|
| 1 | +#include <algorithm> |
| 2 | +#include <functional> |
| 3 | +#include <iostream> |
| 4 | +#include <limits> |
| 5 | +#include <memory> |
| 6 | +#include <stdint.h> |
| 7 | + |
| 8 | +#include "common.h" |
| 9 | + |
| 10 | +template <typename RetTy, typename Ty> |
| 11 | +using Fn2Ty = std::function<RetTy(Ty *, Ty *, unsigned)>; |
| 12 | +template <typename RetTy, typename Ty> |
| 13 | +static void checkVectorFunction(Fn2Ty<RetTy, Ty> ScalarFn, |
| 14 | + Fn2Ty<RetTy, Ty> VectorFn, const char *Name) { |
| 15 | + std::cout << "Checking " << Name << "\n"; |
| 16 | + |
| 17 | + unsigned N = 1000; |
| 18 | + std::unique_ptr<Ty[]> Src1(new Ty[N]); |
| 19 | + std::unique_ptr<Ty[]> Src2(new Ty[N]); |
| 20 | + init_data(Src1, N); |
| 21 | + init_data(Src2, N); |
| 22 | + |
| 23 | + // Test VectorFn with different input data. |
| 24 | + { |
| 25 | + // Check with random inputs. |
| 26 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 27 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 28 | + if (Reference != ToCheck) { |
| 29 | + std::cerr << "Miscompare\n"; |
| 30 | + exit(1); |
| 31 | + } |
| 32 | + } |
| 33 | + |
| 34 | + { |
| 35 | + // Check with Src1 > Src2 for all elements. |
| 36 | + for (unsigned I = 0; I != N; ++I) { |
| 37 | + Src1[I] = std::numeric_limits<Ty>::max(); |
| 38 | + Src2[I] = std::numeric_limits<Ty>::min(); |
| 39 | + } |
| 40 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 41 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 42 | + if (Reference != ToCheck) { |
| 43 | + std::cerr << "Miscompare\n"; |
| 44 | + exit(1); |
| 45 | + } |
| 46 | + } |
| 47 | + |
| 48 | + { |
| 49 | + // Check with Src1 < Src2 for all elements. |
| 50 | + for (unsigned I = 0; I != N; ++I) { |
| 51 | + Src1[I] = std::numeric_limits<Ty>::min(); |
| 52 | + Src2[I] = std::numeric_limits<Ty>::max(); |
| 53 | + } |
| 54 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 55 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 56 | + if (Reference != ToCheck) { |
| 57 | + std::cerr << "Miscompare\n"; |
| 58 | + exit(1); |
| 59 | + } |
| 60 | + } |
| 61 | + |
| 62 | + { |
| 63 | + // Check with only Src1[998] > Src2[998]. |
| 64 | + for (unsigned I = 0; I != N; ++I) |
| 65 | + Src1[I] = Src2[I] = std::numeric_limits<Ty>::min(); |
| 66 | + Src1[998] = std::numeric_limits<Ty>::max(); |
| 67 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 68 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 69 | + if (Reference != ToCheck) { |
| 70 | + std::cerr << "Miscompare\n"; |
| 71 | + exit(1); |
| 72 | + } |
| 73 | + } |
| 74 | + |
| 75 | + { |
| 76 | + // Check with only Src1[0] > Src2[0]. |
| 77 | + for (unsigned I = 0; I != N; ++I) |
| 78 | + Src1[I] = Src2[I] = std::numeric_limits<Ty>::min(); |
| 79 | + Src1[0] = std::numeric_limits<Ty>::max(); |
| 80 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 81 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 82 | + if (Reference != ToCheck) { |
| 83 | + std::cerr << "Miscompare\n"; |
| 84 | + exit(1); |
| 85 | + } |
| 86 | + } |
| 87 | + |
| 88 | + { |
| 89 | + // Check with only Src1[N - 1] > Src2[N - 1]. |
| 90 | + for (unsigned I = 0; I != N; ++I) |
| 91 | + Src1[I] = Src2[I] = std::numeric_limits<Ty>::min(); |
| 92 | + Src1[N - 1] = std::numeric_limits<Ty>::max(); |
| 93 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 94 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 95 | + if (Reference != ToCheck) { |
| 96 | + std::cerr << "Miscompare\n"; |
| 97 | + exit(1); |
| 98 | + } |
| 99 | + } |
| 100 | + |
| 101 | + { |
| 102 | + // Check with only Src1[0] > Src2[0] and Src1[N - 1] > Src2[N - 1]. |
| 103 | + for (unsigned I = 0; I != N; ++I) |
| 104 | + Src1[I] = Src2[I] = std::numeric_limits<Ty>::min(); |
| 105 | + Src1[0] = Src1[N - 1] = std::numeric_limits<Ty>::max(); |
| 106 | + auto Reference = ScalarFn(&Src1[0], &Src2[0], N); |
| 107 | + auto ToCheck = VectorFn(&Src1[0], &Src2[0], N); |
| 108 | + if (Reference != ToCheck) { |
| 109 | + std::cerr << "Miscompare\n"; |
| 110 | + exit(1); |
| 111 | + } |
| 112 | + } |
| 113 | +} |
| 114 | + |
| 115 | +int main(void) { |
| 116 | + rng = std::mt19937(15); |
| 117 | + |
| 118 | +#define DEFINE_ANYOF_LOOP_BODY(TrueVal, FalseVal) \ |
| 119 | + for (unsigned I = 0; I < TC; I++) { \ |
| 120 | + Rdx = A[I] > B[I] ? TrueVal : FalseVal; \ |
| 121 | + } \ |
| 122 | + return Rdx; |
| 123 | + |
| 124 | + { |
| 125 | + // Update 32-bits integer Rdx to 3 once any A[i] > B[i] is true. |
| 126 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 127 | + int32_t Rdx = -1; |
| 128 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ 3, /* FalseVal= */ Rdx)); |
| 129 | + checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn, |
| 130 | + "anyof_icmp_s32_true_update"); |
| 131 | + checkVectorFunction<int32_t, float>(ScalarFn, VectorFn, |
| 132 | + "anyof_fcmp_s32_true_update"); |
| 133 | + } |
| 134 | + |
| 135 | + { |
| 136 | + // Update 16-bits integer Rdx to 3 once any A[i] > B[i] is true. |
| 137 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 138 | + int16_t Rdx = -1; |
| 139 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ 3, /* FalseVal= */ Rdx)); |
| 140 | + checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn, |
| 141 | + "anyof_icmp_s16_true_update"); |
| 142 | + } |
| 143 | + |
| 144 | + { |
| 145 | + // Update 32-bits integer Rdx to 3 once any A[i] > B[i] is false. |
| 146 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 147 | + int32_t Rdx = -1; |
| 148 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ Rdx, /* FalseVal= */ 3)); |
| 149 | + checkVectorFunction<int32_t, int32_t>(ScalarFn, VectorFn, |
| 150 | + "anyof_icmp_s32_false_update"); |
| 151 | + checkVectorFunction<int32_t, float>(ScalarFn, VectorFn, |
| 152 | + "anyof_fcmp_s32_false_update"); |
| 153 | + } |
| 154 | + |
| 155 | + { |
| 156 | + // Update 16-bits integer Rdx to 3 once any A[i] > B[i] is false. |
| 157 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 158 | + int16_t Rdx = -1; |
| 159 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ Rdx, /* FalseVal= */ 3)); |
| 160 | + checkVectorFunction<int16_t, int16_t>(ScalarFn, VectorFn, |
| 161 | + "anyof_icmp_s16_false_update"); |
| 162 | + } |
| 163 | + |
| 164 | + { |
| 165 | + // 32-bits integer Rdx starts from TC, and will be updated to 3 once any |
| 166 | + // A[i] > B[i] is true. |
| 167 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 168 | + uint32_t Rdx = TC; |
| 169 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ 3, /* FalseVal= */ Rdx)); |
| 170 | + checkVectorFunction<uint32_t, uint32_t>(ScalarFn, VectorFn, |
| 171 | + "anyof_icmp_u32_start_TC"); |
| 172 | + checkVectorFunction<uint32_t, float>(ScalarFn, VectorFn, |
| 173 | + "anyof_fcmp_u32_start_TC"); |
| 174 | + } |
| 175 | + |
| 176 | + { |
| 177 | + // 16-bits integer Rdx starts from TC, and will be updated to 3 once any |
| 178 | + // A[i] > B[i] is true. |
| 179 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 180 | + uint16_t Rdx = TC; |
| 181 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ 3, /* FalseVal= */ Rdx)); |
| 182 | + checkVectorFunction<uint16_t, uint16_t>(ScalarFn, VectorFn, |
| 183 | + "anyof_icmp_u16_start_TC"); |
| 184 | + } |
| 185 | + |
| 186 | + { |
| 187 | + // 32-bits integer Rdx starts from 3, and will be updated to TC once any |
| 188 | + // A[i] > B[i] is true. |
| 189 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 190 | + uint32_t Rdx = 3; |
| 191 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ TC, /* FalseVal= */ Rdx)); |
| 192 | + checkVectorFunction<uint32_t, uint32_t>(ScalarFn, VectorFn, |
| 193 | + "anyof_icmp_u32_update_by_TC"); |
| 194 | + checkVectorFunction<uint32_t, float>(ScalarFn, VectorFn, |
| 195 | + "anyof_fcmp_u32_update_by_TC"); |
| 196 | + } |
| 197 | + |
| 198 | + { |
| 199 | + // 16-bits integer Rdx starts from 3, and will be updated to TC once any |
| 200 | + // A[i] > B[i] is true. |
| 201 | + DEFINE_SCALAR_AND_VECTOR_FN2( |
| 202 | + uint16_t Rdx = 3; |
| 203 | + , DEFINE_ANYOF_LOOP_BODY(/* TrueVal= */ TC, /* FalseVal= */ Rdx)); |
| 204 | + checkVectorFunction<uint16_t, uint16_t>(ScalarFn, VectorFn, |
| 205 | + "anyof_icmp_u16_update_by_TC"); |
| 206 | + } |
| 207 | + |
| 208 | + return 0; |
| 209 | +} |
0 commit comments