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.
47 lines
2.6 KiB
Plaintext
47 lines
2.6 KiB
Plaintext
// DEFINE: %{triple} =
|
|
// DEFINE: %{check-unroll} = %flang_fc1 -emit-llvm -O1 -vectorize-loops -funroll-loops -mllvm -force-vector-width=2 -triple %{triple} -o- %s | FileCheck %s --check-prefixes=CHECK,UNROLL
|
|
// DEFINE: %{check-nounroll} = %flang_fc1 -emit-llvm -O1 -vectorize-loops -mllvm -force-vector-width=2 -triple %{triple} -o- %s | FileCheck %s --check-prefixes=CHECK,NO-UNROLL
|
|
|
|
// REDEFINE: %{triple} = aarch64-unknown-linux-gnu
|
|
// RUN: %if aarch64-registered-target %{ %{check-unroll} %}
|
|
// RUN: %if aarch64-registered-target %{ %{check-nounroll} %}
|
|
|
|
// REDEFINE: %{triple} = x86_64-unknown-linux-gnu
|
|
// RUN: %if x86-registered-target %{ %{check-unroll} %}
|
|
// RUN: %if x86-registered-target %{ %{check-nounroll} %}
|
|
|
|
// CHECK-LABEL: @unroll
|
|
// CHECK-SAME: (ptr {{[^%]*}}%[[ARG0:.*]])
|
|
func.func @unroll(%arg0: !fir.ref<!fir.array<1000 x index>> {fir.bindc_name = "a"}) {
|
|
%scope = fir.dummy_scope : !fir.dscope
|
|
%c1000 = arith.constant 1000 : index
|
|
%shape = fir.shape %c1000 : (index) -> !fir.shape<1>
|
|
%a:2 = hlfir.declare %arg0(%shape) dummy_scope %scope {uniq_name = "unrollEa"} : (!fir.ref<!fir.array<1000xindex>>, !fir.shape<1>, !fir.dscope) -> (!fir.ref<!fir.array<1000 x index>>, !fir.ref<!fir.array<1000 x index>>)
|
|
%c1 = arith.constant 1 : index
|
|
fir.do_loop %arg1 = %c1 to %c1000 step %c1 {
|
|
// CHECK: br label %[[BLK:.*]]
|
|
// CHECK: [[BLK]]:
|
|
// CHECK-NEXT: %[[IND:.*]] = phi i64 [ 0, %{{.*}} ], [ %[[NIV:.*]], %[[BLK]] ]
|
|
// CHECK-NEXT: %[[VIND:.*]] = phi <2 x i64> [ <i64 1, i64 2>, %{{.*}} ], [ %[[NVIND:.*]], %[[BLK]] ]
|
|
|
|
// NO-UNROLL-NEXT: %[[GEP:.*]] = getelementptr i64, ptr %[[ARG0]], i64 %[[IND]]
|
|
// NO-UNROLL-NEXT: store <2 x i64> %[[VIND]], ptr %[[GEP]]
|
|
// NO-UNROLL-NEXT: %[[NIV:.*]] = add nuw i64 %{{.*}}, 2
|
|
// NO-UNROLL-NEXT: %[[NVIND]] = add <2 x i64> %[[VIND]], splat (i64 2)
|
|
|
|
// UNROLL-NEXT: %[[VIND1:.*]] = add <2 x i64> %[[VIND]], splat (i64 2)
|
|
// UNROLL-NEXT: %[[GEP0:.*]] = getelementptr i64, ptr %[[ARG0]], i64 %[[IND]]
|
|
// UNROLL-NEXT: %[[GEP1:.*]] = getelementptr i8, ptr %[[GEP0]], i64 16
|
|
// UNROLL-NEXT: store <2 x i64> %[[VIND]], ptr %[[GEP0]]
|
|
// UNROLL-NEXT: store <2 x i64> %[[VIND1]], ptr %[[GEP1]]
|
|
// UNROLL-NEXT: %[[NIV:.*]] = add nuw i64 %[[IND]], 4
|
|
// UNROLL-NEXT: %[[NVIND:.*]] = add <2 x i64> %[[VIND]], splat (i64 4)
|
|
|
|
// CHECK-NEXT: %[[EXIT:.*]] = icmp eq i64 %[[NIV]], 1000
|
|
// CHECK-NEXT: br i1 %[[EXIT]], label %{{.*}}, label %[[BLK]]
|
|
%ai = hlfir.designate %a#0 (%arg1) : (!fir.ref<!fir.array<1000 x index>>, index) -> !fir.ref<index>
|
|
hlfir.assign %arg1 to %ai : index, !fir.ref<index>
|
|
}
|
|
return
|
|
}
|