63 lines · plain
1// REQUIRES: asserts2// RUN: mlir-opt %s --pass-pipeline="builtin.module(llvm.func(sroa))" --split-input-file --mlir-pass-statistics 2>&1 >/dev/null | FileCheck %s3 4// CHECK: SROA5// CHECK-NEXT: (S) 1 destructured slots6// CHECK-NEXT: (S) 2 max subelement number7// CHECK-NEXT: (S) 1 slots with memory benefit8llvm.func @basic() -> i32 {9 %0 = llvm.mlir.constant(1 : i32) : i3210 %1 = llvm.alloca %0 x !llvm.struct<"foo", (i32, i32)> {alignment = 8 : i64} : (i32) -> !llvm.ptr11 %2 = llvm.getelementptr inbounds %1[0, 1] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<"foo", (i32, i32)>12 %3 = llvm.load %2 : !llvm.ptr -> i3213 llvm.return %3 : i3214}15 16// -----17 18// CHECK: SROA19// CHECK-NEXT: (S) 1 destructured slots20// CHECK-NEXT: (S) 2 max subelement number21// CHECK-NEXT: (S) 0 slots with memory benefit22llvm.func @basic_no_memory_benefit() -> i32 {23 %0 = llvm.mlir.constant(1 : i32) : i3224 %1 = llvm.alloca %0 x !llvm.struct<"foo", (i32, i32)> {alignment = 8 : i64} : (i32) -> !llvm.ptr25 %2 = llvm.getelementptr inbounds %1[0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<"foo", (i32, i32)>26 %3 = llvm.getelementptr inbounds %1[0, 1] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<"foo", (i32, i32)>27 %4 = llvm.load %2 : !llvm.ptr -> i3228 %5 = llvm.load %3 : !llvm.ptr -> i3229 %6 = llvm.add %4, %5 : i3230 llvm.return %6 : i3231}32 33// -----34 35// CHECK: SROA36// CHECK-NEXT: (S) 1 destructured slots37// CHECK-NEXT: (S) 10 max subelement number38// CHECK-NEXT: (S) 1 slots with memory benefit39llvm.func @basic_array() -> i32 {40 %0 = llvm.mlir.constant(1 : i32) : i3241 %1 = llvm.alloca %0 x !llvm.array<10 x i32> {alignment = 8 : i64} : (i32) -> !llvm.ptr42 %2 = llvm.getelementptr inbounds %1[0, 2] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<10 x i32>43 %3 = llvm.load %2 : !llvm.ptr -> i3244 llvm.return %3 : i3245}46 47// -----48 49// SROA is applied repeatedly here, peeling off layers of aggregates one after50// the other, four times.51 52// CHECK: SROA53// CHECK-NEXT: (S) 4 destructured slots54// CHECK-NEXT: (S) 10 max subelement number55// CHECK-NEXT: (S) 4 slots with memory benefit56llvm.func @multi_level_direct() -> i32 {57 %0 = llvm.mlir.constant(1 : i32) : i3258 %1 = llvm.alloca %0 x !llvm.struct<"foo", (i32, f64, struct<"bar", (i8, array<10 x array<10 x i32>>, i8)>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr59 %2 = llvm.getelementptr inbounds %1[0, 2, 1, 5, 8] : (!llvm.ptr) -> !llvm.ptr, !llvm.struct<"foo", (i32, f64, struct<"bar", (i8, array<10 x array<10 x i32>>, i8)>)>60 %3 = llvm.load %2 : !llvm.ptr -> i3261 llvm.return %3 : i3262}63