Open
Description
My git version is 90beda2.
Description:
I am experiencing an inconsistent result when executing the same MLIR program with and without the -test-math-polynomial-approximation="enable-avx2".
Steps to Reproduce:
1. MLIR Program (test.mlir):
test.mlir:
module attributes {llvm.data_layout = ""} {
func.func private @printMemrefI64(tensor<*xi64>)
func.func private @printMemrefF32(tensor<*xf32>)
func.func @main() {
%0 = "tosa.const"() <{values = dense<0.000000e+00> : tensor<1xf32>}> : () -> tensor<1xf32>
%1 = "tosa.const"() <{values = dense<-1.023950e+03> : tensor<1x1x6x3xf32>}> : () -> tensor<1x1x6x3xf32>
%2 = "tosa.const"() <{values = dense<3.671420e+03> : tensor<4x4x3x1xf32>}> : () -> tensor<4x4x3x1xf32>
%3 = "tosa.const"() <{values = dense<4.235330e+03> : tensor<3xf32>}> : () -> tensor<3xf32>
%4 = tosa.depthwise_conv2d %1, %2, %3, %0, %0 {acc_type = f32, dilation = array<i64: 1, 2>, pad = array<i64: 1, 2, 2, 2>, stride = array<i64: 2, 1>} : (tensor<1x1x6x3xf32>, tensor<4x4x3x1xf32>, tensor<3xf32>, tensor<1xf32>, tensor<1xf32>) -> tensor<1x1x4x3xf32>
%5 = tosa.cos %4 : (tensor<1x1x4x3xf32>) -> tensor<1x1x4x3xf32>
%cast = tensor.cast %5 : tensor<1x1x4x3xf32> to tensor<*xf32>
call @printMemrefF32(%cast) : (tensor<*xf32>) -> ()
return
}
}
2. Command to Run Without -test-math-polynomial-approximation="enable-avx2":
/path/llvm-project/build/bin/mlir-opt test.mlir -pass-pipeline='builtin.module(func.func(tosa-to-linalg-named,tosa-to-linalg))' | \
/path/llvm-project/build/bin/mlir-opt -tosa-to-arith -one-shot-bufferize="bufferize-function-boundaries" -convert-linalg-to-affine-loops -test-lower-to-llvm | \
/path/llvm-project/build/bin/mlir-runner -e main -entry-point-result=void \
-shared-libs=/path/llvm-project/build/lib/libmlir_runner_utils.so \
-shared-libs=/path/llvm-project/build/lib/libmlir_c_runner_utils.so \
-shared-libs=/path/llvm-project/build/lib/libmlir_async_runtime.so
3. Output Without -test-math-polynomial-approximation="enable-avx2":
[[[[-0.290655, -0.290655, -0.290655],
[-0.290655, -0.290655, -0.290655],
[-0.290655, -0.290655, -0.290655],
[-0.290655, -0.290655, -0.290655]]]]
4. Command to Run With -test-math-polynomial-approximation="enable-avx2":
/path/llvm-project/build/bin/mlir-opt test.mlir -pass-pipeline='builtin.module(func.func(tosa-to-linalg-named,tosa-to-linalg))' | \
/path/llvm-project/build/bin/mlir-opt -tosa-to-arith -one-shot-bufferize="bufferize-function-boundaries" -convert-linalg-to-affine-loops -test-math-polynomial-approximation="enable-avx2" -test-lower-to-llvm | \
/path/llvm-project/build/bin/mlir-runner -e main -entry-point-result=void \
-shared-libs=/path/llvm-project/build/lib/libmlir_runner_utils.so \
-shared-libs=/path/llvm-project/build/lib/libmlir_c_runner_utils.so \
-shared-libs=/path/llvm-project/build/lib/libmlir_async_runtime.so
5. Output With -test-math-polynomial-approximation="enable-avx2":
[[[[-0.841471, -0.841471, -0.841471],
[-0.841471, -0.841471, -0.841471],
[-0.841471, -0.841471, -0.841471],
[-0.841471, -0.841471, -0.841471]]]]
I'm not sure if there is any bug in my program or if the wrong usage of the above passes caused this result.