xref: /llvm-project/mlir/test/Integration/Dialect/LLVMIR/CPU/X86/test-inline-asm-vector.mlir (revision eb206e9ea84eff0a0596fed2de8316d924f946d1)
1// RUN: mlir-opt %s -convert-vector-to-scf -convert-scf-to-cf -convert-vector-to-llvm -convert-func-to-llvm -convert-arith-to-llvm -convert-cf-to-llvm -reconcile-unrealized-casts |  \
2// RUN: mlir-runner -e entry_point_with_all_constants -entry-point-result=void \
3// RUN:   -shared-libs=%mlir_c_runner_utils
4
5module {
6  func.func @function_to_run(%a: vector<8xf32>, %b: vector<8xf32>)  {
7    // CHECK: ( 8, 10, 12, 14, 16, 18, 20, 22 )
8    %r0 = llvm.inline_asm asm_dialect = intel
9        "vaddps $0, $1, $2", "=x,x,x" %a, %b:
10      (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
11    vector.print %r0: vector<8xf32>
12
13    // vblendps implemented with inline_asm.
14    // CHECK: ( 0, 1, 10, 11, 4, 5, 14, 15 )
15    %r1 = llvm.inline_asm asm_dialect = intel
16        "vblendps $0, $1, $2, 0xCC", "=x,x,x" %a, %b:
17      (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
18    vector.print %r1: vector<8xf32>
19
20    // vblendps 0xCC via vector.shuffle (emulates clang intrinsics impl)
21    // CHECK: ( 0, 1, 10, 11, 4, 5, 14, 15 )
22    %r2 = vector.shuffle %a, %b[0, 1, 10, 11, 4, 5, 14, 15]
23      : vector<8xf32>, vector<8xf32>
24    vector.print %r2: vector<8xf32>
25
26    // vblendps 0x33 implemented with inline_asm.
27    // CHECK: ( 8, 9, 2, 3, 12, 13, 6, 7 )
28    %r3 = llvm.inline_asm asm_dialect = intel
29        "vblendps $0, $1, $2, 0x33", "=x,x,x" %a, %b:
30      (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
31    vector.print %r3: vector<8xf32>
32
33    // vblendps 0x33 via vector.shuffle (emulates clang intrinsics impl)
34    // CHECK: ( 8, 9, 2, 3, 12, 13, 6, 7 )
35    %r4 = vector.shuffle %a, %b[8, 9, 2, 3, 12, 13, 6, 7]
36      : vector<8xf32>, vector<8xf32>
37    vector.print %r4: vector<8xf32>
38
39    return
40  }
41
42  // Solely exists to prevent inlining and get the expected assembly.
43  func.func @entry_point(%a: vector<8xf32>, %b: vector<8xf32>)  {
44    func.call @function_to_run(%a, %b) : (vector<8xf32>, vector<8xf32>) -> ()
45    return
46  }
47
48  func.func @entry_point_with_all_constants()  {
49    %a = llvm.mlir.constant(dense<[0.0, 1.0,  2.0,  3.0,  4.0,  5.0,  6.0,  7.0]>
50      : vector<8xf32>) : vector<8xf32>
51    %b = llvm.mlir.constant(dense<[8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0]>
52      : vector<8xf32>) : vector<8xf32>
53    func.call @function_to_run(%a, %b) : (vector<8xf32>, vector<8xf32>) -> ()
54    return
55  }
56}
57