Skip to content

Commit

Permalink
[MLIR][Flang][OpenMP] Make omp.simdloop into a loop wrapper (#87365)
Browse files Browse the repository at this point in the history 
This patch updates the definition of `omp.simdloop` to enforce the
restrictions of a wrapper operation. It has been renamed to `omp.simd`,
to better reflect the naming used in the spec. All uses of "simdloop" in
function names have been updated accordingly.

Some changes to Flang lowering and OpenMP to LLVM IR translation are
introduced to prevent the introduction of compilation/test failures. The
eventual long term solution might be different.
  • Loading branch information
@skatrak
skatrak committed Apr 17, 2024
1 parent f4737a2 commit 3eb0ba3
Show file tree
Hide file tree
Showing 20 changed files with 701 additions and 592 deletions.
167 changes: 110 additions & 57 deletions flang/lib/Lower/OpenMP/OpenMP.cpp
View file

Large diffs are not rendered by default.

101 changes: 54 additions & 47 deletions flang/test/Fir/convert-to-llvm-openmp-and-fir.fir
View file
Original file line number Diff line number Diff line change
Expand Up @@ -180,14 +180,16 @@ func.func @_QPsimd1(%arg0: !fir.ref<i32> {fir.bindc_name = "n"}, %arg1: !fir.ref
omp.parallel {
%1 = fir.alloca i32 {adapt.valuebyref, pinned}
%2 = fir.load %arg0 : !fir.ref<i32>
omp.simdloop for (%arg2) : i32 = (%c1_i32) to (%2) step (%c1_i32) {
fir.store %arg2 to %1 : !fir.ref<i32>
%3 = fir.load %1 : !fir.ref<i32>
%4 = fir.convert %3 : (i32) -> i64
%5 = arith.subi %4, %c1_i64 : i64
%6 = fir.coordinate_of %arg1, %5 : (!fir.ref<!fir.array<?xi32>>, i64) -> !fir.ref<i32>
fir.store %3 to %6 : !fir.ref<i32>
omp.yield
omp.simd {
omp.loop_nest (%arg2) : i32 = (%c1_i32) to (%2) step (%c1_i32) {
fir.store %arg2 to %1 : !fir.ref<i32>
%3 = fir.load %1 : !fir.ref<i32>
%4 = fir.convert %3 : (i32) -> i64
%5 = arith.subi %4, %c1_i64 : i64
%6 = fir.coordinate_of %arg1, %5 : (!fir.ref<!fir.array<?xi32>>, i64) -> !fir.ref<i32>
fir.store %3 to %6 : !fir.ref<i32>
omp.yield
}
}
omp.terminator
}
Expand All @@ -202,8 +204,8 @@ func.func @_QPsimd1(%arg0: !fir.ref<i32> {fir.bindc_name = "n"}, %arg1: !fir.ref
// CHECK: %[[ONE_3:.*]] = llvm.mlir.constant(1 : i64) : i64
// CHECK: %[[I_VAR:.*]] = llvm.alloca %[[ONE_3]] x i32 {pinned} : (i64) -> !llvm.ptr
// CHECK: %[[N:.*]] = llvm.load %[[N_REF]] : !llvm.ptr -> i32
// CHECK: omp.simdloop
// CHECK-SAME: (%[[I:.*]]) : i32 = (%[[ONE_2]]) to (%[[N]]) step (%[[ONE_2]]) {
// CHECK: omp.simd {
// CHECK-NEXT: omp.loop_nest (%[[I:.*]]) : i32 = (%[[ONE_2]]) to (%[[N]]) step (%[[ONE_2]]) {
// CHECK: llvm.store %[[I]], %[[I_VAR]] : i32, !llvm.ptr
// CHECK: %[[I1:.*]] = llvm.load %[[I_VAR]] : !llvm.ptr -> i32
// CHECK: %[[I1_EXT:.*]] = llvm.sext %[[I1]] : i32 to i64
Expand All @@ -212,6 +214,7 @@ func.func @_QPsimd1(%arg0: !fir.ref<i32> {fir.bindc_name = "n"}, %arg1: !fir.ref
// CHECK: llvm.store %[[I1]], %[[ARR_I_REF]] : i32, !llvm.ptr
// CHECK: omp.yield
// CHECK: }
// CHECK: }
// CHECK: omp.terminator
// CHECK: }
// CHECK: llvm.return
Expand Down Expand Up @@ -471,55 +474,59 @@ func.func @_QPomp_target() {

// -----

func.func @_QPsimdloop_with_nested_loop() {
func.func @_QPsimd_with_nested_loop() {
%0 = fir.alloca i32 {adapt.valuebyref}
%1 = fir.alloca !fir.array<10xi32> {bindc_name = "a", uniq_name = "_QFsimdloop_with_nested_loopEa"}
%2 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFsimdloop_with_nested_loopEi"}
%3 = fir.alloca i32 {bindc_name = "j", uniq_name = "_QFsimdloop_with_nested_loopEj"}
%1 = fir.alloca !fir.array<10xi32> {bindc_name = "a", uniq_name = "_QFsimd_with_nested_loopEa"}
%2 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFsimd_with_nested_loopEi"}
%3 = fir.alloca i32 {bindc_name = "j", uniq_name = "_QFsimd_with_nested_loopEj"}
%c1_i32 = arith.constant 1 : i32
%c10_i32 = arith.constant 10 : i32
%c1_i32_0 = arith.constant 1 : i32
omp.simdloop for (%arg0) : i32 = (%c1_i32) to (%c10_i32) inclusive step (%c1_i32_0) {
fir.store %arg0 to %0 : !fir.ref<i32>
%c1_i32_1 = arith.constant 1 : i32
%4 = fir.convert %c1_i32_1 : (i32) -> index
%c10_i32_2 = arith.constant 10 : i32
%5 = fir.convert %c10_i32_2 : (i32) -> index
%c1 = arith.constant 1 : index
%6 = fir.do_loop %arg1 = %4 to %5 step %c1 -> index {
%8 = fir.convert %arg1 : (index) -> i32
fir.store %8 to %3 : !fir.ref<i32>
%9 = fir.load %0 : !fir.ref<i32>
%10 = fir.load %0 : !fir.ref<i32>
%11 = fir.convert %10 : (i32) -> i64
%c1_i64 = arith.constant 1 : i64
%12 = arith.subi %11, %c1_i64 : i64
%13 = fir.coordinate_of %1, %12 : (!fir.ref<!fir.array<10xi32>>, i64) -> !fir.ref<i32>
fir.store %9 to %13 : !fir.ref<i32>
%14 = arith.addi %arg1, %c1 : index
fir.result %14 : index
omp.simd {
omp.loop_nest (%arg0) : i32 = (%c1_i32) to (%c10_i32) inclusive step (%c1_i32_0) {
fir.store %arg0 to %0 : !fir.ref<i32>
%c1_i32_1 = arith.constant 1 : i32
%4 = fir.convert %c1_i32_1 : (i32) -> index
%c10_i32_2 = arith.constant 10 : i32
%5 = fir.convert %c10_i32_2 : (i32) -> index
%c1 = arith.constant 1 : index
%6 = fir.do_loop %arg1 = %4 to %5 step %c1 -> index {
%8 = fir.convert %arg1 : (index) -> i32
fir.store %8 to %3 : !fir.ref<i32>
%9 = fir.load %0 : !fir.ref<i32>
%10 = fir.load %0 : !fir.ref<i32>
%11 = fir.convert %10 : (i32) -> i64
%c1_i64 = arith.constant 1 : i64
%12 = arith.subi %11, %c1_i64 : i64
%13 = fir.coordinate_of %1, %12 : (!fir.ref<!fir.array<10xi32>>, i64) -> !fir.ref<i32>
fir.store %9 to %13 : !fir.ref<i32>
%14 = arith.addi %arg1, %c1 : index
fir.result %14 : index
}
%7 = fir.convert %6 : (index) -> i32
fir.store %7 to %3 : !fir.ref<i32>
omp.yield
}
%7 = fir.convert %6 : (index) -> i32
fir.store %7 to %3 : !fir.ref<i32>
omp.yield
}
return
}

// CHECK-LABEL: llvm.func @_QPsimdloop_with_nested_loop() {
// CHECK-LABEL: llvm.func @_QPsimd_with_nested_loop() {
// CHECK: %[[LOWER:.*]] = llvm.mlir.constant(1 : i32) : i32
// CHECK: %[[UPPER:.*]] = llvm.mlir.constant(10 : i32) : i32
// CHECK: %[[STEP:.*]] = llvm.mlir.constant(1 : i32) : i32
// CHECK: omp.simdloop for (%[[CNT:.*]]) : i32 = (%[[LOWER]]) to (%[[UPPER]]) inclusive step (%[[STEP]]) {
// CHECK: llvm.br ^bb1(%[[VAL_1:.*]], %[[VAL_2:.*]] : i64, i64)
// CHECK: ^bb1(%[[VAL_3:.*]]: i64, %[[VAL_4:.*]]: i64):
// CHECK: %[[VAL_5:.*]] = llvm.mlir.constant(0 : index) : i64
// CHECK: %[[VAL_6:.*]] = llvm.icmp "sgt" %[[VAL_4]], %[[VAL_5]] : i64
// CHECK: llvm.cond_br %[[VAL_6]], ^bb2, ^bb3
// CHECK: ^bb2:
// CHECK: llvm.br ^bb1(%[[VAL_7:.*]], %[[VAL_8:.*]] : i64, i64)
// CHECK: ^bb3:
// CHECK: omp.yield
// CHECK: omp.simd {
// CHECK-NEXT: omp.loop_nest (%[[CNT:.*]]) : i32 = (%[[LOWER]]) to (%[[UPPER]]) inclusive step (%[[STEP]]) {
// CHECK: llvm.br ^bb1(%[[VAL_1:.*]], %[[VAL_2:.*]] : i64, i64)
// CHECK: ^bb1(%[[VAL_3:.*]]: i64, %[[VAL_4:.*]]: i64):
// CHECK: %[[VAL_5:.*]] = llvm.mlir.constant(0 : index) : i64
// CHECK: %[[VAL_6:.*]] = llvm.icmp "sgt" %[[VAL_4]], %[[VAL_5]] : i64
// CHECK: llvm.cond_br %[[VAL_6]], ^bb2, ^bb3
// CHECK: ^bb2:
// CHECK: llvm.br ^bb1(%[[VAL_7:.*]], %[[VAL_8:.*]] : i64, i64)
// CHECK: ^bb3:
// CHECK: omp.yield
// CHECK: }
// CHECK: }
// CHECK: llvm.return
// CHECK: }
Expand Down
23 changes: 11 additions & 12 deletions flang/test/Lower/OpenMP/FIR/if-clause.f90
View file
Original file line number Diff line number Diff line change
Expand Up @@ -116,15 +116,15 @@ program main
do i = 1, 10
end do
!$omp end parallel do simd

! CHECK: omp.parallel
! CHECK-SAME: if({{.*}})
! CHECK: omp.wsloop
!$omp parallel do simd if(parallel: .true.) if(simd: .false.)
do i = 1, 10
end do
!$omp end parallel do simd

! CHECK: omp.parallel
! CHECK-SAME: if({{.*}})
! CHECK: omp.wsloop
Expand All @@ -134,7 +134,7 @@ program main
do i = 1, 10
end do
!$omp end parallel do simd

! CHECK: omp.parallel
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
Expand All @@ -147,22 +147,22 @@ program main
! ----------------------------------------------------------------------------
! SIMD
! ----------------------------------------------------------------------------
! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
!$omp simd
do i = 1, 10
end do
!$omp end simd

! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-SAME: if({{.*}})
!$omp simd if(.true.)
do i = 1, 10
end do
!$omp end simd

! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-SAME: if({{.*}})
!$omp simd if(simd: .true.)
do i = 1, 10
Expand Down Expand Up @@ -281,7 +281,6 @@ program main
end do
!$omp end target parallel do


! CHECK: omp.target
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
Expand Down Expand Up @@ -360,7 +359,7 @@ program main
! CHECK: omp.target
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
!$omp target simd
Expand All @@ -370,7 +369,7 @@ program main

! CHECK: omp.target
! CHECK-SAME: if({{.*}})
! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-SAME: if({{.*}})
!$omp target simd if(.true.)
do i = 1, 10
Expand All @@ -379,7 +378,7 @@ program main

! CHECK: omp.target
! CHECK-SAME: if({{.*}})
! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-SAME: if({{.*}})
!$omp target simd if(target: .true.) if(simd: .false.)
do i = 1, 10
Expand All @@ -388,7 +387,7 @@ program main

! CHECK: omp.target
! CHECK-SAME: if({{.*}})
! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
!$omp target simd if(target: .true.)
Expand All @@ -399,7 +398,7 @@ program main
! CHECK: omp.target
! CHECK-NOT: if({{.*}})
! CHECK-SAME: {
! CHECK: omp.simdloop
! CHECK: omp.simd
! CHECK-SAME: if({{.*}})
!$omp target simd if(simd: .true.)
do i = 1, 10
Expand Down
2 changes: 1 addition & 1 deletion flang/test/Lower/OpenMP/FIR/loop-combined.f90
View file
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,7 @@ program main
! TARGET SIMD
! ----------------------------------------------------------------------------
! CHECK: omp.target
! CHECK: omp.simdloop
! CHECK: omp.simd
!$omp target simd
do i = 1, 10
end do
Expand Down
3 changes: 2 additions & 1 deletion flang/test/Lower/OpenMP/FIR/parallel-private-clause.f90
View file
Original file line number Diff line number Diff line change
Expand Up @@ -361,7 +361,8 @@ subroutine simd_loop_1
! FIRDialect: %[[UB:.*]] = arith.constant 9 : i32
! FIRDialect: %[[STEP:.*]] = arith.constant 1 : i32

! FIRDialect: omp.simdloop for (%[[I:.*]]) : i32 = (%[[LB]]) to (%[[UB]]) inclusive step (%[[STEP]]) {
! FIRDialect: omp.simd {
! FIRDialect-NEXT: omp.loop_nest (%[[I:.*]]) : i32 = (%[[LB]]) to (%[[UB]]) inclusive step (%[[STEP]]) {
!$OMP SIMD PRIVATE(r)
do i=1, 9
! FIRDialect: fir.store %[[I]] to %[[LOCAL:.*]] : !fir.ref<i32>
Expand Down

0 comments on commit 3eb0ba3

Please sign in to comment.