Update quantized matmul tests to DQ/Q format supported by fx_importer

[ubfx]

Continuation of #3809 for the matmul tests.

[ubfx]

linalg IR of one of the tests, showing the expected integer matmul op:

#map = affine_map<(d0, d1) -> (d0, d1)>
module {
  ml_program.global private mutable @global_seed(dense<0> : tensor<i64>) : tensor<i64>
  func.func @AtenMmQuint8(%arg0: tensor<3x4xi8>, %arg1: tensor<4x3xi8>) -> tensor<3x3xf32> {
    %c0_i32 = arith.constant 0 : i32
    %c-128_i8 = arith.constant -128 : i8
    %c0_i8 = arith.constant 0 : i8
    %c-1_i8 = arith.constant -1 : i8
    %c-63_i32 = arith.constant -63 : i32
    %c32_i32 = arith.constant 32 : i32
    %cst = arith.constant 4.278500e-03 : f64
    %0 = tensor.empty() : tensor<3x4xi8>
    %1 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel"]} ins(%arg0 : tensor<3x4xi8>) outs(%0 : tensor<3x4xi8>) {
    ^bb0(%in: i8, %out: i8):
      %11 = arith.cmpi ult, %in, %c0_i8 : i8
      %12 = arith.select %11, %c0_i8, %in : i8
      %13 = arith.cmpi uge, %12, %c-1_i8 : i8
      %14 = arith.select %13, %c-1_i8, %12 : i8
      linalg.yield %14 : i8
    } -> tensor<3x4xi8>
    %2 = tensor.empty() : tensor<4x3xi8>
    %3 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel"]} ins(%arg1 : tensor<4x3xi8>) outs(%2 : tensor<4x3xi8>) {
    ^bb0(%in: i8, %out: i8):
      %11 = arith.cmpi ult, %in, %c0_i8 : i8
      %12 = arith.select %11, %c0_i8, %in : i8
      %13 = arith.cmpi uge, %12, %c-1_i8 : i8
      %14 = arith.select %13, %c-1_i8, %12 : i8
      linalg.yield %14 : i8
    } -> tensor<4x3xi8>
    %4 = tensor.empty() : tensor<3x3xi32>
    %5 = linalg.fill ins(%c0_i32 : i32) outs(%4 : tensor<3x3xi32>) -> tensor<3x3xi32>
    %6 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel"]} ins(%1 : tensor<3x4xi8>) outs(%0 : tensor<3x4xi8>) {
    ^bb0(%in: i8, %out: i8):
      %11 = arith.addi %in, %c-128_i8 : i8
      linalg.yield %11 : i8
    } -> tensor<3x4xi8>
    %7 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel"]} ins(%3 : tensor<4x3xi8>) outs(%2 : tensor<4x3xi8>) {
    ^bb0(%in: i8, %out: i8):
      %11 = arith.addi %in, %c-128_i8 : i8
      linalg.yield %11 : i8
    } -> tensor<4x3xi8>
    %8 = linalg.quantized_matmul ins(%6, %7, %c-63_i32, %c32_i32 : tensor<3x4xi8>, tensor<4x3xi8>, i32, i32) outs(%5 : tensor<3x3xi32>) -> tensor<3x3xi32>
    %9 = tensor.empty() : tensor<3x3xf32>
    %10 = linalg.generic {indexing_maps = [#map, #map], iterator_types = ["parallel", "parallel"]} ins(%8 : tensor<3x3xi32>) outs(%9 : tensor<3x3xf32>) {
    ^bb0(%in: i32, %out: f32):
      %11 = arith.sitofp %in : i32 to f32
      %12 = arith.truncf %cst : f64 to f32
      %13 = arith.mulf %11, %12 : f32
      linalg.yield %13 : f32
    } -> tensor<3x3xf32>
    return %10 : tensor<3x3xf32>
  }
}