// RUN: mlir-opt -allow-unregistered-dialect %s -pass-pipeline='builtin.module(func.func(affine-loop-fusion))' -split-input-file | FileCheck %s // Part II of fusion tests in mlir/test/Transforms/loop-fusion=2.mlir. // Part III of fusion tests in mlir/test/Transforms/loop-fusion-3.mlir // Part IV of fusion tests in mlir/test/Transforms/loop-fusion-4.mlir // TODO: Add more tests: // *) Add nested fusion test cases when non-constant loop bound support is // added to iteration domain in dependence check. // *) Add a test w/ floordiv/ceildiv/mod when supported in dependence check. // *) Add tests which check fused computation slice indexing and loop bounds. // TODO: Test clean up: move memref allocs to func args. // ----- // CHECK-LABEL: func @should_fuse_raw_dep_for_locality() { func.func @should_fuse_raw_dep_for_locality() { %m = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %m[%i1] : memref<10xf32> } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_reduction_to_pointwise() { func.func @should_fuse_reduction_to_pointwise() { %a = memref.alloc() : memref<10x10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.for %i1 = 0 to 10 { %v0 = affine.load %b[%i0] : memref<10xf32> %v1 = affine.load %a[%i0, %i1] : memref<10x10xf32> %v3 = arith.addf %v0, %v1 : f32 affine.store %v3, %b[%i0] : memref<10xf32> } } affine.for %i2 = 0 to 10 { %v4 = affine.load %b[%i2] : memref<10xf32> affine.store %v4, %c[%i2] : memref<10xf32> } // Should fuse in entire inner loop on %i1 from source loop nest, as %i1 // is not used in the access function of the store/load on %b. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32> // CHECK-NEXT: arith.addf %{{.*}}, %{{.*}} : f32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-DAG: [[$MAP_SHIFT_MINUS_ONE_R1:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 - 1)> // CHECK-DAG: [[$MAP_SHIFT_D0_BY_ONE:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (d0 + 1)> // CHECK-DAG: [[$MAP_SHIFT_D1_BY_ONE:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (d1 + 1)> // CHECK-LABEL: func @should_fuse_loop_nests_with_shifts() { func.func @should_fuse_loop_nests_with_shifts() { %a = memref.alloc() : memref<10x10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 9 { affine.for %i1 = 0 to 9 { affine.store %cf7, %a[%i0 + 1, %i1 + 1] : memref<10x10xf32> } } affine.for %i2 = 1 to 10 { affine.for %i3 = 1 to 10 { %v0 = affine.load %a[%i2, %i3] : memref<10x10xf32> } } // Source slice affine apply sequence: // *) First two affine apply's map from the dst to src iteration space. // *) Third affine apply is access function around src store. // *) Fourth affine apply shifts the stores access function by '-1', because // of the offset induced by reducing the memref shape from 10x10 to 9x9. // *) Fifth affine apply shifts the loads access function by '-1', because // of the offset induced by reducing the memref shape from 10x10 to 9x9. // NOTE: Should create a private memref with reduced shape 9x9xf32. // CHECK: affine.for %{{.*}} = 1 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 1 to 10 { // CHECK-NEXT: %[[I:.*]] = affine.apply [[$MAP_SHIFT_MINUS_ONE_R1]](%{{.*}}) // CHECK-NEXT: %[[J:.*]] = affine.apply [[$MAP_SHIFT_MINUS_ONE_R1]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP_SHIFT_D0_BY_ONE]](%[[I]], %[[J]]) // CHECK-NEXT: affine.apply [[$MAP_SHIFT_D1_BY_ONE]](%[[I]], %[[J]]) // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, 0] : memref<1x1xf32> // CHECK-NEXT: affine.load %{{.*}}[0, 0] : memref<1x1xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_loop_nest() { func.func @should_fuse_loop_nest() { %a = memref.alloc() : memref<10x10xf32> %b = memref.alloc() : memref<10x10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.for %i1 = 0 to 10 { affine.store %cf7, %a[%i0, %i1] : memref<10x10xf32> } } affine.for %i2 = 0 to 10 { affine.for %i3 = 0 to 10 { %v0 = affine.load %a[%i3, %i2] : memref<10x10xf32> affine.store %v0, %b[%i2, %i3] : memref<10x10xf32> } } affine.for %i4 = 0 to 10 { affine.for %i5 = 0 to 10 { %v1 = affine.load %b[%i4, %i5] : memref<10x10xf32> } } // Expecting private memref for '%a' first, then private memref for '%b'. // CHECK-DAG: [[NEWA:%[0-9a-zA-Z_]+]] = memref.alloc() : memref<1x1xf32> // CHECK-DAG: [[NEWB:%[0-9a-zA-Z_]+]] = memref.alloc() : memref<1x1xf32> // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, [[NEWA]][0, 0] : memref<1x1xf32> // CHECK-NEXT: affine.load [[NEWA]][0, 0] : memref<1x1xf32> // CHECK-NEXT: affine.store %{{.*}}, [[NEWB]][0, 0] : memref<1x1xf32> // CHECK-NEXT: affine.load [[NEWB]][0, 0] : memref<1x1xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_across_intermediate_loop_with_no_deps() { func.func @should_fuse_across_intermediate_loop_with_no_deps() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { %v0 = affine.load %a[%i0] : memref<10xf32> affine.store %v0, %b[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %c[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v1 = affine.load %b[%i2] : memref<10xf32> } // Should fuse first loop (past second loop with no dependences) into third. // Note that fusion creates a private memref '%2' for the fused loop nest. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_all_loops() { func.func @should_fuse_all_loops() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 // Set up flow dependences from first and second loops to third. affine.for %i0 = 0 to 10 { affine.store %cf7, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %b[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v0 = affine.load %a[%i2] : memref<10xf32> %v1 = affine.load %b[%i2] : memref<10xf32> } // Should fuse first and second loops into third. // Expecting private memref for '%a' first, then private memref for '%b'. // CHECK-DAG: [[NEWA:%[0-9a-zA-Z_]+]] = memref.alloc() : memref<1xf32> // CHECK-DAG: [[NEWB:%[0-9a-zA-Z_]+]] = memref.alloc() : memref<1xf32> // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, [[NEWA]][0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, [[NEWB]][0] : memref<1xf32> // CHECK-NEXT: affine.load [[NEWA]][0] : memref<1xf32> // CHECK-NEXT: affine.load [[NEWB]][0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_first_and_second_loops() { func.func @should_fuse_first_and_second_loops() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %a[%i1] : memref<10xf32> affine.store %cf7, %b[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v1 = affine.load %c[%i2] : memref<10xf32> } // Should fuse first loop into the second (last loop should not be fused). // Should create private memref '%2' for fused scf. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_not_fuse_would_create_cycle() { func.func @should_not_fuse_would_create_cycle() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 // Set up the following dependences: // 1) loop0 -> loop1 on memref '%{{.*}}' // 2) loop0 -> loop2 on memref '%{{.*}}' // 3) loop1 -> loop2 on memref '%{{.*}}' affine.for %i0 = 0 to 10 { %v0 = affine.load %a[%i0] : memref<10xf32> affine.store %cf7, %b[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %a[%i1] : memref<10xf32> %v1 = affine.load %c[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v2 = affine.load %b[%i2] : memref<10xf32> affine.store %cf7, %c[%i2] : memref<10xf32> } // Should not fuse: fusing loop first loop into last would create a cycle. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_producer_consumer() { func.func @should_fuse_producer_consumer() { %m = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %m[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v1 = affine.load %m[%i2] : memref<10xf32> } // Fusing loop %i0 to %i2 would violate the WAW dependence between %i0 and // %i1, but OK to fuse %i1 into %i2. // TODO: When the fusion pass is run to a fixed-point, it should // fuse all three of these loop nests. // CHECK: memref.alloc() : memref<1xf32> // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_and_move_to_preserve_war_dep() { func.func @should_fuse_and_move_to_preserve_war_dep() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { %v0 = affine.load %a[%i0] : memref<10xf32> affine.store %v0, %b[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %a[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v1 = affine.load %b[%i2] : memref<10xf32> } // Loops '%i1' and '%i2' have no dependences. We can fuse a slice of '%i0' // into '%i2' if we move the fused loop nest before '%i1', which preserves // the WAR dependence from load '%a' in '%i0' to the store '%a' in loop '%i1'. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_if_top_level_access() { func.func @should_fuse_if_top_level_access() { %m = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %m[%i1] : memref<10xf32> } %c0 = arith.constant 4 : index %v1 = affine.load %m[%c0] : memref<10xf32> // Top-level load to '%m' should prevent creating a private memref but // loop nests should be fused and '%i0' should be removed. // CHECK: %[[m:.*]] = memref.alloc() : memref<10xf32> // CHECK-NOT: memref.alloc // CHECK: affine.for %[[i1:.*]] = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %[[m]][%[[i1]]] : memref<10xf32> // CHECK-NEXT: affine.load %[[m]][%[[i1]]] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.load %[[m]][%{{.*}}] : memref<10xf32> return } // ----- // CHECK-LABEL: func @should_fuse_but_not_remove_src() { func.func @should_fuse_but_not_remove_src() { %m = memref.alloc() : memref<100xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 100 { affine.store %cf7, %m[%i0] : memref<100xf32> } affine.for %i1 = 0 to 17 { %v0 = affine.load %m[%i1] : memref<100xf32> } %v1 = affine.load %m[99] : memref<100xf32> // Loop '%i0' and '%i1' should be fused but '%i0' shouldn't be removed to // preserve the dependence with the top-level access. // CHECK: affine.for %{{.*}} = 0 to 100 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<100xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 17 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.load %{{.*}}[99] : memref<100xf32> // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_no_top_level_access() { func.func @should_fuse_no_top_level_access() { %m = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %m[%i1] : memref<10xf32> } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- #set0 = affine_set<(d0) : (1 == 0)> // CHECK-LABEL: func @should_not_fuse_if_op_at_top_level() { func.func @should_not_fuse_if_op_at_top_level() { %m = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %m[%i1] : memref<10xf32> } %c0 = arith.constant 4 : index affine.if #set0(%c0) { } // Top-level IfOp should prevent fusion. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } return } // ----- #set0 = affine_set<(d0) : (1 == 0)> // CHECK-LABEL: func @should_not_fuse_if_op_in_loop_nest() { func.func @should_not_fuse_if_op_in_loop_nest() { %m = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 %c4 = arith.constant 4 : index affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.if #set0(%c4) { } %v0 = affine.load %m[%i1] : memref<10xf32> } // IfOp in ForOp should prevent fusion. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.if #set(%{{.*}}) { // CHECK-NEXT: } // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } return } // ----- #set = affine_set<(d0) : (d0 - 1 >= 0)> // CHECK-LABEL: func @should_fuse_if_op_in_loop_nest_not_sandwiched() -> memref<10xf32> { func.func @should_fuse_if_op_in_loop_nest_not_sandwiched() -> memref<10xf32> { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %a[%i1] : memref<10xf32> affine.store %v0, %b[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { affine.if #set(%i2) { %v0 = affine.load %b[%i2] : memref<10xf32> } } // IfOp in ForOp should not prevent fusion if it does not in between the // source and dest ForOp ops. // CHECK: affine.for // CHECK-NEXT: affine.store // CHECK-NEXT: affine.load // CHECK-NEXT: affine.store // CHECK: affine.for // CHECK-NEXT: affine.if // CHECK-NEXT: affine.load // CHECK-NOT: affine.for // CHECK: return return %a : memref<10xf32> } // ----- #set = affine_set<(d0) : (d0 - 1 >= 0)> // CHECK-LABEL: func @should_not_fuse_if_op_in_loop_nest_between_src_and_dest() -> memref<10xf32> { func.func @should_not_fuse_if_op_in_loop_nest_between_src_and_dest() -> memref<10xf32> { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.if #set(%i1) { affine.store %cf7, %a[%i1] : memref<10xf32> } } affine.for %i3 = 0 to 10 { %v0 = affine.load %a[%i3] : memref<10xf32> affine.store %v0, %b[%i3] : memref<10xf32> } return %b : memref<10xf32> // IfOp in ForOp which modifies the memref should prevent fusion if it is in // between the source and dest ForOp. // CHECK: affine.for // CHECK-NEXT: affine.store // CHECK: affine.for // CHECK-NEXT: affine.if // CHECK-NEXT: affine.store // CHECK: affine.for // CHECK-NEXT: affine.load // CHECK-NEXT: affine.store // CHECK: return } // ----- // CHECK-LABEL: func @permute_and_fuse() { func.func @permute_and_fuse() { %m = memref.alloc() : memref<10x20x30xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.for %i1 = 0 to 20 { affine.for %i2 = 0 to 30 { affine.store %cf7, %m[%i0, %i1, %i2] : memref<10x20x30xf32> } } } affine.for %i3 = 0 to 30 { affine.for %i4 = 0 to 10 { affine.for %i5 = 0 to 20 { %v0 = affine.load %m[%i4, %i5, %i3] : memref<10x20x30xf32> "foo"(%v0) : (f32) -> () } } } // CHECK: affine.for %{{.*}} = 0 to 30 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 20 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, 0, 0] : memref<1x1x1xf32> // CHECK-NEXT: affine.load %{{.*}}[0, 0, 0] : memref<1x1x1xf32> // CHECK-NEXT: "foo"(%{{.*}}) : (f32) -> () // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-DAG: [[$MAP0:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (d0 * 4 + d1)> // CHECK-DAG: [[$MAP1:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 floordiv 4)> // CHECK-DAG: [[$MAP2:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 mod 4)> // Reshape from a 64 x f32 to 16 x 4 x f32. // CHECK-LABEL: func @fuse_reshape_64_16_4 func.func @fuse_reshape_64_16_4(%in : memref<64xf32>) { %out = memref.alloc() : memref<16x4xf32> affine.for %i0 = 0 to 64 { %v = affine.load %in[%i0] : memref<64xf32> affine.store %v, %out[%i0 floordiv 4, %i0 mod 4] : memref<16x4xf32> } affine.for %i1 = 0 to 16 { affine.for %i2 = 0 to 4 { %w = affine.load %out[%i1, %i2] : memref<16x4xf32> "foo"(%w) : (f32) -> () } } return // CHECK: affine.for %{{.*}} = // CHECK-NEXT: affine.for %{{.*}} = // CHECK-NOT: for // CHECK: } // CHECK-NEXT: } // CHECK-NEXT: return } // ----- // CHECK-DAG: [[$MAP0:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 floordiv 4)> // CHECK-DAG: [[$MAP1:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 mod 4)> // CHECK-DAG: [[$MAP2:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (d0 * 4 + d1)> // Reshape a 16x4xf32 to 64xf32. // CHECK-LABEL: func @fuse_reshape_16_4_64 func.func @fuse_reshape_16_4_64() { %in = memref.alloc() : memref<16x4xf32> %out = memref.alloc() : memref<64xf32> affine.for %i0 = 0 to 16 { affine.for %i1 = 0 to 4 { %v = affine.load %in[%i0, %i1] : memref<16x4xf32> affine.store %v, %out[4*%i0 + %i1] : memref<64xf32> } } affine.for %i2 = 0 to 64 { %w = affine.load %out[%i2] : memref<64xf32> "foo"(%w) : (f32) -> () } // CHECK: affine.for %{{.*}} = 0 to 64 { // CHECK-NEXT: affine.apply [[$MAP0]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP1]](%{{.*}}) // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<16x4xf32> // CHECK-NEXT: affine.apply [[$MAP2]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: "foo"(%{{.*}}) : (f32) -> () // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // All three loop nests below (6-d one, 2-d one, 2-d one is fused into a single // 2-d loop nest). func.func @R6_to_R2_reshape_square() -> memref<64x9xi32> { %in = memref.alloc() : memref<2x2x3x3x16x1xi32> %out = memref.alloc() : memref<64x9xi32> %live_out = memref.alloc() : memref<64x9xi32> // Initialize input. affine.for %i0 = 0 to 2 { affine.for %i1 = 0 to 2 { affine.for %i2 = 0 to 3 { affine.for %i3 = 0 to 3 { affine.for %i4 = 0 to 16 { affine.for %i5 = 0 to 1 { %val = "foo"(%i0, %i1, %i2, %i3, %i4, %i5) : (index, index, index, index, index, index) -> i32 affine.store %val, %in[%i0, %i1, %i2, %i3, %i4, %i5] : memref<2x2x3x3x16x1xi32> } } } } } } affine.for %ii = 0 to 64 { affine.for %jj = 0 to 9 { // Convert output coordinates to linear index. %a0 = affine.apply affine_map<(d0, d1) -> (d0 * 9 + d1)> (%ii, %jj) %0 = affine.apply affine_map<(d0) -> (d0 floordiv (2 * 3 * 3 * 16 * 1))>(%a0) %1 = affine.apply affine_map<(d0) -> ((d0 mod 288) floordiv (3 * 3 * 16 * 1))>(%a0) %2 = affine.apply affine_map<(d0) -> (((d0 mod 288) mod 144) floordiv (3 * 16 * 1))>(%a0) %3 = affine.apply affine_map<(d0) -> ((((d0 mod 288) mod 144) mod 48) floordiv (16 * 1))>(%a0) %4 = affine.apply affine_map<(d0) -> ((((d0 mod 288) mod 144) mod 48) mod 16)>(%a0) %5 = affine.apply affine_map<(d0) -> (((((d0 mod 144) mod 144) mod 48) mod 16) mod 1)>(%a0) %v = affine.load %in[%0, %1, %2, %3, %4, %5] : memref<2x2x3x3x16x1xi32> affine.store %v, %out[%ii, %jj] : memref<64x9xi32> } } affine.for %i = 0 to 64 { affine.for %j = 0 to 9 { %a = affine.load %out[%i, %j] : memref<64x9xi32> %b = arith.muli %a, %a : i32 affine.store %b, %live_out[%i, %j] : memref<64x9xi32> } } return %live_out : memref<64x9xi32> } // Everything above is fused to a single 2-d loop nest, and the 6-d tensor %in // is eliminated if -memref-dataflow-opt is also supplied. // // CHECK-DAG: [[$MAP0:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> ((d0 * 9 + d1) floordiv 288)> // CHECK-DAG: [[$MAP1:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (((d0 * 9 + d1) mod 288) floordiv 144)> // CHECK-DAG: [[$MAP2:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (((d0 * 9 + d1) mod 144) floordiv 48)> // CHECK-DAG: [[$MAP3:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (((d0 * 9 + d1) mod 48) floordiv 16)> // CHECK-DAG: [[$MAP4:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> ((d0 * 9 + d1) mod 16)> // CHECK-DAG: [[$MAP11:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (d0 * 9 + d1)> // CHECK-DAG: [[$MAP12:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 floordiv 288)> // CHECK-DAG: [[$MAP13:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> ((d0 mod 288) floordiv 144)> // CHECK-DAG: [[$MAP14:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> ((d0 mod 144) floordiv 48)> // CHECK-DAG: [[$MAP15:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> ((d0 mod 48) floordiv 16)> // CHECK-DAG: [[$MAP16:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 mod 16)> // CHECK-DAG: [[$MAP17:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (0)> // // CHECK-LABEL: func @R6_to_R2_reshape // CHECK: memref.alloc() : memref<1x2x3x3x16x1xi32> // CHECK: memref.alloc() : memref<1x1xi32> // CHECK: memref.alloc() : memref<64x9xi32> // CHECK-NEXT: affine.for %{{.*}} = 0 to 64 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 9 { // CHECK-NEXT: affine.apply [[$MAP0]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.apply [[$MAP1]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.apply [[$MAP2]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.apply [[$MAP3]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.apply [[$MAP4]](%{{.*}}, %{{.*}}) // CHECK-NEXT: "foo"(%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}) : (index, index, index, index, index, index) -> i32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, ((%{{.*}} * 9 + %{{.*}}) mod 288) floordiv 144, ((%{{.*}} * 9 + %{{.*}}) mod 144) floordiv 48, ((%{{.*}} * 9 + %{{.*}}) mod 48) floordiv 16, (%{{.*}} * 9 + %{{.*}}) mod 16, 0] : memref<1x2x3x3x16x1xi32> // CHECK-NEXT: affine.apply [[$MAP11]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.apply [[$MAP12]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP13]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP14]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP15]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP16]](%{{.*}}) // CHECK-NEXT: affine.apply [[$MAP17]](%{{.*}}) // CHECK-NEXT: affine.load %{{.*}}[0, ((%{{.*}} * 9 + %{{.*}}) mod 288) floordiv 144, ((%{{.*}} * 9 + %{{.*}}) mod 144) floordiv 48, ((%{{.*}} * 9 + %{{.*}}) mod 48) floordiv 16, (%{{.*}} * 9 + %{{.*}}) mod 16, 0] : memref<1x2x3x3x16x1xi32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, 0] : memref<1x1xi32> // CHECK-NEXT: affine.load %{{.*}}[0, 0] : memref<1x1xi32> // CHECK-NEXT: arith.muli %{{.*}}, %{{.*}} : i32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}, %{{.*}}] : memref<64x9xi32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return %{{.*}} : memref<64x9xi32> // ----- // CHECK-LABEL: func @fuse_symbolic_bounds func.func @fuse_symbolic_bounds(%M : index, %N : index) { %N_plus_5 = affine.apply affine_map<(d0) -> (d0 + 5)>(%N) %m = memref.alloc(%M, %N_plus_5) : memref %c0 = arith.constant 0.0 : f32 %s = arith.constant 5 : index affine.for %i0 = 0 to %M { affine.for %i1 = 0 to affine_map<(d0) -> (d0 + 5)> (%N) { affine.store %c0, %m[%i0, %i1] : memref } } affine.for %i2 = 0 to %M { affine.for %i3 = 0 to %N { %v = affine.load %m[%i2, %i3 + symbol(%s)] : memref } } return } // ----- // CHECK-LABEL: func @should_fuse_reduction_at_depth_of_one func.func @should_fuse_reduction_at_depth_of_one() { %a = memref.alloc() : memref<10x100xf32> %b = memref.alloc() : memref<10xf32> affine.for %i0 = 0 to 10 { affine.for %i1 = 0 to 100 { %v0 = affine.load %b[%i0] : memref<10xf32> %v1 = affine.load %a[%i0, %i1] : memref<10x100xf32> %v2 = "maxf"(%v0, %v1) : (f32, f32) -> f32 affine.store %v2, %b[%i0] : memref<10xf32> } } affine.for %i2 = 0 to 10 { affine.for %i3 = 0 to 100 { %v3 = affine.load %b[%i2] : memref<10xf32> %v4 = affine.load %a[%i2, %i3] : memref<10x100xf32> %v5 = arith.subf %v4, %v3 : f32 affine.store %v5, %b[%i2] : memref<10xf32> } } // This test should fuse the src reduction loop at depth 1 in the destination // loop nest, which improves locality and enables subsequence passes to // decrease the reduction memref size and possibly place it in a faster // memory space. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 100 { // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x100xf32> // CHECK-NEXT: "maxf"(%{{.*}}, %{{.*}}) : (f32, f32) -> f32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 100 { // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x100xf32> // CHECK-NEXT: arith.subf %{{.*}}, %{{.*}} : f32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_at_src_depth1_and_dst_depth1 func.func @should_fuse_at_src_depth1_and_dst_depth1() { %a = memref.alloc() : memref<100x16xf32> %b = memref.alloc() : memref<100x16xf32> affine.for %i0 = 0 to 100 { affine.for %i1 = 0 to 16 { %v0 = affine.load %a[%i0, %i1] : memref<100x16xf32> "op0"(%v0) : (f32) -> () } affine.for %i2 = 0 to 16 { %v1 = "op1"() : () -> (f32) affine.store %v1, %b[%i0, %i2] : memref<100x16xf32> } } affine.for %i3 = 0 to 100 { affine.for %i4 = 0 to 16 { %v2 = affine.load %b[%i3, %i4] : memref<100x16xf32> "op2"(%v2) : (f32) -> () } } // We can slice iterations of the '%i0' and '%i1' loops in the source // loop nest, but slicing at depth 2 and inserting the slice in the // destination loop nest at depth2 causes extra computation. Instead, // the fusion algorithm should detect that the source loop should be sliced // at depth 1 and the slice should be inserted at depth 1. // CHECK: affine.for %{{.*}} = 0 to 100 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<100x16xf32> // CHECK-NEXT: "op0"(%{{.*}}) : (f32) -> () // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: %{{.*}} = "op1"() : () -> f32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, %{{.*}}] : memref<1x16xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.load %{{.*}}[0, %{{.*}}] : memref<1x16xf32> // CHECK-NEXT: "op2"(%{{.*}}) : (f32) -> () // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK: [[$MAP0:#map[0-9]*]] = affine_map<(d0, d1) -> (d0 * 10 + d1)> // CHECK-LABEL: func @should_fuse_src_depth1_at_dst_depth2 func.func @should_fuse_src_depth1_at_dst_depth2() { %a = memref.alloc() : memref<100xf32> %c0 = arith.constant 0.0 : f32 affine.for %i0 = 0 to 100 { affine.store %c0, %a[%i0] : memref<100xf32> } affine.for %i1 = 0 to 10 { affine.for %i2 = 0 to 10 { %a0 = affine.apply affine_map<(d0, d1) -> (d0 * 10 + d1)> (%i1, %i2) %v0 = affine.load %a[%a0] : memref<100xf32> } } // The source loop nest slice loop bound is a function of both destination // loop IVs, so we should slice at depth 1 and insert the slice at depth 2. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.apply [[$MAP0]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.apply [[$MAP0]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @fusion_at_depth0_not_currently_supported func.func @fusion_at_depth0_not_currently_supported() { %0 = memref.alloc() : memref<10xf32> %c0 = arith.constant 0 : index %cst = arith.constant 0.000000e+00 : f32 affine.for %i0 = 0 to 10 { affine.store %cst, %0[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %1 = affine.load %0[%c0] : memref<10xf32> } // NOTE: Should shrink memref size to 1 element access by load in dst loop // nest, and make the store in the slice store to the same element. // CHECK-DAG: memref.alloc() : memref<1xf32> // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_deep_loop_nests func.func @should_fuse_deep_loop_nests() { %0 = memref.alloc() : memref<2x2x3x3x16x10xf32, 2> %1 = memref.alloc() : memref<2x2x3x3x16x10xf32, 2> %2 = memref.alloc() : memref<3x3x3x3x16x10xf32, 2> %c0 = arith.constant 0 : index %c1 = arith.constant 1 : index %c1_0 = arith.constant 1 : index %cst = arith.constant 0.000000e+00 : f32 affine.for %i0 = 0 to 2 { affine.for %i1 = 0 to 2 { affine.for %i2 = 0 to 3 { affine.for %i3 = 0 to 3 { affine.for %i4 = 0 to 16 { affine.for %i5 = 0 to 10 { %3 = affine.load %0[%i0, %i1, %i2, %i3, %i4, %i5] : memref<2x2x3x3x16x10xf32, 2> } } affine.for %i6 = 0 to 16 { affine.for %i7 = 0 to 10 { affine.store %cst, %1[%i0, %i1, %i2, %i3, %i6, %i7] : memref<2x2x3x3x16x10xf32, 2> } } } } } } affine.for %i8 = 0 to 3 { affine.for %i9 = 0 to 3 { affine.for %i10 = 0 to 2 { affine.for %i11 = 0 to 2 { affine.for %i12 = 0 to 3 { affine.for %i13 = 0 to 3 { affine.for %i14 = 0 to 2 { affine.for %i15 = 0 to 2 { affine.for %i16 = 0 to 16 { affine.for %i17 = 0 to 10 { %5 = affine.load %0[%i14, %i15, %i12, %i13, %i16, %i17] : memref<2x2x3x3x16x10xf32, 2> } } affine.for %i18 = 0 to 16 { affine.for %i19 = 0 to 10 { %6 = affine.load %1[%i10, %i11, %i8, %i9, %i18, %i19] : memref<2x2x3x3x16x10xf32, 2> } } } } } } } } } } // The first four loops of the source loop nest can be sliced with iteration // bounds which are a function of the first four loops of destination loop nest, // where the destination loops nests have been interchanged. // CHECK-DAG: memref.alloc() : memref<1x1x1x1x16x10xf32, 2> // CHECK: affine.for %{{.*}} = 0 to 3 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 3 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 2 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 2 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 3 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 3 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}] : memref<2x2x3x3x16x10xf32, 2> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, 0, 0, 0, %{{.*}}, %{{.*}}] : memref<1x1x1x1x16x10xf32, 2> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 2 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 2 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}] : memref<2x2x3x3x16x10xf32, 2> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[0, 0, 0, 0, %{{.*}}, %{{.*}}] : memref<1x1x1x1x16x10xf32, 2> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_at_depth1_and_reduce_slice_trip_count func.func @should_fuse_at_depth1_and_reduce_slice_trip_count() { %a = memref.alloc() : memref<4x256xf32> %b = memref.alloc() : memref<4x256xf32> %c0 = arith.constant 0 : index %cf0 = arith.constant 0.0 : f32 affine.for %i0 = 0 to 4 { affine.for %i1 = 0 to 256 { %v0 = affine.load %b[%i0, %i1] : memref<4x256xf32> } affine.for %i2 = 0 to 256 { affine.store %cf0, %a[%i0, %i2] : memref<4x256xf32> } } affine.for %d0 = 0 to 4 { affine.for %d1 = 0 to 16 { %v1 = affine.load %a[%d0, %d1] : memref<4x256xf32> } } // The cost of fusing at depth 2 is greater than the cost of fusing at depth 1 // for two reasons: // 1) Inserting the unsliceable src loop %i1 to a higher depth removes // redundant computation and reduces costs. // 2) Inserting the sliceable src loop %i2 at depth 1, we can still reduce // its trip count to 16 (from 256) reducing costs. // NOTE: the size of the private memref created for the fused loop nest // is reduced from the original shape from 4x256 to 4x16 because of the // data accessed by the load. // CHECK-DAG: memref.alloc() : memref<1x16xf32> // CHECK: affine.for %{{.*}} = 0 to 4 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 256 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<4x256xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, %{{.*}}] : memref<1x16xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 16 { // CHECK-NEXT: affine.load %{{.*}}[0, %{{.*}}] : memref<1x16xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_at_depth1_with_trip_count_20 func.func @should_fuse_at_depth1_with_trip_count_20() { %a = memref.alloc() : memref<100xf32> %c0 = arith.constant 0 : index %cf0 = arith.constant 0.0 : f32 affine.for %i0 = 0 to 100 { affine.store %cf0, %a[%i0]: memref<100xf32> } affine.for %i1 = 0 to 5 { affine.for %i2 = 0 to 10 { %v0 = affine.load %a[%i2]: memref<100xf32> } affine.for %i3 = 0 to 10 { affine.for %i4 = 0 to 20 { %v1 = affine.load %a[%i4]: memref<100xf32> } } } // NOTE: The size of the private memref created for fusion is shrunk to 20xf32 // CHECK-DAG: memref.alloc() : memref<20xf32> // CHECK: affine.for %{{.*}} = 0 to 5 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 20 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<20xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<20xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 20 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<20xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_at_depth1_with_trip_count_19 func.func @should_fuse_at_depth1_with_trip_count_19() { %a = memref.alloc() : memref<100xf32> %c0 = arith.constant 0 : index %cf0 = arith.constant 0.0 : f32 affine.for %i0 = 0 to 100 { affine.store %cf0, %a[%i0]: memref<100xf32> } affine.for %i1 = 0 to 5 { affine.for %i2 = 0 to 19 { %v0 = affine.load %a[%i2]: memref<100xf32> } affine.for %i3 = 0 to 10 { affine.for %i4 = 0 to 10 { %v1 = affine.load %a[%i4]: memref<100xf32> } } } // NOTE: The size of the private memref created for fusion is shrunk to 19xf32 // CHECK-DAG: memref.alloc() : memref<19xf32> // CHECK: affine.for %{{.*}} = 0 to 5 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 19 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<19xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 19 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<19xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<19xf32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_with_private_memref() { func.func @should_fuse_with_private_memref() { %m = memref.alloc() : memref<100xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 100 { affine.store %cf7, %m[%i0] : memref<100xf32> } affine.for %i1 = 0 to 17 { %v0 = affine.load %m[%i1] : memref<100xf32> } affine.for %i2 = 0 to 82 { %v1 = affine.load %m[%i2] : memref<100xf32> } // Should create a new private memref. // CHECK-DAG: memref.alloc() : memref<1xf32> // CHECK: affine.for %{{.*}} = 0 to 17 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_live_out_arg_but_preserve_src_loop(%{{.*}}: memref<10xf32>) { func.func @should_fuse_live_out_arg_but_preserve_src_loop(%arg0: memref<10xf32>) { %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %arg0[%i0] : memref<10xf32> } affine.for %i1 = 0 to 9 { %v0 = affine.load %arg0[%i1] : memref<10xf32> } // This tests that the loop nest '%i0' should not be removed after fusion // because it writes to memref argument '%arg0', and its read region // does not cover its write region (so fusion would shrink the write region // in the fused loop nest, so complete live out data region would not // be written). // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}} : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 9 { // CHECK-NEXT: affine.store %{{.*}} : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}} : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_live_out_arg(%{{.*}}: memref<10xf32>) { func.func @should_fuse_live_out_arg(%arg0: memref<10xf32>) { %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %arg0[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %arg0[%i1] : memref<10xf32> } // The read/write regions for memref '%{{.*}}' are the same for both // loops, so they should fuse. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_escaping_memref_but_preserve_src_loop() -> memref<10xf32> func.func @should_fuse_escaping_memref_but_preserve_src_loop() -> memref<10xf32> { %cf7 = arith.constant 7.0 : f32 %m = memref.alloc() : memref<10xf32> affine.for %i0 = 0 to 10 { affine.store %cf7, %m[%i0] : memref<10xf32> } affine.for %i1 = 0 to 9 { %v0 = affine.load %m[%i1] : memref<10xf32> } // This tests that the loop nest '%i0' should not be removed after fusion // because it writes to memref '%m', which is returned by the function, and // the '%i1' memory region does not cover '%i0' memory region. // CHECK-DAG: memref.alloc() : memref<1xf32> // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}} : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 9 { // CHECK-NEXT: affine.store %{{.*}} : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}} : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return %{{.*}} : memref<10xf32> return %m : memref<10xf32> } // ----- // This should fuse with the %in becoming a 1x1x1. func.func @R3_to_R2_reshape() { %in = memref.alloc() : memref<2x3x16xi32> %c0 = arith.constant 0 : index affine.for %i0 = 0 to 2 { affine.for %i1 = 0 to 3 { affine.for %i2 = 0 to 16 { %val = "foo"(%i0, %i1, %i2) : (index, index, index) -> i32 affine.store %val, %in[%i0, %i1, %i2] : memref<2x3x16xi32> } } } affine.for %ii = 0 to 32 { affine.for %jj = 0 to 3 { %a0 = affine.apply affine_map<(d0, d1) -> (d0 * 3 + d1)> (%ii, %jj) %idx = affine.apply affine_map<(d0) -> (d0 floordiv (3 * 16))> (%a0) %v = affine.load %in[%idx, %jj, %c0] : memref<2x3x16xi32> } } return } // CHECK-DAG: [[$MAP0:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> ((d0 * 3 + d1) floordiv 48)> // CHECK-DAG: [[$MAP1:#map[0-9a-zA-Z_]*]] = affine_map<(d0, d1) -> (d0 * 3 + d1)> // CHECK-DAG: [[$MAP2:#map[0-9a-zA-Z_]*]] = affine_map<(d0) -> (d0 floordiv 48)> // CHECK-LABEL: func @R3_to_R2_reshape() // CHECK-DAG: memref.alloc() : memref<1x1x1xi32> // CHECK: affine.for %{{.*}} = 0 to 32 { // CHECK-NEXT: affine.for %{{.*}} = 0 to 3 { // CHECK-NEXT: affine.apply [[$MAP0]](%{{.*}}, %{{.*}}) // CHECK-NEXT: "foo"(%{{.*}}, %{{.*}}, %{{.*}}) : (index, index, index) -> i32 // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0, 0, 0] : memref<1x1x1xi32> // CHECK-NEXT: affine.apply [[$MAP1]](%{{.*}}, %{{.*}}) // CHECK-NEXT: affine.apply [[$MAP2]](%{{.*}}) // CHECK-NEXT: affine.load %{{.*}}[0, 0, 0] : memref<1x1x1xi32> // CHECK-NEXT: } // CHECK-NEXT: } // CHECK-NEXT: return // ----- func.func @should_fuse_multi_output_producer() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { affine.store %cf7, %a[%i0] : memref<10xf32> affine.store %cf7, %b[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { %v0 = affine.load %a[%i1] : memref<10xf32> %v1 = affine.load %b[%i1] : memref<10xf32> } // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @fusion_preventing_deps_on_middle_loop() { func.func @fusion_preventing_deps_on_middle_loop() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { %v0 = affine.load %a[%i0] : memref<10xf32> affine.store %v0, %b[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %a[%i1] : memref<10xf32> %v1 = affine.load %c[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v2 = affine.load %b[%i2] : memref<10xf32> affine.store %v2, %c[%i2] : memref<10xf32> } // Loops '%i0' and '%i2' cannot fuse along producer/consumer edge on memref // '%b', because of the WAR dep from '%i0' to '%i1' on memref '%a' and // because of the WAR dep from '%i1' to '%i2' on memref '%c'. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_and_move_to_preserve_war_dep() { func.func @should_fuse_and_move_to_preserve_war_dep() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { %v0 = affine.load %b[%i0] : memref<10xf32> affine.store %v0, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 3 { %v2 = affine.load %c[%i1] : memref<10xf32> } affine.for %i2 = 0 to 5 { affine.store %cf7, %b[%i2] : memref<10xf32> } affine.for %i3 = 0 to 10 { %v1 = affine.load %a[%i3] : memref<10xf32> affine.store %cf7, %c[%i3] : memref<10xf32> } // Dependence graph: // // %i0 --------- // | | // --- %i1 | %b | %a // | | | // %c | %i2 <-- | // | | // --> %i3 <-------- // // It is possible to fuse loop '%i0' into '%i3' and preserve dependences // if the fused loop nest is inserted between loops '%i1' and '%i2'. // CHECK-DAG: memref.alloc() : memref<1xf32> // CHECK: affine.for %{{.*}} = 0 to 3 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 5 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @fusion_preventing_dep_on_constant() { func.func @fusion_preventing_dep_on_constant() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 affine.for %i0 = 0 to 10 { %v0 = affine.load %b[%i0] : memref<10xf32> affine.store %cf7, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %b[%i1] : memref<10xf32> } %cf11 = arith.constant 11.0 : f32 affine.for %i2 = 0 to 10 { %v2 = affine.load %a[%i2] : memref<10xf32> affine.store %cf11, %c[%i2] : memref<10xf32> } // Loops '%i0' and '%i2' cannot fuse along producer/consumer edge on memref // '%a', because of the WAR dep from '%i0' to '%i1' on memref '%b' and // because of the SSA value dep from '%cf11' def to use in '%i2'. // CHECK: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: %{{.*}} = arith.constant 1.100000e+01 : f32 // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: func @should_fuse_and_preserve_dep_on_constant() { func.func @should_fuse_and_preserve_dep_on_constant() { %a = memref.alloc() : memref<10xf32> %b = memref.alloc() : memref<10xf32> %c = memref.alloc() : memref<10xf32> %cf7 = arith.constant 7.0 : f32 %cf11 = arith.constant 11.0 : f32 affine.for %i0 = 0 to 10 { %v0 = affine.load %b[%i0] : memref<10xf32> affine.store %cf7, %a[%i0] : memref<10xf32> } affine.for %i1 = 0 to 10 { affine.store %cf7, %b[%i1] : memref<10xf32> } affine.for %i2 = 0 to 10 { %v2 = affine.load %a[%i2] : memref<10xf32> affine.store %cf11, %c[%i2] : memref<10xf32> } // Loops '%i0' and '%i2' can fuse along producer/consumer edge on memref // '%a', and preserve the WAR dep from '%i0' to '%i1' on memref '%b', and // the SSA value dep from '%cf11' def to use in '%i2'. // CHECK: arith.constant 1.100000e+01 : f32 // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.load %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.load %{{.*}}[0] : memref<1xf32> // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: affine.for %{{.*}} = 0 to 10 { // CHECK-NEXT: affine.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32> // CHECK-NEXT: } // CHECK-NEXT: return return } // ----- // CHECK-LABEL: @producer_consumer_with_outmost_user func.func @producer_consumer_with_outmost_user(%arg0 : f16) { %c0 = arith.constant 0 : index %src = memref.alloc() : memref %dst = memref.alloc() : memref %tag = memref.alloc() : memref<1xi32> affine.for %arg1 = 4 to 6 { affine.for %arg2 = 0 to 1 { %0 = arith.addf %arg0, %arg0 : f16 affine.store %0, %src[] : memref } affine.for %arg3 = 0 to 1 { %0 = affine.load %src[] : memref } } affine.dma_start %src[], %dst[], %tag[%c0], %c0 : memref, memref, memref<1xi32> // CHECK: %[[CST_INDEX:.*]] = arith.constant 0 : index // CHECK: %[[DMA_SRC:.*]] = memref.alloc() : memref // CHECK: %[[DMA_DST:.*]] = memref.alloc() : memref // CHECK: %[[DMA_TAG:.*]] = memref.alloc() : memref<1xi32> // CHECK: affine.for %arg1 = 4 to 6 // CHECK-NEXT: affine.for %arg2 = 0 to 1 // CHECK-NEXT: %[[RESULT_ADD:.*]] = arith.addf %arg0, %arg0 : f16 // CHECK-NEXT: affine.store %[[RESULT_ADD]], %[[DMA_SRC]][] : memref // CHECK-NEXT: affine.load %[[DMA_SRC]][] : memref // CHECK: affine.dma_start %[[DMA_SRC]][], %[[DMA_DST]][], %[[DMA_TAG]][%[[CST_INDEX]]], %[[CST_INDEX]] : memref, memref, memref<1xi32> // CHECK-NEXT: return return } // Add further tests in mlir/test/Transforms/loop-fusion-4.mlir