f1ea831dbc
The changes are needed to get leslie3d same performance with HLFIR as with FIR lowering. The two module allocatable variables cannot alias, so the optimized bufferization should be able to elide the temporary and inline the assignment loop.
100 lines
5.6 KiB
Plaintext
100 lines
5.6 KiB
Plaintext
// RUN: fir-opt %s -pass-pipeline='builtin.module(func.func(test-fir-alias-analysis))' 2>&1 | FileCheck %s
|
|
|
|
// leslie3d case with two allocatable module variables
|
|
// that cannot alias:
|
|
// module les3d_data
|
|
// implicit real*8 (a-h,o-z)
|
|
// integer imax, jmax, kmax
|
|
// double precision,allocatable,dimension(:,:,:,:,:) :: q
|
|
// double precision,allocatable,dimension(:,:,:,:) :: du
|
|
// end module les3d_data
|
|
// subroutine update()
|
|
// use les3d_data
|
|
// implicit real*8(a-h,o-z)
|
|
// i2 = imax - 1
|
|
// do k = 1, kmax - 1
|
|
// do j = 1, jmax - 1
|
|
// q(1:i2,j,k,1,m) = (q(1:i2,j,k,1,m) + du(1:i2,j,k,1))
|
|
// end do
|
|
// end do
|
|
// end subroutine update
|
|
|
|
// CHECK: allocatable_mod1#0 <-> allocatable_mod2#0: NoAlias
|
|
func.func @_QPupdate() {
|
|
%c1 = arith.constant 1 : index
|
|
%c0 = arith.constant 0 : index
|
|
%c1_i32 = arith.constant 1 : i32
|
|
%0 = fir.address_of(@_QMles3d_dataEdu) : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?xf64>>>>
|
|
%1:2 = hlfir.declare %0 {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMles3d_dataEdu"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?xf64>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?xf64>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?xf64>>>>)
|
|
%2 = fir.alloca i32 {bindc_name = "i2", uniq_name = "_QFupdateEi2"}
|
|
%3:2 = hlfir.declare %2 {uniq_name = "_QFupdateEi2"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%4 = fir.address_of(@_QMles3d_dataEimax) : !fir.ref<i32>
|
|
%5:2 = hlfir.declare %4 {uniq_name = "_QMles3d_dataEimax"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%6 = fir.alloca i32 {bindc_name = "j", uniq_name = "_QFupdateEj"}
|
|
%7:2 = hlfir.declare %6 {uniq_name = "_QFupdateEj"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%8 = fir.address_of(@_QMles3d_dataEjmax) : !fir.ref<i32>
|
|
%9:2 = hlfir.declare %8 {uniq_name = "_QMles3d_dataEjmax"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%10 = fir.alloca i32 {bindc_name = "k", uniq_name = "_QFupdateEk"}
|
|
%11:2 = hlfir.declare %10 {uniq_name = "_QFupdateEk"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%12 = fir.address_of(@_QMles3d_dataEkmax) : !fir.ref<i32>
|
|
%13:2 = hlfir.declare %12 {uniq_name = "_QMles3d_dataEkmax"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%14 = fir.alloca i32 {bindc_name = "m", uniq_name = "_QFupdateEm"}
|
|
%15:2 = hlfir.declare %14 {uniq_name = "_QFupdateEm"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
|
|
%16 = fir.address_of(@_QMles3d_dataEq) : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?x?xf64>>>>
|
|
%17:2 = hlfir.declare %16 {fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMles3d_dataEq"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?x?xf64>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?x?xf64>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?x?xf64>>>>)
|
|
%18 = fir.load %5#0 : !fir.ref<i32>
|
|
%19 = arith.subi %18, %c1_i32 : i32
|
|
hlfir.assign %19 to %3#0 : i32, !fir.ref<i32>
|
|
%20 = fir.load %13#0 : !fir.ref<i32>
|
|
%21 = arith.subi %20, %c1_i32 : i32
|
|
%22 = fir.convert %21 : (i32) -> index
|
|
%23 = fir.convert %c1 : (index) -> i32
|
|
%24:2 = fir.do_loop %arg0 = %c1 to %22 step %c1 iter_args(%arg1 = %23) -> (index, i32) {
|
|
fir.store %arg1 to %11#1 : !fir.ref<i32>
|
|
%25 = fir.load %9#0 : !fir.ref<i32>
|
|
%26 = arith.subi %25, %c1_i32 : i32
|
|
%27 = fir.convert %26 : (i32) -> index
|
|
%28:2 = fir.do_loop %arg2 = %c1 to %27 step %c1 iter_args(%arg3 = %23) -> (index, i32) {
|
|
fir.store %arg3 to %7#1 : !fir.ref<i32>
|
|
%32 = fir.load %17#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?x?xf64>>>>
|
|
%33 = fir.load %3#0 : !fir.ref<i32>
|
|
%34 = fir.convert %33 : (i32) -> index
|
|
%35 = arith.cmpi sgt, %34, %c0 : index
|
|
%36 = arith.select %35, %34, %c0 : index
|
|
%37 = fir.load %7#0 : !fir.ref<i32>
|
|
%38 = fir.convert %37 : (i32) -> i64
|
|
%39 = fir.load %11#0 : !fir.ref<i32>
|
|
%40 = fir.convert %39 : (i32) -> i64
|
|
%41 = fir.load %15#0 : !fir.ref<i32>
|
|
%42 = fir.convert %41 : (i32) -> i64
|
|
%43 = fir.shape %36 : (index) -> !fir.shape<1>
|
|
%44 = hlfir.designate %32 (%c1:%34:%c1, %38, %40, %c1, %42) shape %43 {test.ptr = "allocatable_mod1"} : (!fir.box<!fir.heap<!fir.array<?x?x?x?x?xf64>>>, index, index, index, i64, i64, index, i64, !fir.shape<1>) -> !fir.box<!fir.array<?xf64>>
|
|
%45 = fir.load %1#0 : !fir.ref<!fir.box<!fir.heap<!fir.array<?x?x?x?xf64>>>>
|
|
%46 = hlfir.designate %45 (%c1:%34:%c1, %38, %40, %c1) shape %43 {test.ptr = "allocatable_mod2"} : (!fir.box<!fir.heap<!fir.array<?x?x?x?xf64>>>, index, index, index, i64, i64, index, !fir.shape<1>) -> !fir.box<!fir.array<?xf64>>
|
|
%47 = hlfir.elemental %43 unordered : (!fir.shape<1>) -> !hlfir.expr<?xf64> {
|
|
^bb0(%arg4: index):
|
|
%51 = hlfir.designate %44 (%arg4) : (!fir.box<!fir.array<?xf64>>, index) -> !fir.ref<f64>
|
|
%52 = hlfir.designate %46 (%arg4) : (!fir.box<!fir.array<?xf64>>, index) -> !fir.ref<f64>
|
|
%53 = fir.load %51 : !fir.ref<f64>
|
|
%54 = fir.load %52 : !fir.ref<f64>
|
|
%55 = arith.addf %53, %54 fastmath<fast> : f64
|
|
%56 = hlfir.no_reassoc %55 : f64
|
|
hlfir.yield_element %56 : f64
|
|
}
|
|
hlfir.assign %47 to %44 : !hlfir.expr<?xf64>, !fir.box<!fir.array<?xf64>>
|
|
hlfir.destroy %47 : !hlfir.expr<?xf64>
|
|
%48 = arith.addi %arg2, %c1 : index
|
|
%49 = fir.load %7#1 : !fir.ref<i32>
|
|
%50 = arith.addi %49, %23 : i32
|
|
fir.result %48, %50 : index, i32
|
|
}
|
|
fir.store %28#1 to %7#1 : !fir.ref<i32>
|
|
%29 = arith.addi %arg0, %c1 : index
|
|
%30 = fir.load %11#1 : !fir.ref<i32>
|
|
%31 = arith.addi %30, %23 : i32
|
|
fir.result %29, %31 : index, i32
|
|
}
|
|
fir.store %24#1 to %11#1 : !fir.ref<i32>
|
|
return
|
|
}
|