[mlir] Inconsistent output when executing MLIR program with and without --test-math-polynomial-approximation

[anonymoususer-1]

git version: adf892d

system: Ubuntu 18.04.6 LTS

Description:

I am experiencing an inconsistent result when executing the same MLIR program with and without the --test-math-polynomial-approximation. I am not sure whether the semantic of --test-math-polynomial-approximation should affect the results.

Steps to Reproduce:

1. MLIR Program (a.mlir):

a.mlir:

module {
  func.func private @printMemrefI32(tensor<*xi32>)
  func.func private @printMemrefF32(tensor<*xf32>)
  func.func @main() {
    %0 = "tosa.const"() <{value = dense<[0, 2, 1]> : tensor<3xi32>}> : () -> tensor<3xi32>
    %1 = "tosa.const"() <{value = dense<4068> : tensor<1x4x3xi32>}> : () -> tensor<1x4x3xi32>
    %2 = "tosa.const"() <{value = dense<-1976> : tensor<1x4x3xi32>}> : () -> tensor<1x4x3xi32>
    %3 = tosa.maximum %2, %1 : (tensor<1x4x3xi32>, tensor<1x4x3xi32>) -> tensor<1x4x3xi32>
    %4 = tosa.transpose %3, %0 : (tensor<1x4x3xi32>, tensor<3xi32>) -> tensor<1x3x4xi32>
    %5 = tosa.matmul %2, %4 : (tensor<1x4x3xi32>, tensor<1x3x4xi32>) -> tensor<1x4x4xi32>
    %6 = tosa.cast %5 : (tensor<1x4x4xi32>) -> tensor<1x4x4xf32>
    %7 = tosa.floor %6 : (tensor<1x4x4xf32>) -> tensor<1x4x4xf32>
    %8 = tosa.sin %7 : (tensor<1x4x4xf32>) -> tensor<1x4x4xf32>
    %cast = tensor.cast %8 : tensor<1x4x4xf32> to tensor<*xf32>
    call @printMemrefF32(%cast) : (tensor<*xf32>) -> ()
    return
  }
}

2. Command to Run Without --test-math-polynomial-approximation:

 /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt a.mlir -pass-pipeline="builtin.module(func.func(tosa-to-linalg-named,tosa-to-linalg))" |  /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt -tosa-to-arith -one-shot-bufferize="bufferize-function-boundaries" \
-convert-linalg-to-affine-loops  -finalize-memref-to-llvm   -lower-affine  -finalize-memref-to-llvm   -convert-arith-to-llvm \
-convert-math-to-llvm  -convert-scf-to-cf  -convert-arith-to-llvm  -convert-func-to-llvm  -convert-func-to-llvm  \
-reconcile-unrealized-casts | /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 --test-math-polynomial-approximation:

[[[-0.0299986,    -0.0299986,    -0.0299986,    -0.0299986],
  [-0.0299986,    -0.0299986,    -0.0299986,    -0.0299986],
  [-0.0299986,    -0.0299986,    -0.0299986,    -0.0299986],
  [-0.0299986,    -0.0299986,    -0.0299986,    -0.0299986]]]

4. Command to Run With --test-math-polynomial-approximation:

 /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt a.mlir -pass-pipeline="builtin.module(func.func(tosa-to-linalg-named,tosa-to-linalg))"  |  /data/szy/MLIR/llvm-release/llvm-project/build/bin/mlir-opt -tosa-to-arith  \
--test-math-polynomial-approximation   -one-shot-bufferize="bufferize-function-boundaries"  \
-convert-linalg-to-affine-loops  -finalize-memref-to-llvm   -lower-affine  -finalize-memref-to-llvm   \
-convert-arith-to-llvm  -convert-math-to-llvm  -convert-scf-to-cf  -convert-arith-to-llvm  -convert-func-to-llvm  \
-convert-func-to-llvm  -reconcile-unrealized-casts | /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 --test-math-polynomial-approximation:

[[[1,    1,    1,    1],
  [1,    1,    1,    1],
  [1,    1,    1,    1],
  [1,    1,    1,    1]]]

[anonymoususer-1]

This test case actually computes the result of math.sin -2.41151e+07
The only difference between the two commands is that the second one includes --test-math-polynomial-approximation, whereas the first does not.
I found that the --test-math-polynomial-approximation pass attempts to transform math.sin into a series of operations.
Hi @asaadaldien , I notice that you have implemented this pass. I am sorry to disturb you, but could you please take a moment to review this problem? Thank you!