[mlir] Inconsistent results for linalg.generic

[AnonymousBugreporter1]

I have the following MLIR program:
test.mlir:

#map = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
module {
  func.func @main() -> i32 {
    %c0_i32 = arith.constant 0 : i32
    %idx0 = index.constant 0
    %c1_i32 = arith.constant 1 : i32
    %alloc = memref.alloc() : memref<16x1x16xi32>
    linalg.fill ins(%c1_i32 : i32) outs(%alloc : memref<16x1x16xi32>)
    linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel"]} ins(%alloc : memref<16x1x16xi32>) outs(%alloc : memref<16x1x16xi32>) {
    ^bb0(%in: i32, %out: i32):
      linalg.yield %c0_i32 : i32
    }
    %0 = memref.load %alloc[%idx0, %idx0, %idx0] : memref<16x1x16xi32>
    vector.print %0 : i32
    return %0 : i32
  }
}

When I ran

/data/tmp/v0302/llvm-project/build/bin/mlir-opt \
--lower-affine --convert-index-to-llvm --convert-linalg-to-loops --convert-scf-to-cf --convert-arith-to-llvm --convert-vector-to-llvm --finalize-memref-to-llvm --convert-cf-to-llvm --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v0302/llvm-project/build/bin/mlir-runner --entry-point-result=i32 --shared-libs=/data/tmp/v0302/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v0302/llvm-project/build/lib/libmlir_c_runner_utils.so

on the program, I got the result of:

0
0

However, when I ran

/data/tmp/v0302/llvm-project/build/bin/mlir-opt \
--canonicalize \
--lower-affine --convert-index-to-llvm --convert-linalg-to-loops --convert-scf-to-cf --convert-arith-to-llvm --convert-vector-to-llvm --finalize-memref-to-llvm --convert-cf-to-llvm --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v0302/llvm-project/build/bin/mlir-runner --entry-point-result=i32 --shared-libs=/data/tmp/v0302/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v0302/llvm-project/build/lib/libmlir_c_runner_utils.so

on the program, I got the result of:

1
1

The above two results seem to be inconsistent. I'm not sure if there is any bug in my program or if the wrong usage of the above passes caused these results.

My git version is b6d5fa0.

[llvmbot]

@llvm/issue-subscribers-mlir

Author: None (wangyongj1a)

I have the following MLIR program: test.mlir: ``` #map = affine_map<(d0, d1, d2) -> (d0, d1, d2)> module { func.func @main() -> i32 { %c0_i32 = arith.constant 0 : i32 %idx0 = index.constant 0 %c1_i32 = arith.constant 1 : i32 %alloc = memref.alloc() : memref<16x1x16xi32> linalg.fill ins(%c1_i32 : i32) outs(%alloc : memref<16x1x16xi32>) linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel"]} ins(%alloc : memref<16x1x16xi32>) outs(%alloc : memref<16x1x16xi32>) { ^bb0(%in: i32, %out: i32): linalg.yield %c0_i32 : i32 } %0 = memref.load %alloc[%idx0, %idx0, %idx0] : memref<16x1x16xi32> vector.print %0 : i32 return %0 : i32 } } ``` When I ran ``` /data/tmp/v0302/llvm-project/build/bin/mlir-opt \ --lower-affine --convert-index-to-llvm --convert-linalg-to-loops --convert-scf-to-cf --convert-arith-to-llvm --convert-vector-to-llvm --finalize-memref-to-llvm --convert-cf-to-llvm --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v0302/llvm-project/build/bin/mlir-runner --entry-point-result=i32 --shared-libs=/data/tmp/v0302/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v0302/llvm-project/build/lib/libmlir_c_runner_utils.so ``` on the program, I got the result of: ``` 0 0 ``` However, when I ran ``` /data/tmp/v0302/llvm-project/build/bin/mlir-opt \ --canonicalize \ --lower-affine --convert-index-to-llvm --convert-linalg-to-loops --convert-scf-to-cf --convert-arith-to-llvm --convert-vector-to-llvm --finalize-memref-to-llvm --convert-cf-to-llvm --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v0302/llvm-project/build/bin/mlir-runner --entry-point-result=i32 --shared-libs=/data/tmp/v0302/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v0302/llvm-project/build/lib/libmlir_c_runner_utils.so ``` on the program, I got the result of: ``` 1 1 ``` The above two results seem to be inconsistent. I'm not sure if there is any bug in my program or if the wrong usage of the above passes caused these results.

My git version is b6d5fa0.

[llvmbot]

@llvm/issue-subscribers-mlir-linalg

Author: None (wangyongj1a)

I have the following MLIR program: test.mlir: ``` #map = affine_map<(d0, d1, d2) -> (d0, d1, d2)> module { func.func @main() -> i32 { %c0_i32 = arith.constant 0 : i32 %idx0 = index.constant 0 %c1_i32 = arith.constant 1 : i32 %alloc = memref.alloc() : memref<16x1x16xi32> linalg.fill ins(%c1_i32 : i32) outs(%alloc : memref<16x1x16xi32>) linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel", "parallel"]} ins(%alloc : memref<16x1x16xi32>) outs(%alloc : memref<16x1x16xi32>) { ^bb0(%in: i32, %out: i32): linalg.yield %c0_i32 : i32 } %0 = memref.load %alloc[%idx0, %idx0, %idx0] : memref<16x1x16xi32> vector.print %0 : i32 return %0 : i32 } } ``` When I ran ``` /data/tmp/v0302/llvm-project/build/bin/mlir-opt \ --lower-affine --convert-index-to-llvm --convert-linalg-to-loops --convert-scf-to-cf --convert-arith-to-llvm --convert-vector-to-llvm --finalize-memref-to-llvm --convert-cf-to-llvm --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v0302/llvm-project/build/bin/mlir-runner --entry-point-result=i32 --shared-libs=/data/tmp/v0302/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v0302/llvm-project/build/lib/libmlir_c_runner_utils.so ``` on the program, I got the result of: ``` 0 0 ``` However, when I ran ``` /data/tmp/v0302/llvm-project/build/bin/mlir-opt \ --canonicalize \ --lower-affine --convert-index-to-llvm --convert-linalg-to-loops --convert-scf-to-cf --convert-arith-to-llvm --convert-vector-to-llvm --finalize-memref-to-llvm --convert-cf-to-llvm --convert-func-to-llvm --convert-arith-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v0302/llvm-project/build/bin/mlir-runner --entry-point-result=i32 --shared-libs=/data/tmp/v0302/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v0302/llvm-project/build/lib/libmlir_c_runner_utils.so ``` on the program, I got the result of: ``` 1 1 ``` The above two results seem to be inconsistent. I'm not sure if there is any bug in my program or if the wrong usage of the above passes caused these results.

My git version is b6d5fa0.

[IanWood1]

Looks like there was a bug causing --canonicalize to erase the linalg.generic. Fix here: #130000