// RUN: mlir-opt %s -test-linalg-hoisting=test-hoist-view-allocs -allow-unregistered-dialect | FileCheck %s // RUN: mlir-opt %s -test-linalg-hoisting=test-hoist-redundant-transfers -allow-unregistered-dialect | FileCheck %s --check-prefix=VECTOR_TRANSFERS // CHECK-LABEL: func @hoist_allocs( // CHECK-SAME: %[[VAL:[a-zA-Z0-9]*]]: index, // CHECK-SAME: %[[LB:[a-zA-Z0-9]*]]: index, // CHECK-SAME: %[[UB:[a-zA-Z0-9]*]]: index, // CHECK-SAME: %[[STEP:[a-zA-Z0-9]*]]: index, // CHECK-SAME: %[[CMP:[a-zA-Z0-9]*]]: i1 func @hoist_allocs(%val: index, %lb : index, %ub : index, %step: index, %cmp: i1) { // CHECK-DAG: alloca(%[[VAL]]) : memref // CHECK-DAG: %[[A0:.*]] = alloc(%[[VAL]]) : memref // CHECK: scf.for %[[I:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] { // CHECK: alloca(%[[I]]) : memref // CHECK: %[[A1:.*]] = alloc(%[[I]]) : memref // CHECK: scf.for %[[J:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] { // CHECK-DAG: alloca(%[[J]]) : memref // CHECK-DAG: %[[A2:.*]] = alloc(%[[J]]) : memref // CHECK: scf.for %[[K:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] { scf.for %i = %lb to %ub step %step { scf.for %j = %lb to %ub step %step { scf.for %k = %lb to %ub step %step { // Hoist allocs / deallocs outermost, keep view/subview below k. %sa0 = alloca(%val) : memref %a0 = alloc(%val) : memref // CHECK: std.view %[[A0]][%[[LB]]][] : memref to memref<16xf32> // CHECK: subview %[[A0]][0] [42] [1] : memref to memref<42xi8> %v0 = view %a0[%lb][] : memref to memref<16 x f32> %sv0 = subview %a0[0][42][1] : memref to memref<42 x i8> dealloc %a0 : memref // Hoist below i. %sa1 = alloca(%i) : memref %a1 = alloc(%i) : memref dealloc %a1 : memref // Hoist below j. %sa2 = alloca(%j) : memref %a2 = alloc(%j) : memref dealloc %a2 : memref // Don't hoist since k innermost. // CHECK: alloca(%[[K]]) : memref // CHECK: %[[A3:.*]] = alloc(%[[K]]) : memref // CHECK: dealloc %[[A3]] : memref %sa3 = alloca(%k) : memref %a3 = alloc(%k) : memref dealloc %a3 : memref // No hoisting due to control flow. // CHECK: scf.if %[[CMP]] { // CHECK: alloca(%[[VAL]]) : memref // CHECK: %[[A4:.*]] = alloc(%[[VAL]]) : memref // CHECK: dealloc %[[A4]] : memref scf.if %cmp { %sa4 = alloca(%val) : memref %a4 = alloc(%val) : memref dealloc %a4 : memref } // No hoisting due to load/store. // CHECK: %[[SA5:.*]] = alloca(%[[VAL]]) : memref // CHECK: %[[A5:.*]] = alloc(%[[VAL]]) : memref // CHECK: load %[[A5]][%[[LB]]] : memref // CHECK: store %{{.*}}, %[[SA5]][%[[LB]]] : memref // CHECK: dealloc %[[A5]] : memref %sa5 = alloca(%val) : memref %a5 = alloc(%val) : memref %v5 = load %a5[%lb] : memref store %v5, %sa5[%lb] : memref dealloc %a5 : memref } } } // CHECK: } // CHECK: dealloc %[[A2]] : memref // CHECK: } // CHECK: dealloc %[[A1]] : memref // CHECK: } // CHECK: dealloc %[[A0]] : memref return } // VECTOR_TRANSFERS-LABEL: func @hoist_vector_transfer_pairs( // VECTOR_TRANSFERS-SAME: %[[MEMREF0:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF1:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF2:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF3:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF4:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF5:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[VAL:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[LB:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[UB:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[STEP:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[CMP:[a-zA-Z0-9]*]]: i1 func @hoist_vector_transfer_pairs( %memref0: memref, %memref1: memref, %memref2: memref, %memref3: memref, %memref4: memref, %memref5: memref, %val: index, %lb : index, %ub : index, %step: index, %cmp: i1) { %c0 = constant 0 : index %cst = constant 0.0 : f32 // VECTOR_TRANSFERS: vector.transfer_read %{{.*}} : memref, vector<1xf32> // VECTOR_TRANSFERS: scf.for %[[I:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] iter_args({{.*}}) -> (vector<1xf32>) { // VECTOR_TRANSFERS: vector.transfer_read %{{.*}} : memref, vector<2xf32> // VECTOR_TRANSFERS: scf.for %[[J:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] iter_args({{.*}}) -> (vector<1xf32>, vector<2xf32>) { // VECTOR_TRANSFERS: vector.transfer_read %{{.*}} : memref, vector<3xf32> // VECTOR_TRANSFERS: vector.transfer_read %{{.*}} : memref, vector<4xf32> // VECTOR_TRANSFERS: "some_crippling_use"(%[[MEMREF4]]) : (memref) -> () // VECTOR_TRANSFERS: vector.transfer_read %{{.*}} : memref, vector<5xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<1xf32>) -> vector<1xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<2xf32>) -> vector<2xf32> // VECTOR_TRANSFERS: "some_use"(%[[MEMREF2]]) : (memref) -> vector<3xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<4xf32>) -> vector<4xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<5xf32>) -> vector<5xf32> // VECTOR_TRANSFERS: vector.transfer_write %{{.*}} : vector<3xf32>, memref // VECTOR_TRANSFERS: vector.transfer_write %{{.*}} : vector<4xf32>, memref // VECTOR_TRANSFERS: vector.transfer_write %{{.*}} : vector<5xf32>, memref // VECTOR_TRANSFERS: "some_crippling_use"(%[[MEMREF3]]) : (memref) -> () // VECTOR_TRANSFERS: scf.yield {{.*}} : vector<1xf32>, vector<2xf32> // VECTOR_TRANSFERS: } // VECTOR_TRANSFERS: vector.transfer_write %{{.*}} : vector<2xf32>, memref // VECTOR_TRANSFERS: scf.yield {{.*}} : vector<1xf32> // VECTOR_TRANSFERS: } // VECTOR_TRANSFERS: vector.transfer_write %{{.*}} : vector<1xf32>, memref scf.for %i = %lb to %ub step %step { scf.for %j = %lb to %ub step %step { %r0 = vector.transfer_read %memref1[%c0, %c0], %cst: memref, vector<1xf32> %r1 = vector.transfer_read %memref0[%i, %i], %cst: memref, vector<2xf32> %r2 = vector.transfer_read %memref2[%c0, %c0], %cst: memref, vector<3xf32> %r3 = vector.transfer_read %memref3[%c0, %c0], %cst: memref, vector<4xf32> "some_crippling_use"(%memref4) : (memref) -> () %r4 = vector.transfer_read %memref4[%c0, %c0], %cst: memref, vector<5xf32> %r5 = vector.transfer_read %memref5[%c0, %c0], %cst: memref, vector<6xf32> "some_crippling_use"(%memref5) : (memref) -> () %u0 = "some_use"(%r0) : (vector<1xf32>) -> vector<1xf32> %u1 = "some_use"(%r1) : (vector<2xf32>) -> vector<2xf32> %u2 = "some_use"(%memref2) : (memref) -> vector<3xf32> %u3 = "some_use"(%r3) : (vector<4xf32>) -> vector<4xf32> %u4 = "some_use"(%r4) : (vector<5xf32>) -> vector<5xf32> %u5 = "some_use"(%r5) : (vector<6xf32>) -> vector<6xf32> vector.transfer_write %u0, %memref1[%c0, %c0] : vector<1xf32>, memref vector.transfer_write %u1, %memref0[%i, %i] : vector<2xf32>, memref vector.transfer_write %u2, %memref2[%c0, %c0] : vector<3xf32>, memref vector.transfer_write %u3, %memref3[%c0, %c0] : vector<4xf32>, memref vector.transfer_write %u4, %memref4[%c0, %c0] : vector<5xf32>, memref vector.transfer_write %u5, %memref5[%c0, %c0] : vector<6xf32>, memref "some_crippling_use"(%memref3) : (memref) -> () } } return } // VECTOR_TRANSFERS-LABEL: func @hoist_vector_transfer_pairs_disjoint( // VECTOR_TRANSFERS-SAME: %[[MEMREF0:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF1:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF2:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[MEMREF3:[a-zA-Z0-9]*]]: memref, // VECTOR_TRANSFERS-SAME: %[[VAL:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[LB:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[UB:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[STEP:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[RANDOM:[a-zA-Z0-9]*]]: index, // VECTOR_TRANSFERS-SAME: %[[CMP:[a-zA-Z0-9]*]]: i1 func @hoist_vector_transfer_pairs_disjoint( %memref0: memref, %memref1: memref, %memref2: memref, %memref3: memref, %val: index, %lb : index, %ub : index, %step: index, %random_index : index, %cmp: i1) { %c0 = constant 0 : index %c1 = constant 1 : index %c3 = constant 3 : index %cst = constant 0.0 : f32 // VECTOR_TRANSFERS: vector.transfer_read %[[MEMREF2]]{{.*}} : memref, vector<3xf32> // VECTOR_TRANSFERS: vector.transfer_read %[[MEMREF2]]{{.*}} : memref, vector<3xf32> // VECTOR_TRANSFERS: vector.transfer_read %[[MEMREF3]]{{.*}} : memref, vector<4xf32> // VECTOR_TRANSFERS: vector.transfer_read %[[MEMREF3]]{{.*}} : memref, vector<4xf32> // VECTOR_TRANSFERS: scf.for %[[I:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] iter_args({{.*}}) -> // VECTOR_TRANSFERS-SAME: (vector<3xf32>, vector<3xf32>, vector<4xf32>, vector<4xf32>) { // VECTOR_TRANSFERS: scf.for %[[J:.*]] = %[[LB]] to %[[UB]] step %[[STEP]] iter_args({{.*}}) -> // VECTOR_TRANSFERS-SAME: (vector<3xf32>, vector<3xf32>, vector<4xf32>, vector<4xf32>) { // VECTOR_TRANSFERS: vector.transfer_read %[[MEMREF1]]{{.*}} : memref, vector<2xf32> // VECTOR_TRANSFERS: vector.transfer_read %[[MEMREF1]]{{.*}} : memref, vector<2xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<2xf32>) -> vector<2xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<2xf32>) -> vector<2xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<3xf32>) -> vector<3xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<3xf32>) -> vector<3xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<4xf32>) -> vector<4xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<4xf32>) -> vector<4xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<2xf32>) -> vector<2xf32> // VECTOR_TRANSFERS: "some_use"(%{{.*}}) : (vector<2xf32>) -> vector<2xf32> // VECTOR_TRANSFERS: vector.transfer_write %{{.*}}, %[[MEMREF1]]{{.*}} : vector<2xf32>, memref // VECTOR_TRANSFERS: vector.transfer_write %{{.*}}, %[[MEMREF1]]{{.*}} : vector<2xf32>, memref // VECTOR_TRANSFERS: scf.yield {{.*}} : vector<3xf32>, vector<3xf32>, vector<4xf32>, vector<4xf32> // VECTOR_TRANSFERS: } // VECTOR_TRANSFERS: scf.yield {{.*}} : vector<3xf32>, vector<3xf32>, vector<4xf32>, vector<4xf32> // VECTOR_TRANSFERS: } // VECTOR_TRANSFERS: vector.transfer_write %{{.*}}, %[[MEMREF3]]{{.*}} : vector<4xf32>, memref // VECTOR_TRANSFERS: vector.transfer_write %{{.*}}, %[[MEMREF3]]{{.*}} : vector<4xf32>, memref // VECTOR_TRANSFERS: vector.transfer_write %{{.*}}, %[[MEMREF2]]{{.*}} : vector<3xf32>, memref // VECTOR_TRANSFERS: vector.transfer_write %{{.*}}, %[[MEMREF2]]{{.*}} : vector<3xf32>, memref scf.for %i = %lb to %ub step %step { scf.for %j = %lb to %ub step %step { %r00 = vector.transfer_read %memref1[%c0, %c0], %cst: memref, vector<2xf32> %r01 = vector.transfer_read %memref1[%c0, %c1], %cst: memref, vector<2xf32> %r20 = vector.transfer_read %memref2[%c0, %c0], %cst: memref, vector<3xf32> %r21 = vector.transfer_read %memref2[%c0, %c3], %cst: memref, vector<3xf32> %r30 = vector.transfer_read %memref3[%c0, %random_index], %cst: memref, vector<4xf32> %r31 = vector.transfer_read %memref3[%c1, %random_index], %cst: memref, vector<4xf32> %r10 = vector.transfer_read %memref0[%i, %i], %cst: memref, vector<2xf32> %r11 = vector.transfer_read %memref0[%random_index, %random_index], %cst: memref, vector<2xf32> %u00 = "some_use"(%r00) : (vector<2xf32>) -> vector<2xf32> %u01 = "some_use"(%r01) : (vector<2xf32>) -> vector<2xf32> %u20 = "some_use"(%r20) : (vector<3xf32>) -> vector<3xf32> %u21 = "some_use"(%r21) : (vector<3xf32>) -> vector<3xf32> %u30 = "some_use"(%r30) : (vector<4xf32>) -> vector<4xf32> %u31 = "some_use"(%r31) : (vector<4xf32>) -> vector<4xf32> %u10 = "some_use"(%r10) : (vector<2xf32>) -> vector<2xf32> %u11 = "some_use"(%r11) : (vector<2xf32>) -> vector<2xf32> vector.transfer_write %u00, %memref1[%c0, %c0] : vector<2xf32>, memref vector.transfer_write %u01, %memref1[%c0, %c1] : vector<2xf32>, memref vector.transfer_write %u20, %memref2[%c0, %c0] : vector<3xf32>, memref vector.transfer_write %u21, %memref2[%c0, %c3] : vector<3xf32>, memref vector.transfer_write %u30, %memref3[%c0, %random_index] : vector<4xf32>, memref vector.transfer_write %u31, %memref3[%c1, %random_index] : vector<4xf32>, memref vector.transfer_write %u10, %memref0[%i, %i] : vector<2xf32>, memref vector.transfer_write %u11, %memref0[%random_index, %random_index] : vector<2xf32>, memref } } return }