Compute the Eulerian number

Jelly, 8 bytes

Œ!Z>2\Sċ

Try it online! (takes a while) or verify the smaller test cases.

How it works

Œ!Z>2\Sċ  Main link. Arguments: n, m

Œ!        Generate the matrix of all permutations of [1, ..., n].
  Z       Zip/transpose, placing the permutations in the columns.
   >2\    Compare columns pairwise with vectorizing greater-than.
          This generates a 1 in the column for each rise in that permutation.
      S   Compute the vectorizing sum of the columns, counting the number of rises.
       ċ  Count how many times m appears in the computed counts.

JavaScript (ES6), 50 46 45 bytes

f=(n,m,d=n-m)=>m?d&&f(--n,m)*++m+f(n,m-2)*d:1

Based on the recursive formula:

A(n, m) = (n - m)A(n - 1, m - 1) + (m + 1)A(n - 1, m)    

Test cases

f=(n,m,d=n-m)=>m?d&&f(--n,m)*++m+f(n,m-2)*d:1

console.log(f( 0,  0));  // 1
console.log(f( 1,  0));  // 1
console.log(f( 1,  1));  // 0
console.log(f( 2,  0));  // 1
console.log(f( 2,  1));  // 1
console.log(f( 2,  2));  // 0
console.log(f( 3,  0));  // 1
console.log(f( 3,  1));  // 4
console.log(f( 3,  2));  // 1
console.log(f( 3,  3));  // 0
console.log(f( 4,  0));  // 1
console.log(f( 4,  1));  // 11
console.log(f( 4,  2));  // 11
console.log(f( 4,  3));  // 1
console.log(f( 4,  4));  // 0
console.log(f( 5,  1));  // 26
console.log(f( 7,  4));  // 1191
console.log(f( 9,  5));  // 88234
console.log(f(10,  5));  // 1310354
console.log(f(10,  7));  // 47840
console.log(f(10, 10));  // 0
console.log(f(12,  2));  // 478271
console.log(f(15,  6));  // 311387598411
console.log(f(17,  1));  // 131054
console.log(f(20, 16));  // 1026509354985
console.log(f(42, 42));  // 0


CJam (21 19 bytes - or 18 if argument order is free)

{\e!f{2ew::>1b=}1b}

This is an anonymous block (function) which takes n m on the stack. (If it's permitted to take m n on the stack then the \ can be saved). It computes all permutations and filters, so the online test suite must be rather limited.

Thanks to Martin for pointing out an approximation to filter-with-parameter.

Dissection

{        e# Define a block. Stack: n m
  \      e#   Flip the stack to give m n
  e!f{   e#   Generate permutations of [0 .. n-1] and map with parameter m
    2ew  e#     Stack: m perm; generate the list of n-1 pairs of consecutive
         e#     elements of perm
    ::>  e#     Map each pair to 1 if it's a rise and 0 if it's a fall
    1b   e#     Count the falls
    =    e#     Map to 1 if there are m falls and 0 otherwise
  }
  1b     e#   Count the permutations with m falls
}

Note that the Eulerian numbers are symmetric: E(n, m) = E(n, n-m), so it's irrelevant whether you count falls or rises.

Efficiently: 32 bytes

{1a@{0\+_ee::*(;\W%ee::*W%.+}*=}

Online test suite.

This implements the recurrence on whole rows.

{          e# Define a block. Stack: n m
  1a@      e#   Push the row for n=0: [1]; and rotate n to top of stack
  {        e#   Repeat n times:
           e#     Stack: m previous-row
    0\+_   e#     Prepend a 0 to the row and duplicate
    ee::*  e#     Multiply each element by its index
           e#     This gives A[j] = j * E(i-1, j-1)
    (;     e#     Pop the first element, so that A[j] = (j+1) * E(i-1, j)
    \W%    e#     Get the other copy of the previous row and reverse it
    ee::*  e#     Multiply each element by its index
           e#     This gives B[j] = j * E(i-1, i-1-j)
    W%     e#     Reverse again, giving B[j] = (i-j) * E(i-1, j-1)
    .+     e#     Pointwise addition
  }*
  =        e#   Extract the element at index j
}