Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

transformations: stencil bufferization polish #3001

Merged
merged 18 commits into from
Aug 17, 2024
Merged

transformations: stencil bufferization polish #3001

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
10154e2
Finer analysis in stencil bufferization and related side-effect updat…
PapyChacal Aug 7, 2024
f0406e8
Try out more progressive bufferization on apply.
PapyChacal Aug 7, 2024
65ae4b6
Fix up constructor.
PapyChacal Aug 8, 2024
dbf0fa1
Unused copypasta cleanup.
PapyChacal Aug 9, 2024
b07b52b
Comments cleanup.
PapyChacal Aug 9, 2024
14ab642
Update doc.
PapyChacal Aug 9, 2024
aeee21a
Rewrite AppltStoreFoldPattern in terms of apply, not store.
PapyChacal Aug 9, 2024
2a3df6f
Fold all stores if multiple.
PapyChacal Aug 9, 2024
4a99a6f
Merge branch 'main' into emilien/better-bufferization
PapyChacal Aug 9, 2024
31c4b4b
Comments.
PapyChacal Aug 9, 2024
f7ea30b
Update docstrings.
PapyChacal Aug 12, 2024
737448f
Merge branch 'main' into emilien/better-bufferization
PapyChacal Aug 12, 2024
fe246fe
Merge branch 'main' into emilien/better-bufferization
PapyChacal Aug 15, 2024
d27e544
Update xdsl/dialects/stencil.py
PapyChacal Aug 16, 2024
556920c
Merge branch 'main' into emilien/better-bufferization
PapyChacal Aug 16, 2024
01d874b
Empty tuple
PapyChacal Aug 16, 2024
8b0b94f
Merge branch 'main' into emilien/better-bufferization
PapyChacal Aug 16, 2024
a914a27
Merge branch 'main' into emilien/better-bufferization
PapyChacal Aug 17, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
115 changes: 115 additions & 0 deletions tests/filecheck/dialects/stencil/oec-kernels/fvtp2d_qi.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
// RUN: XDSL_ROUNDTRIP
// RUN: xdsl-opt %s -p stencil-storage-materialization,stencil-shape-inference | filecheck %s --check-prefix SHAPE
// RUN: xdsl-opt %s -p stencil-storage-materialization,stencil-shape-inference,convert-stencil-to-ll-mlir | filecheck %s --check-prefix MLIR
// RUN: xdsl-opt %s -p stencil-storage-materialization,stencil-shape-inference,stencil-bufferize | filecheck %s --check-prefix BUFF

func.func @fvtp2d_qi(%arg0: !stencil.field<?x?x?xf64>, %arg1: !stencil.field<?x?x?xf64>, %arg2: !stencil.field<?x?x?xf64>, %arg3: !stencil.field<?x?x?xf64>, %arg4: !stencil.field<?x?x?xf64>, %arg5: !stencil.field<?x?x?xf64>, %arg6: !stencil.field<?x?x?xf64>) attributes {stencil.program} {
%0 = stencil.cast %arg0 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
Expand Down Expand Up @@ -547,3 +548,117 @@ func.func @fvtp2d_qi(%arg0: !stencil.field<?x?x?xf64>, %arg1: !stencil.field<?x?
// MLIR-NEXT: memref.dealloc %arg8 : memref<64x67x64xf64, strided<[4288, 64, 1], offset: 64>>
// MLIR-NEXT: func.return
// MLIR-NEXT: }

// BUFF: func.func @fvtp2d_qi(%arg0 : !stencil.field<?x?x?xf64>, %arg1 : !stencil.field<?x?x?xf64>, %arg2 : !stencil.field<?x?x?xf64>, %arg3 : !stencil.field<?x?x?xf64>, %arg4 : !stencil.field<?x?x?xf64>, %arg5 : !stencil.field<?x?x?xf64>, %arg6 : !stencil.field<?x?x?xf64>) attributes {"stencil.program"}{
// BUFF-NEXT: %0 = stencil.alloc : !stencil.field<[0,64]x[0,65]x[0,64]xf64>
// BUFF-NEXT: %1 = stencil.alloc : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %2 = stencil.alloc : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %3 = stencil.alloc : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %4 = stencil.alloc : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %5 = stencil.alloc : !stencil.field<[0,64]x[-1,66]x[0,64]xf64>
// BUFF-NEXT: %6 = stencil.cast %arg0 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %7 = stencil.cast %arg1 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %8 = stencil.cast %arg2 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %9 = stencil.cast %arg3 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %10 = stencil.cast %arg4 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %11 = stencil.cast %arg5 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %12 = stencil.cast %arg6 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: stencil.apply(%arg7 = %6 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) outs (%5 : !stencil.field<[0,64]x[-1,66]x[0,64]xf64>) {
// BUFF-NEXT: %cst = arith.constant 1.000000e+00 : f64
// BUFF-NEXT: %cst_1 = arith.constant 7.000000e+00 : f64
// BUFF-NEXT: %cst_2 = arith.constant 1.200000e+01 : f64
// BUFF-NEXT: %13 = arith.divf %cst_1, %cst_2 : f64
// BUFF-NEXT: %14 = arith.divf %cst, %cst_2 : f64
// BUFF-NEXT: %15 = stencil.access %arg7[0, -1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %16 = stencil.access %arg7[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %17 = arith.addf %15, %16 : f64
// BUFF-NEXT: %18 = stencil.access %arg7[0, -2, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %19 = stencil.access %arg7[0, 1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %20 = arith.addf %18, %19 : f64
// BUFF-NEXT: %21 = arith.mulf %13, %17 : f64
// BUFF-NEXT: %22 = arith.mulf %14, %20 : f64
// BUFF-NEXT: %23 = arith.addf %21, %22 : f64
// BUFF-NEXT: %24 = stencil.store_result %23 : !stencil.result<f64>
// BUFF-NEXT: stencil.return %24 : !stencil.result<f64>
// BUFF-NEXT: } to <[0, -1, 0], [64, 66, 64]>
// BUFF-NEXT: stencil.apply(%arg7 = %6 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %arg8 = %5 : !stencil.field<[0,64]x[-1,66]x[0,64]xf64>) outs (%4 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>, %3 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>, %2 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>, %1 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>) {
// BUFF-NEXT: %cst = arith.constant 0.000000e+00 : f64
// BUFF-NEXT: %cst_1 = arith.constant 1.000000e+00 : f64
// BUFF-NEXT: %13 = stencil.access %arg8[0, 0, 0] : !stencil.field<[0,64]x[-1,66]x[0,64]xf64>
// BUFF-NEXT: %14 = stencil.access %arg7[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %15 = arith.subf %13, %14 : f64
// BUFF-NEXT: %16 = stencil.access %arg8[0, 1, 0] : !stencil.field<[0,64]x[-1,66]x[0,64]xf64>
// BUFF-NEXT: %17 = arith.subf %16, %14 : f64
// BUFF-NEXT: %18 = arith.addf %15, %17 : f64
// BUFF-NEXT: %19 = arith.mulf %15, %17 : f64
// BUFF-NEXT: %20 = arith.cmpf olt, %19, %cst : f64
// BUFF-NEXT: %21 = arith.select %20, %cst_1, %cst : f64
// BUFF-NEXT: %22 = stencil.store_result %15 : !stencil.result<f64>
// BUFF-NEXT: %23 = stencil.store_result %17 : !stencil.result<f64>
// BUFF-NEXT: %24 = stencil.store_result %18 : !stencil.result<f64>
// BUFF-NEXT: %25 = stencil.store_result %21 : !stencil.result<f64>
// BUFF-NEXT: stencil.return %22, %23, %24, %25 : !stencil.result<f64>, !stencil.result<f64>, !stencil.result<f64>, !stencil.result<f64>
// BUFF-NEXT: } to <[0, -1, 0], [64, 65, 64]>
// BUFF-NEXT: stencil.apply(%arg7 = %6 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %arg8 = %7 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %arg9 = %4 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>, %arg10 = %3 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>, %arg11 = %2 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>, %arg12 = %1 : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>) outs (%12 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) {
// BUFF-NEXT: %cst = arith.constant 0.000000e+00 : f64
// BUFF-NEXT: %cst_1 = arith.constant 1.000000e+00 : f64
// BUFF-NEXT: %13 = stencil.access %arg12[0, -1, 0] : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %14 = arith.cmpf oeq, %13, %cst : f64
// BUFF-NEXT: %15 = arith.select %14, %cst_1, %cst : f64
// BUFF-NEXT: %16 = stencil.access %arg12[0, 0, 0] : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %17 = arith.mulf %16, %15 : f64
// BUFF-NEXT: %18 = arith.addf %13, %17 : f64
// BUFF-NEXT: %19 = stencil.access %arg8[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %20 = arith.cmpf ogt, %19, %cst : f64
// BUFF-NEXT: %21 = "scf.if"(%20) ({
// BUFF-NEXT: %22 = stencil.access %arg10[0, -1, 0] : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %23 = stencil.access %arg11[0, -1, 0] : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %24 = arith.mulf %19, %23 : f64
// BUFF-NEXT: %25 = arith.subf %22, %24 : f64
// BUFF-NEXT: %26 = arith.subf %cst_1, %19 : f64
// BUFF-NEXT: %27 = arith.mulf %26, %25 : f64
// BUFF-NEXT: scf.yield %27 : f64
// BUFF-NEXT: }, {
// BUFF-NEXT: %28 = stencil.access %arg9[0, 0, 0] : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %29 = stencil.access %arg11[0, 0, 0] : !stencil.field<[0,64]x[-1,65]x[0,64]xf64>
// BUFF-NEXT: %30 = arith.mulf %19, %29 : f64
// BUFF-NEXT: %31 = arith.addf %28, %30 : f64
// BUFF-NEXT: %32 = arith.addf %cst_1, %19 : f64
// BUFF-NEXT: %33 = arith.mulf %32, %31 : f64
// BUFF-NEXT: scf.yield %33 : f64
// BUFF-NEXT: }) : (i1) -> f64
// BUFF-NEXT: %34 = arith.mulf %21, %18 : f64
// BUFF-NEXT: %35 = "scf.if"(%20) ({
// BUFF-NEXT: %36 = stencil.access %arg7[0, -1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %37 = arith.addf %36, %34 : f64
// BUFF-NEXT: scf.yield %37 : f64
// BUFF-NEXT: }, {
// BUFF-NEXT: %38 = stencil.access %arg7[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %39 = arith.addf %38, %34 : f64
// BUFF-NEXT: scf.yield %39 : f64
// BUFF-NEXT: }) : (i1) -> f64
// BUFF-NEXT: %40 = stencil.store_result %35 : !stencil.result<f64>
// BUFF-NEXT: stencil.return %40 : !stencil.result<f64>
// BUFF-NEXT: } to <[0, 0, 0], [64, 65, 64]>
// BUFF-NEXT: stencil.apply(%arg7 = %9 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %arg8 = %12 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) outs (%0 : !stencil.field<[0,64]x[0,65]x[0,64]xf64>) {
// BUFF-NEXT: %13 = stencil.access %arg7[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %14 = stencil.access %arg8[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %15 = arith.mulf %13, %14 : f64
// BUFF-NEXT: %16 = stencil.store_result %15 : !stencil.result<f64>
// BUFF-NEXT: stencil.return %16 : !stencil.result<f64>
// BUFF-NEXT: } to <[0, 0, 0], [64, 65, 64]>
// BUFF-NEXT: stencil.apply(%arg7 = %6 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %arg8 = %10 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %arg9 = %0 : !stencil.field<[0,64]x[0,65]x[0,64]xf64>, %arg10 = %8 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) outs (%11 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) {
// BUFF-NEXT: %13 = stencil.access %arg7[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %14 = stencil.access %arg8[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %15 = arith.mulf %13, %14 : f64
// BUFF-NEXT: %16 = stencil.access %arg9[0, 0, 0] : !stencil.field<[0,64]x[0,65]x[0,64]xf64>
// BUFF-NEXT: %17 = stencil.access %arg9[0, 1, 0] : !stencil.field<[0,64]x[0,65]x[0,64]xf64>
// BUFF-NEXT: %18 = arith.subf %16, %17 : f64
// BUFF-NEXT: %19 = arith.addf %15, %18 : f64
// BUFF-NEXT: %20 = stencil.access %arg10[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// BUFF-NEXT: %21 = arith.divf %19, %20 : f64
// BUFF-NEXT: %22 = stencil.store_result %21 : !stencil.result<f64>
// BUFF-NEXT: stencil.return %22 : !stencil.result<f64>
// BUFF-NEXT: } to <[0, 0, 0], [64, 64, 64]>
// BUFF-NEXT: func.return
// BUFF-NEXT: }
75 changes: 26 additions & 49 deletions tests/filecheck/transforms/stencil-bufferize.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,43 +73,43 @@ func.func @copy_1d(%0 : !stencil.field<?xf64>, %out : !stencil.field<?xf64>) {
// CHECK-NEXT: func.return
// CHECK-NEXT: }

func.func @copy_2d(%0 : !stencil.field<?x?xf64>) {
func.func @copy_2d(%0 : !stencil.field<?x?xf64>, %out : !stencil.field<[-4,68]x[-4,68]xf64>) {
%1 = stencil.cast %0 : !stencil.field<?x?xf64> -> !stencil.field<[-4,68]x[-4,68]xf64>
%2 = stencil.load %1 : !stencil.field<[-4,68]x[-4,68]xf64> -> !stencil.temp<[-1,64]x[0,68]xf64>
%3 = stencil.apply(%4 = %2 : !stencil.temp<[-1,64]x[0,68]xf64>) -> (!stencil.temp<[0,64]x[0,68]xf64>) {
%5 = stencil.access %4[-1, 0] : !stencil.temp<[-1,64]x[0,68]xf64>
stencil.return %5 : f64
}
stencil.store %3 to %out (<[0, 0], [64, 68]>) : !stencil.temp<[0,64]x[0,68]xf64> to !stencil.field<[-4,68]x[-4,68]xf64>
func.return
}

// CHECK: func.func @copy_2d(%0 : !stencil.field<?x?xf64>) {
// CHECK: func.func @copy_2d(%0 : !stencil.field<?x?xf64>, %out : !stencil.field<[-4,68]x[-4,68]xf64>) {
// CHECK-NEXT: %1 = stencil.cast %0 : !stencil.field<?x?xf64> -> !stencil.field<[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %2 = stencil.alloc : !stencil.field<[0,64]x[0,68]xf64>
// CHECK-NEXT: stencil.apply(%3 = %1 : !stencil.field<[-4,68]x[-4,68]xf64>) outs (%2 : !stencil.field<[0,64]x[0,68]xf64>) {
// CHECK-NEXT: %4 = stencil.access %3[-1, 0] : !stencil.field<[-4,68]x[-4,68]xf64>
// CHECK-NEXT: stencil.return %4 : f64
// CHECK-NEXT: stencil.apply(%2 = %1 : !stencil.field<[-4,68]x[-4,68]xf64>) outs (%out : !stencil.field<[-4,68]x[-4,68]xf64>) {
// CHECK-NEXT: %3 = stencil.access %2[-1, 0] : !stencil.field<[-4,68]x[-4,68]xf64>
// CHECK-NEXT: stencil.return %3 : f64
// CHECK-NEXT: } to <[0, 0], [64, 68]>
// CHECK-NEXT: func.return
// CHECK-NEXT: }


func.func @copy_3d(%0 : !stencil.field<?x?x?xf64>) {
func.func @copy_3d(%0 : !stencil.field<?x?x?xf64>, %out : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) {
%1 = stencil.cast %0 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64>
%2 = stencil.load %1 : !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64> -> !stencil.temp<[-1,64]x[0,64]x[0,69]xf64>
%3 = stencil.apply(%4 = %2 : !stencil.temp<[-1,64]x[0,64]x[0,69]xf64>) -> (!stencil.temp<[0,64]x[0,64]x[0,68]xf64>) {
%5 = stencil.access %4[-1, 0, 1] : !stencil.temp<[-1,64]x[0,64]x[0,69]xf64>
stencil.return %5 : f64
}
stencil.store %3 to %out (<[0, 0, 0], [64, 64, 68]>) : !stencil.temp<[0,64]x[0,64]x[0,68]xf64> to !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
func.return
}

// CHECK: func.func @copy_3d(%0 : !stencil.field<?x?x?xf64>) {
// CHECK: func.func @copy_3d(%0 : !stencil.field<?x?x?xf64>, %out : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) {
// CHECK-NEXT: %1 = stencil.cast %0 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64>
// CHECK-NEXT: %2 = stencil.alloc : !stencil.field<[0,64]x[0,64]x[0,68]xf64>
// CHECK-NEXT: stencil.apply(%3 = %1 : !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64>) outs (%2 : !stencil.field<[0,64]x[0,64]x[0,68]xf64>) {
// CHECK-NEXT: %4 = stencil.access %3[-1, 0, 1] : !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64>
// CHECK-NEXT: stencil.return %4 : f64
// CHECK-NEXT: stencil.apply(%2 = %1 : !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64>) outs (%out : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) {
// CHECK-NEXT: %3 = stencil.access %2[-1, 0, 1] : !stencil.field<[-4,68]x[-4,70]x[-4,72]xf64>
// CHECK-NEXT: stencil.return %3 : f64
// CHECK-NEXT: } to <[0, 0, 0], [64, 64, 68]>
// CHECK-NEXT: func.return
// CHECK-NEXT: }
Expand Down Expand Up @@ -140,20 +140,19 @@ func.func @offsets(%0 : !stencil.field<?x?x?xf64>, %1 : !stencil.field<?x?x?xf64
// CHECK: func.func @offsets(%0 : !stencil.field<?x?x?xf64>, %1 : !stencil.field<?x?x?xf64>, %2 : !stencil.field<?x?x?xf64>) {
// CHECK-NEXT: %3 = stencil.cast %0 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %4 = stencil.cast %1 : !stencil.field<?x?x?xf64> -> !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %5 = stencil.alloc : !stencil.field<[0,64]x[0,64]x[0,64]xf64>
// CHECK-NEXT: stencil.apply(%6 = %3 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) outs (%4 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>, %5 : !stencil.field<[0,64]x[0,64]x[0,64]xf64>) {
// CHECK-NEXT: %7 = stencil.access %6[-1, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %8 = stencil.access %6[1, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %9 = stencil.access %6[0, 1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %10 = stencil.access %6[0, -1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %11 = stencil.access %6[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %12 = arith.addf %7, %8 : f64
// CHECK-NEXT: %13 = arith.addf %9, %10 : f64
// CHECK-NEXT: %14 = arith.addf %12, %13 : f64
// CHECK-NEXT: stencil.apply(%5 = %3 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) outs (%4 : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>) {
// CHECK-NEXT: %6 = stencil.access %5[-1, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %7 = stencil.access %5[1, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %8 = stencil.access %5[0, 1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %9 = stencil.access %5[0, -1, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %10 = stencil.access %5[0, 0, 0] : !stencil.field<[-4,68]x[-4,68]x[-4,68]xf64>
// CHECK-NEXT: %11 = arith.addf %6, %7 : f64
// CHECK-NEXT: %12 = arith.addf %8, %9 : f64
// CHECK-NEXT: %13 = arith.addf %11, %12 : f64
// CHECK-NEXT: %cst = arith.constant -4.000000e+00 : f64
// CHECK-NEXT: %15 = arith.mulf %11, %cst : f64
// CHECK-NEXT: %16 = arith.addf %15, %14 : f64
// CHECK-NEXT: stencil.return %16, %15 : f64, f64
// CHECK-NEXT: %14 = arith.mulf %10, %cst : f64
// CHECK-NEXT: %15 = arith.addf %14, %13 : f64
// CHECK-NEXT: stencil.return %15 : f64
// CHECK-NEXT: } to <[0, 0, 0], [64, 64, 64]>
// CHECK-NEXT: func.return
// CHECK-NEXT: }
Expand Down Expand Up @@ -342,31 +341,9 @@ func.func @stencil_init_index_offset(%0 : !stencil.field<[0,64]x[0,64]x[0,64]xin
// CHECK-NEXT: func.return
// CHECK-NEXT: }

func.func @store_result_lowering(%arg0 : f64) {
%0, %1 = stencil.apply(%arg1 = %arg0 : f64) -> (!stencil.temp<[0,7]x[0,7]x[0,7]xf64>, !stencil.temp<[0,7]x[0,7]x[0,7]xf64>) {
%2 = stencil.store_result %arg1 : !stencil.result<f64>
%3 = stencil.store_result %arg1 : !stencil.result<f64>
stencil.return %2, %3 : !stencil.result<f64>, !stencil.result<f64>
}
%2 = stencil.buffer %1 : !stencil.temp<[0,7]x[0,7]x[0,7]xf64> -> !stencil.temp<[0,7]x[0,7]x[0,7]xf64>
%3 = stencil.buffer %0 : !stencil.temp<[0,7]x[0,7]x[0,7]xf64> -> !stencil.temp<[0,7]x[0,7]x[0,7]xf64>
func.return
}

// CHECK: func.func @store_result_lowering(%arg0 : f64) {
// CHECK-NEXT: %0 = stencil.alloc : !stencil.field<[0,7]x[0,7]x[0,7]xf64>
// CHECK-NEXT: %1 = stencil.alloc : !stencil.field<[0,7]x[0,7]x[0,7]xf64>
// CHECK-NEXT: stencil.apply(%arg1 = %arg0 : f64) outs (%0 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>, %1 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>) {
// CHECK-NEXT: %2 = stencil.store_result %arg1 : !stencil.result<f64>
// CHECK-NEXT: %3 = stencil.store_result %arg1 : !stencil.result<f64>
// CHECK-NEXT: stencil.return %2, %3 : !stencil.result<f64>, !stencil.result<f64>
// CHECK-NEXT: } to <[0, 0, 0], [7, 7, 7]>
// CHECK-NEXT: func.return
// CHECK-NEXT: }

func.func @if_lowering(%arg0 : f64, %b0 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>, %b1 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>) attributes {"stencil.program"}{
%0, %1 = stencil.apply(%arg1 = %arg0 : f64) -> (!stencil.temp<[0,7]x[0,7]x[0,7]xf64>, !stencil.temp<[0,7]x[0,7]x[0,7]xf64>) {
%true = "test.op"() : () -> i1
%true = "test.pureop"() : () -> i1
%2, %3 = "scf.if"(%true) ({
%4 = stencil.store_result %arg1 : !stencil.result<f64>
scf.yield %4, %arg1 : !stencil.result<f64>, f64
Expand All @@ -384,7 +361,7 @@ func.func @if_lowering(%arg0 : f64, %b0 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>,

// CHECK: func.func @if_lowering(%arg0 : f64, %b0 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>, %b1 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>) attributes {"stencil.program"}{
// CHECK-NEXT: stencil.apply(%arg1 = %arg0 : f64) outs (%b0 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>, %b1 : !stencil.field<[0,7]x[0,7]x[0,7]xf64>) {
// CHECK-NEXT: %true = "test.op"() : () -> i1
// CHECK-NEXT: %true = "test.pureop"() : () -> i1
// CHECK-NEXT: %0, %1 = "scf.if"(%true) ({
// CHECK-NEXT: %2 = stencil.store_result %arg1 : !stencil.result<f64>
// CHECK-NEXT: scf.yield %2, %arg1 : !stencil.result<f64>, f64
Expand Down
Loading
Loading