1// RUN: mlir-opt %s -test-lower-to-llvm | \ 2// RUN: mlir-runner -e entry -entry-point-result=void \ 3// RUN: -shared-libs=%mlir_c_runner_utils | \ 4// RUN: FileCheck %s 5 6func.func @expand16(%base: memref<?xf32>, 7 %mask: vector<16xi1>, 8 %pass_thru: vector<16xf32>) -> vector<16xf32> { 9 %c0 = arith.constant 0: index 10 %e = vector.expandload %base[%c0], %mask, %pass_thru 11 : memref<?xf32>, vector<16xi1>, vector<16xf32> into vector<16xf32> 12 return %e : vector<16xf32> 13} 14 15func.func @expand16_at8(%base: memref<?xf32>, 16 %mask: vector<16xi1>, 17 %pass_thru: vector<16xf32>) -> vector<16xf32> { 18 %c8 = arith.constant 8: index 19 %e = vector.expandload %base[%c8], %mask, %pass_thru 20 : memref<?xf32>, vector<16xi1>, vector<16xf32> into vector<16xf32> 21 return %e : vector<16xf32> 22} 23 24func.func @entry() { 25 // Set up memory. 26 %c0 = arith.constant 0: index 27 %c1 = arith.constant 1: index 28 %c16 = arith.constant 16: index 29 %A = memref.alloc(%c16) : memref<?xf32> 30 scf.for %i = %c0 to %c16 step %c1 { 31 %i32 = arith.index_cast %i : index to i32 32 %fi = arith.sitofp %i32 : i32 to f32 33 memref.store %fi, %A[%i] : memref<?xf32> 34 } 35 36 // Set up pass thru vector. 37 %u = arith.constant -7.0: f32 38 %v = arith.constant 7.7: f32 39 %pass = vector.broadcast %u : f32 to vector<16xf32> 40 41 // Set up masks. 42 %f = arith.constant 0: i1 43 %t = arith.constant 1: i1 44 %none = vector.constant_mask [0] : vector<16xi1> 45 %all = vector.constant_mask [16] : vector<16xi1> 46 %some1 = vector.constant_mask [4] : vector<16xi1> 47 %0 = vector.insert %f, %some1[0] : i1 into vector<16xi1> 48 %1 = vector.insert %t, %0[7] : i1 into vector<16xi1> 49 %2 = vector.insert %t, %1[11] : i1 into vector<16xi1> 50 %3 = vector.insert %t, %2[13] : i1 into vector<16xi1> 51 %some2 = vector.insert %t, %3[15] : i1 into vector<16xi1> 52 %some3 = vector.insert %f, %some2[2] : i1 into vector<16xi1> 53 54 // 55 // Expanding load tests. 56 // 57 58 %e1 = call @expand16(%A, %none, %pass) 59 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 60 vector.print %e1 : vector<16xf32> 61 // CHECK: ( -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7 ) 62 63 %e2 = call @expand16(%A, %all, %pass) 64 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 65 vector.print %e2 : vector<16xf32> 66 // CHECK-NEXT: ( 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 ) 67 68 %e3 = call @expand16(%A, %some1, %pass) 69 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 70 vector.print %e3 : vector<16xf32> 71 // CHECK-NEXT: ( 0, 1, 2, 3, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7 ) 72 73 %e4 = call @expand16(%A, %some2, %pass) 74 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 75 vector.print %e4 : vector<16xf32> 76 // CHECK-NEXT: ( -7, 0, 1, 2, -7, -7, -7, 3, -7, -7, -7, 4, -7, 5, -7, 6 ) 77 78 %e5 = call @expand16(%A, %some3, %pass) 79 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 80 vector.print %e5 : vector<16xf32> 81 // CHECK-NEXT: ( -7, 0, -7, 1, -7, -7, -7, 2, -7, -7, -7, 3, -7, 4, -7, 5 ) 82 83 %4 = vector.insert %v, %pass[1] : f32 into vector<16xf32> 84 %5 = vector.insert %v, %4[2] : f32 into vector<16xf32> 85 %alt_pass = vector.insert %v, %5[14] : f32 into vector<16xf32> 86 %e6 = call @expand16(%A, %some3, %alt_pass) 87 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 88 vector.print %e6 : vector<16xf32> 89 // CHECK-NEXT: ( -7, 0, 7.7, 1, -7, -7, -7, 2, -7, -7, -7, 3, -7, 4, 7.7, 5 ) 90 91 %e7 = call @expand16_at8(%A, %some1, %pass) 92 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>) 93 vector.print %e7 : vector<16xf32> 94 // CHECK-NEXT: ( 8, 9, 10, 11, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7 ) 95 96 memref.dealloc %A : memref<?xf32> 97 return 98} 99