163 lines · plain
1// RUN: mlir-opt %s -test-loop-fusion=test-loop-fusion-slice-computation -split-input-file -verify-diagnostics | FileCheck %s2 3// -----4 5// CHECK-LABEL: func @slice_depth1_loop_nest() {6func.func @slice_depth1_loop_nest() {7 %0 = memref.alloc() : memref<100xf32>8 %cst = arith.constant 7.000000e+00 : f329 affine.for %i0 = 0 to 16 {10 // expected-remark@-1 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 1) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] )}}11 affine.store %cst, %0[%i0] : memref<100xf32>12 }13 affine.for %i1 = 0 to 5 {14 // expected-remark@-1 {{slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] )}}15 %1 = affine.load %0[%i1] : memref<100xf32>16 }17 return18}19 20// -----21 22// CHECK-LABEL: func @forward_slice_slice_depth1_loop_nest() {23func.func @forward_slice_slice_depth1_loop_nest() {24 %0 = memref.alloc() : memref<100xf32>25 %cst = arith.constant 7.000000e+00 : f3226 affine.for %i0 = 0 to 5 {27 // expected-remark@-1 {{slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 1) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] )}}28 affine.store %cst, %0[%i0] : memref<100xf32>29 }30 affine.for %i1 = 0 to 16 {31 // expected-remark@-1 {{Incorrect slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] )}}32 %1 = affine.load %0[%i1] : memref<100xf32>33 }34 return35}36 37// -----38 39// Loop %i0 writes to locations [2, 17] and loop %i0 reads from locations [3, 6]40// Slice loop bounds should be adjusted such that the load/store are for the41// same location.42// CHECK-LABEL: func @slice_depth1_loop_nest_with_offsets() {43func.func @slice_depth1_loop_nest_with_offsets() {44 %0 = memref.alloc() : memref<100xf32>45 %cst = arith.constant 7.000000e+00 : f3246 affine.for %i0 = 0 to 16 {47 // expected-remark@-1 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 2) loop bounds: [(d0) -> (d0 + 3), (d0) -> (d0 + 4)] )}}48 %a0 = affine.apply affine_map<(d0) -> (d0 + 2)>(%i0)49 affine.store %cst, %0[%a0] : memref<100xf32>50 }51 affine.for %i1 = 4 to 8 {52 // expected-remark@-1 {{slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0 - 3), (d0) -> (d0 - 2)] )}}53 %a1 = affine.apply affine_map<(d0) -> (d0 - 1)>(%i1)54 %1 = affine.load %0[%a1] : memref<100xf32>55 }56 return57}58 59// -----60 61// Slices at loop depth 1 should only slice the loop bounds of the first scf.62// Slices at loop depth 2 should slice loop bounds of both loops.63// CHECK-LABEL: func @slice_depth2_loop_nest() {64func.func @slice_depth2_loop_nest() {65 %0 = memref.alloc() : memref<100x100xf32>66 %cst = arith.constant 7.000000e+00 : f3267 affine.for %i0 = 0 to 16 {68 // expected-remark@-1 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 1) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (8)] )}}69 // expected-remark@-2 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 2 : insert point: (2, 1) loop bounds: [(d0, d1) -> (d0), (d0, d1) -> (d0 + 1)] [(d0, d1) -> (d1), (d0, d1) -> (d1 + 1)] )}}70 affine.for %i1 = 0 to 16 {71 affine.store %cst, %0[%i0, %i1] : memref<100x100xf32>72 }73 }74 affine.for %i2 = 0 to 10 {75 // expected-remark@-1 {{slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (8)] )}}76 // expected-remark@-2 {{slice ( src loop: 0, dst loop: 1, depth: 2 : insert point: (2, 0) loop bounds: [(d0, d1) -> (d0), (d0, d1) -> (d0 + 1)] [(d0, d1) -> (d1), (d0, d1) -> (d1 + 1)] )}}77 affine.for %i3 = 0 to 8 {78 %1 = affine.load %0[%i2, %i3] : memref<100x100xf32>79 }80 }81 return82}83 84// -----85 86// The load at depth 1 in loop nest %i2 prevents slicing loop nest %i0 at depths87// greater than 1. However, loop nest %i2 can be sliced into loop nest %i0 at88// depths 1 and 2 because the dependent store in loop nest %i0 is at depth 2.89// CHECK-LABEL: func @slice_depth2_loop_nest_two_loads() {90func.func @slice_depth2_loop_nest_two_loads() {91 %0 = memref.alloc() : memref<100x100xf32>92 %c0 = arith.constant 0 : index93 %cst = arith.constant 7.000000e+00 : f3294 affine.for %i0 = 0 to 16 {95 // expected-remark@-1 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 1) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (8)] )}}96 // expected-remark@-2 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 2 : insert point: (2, 1) loop bounds: [(d0, d1) -> (d0), (d0, d1) -> (d0 + 1)] [(d0, d1) -> (0), (d0, d1) -> (16)] )}}97 affine.for %i1 = 0 to 16 {98 affine.store %cst, %0[%i0, %i1] : memref<100x100xf32>99 }100 }101 affine.for %i2 = 0 to 10 {102 // expected-remark@-1 {{slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (8)] )}}103 affine.for %i3 = 0 to 8 {104 %1 = affine.load %0[%i2, %i3] : memref<100x100xf32>105 }106 %2 = affine.load %0[%i2, %c0] : memref<100x100xf32>107 }108 return109}110 111// -----112 113// The store at depth 1 in loop nest %i0 prevents slicing loop nest %i2 at114// depths greater than 1 into loop nest %i0. However, loop nest %i0 can be115// sliced into loop nest %i2 at depths 1 and 2 because the dependent load in116// loop nest %i2 is at depth 2.117// CHECK-LABEL: func @slice_depth2_loop_nest_two_stores() {118func.func @slice_depth2_loop_nest_two_stores() {119 %0 = memref.alloc() : memref<100x100xf32>120 %c0 = arith.constant 0 : index121 %cst = arith.constant 7.000000e+00 : f32122 affine.for %i0 = 0 to 16 {123 // expected-remark@-1 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 2) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (8)] )}}124 affine.for %i1 = 0 to 16 {125 affine.store %cst, %0[%i0, %i1] : memref<100x100xf32>126 }127 affine.store %cst, %0[%i0, %c0] : memref<100x100xf32>128 }129 affine.for %i2 = 0 to 10 {130 // expected-remark@-1 {{slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (16)] )}}131 // expected-remark@-2 {{slice ( src loop: 0, dst loop: 1, depth: 2 : insert point: (2, 0) loop bounds: [(d0, d1) -> (d0), (d0, d1) -> (d0 + 1)] [(d0, d1) -> (0), (d0, d1) -> (16)] )}}132 affine.for %i3 = 0 to 8 {133 %1 = affine.load %0[%i2, %i3] : memref<100x100xf32>134 }135 }136 return137}138 139// -----140 141// Test loop nest which has a smaller outer trip count than its inner scf.142// CHECK-LABEL: func @slice_loop_nest_with_smaller_outer_trip_count() {143func.func @slice_loop_nest_with_smaller_outer_trip_count() {144 %0 = memref.alloc() : memref<100x100xf32>145 %c0 = arith.constant 0 : index146 %cst = arith.constant 7.000000e+00 : f32147 affine.for %i0 = 0 to 16 {148 // expected-remark@-1 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 1 : insert point: (1, 1) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (10)] )}}149 // expected-remark@-2 {{Incorrect slice ( src loop: 1, dst loop: 0, depth: 2 : insert point: (2, 1) loop bounds: [(d0, d1) -> (d0), (d0, d1) -> (d0 + 1)] [(d0, d1) -> (d1), (d0, d1) -> (d1 + 1)] )}}150 affine.for %i1 = 0 to 16 {151 affine.store %cst, %0[%i0, %i1] : memref<100x100xf32>152 }153 }154 affine.for %i2 = 0 to 8 {155 // expected-remark@-1 {{slice ( src loop: 0, dst loop: 1, depth: 1 : insert point: (1, 0) loop bounds: [(d0) -> (d0), (d0) -> (d0 + 1)] [(d0) -> (0), (d0) -> (10)] )}}156 // expected-remark@-2 {{slice ( src loop: 0, dst loop: 1, depth: 2 : insert point: (2, 0) loop bounds: [(d0, d1) -> (d0), (d0, d1) -> (d0 + 1)] [(d0, d1) -> (d1), (d0, d1) -> (d1 + 1)] )}}157 affine.for %i3 = 0 to 10 {158 %1 = affine.load %0[%i2, %i3] : memref<100x100xf32>159 }160 }161 return162}163