82 lines · plain
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 %s5 6func.func @maskedload16(%base: memref<?xf32>, %mask: vector<16xi1>,7 %pass_thru: vector<16xf32>) -> vector<16xf32> {8 %c0 = arith.constant 0: index9 %ld = vector.maskedload %base[%c0], %mask, %pass_thru10 : memref<?xf32>, vector<16xi1>, vector<16xf32> into vector<16xf32>11 return %ld : vector<16xf32>12}13 14func.func @maskedload16_at8(%base: memref<?xf32>, %mask: vector<16xi1>,15 %pass_thru: vector<16xf32>) -> vector<16xf32> {16 %c8 = arith.constant 8: index17 %ld = vector.maskedload %base[%c8], %mask, %pass_thru18 : memref<?xf32>, vector<16xi1>, vector<16xf32> into vector<16xf32>19 return %ld : vector<16xf32>20}21 22func.func @entry() {23 // Set up memory.24 %c0 = arith.constant 0: index25 %c1 = arith.constant 1: index26 %c16 = arith.constant 16: index27 %A = memref.alloc(%c16) : memref<?xf32>28 scf.for %i = %c0 to %c16 step %c1 {29 %i32 = arith.index_cast %i : index to i3230 %fi = arith.sitofp %i32 : i32 to f3231 memref.store %fi, %A[%i] : memref<?xf32>32 }33 34 // Set up pass thru vector.35 %u = arith.constant -7.0: f3236 %pass = vector.broadcast %u : f32 to vector<16xf32>37 38 // Set up masks.39 %f = arith.constant 0: i140 %t = arith.constant 1: i141 %none = vector.constant_mask [0] : vector<16xi1>42 %all = vector.constant_mask [16] : vector<16xi1>43 %some = vector.constant_mask [8] : vector<16xi1>44 %0 = vector.insert %f, %some[0] : i1 into vector<16xi1>45 %1 = vector.insert %t, %0[13] : i1 into vector<16xi1>46 %2 = vector.insert %t, %1[14] : i1 into vector<16xi1>47 %other = vector.insert %t, %2[14] : i1 into vector<16xi1>48 49 //50 // Masked load tests.51 //52 53 %l1 = call @maskedload16(%A, %none, %pass)54 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>)55 vector.print %l1 : vector<16xf32>56 // CHECK: ( -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7, -7 )57 58 %l2 = call @maskedload16(%A, %all, %pass)59 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>)60 vector.print %l2 : vector<16xf32>61 // CHECK: ( 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 )62 63 %l3 = call @maskedload16(%A, %some, %pass)64 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>)65 vector.print %l3 : vector<16xf32>66 // CHECK: ( 0, 1, 2, 3, 4, 5, 6, 7, -7, -7, -7, -7, -7, -7, -7, -7 )67 68 %l4 = call @maskedload16(%A, %other, %pass)69 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>)70 vector.print %l4 : vector<16xf32>71 // CHECK: ( -7, 1, 2, 3, 4, 5, 6, 7, -7, -7, -7, -7, -7, 13, 14, -7 )72 73 %l5 = call @maskedload16_at8(%A, %some, %pass)74 : (memref<?xf32>, vector<16xi1>, vector<16xf32>) -> (vector<16xf32>)75 vector.print %l5 : vector<16xf32>76 // CHECK: ( 8, 9, 10, 11, 12, 13, 14, 15, -7, -7, -7, -7, -7, -7, -7, -7 )77 78 memref.dealloc %A : memref<?xf32>79 return80}81 82