> {fir.bindc_name = "nocap"}) attributes {fir.bindc_name = "arg_nocapture_bindc", fir.proc_attrs = #fir.proc_attrs<bind_c>} {
|
|
%0 = fir.dummy_scope : !fir.dscope
|
|
%1 = fir.declare %arg0 dummy_scope %0 {fortran_attrs = #fir.var_attrs<target>, uniq_name = "_QFarg_nocapture_bindcEtar"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
|
|
%2 = fir.declare %arg1 dummy_scope %0 {uniq_name = "_QFarg_nocapture_bindcEnocap"} : (!fir.ref<i32>, !fir.dscope) -> !fir.ref<i32>
|
|
return
|
|
}
|
|
// CHECK-LABEL: func.func @arg_nocapture_bindc(
|
|
// CHECK-NOT: llvm.nocapture
|
|
|
|
|
|
// If a function declaration is from a module and is not bind(c), and if the dummy argument doesn't have
|
|
// the target, asynchronous, volatile, or pointer attribute, then add llvm.nocapture to the dummy argument.
|
|
|
|
func.func private @_QMarg_modPcheck_args(!fir.ref<i32> {fir.target}, !fir.ref<i32> {fir.asynchronous}, !fir.ref<i32> {fir.volatile}, !fir.ref<!fir.box<!fir.ptr<i32>>>, !fir.ref<i32>, !fir.boxchar<1>, !fir.ref<complex<f32>>)
|
|
// CHECK-LABEL: func.func private @_QMarg_modPcheck_args(
|
|
// CHECK-SAME: !fir.ref<i32> {fir.target},
|
|
// CHECK-SAME: !fir.ref<i32> {fir.asynchronous},
|
|
// CHECK-SAME: !fir.ref<i32> {fir.volatile},
|
|
// CHECK-SAME: !fir.ref<!fir.box<!fir.ptr<i32>>> {llvm.nocapture},
|
|
// CHECK-SAME: !fir.ref<i32> {llvm.nocapture},
|
|
// CHECK-SAME: !fir.boxchar<1>,
|
|
// CHECK-SAME: !fir.ref<complex<f32>> {llvm.nocapture})
|
|
|
|
// Test !fir.box arguments:
|
|
// CHECK-LABEL: func.func private @test_box(
|
|
// CHECK-SAME: %[[ARG0:.*]]: !fir.box<i32> {llvm.nocapture}) {
|
|
func.func private @test_box(%arg0: !fir.box<i32>) {
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func private @test_box_target(
|
|
// CHECK-SAME: %[[ARG0:.*]]: !fir.box<i32> {fir.target, llvm.nocapture}) {
|
|
func.func private @test_box_target(%arg0: !fir.box<i32> {fir.target}) {
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func private @test_box_volatile(
|
|
// CHECK-SAME: %[[ARG0:.*]]: !fir.box<i32> {fir.volatile, llvm.nocapture}) {
|
|
func.func private @test_box_volatile(%arg0: !fir.box<i32> {fir.volatile}) {
|
|
return
|
|
}
|
|
|
|
// CHECK-LABEL: func.func private @test_box_asynchronous(
|
|
// CHECK-SAME: %[[ARG0:.*]]: !fir.box<i32> {fir.asynchronous, llvm.nocapture}) {
|
|
func.func private @test_box_asynchronous(%arg0: !fir.box<i32> {fir.asynchronous}) {
|
|
return
|
|
}
|