Open
Description
git version: f0c61d2
system: Ubuntu 18.04.6 LTS
Description:
I am experiencing an inconsistent result when executing the same MLIR program with and without --int-range-optimizations.
Steps to Reproduce:
1. MLIR Program (a.mlir):
a.mlir:
module {
func.func @main() {
%false = arith.constant false
%c8_i32 = arith.constant 8 : i32
%20 = arith.trunci %c8_i32 : i32 to i6
%35 = arith.constant 35 : i63
%36 = arith.constant 36 : i64
%sum_2, %overflow_3 = arith.addui_extended %false, %false : i1, i1
cf.cond_br %false, ^bb1, ^bb2
^bb1: // pred: ^bb0
cf.br ^bb3(%35, %36, %20 : i63, i64, i6)
^bb2: // pred: ^bb0
%c4_i64 = arith.constant 4 : i64
cf.br ^bb3(%35, %c4_i64, %20 : i63, i64, i6)
^bb3(%73: i63, %74: i64, %75: i6): // 2 preds: ^bb1, ^bb2
cf.br ^bb4
^bb4: // pred: ^bb3
cf.cond_br %sum_2, ^bb5, ^bb6
^bb5: // pred: ^bb4
%c-7668919674129928000_i64 = arith.constant -7668919674129928000 : i64
cf.br ^bb7(%c-7668919674129928000_i64 : i64)
^bb6: // pred: ^bb4
%c52_i64 = arith.constant 52 : i64
cf.br ^bb7(%c52_i64 : i64)
^bb7(%76: i64): // 2 preds: ^bb5, ^bb6
cf.br ^bb8
^bb8: // pred: ^bb7
vector.print str "%266="
vector.print %76 : i64
return
}
}
2. Command to Run without --int-range-optimizations :
/data/szy/workspace/llvm-release/llvm-project/build/bin/mlir-opt /data/szy/workspace/mlir-inconsistent/a.mlir \
-convert-arith-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm -convert-cf-to-llvm \
-convert-func-to-llvm -reconcile-unrealized-casts \
| timeout 10 /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-cpu-runner -e main \
-entry-point-result=void --shared-libs=/data/szy/MLIR/llvm-release/llvm-project/build/lib/libmlir_c_runner_utils.so \
--shared-libs=/data/szy/MLIR/llvm-release/llvm-project/build/lib/libmlir_runner_utils.so \
--shared-libs=/data/szy/MLIR/llvm-release/llvm-project/build/lib/libmlir_async_runtime.so
3. Output without --int-range-optimizations ::
%266=52
4. Command to Run with --int-range-optimizations :
/data/szy/workspace/llvm-release/llvm-project/build/bin/mlir-opt /data/szy/workspace/mlir-inconsistent/a.mlir \
--int-range-optimizations -convert-arith-to-llvm -convert-vector-to-llvm -convert-arith-to-llvm \
-convert-cf-to-llvm -convert-func-to-llvm -reconcile-unrealized-casts \
| timeout 10 /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-cpu-runner -e main \
-entry-point-result=void --shared-libs=/data/szy/MLIR/llvm-release/llvm-project/build/lib/libmlir_c_runner_utils.so \
--shared-libs=/data/szy/MLIR/llvm-release/llvm-project/build/lib/libmlir_runner_utils.so \
--shared-libs=/data/szy/MLIR/llvm-release/llvm-project/build/lib/libmlir_async_runtime.so
5. Output with --int-range-optimizations :
%266=4
6. Analysis for this case :
After running with --int-range-optimizations, a.mlir becomes as follows:
It directly returns %c4_i64 (4)
module {
func.func @main() {
%c4_i64 = arith.constant 4 : i64
%false = arith.constant false
cf.cond_br %false, ^bb1, ^bb1
^bb1: // 2 preds: ^bb0, ^bb0
cf.br ^bb2
^bb2: // pred: ^bb1
cf.br ^bb3
^bb3: // pred: ^bb2
cf.cond_br %false, ^bb4, ^bb4
^bb4: // 2 preds: ^bb3, ^bb3
cf.br ^bb5
^bb5: // pred: ^bb4
cf.br ^bb6
^bb6: // pred: ^bb5
vector.print str "%266="
vector.print %c4_i64 : i64
return
}
}