90ba33099c
This patch canonicalizes getelementptr instructions with constant indices to use the `i8` source element type. This makes it easier for optimizations to recognize that two GEPs are identical, because they don't need to see past many different ways to express the same offset. This is a first step towards https://discourse.llvm.org/t/rfc-replacing-getelementptr-with-ptradd/68699. This is limited to constant GEPs only for now, as they have a clear canonical form, while we're not yet sure how exactly to deal with variable indices. The test llvm/test/Transforms/PhaseOrdering/switch_with_geps.ll gives two representative examples of the kind of optimization improvement we expect from this change. In the first test SimplifyCFG can now realize that all switch branches are actually the same. In the second test it can convert it into simple arithmetic. These are representative of common optimization failures we see in Rust. Fixes https://github.com/llvm/llvm-project/issues/69841.
34 lines
1.4 KiB
Plaintext
34 lines
1.4 KiB
Plaintext
// Test that the HLFIR pipeline does not call MLIR canonicalizer with block
|
|
// merging enabled (moving fir.shape to block argument would cause failures
|
|
// when translating the FIR to LLVM).
|
|
// RUN: %flang_fc1 %s -flang-experimental-hlfir -emit-llvm -O2 -o - | FileCheck %s
|
|
|
|
func.func @no_shape_merge(%cdt: i1, %from: !fir.ref<!fir.array<?xf64>>, %to : !fir.ref<f64>) {
|
|
%c10 = arith.constant 10 : index
|
|
%c20 = arith.constant 20 : index
|
|
%c5 = arith.constant 5 : index
|
|
%shape1 = fir.shape %c10 : (index) -> !fir.shape<1>
|
|
%shape2 = fir.shape %c20 : (index) -> !fir.shape<1>
|
|
cf.cond_br %cdt, ^bb1, ^bb2
|
|
^bb1: // pred: ^bb0
|
|
%coor1 = fir.array_coor %from(%shape1) %c5 : (!fir.ref<!fir.array<?xf64>>, !fir.shape<1>, index) -> !fir.ref<f64>
|
|
%load1 = fir.load %coor1 : !fir.ref<f64>
|
|
fir.store %load1 to %to : !fir.ref<f64>
|
|
cf.br ^bb3
|
|
^bb2: // pred: ^bb0
|
|
%coor2 = fir.array_coor %from(%shape2) %c5 : (!fir.ref<!fir.array<?xf64>>, !fir.shape<1>, index) -> !fir.ref<f64>
|
|
%load2 = fir.load %coor2 : !fir.ref<f64>
|
|
fir.store %load2 to %to : !fir.ref<f64>
|
|
cf.br ^bb3
|
|
^bb3: // pred: ^bb1, ^bb2
|
|
return
|
|
}
|
|
|
|
// Note: block merging happens in the output below, but after FIR codegen.
|
|
|
|
// CHECK-LABEL: define void @no_shape_merge(
|
|
// CHECK: %[[GEP:.*]] = getelementptr i8, ptr %{{.*}}
|
|
// CHECK: %[[LOAD:.*]] = load double, ptr %[[GEP]]
|
|
// CHECK: store double %[[LOAD]], ptr %{{.*}}
|
|
// CHECK: ret void
|