diffsl 0.9.4

A compiler for a domain-specific language for ordinary differential equations (ODE).
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151

// RUN: %eopt --split-input-file --enzyme --canonicalize --remove-unnecessary-enzyme-ops --enzyme-simplify-math --lower-llvm-ext --canonicalize %s | FileCheck %s

module {
  llvm.func @square(%x: f64) -> f64 {
    %next = arith.mulf %x, %x : f64
    llvm.return %next : f64
  }

  func.func @dsquare(%x: f64, %dr: f64) -> f64 {
    %dx = enzyme.autodiff @square(%x, %dr)
      {
        activity=[#enzyme<activity enzyme_active>],
        ret_activity=[#enzyme<activity enzyme_activenoneed>]
      } : (f64, f64) -> f64
    return %dx : f64
  }

// CHECK: llvm.func @diffesquare(%arg0: f64, %arg1: f64) -> f64 attributes {sym_visibility = "private"} {
// CHECK-NEXT:    %0 = arith.mulf %arg1, %arg0 : f64
// CHECK-NEXT:    %1 = arith.mulf %arg1, %arg0 : f64
// CHECK-NEXT:    %2 = arith.addf %0, %1 : f64
// CHECK-NEXT:    llvm.return %2 : f64
// CHECK-NEXT:  }
}

// -----

module {
llvm.func @multireturn(%x: f64, %y: f64) -> f64 {
  %0 = arith.mulf %x, %y : f64
  llvm.return %0 : f64
}

func.func @dmultireturn(%x: f64, %y: f64, %dr: f64) -> (f64, f64) {
  %res = enzyme.autodiff @multireturn(%x, %y, %dr)
    {
      activity=[#enzyme<activity enzyme_active>, #enzyme<activity enzyme_active>],
      ret_activity=[#enzyme<activity enzyme_activenoneed>]
    } : (f64, f64, f64) -> !llvm.struct<(f64, f64)>
  %fst = llvm.extractvalue %res[0] : !llvm.struct<(f64, f64)>
  %snd = llvm.extractvalue %res[1] : !llvm.struct<(f64, f64)>
  return %fst, %snd : f64, f64
}
}

// CHECK: llvm.func @diffemultireturn(%arg0: f64, %arg1: f64, %arg2: f64) -> !llvm.struct<(f64, f64)> attributes {sym_visibility = "private"} {
// CHECK-NEXT:    %0 = llvm.mlir.poison : !llvm.struct<(f64, f64)>
// CHECK-NEXT:    %1 = arith.mulf %arg2, %arg1 : f64
// CHECK-NEXT:    %2 = arith.mulf %arg2, %arg0 : f64
// CHECK-NEXT:    %3 = llvm.insertvalue %1, %0[0] : !llvm.struct<(f64, f64)>
// CHECK-NEXT:    %4 = llvm.insertvalue %2, %3[1] : !llvm.struct<(f64, f64)>
// CHECK-NEXT:    llvm.return %4 : !llvm.struct<(f64, f64)>
// CHECK-NEXT:  }

// -----

module {
llvm.func @loadstore(%a: !llvm.ptr, %b: f32) -> f32 {
  %sz = arith.constant 32 : i64
  llvm_ext.ptr_size_hint %a, %sz : !llvm.ptr, i64
  llvm.store %b, %a : f32, !llvm.ptr
  %0 = llvm.load %a : !llvm.ptr -> f32
  llvm.return %0 : f32
}

func.func @dloadstore(%a: !llvm.ptr, %da: !llvm.ptr, %b: f32, %dres: f32) -> f32 {
  %res = enzyme.autodiff @loadstore(%a, %da, %b, %dres)
    {
      activity=[#enzyme<activity enzyme_dup>, #enzyme<activity enzyme_active>],
      ret_activity=[#enzyme<activity enzyme_activenoneed>]
    } : (!llvm.ptr, !llvm.ptr, f32, f32) -> f32
  return %res : f32
}
}

// CHECK:  llvm.func @diffeloadstore(%[[a:.+]]: !llvm.ptr, %[[da:.+]]: !llvm.ptr, %[[b:.+]]: f32, %[[dres:.+]]: f32) -> f32 attributes {sym_visibility = "private"} {
// CHECK-NEXT:    %[[zero:.+]] = arith.constant 0.000000e+00 : f32
// CHECK-NEXT:    llvm.store %[[b]], %[[a]] : f32, !llvm.ptr
// CHECK-NEXT:    %[[daval1:.+]] = llvm.load %[[da]] : !llvm.ptr -> f32
// CHECK-NEXT:    %[[daval2:.+]] = arith.addf %[[daval1]], %[[dres]] : f32
// CHECK-NEXT:    llvm.store %[[daval2]], %[[da]] : f32, !llvm.ptr
// CHECK-NEXT:    %[[daval3:.+]] = llvm.load %[[da]] : !llvm.ptr -> f32
// CHECK-NEXT:    llvm.store %[[zero]], %[[da]] : f32, !llvm.ptr
// CHECK-NEXT:    llvm.return %[[daval3]] : f32
// CHECK-NEXT:  }

// -----

module {
llvm.func @f_iter(%a: !llvm.ptr) -> f32 {
  %lb = arith.constant 0 : index
  %ub = arith.constant 9 : index
  %step = arith.constant 1 : index

  %prod_0 = arith.constant 0.00 : f32

  %prod = scf.for %iv = %lb to %ub step %step
      iter_args(%prod_iter = %prod_0) -> f32 {
    %i = arith.index_cast %iv : index to i32
    %ptr = llvm.getelementptr %a[%i] : (!llvm.ptr, i32) -> !llvm.ptr, f32
    %val = llvm.load %ptr : !llvm.ptr -> f32
    %prod_next = arith.mulf %val, %prod_iter : f32
    scf.yield %prod_next : f32
  }

  llvm.return %prod : f32
}
func.func @f_iter_autodiff(%a: !llvm.ptr, %da: !llvm.ptr, %dres: f32) {
  enzyme.autodiff @f_iter(%a, %da, %dres)
    {
      activity=[#enzyme<activity enzyme_dup>],
      ret_activity=[#enzyme<activity enzyme_activenoneed>]
    } : (!llvm.ptr, !llvm.ptr, f32) -> ()
  return
}
}

// CHECK:  llvm.func @diffef_iter(%arg0: !llvm.ptr, %arg1: !llvm.ptr, %arg2: f32) attributes {sym_visibility = "private"} {
// CHECK-NEXT:    %c8 = arith.constant 8 : index
// CHECK-NEXT:    %c1 = arith.constant 1 : index
// CHECK-NEXT:    %c9 = arith.constant 9 : index
// CHECK-NEXT:    %c0 = arith.constant 0 : index
// CHECK-NEXT:    %cst = arith.constant 0.000000e+00 : f32
// CHECK-NEXT:    %alloc = memref.alloc() : memref<9xf32>
// CHECK-NEXT:    %alloc_0 = memref.alloc() : memref<9xf32>
// CHECK-NEXT:    %0 = scf.for %arg3 = %c0 to %c9 step %c1 iter_args(%arg4 = %cst) -> (f32) {
// CHECK-NEXT:      memref.store %arg4, %alloc_0[%arg3] : memref<9xf32>
// CHECK-NEXT:      %2 = arith.index_cast %arg3 : index to i32
// CHECK-NEXT:      %3 = llvm.getelementptr %arg0[%2] : (!llvm.ptr, i32) -> !llvm.ptr, f32
// CHECK-NEXT:      %4 = llvm.load %3 : !llvm.ptr -> f32
// CHECK-NEXT:      memref.store %4, %alloc[%arg3] : memref<9xf32>
// CHECK-NEXT:      %5 = arith.mulf %4, %arg4 : f32
// CHECK-NEXT:      scf.yield %5 : f32
// CHECK-NEXT:    }
// CHECK-NEXT:    %1 = scf.for %arg3 = %c0 to %c9 step %c1 iter_args(%arg4 = %arg2) -> (f32) {
// CHECK-NEXT:      %2 = arith.subi %c8, %arg3 : index
// CHECK-NEXT:      %3 = memref.load %alloc[%2] : memref<9xf32>
// CHECK-NEXT:      %4 = memref.load %alloc_0[%2] : memref<9xf32>
// CHECK-NEXT:      %5 = arith.index_cast %arg3 : index to i32
// CHECK-NEXT:      %6 = llvm.getelementptr %arg1[%5] : (!llvm.ptr, i32) -> !llvm.ptr, f32
// CHECK-NEXT:      %7 = arith.mulf %arg4, %4 : f32
// CHECK-NEXT:      %8 = arith.mulf %arg4, %3 : f32
// CHECK-NEXT:      %9 = llvm.load %6 : !llvm.ptr -> f32
// CHECK-NEXT:      %10 = arith.addf %9, %7 : f32
// CHECK-NEXT:      llvm.store %10, %6 : f32, !llvm.ptr
// CHECK-NEXT:      scf.yield %8 : f32
// CHECK-NEXT:    }
// CHECK-NEXT:    memref.dealloc %alloc_0 : memref<9xf32>
// CHECK-NEXT:    memref.dealloc %alloc : memref<9xf32>
// CHECK-NEXT:    llvm.return
// CHECK-NEXT:  }