| /external/tensorflow/tensorflow/compiler/xla/tests/ |
| D | bfloat16_test.cc | 56 auto x = builder.ConstantR0<bfloat16>(static_cast<bfloat16>(2.0f)); in XLA_TEST_F()
57 auto y = builder.ConstantR0<bfloat16>(static_cast<bfloat16>(1.0f)); in XLA_TEST_F()
60 ComputeAndCompareR0<bfloat16>(&builder, static_cast<bfloat16>(3.0f), {}, in XLA_TEST_F()
66 auto x = builder.ConstantR0<bfloat16>(static_cast<bfloat16>(4.0f)); in XLA_TEST_F()
69 ComputeAndCompareR0<bfloat16>(&builder, static_cast<bfloat16>(1.387f), {}, in XLA_TEST_F()
75 builder.Neg(builder.ConstantR0<bfloat16>(static_cast<bfloat16>(2.1f))); in XLA_TEST_F()
77 ComputeAndCompareR0<bfloat16>(&builder, static_cast<bfloat16>(-2.1f), {}, in XLA_TEST_F()
85 auto operand = builder.ConstantR4FromArray4D<bfloat16>( in XLA_TEST_F()
86 {{{{static_cast<bfloat16>(1.f)}, {static_cast<bfloat16>(2.f)}}, in XLA_TEST_F()
87 {{static_cast<bfloat16>(3.f)}, {static_cast<bfloat16>(4.f)}}}, in XLA_TEST_F()
[all …]
|
| D | matrix_ops_simple_test.cc | 232 Array2D<bfloat16> lhs( in TEST_P()
233 {{bfloat16(1.0f), bfloat16(2.0f)}, {bfloat16(3.0), bfloat16(4.0)}}); in TEST_P()
234 Array2D<bfloat16> rhs( in TEST_P()
235 {{bfloat16(10.0f), bfloat16(11.0f)}, {bfloat16(12.0f), bfloat16(13.0f)}}); in TEST_P()
241 auto prim_type = primitive_util::NativeToPrimitiveType<bfloat16>(); in TEST_P()
250 *Literal::CreateR2FromArray2DWithLayout<bfloat16>( in TEST_P()
255 *Literal::CreateR2FromArray2DWithLayout<bfloat16>( in TEST_P()
266 Array2D<bfloat16> expected; in TEST_P()
270 expected = Array2D<bfloat16>( in TEST_P()
271 {{bfloat16(47), bfloat16(52)}, {bfloat16(71), bfloat16(78)}}); in TEST_P()
[all …]
|
| D | prng_test.cc | 92 UniformTest<bfloat16>(static_cast<bfloat16>(low), in XLA_TEST_F()
93 static_cast<bfloat16>(high), {}, /*seed=*/seed); in XLA_TEST_F()
96 UniformTest<bfloat16>(static_cast<bfloat16>(32.75), in XLA_TEST_F()
97 static_cast<bfloat16>(33), {}, /*seed=*/seed); in XLA_TEST_F()
98 UniformTest<bfloat16>(static_cast<bfloat16>(32.50), in XLA_TEST_F()
99 static_cast<bfloat16>(32.75), {}, /*seed=*/seed); in XLA_TEST_F()
100 UniformTest<bfloat16>(static_cast<bfloat16>(-33.00), in XLA_TEST_F()
101 static_cast<bfloat16>(-32.75), {}, /*seed=*/seed); in XLA_TEST_F()
102 UniformTest<bfloat16>(static_cast<bfloat16>(-32.75), in XLA_TEST_F()
103 static_cast<bfloat16>(-32.50), {}, /*seed=*/seed); in XLA_TEST_F()
[all …]
|
| D | dynamic_ops_test.cc | 205 XLA_TEST_F(DynamicSliceTest, Int32R1BF16) { TestR1<int32, bfloat16>(); } in XLA_TEST_F()
211 XLA_TEST_F(DynamicSliceTest, Int32R2BF16) { TestR2<int32, bfloat16>(); } in XLA_TEST_F()
217 XLA_TEST_F(DynamicSliceTest, Int32R3BF16) { TestR3<int32, bfloat16>(); } in XLA_TEST_F()
442 if (std::is_same<bfloat16, T>::value) { in RunR3Contiguous()
505 TestR1<int32, bfloat16>(); in XLA_TEST_F()
513 TestR2<int32, bfloat16>(); in XLA_TEST_F()
521 TestR3<int32, bfloat16>(); in XLA_TEST_F()
528 TestWrap<int32, bfloat16>(); in XLA_TEST_F()
603 RunR3Contiguous<bfloat16>(operand_shape, /*index=*/1, /*size=*/1); in XLA_TEST_F()
615 RunR3Contiguous<bfloat16>(operand_shape, /*index=*/1, /*size=*/2); in XLA_TEST_F()
[all …]
|
| D | literal_test_util.cc | 149 return ConvertType<bfloat16, float>(literal); in ConvertBF16ToF32()
154 return ConvertType<float, bfloat16>(literal); in ConvertF32ToBF16()
208 ::testing::AssertionResult CompareEqual<bfloat16>(bfloat16 lhs, bfloat16 rhs) { in CompareEqual()
209 return CompareFloatsBitwiseEqual<bfloat16, uint16>(lhs, rhs); in CompareEqual()
302 match = ExpectLiteralsEqual<bfloat16>(expected, actual, &multi_index, 0); in Equal()
389 ExpectLiteralsNear<bfloat16>(expected, actual, 0); in ExpectNear()
657 bool NearComparator::ExpectValuesNear<bfloat16>(bfloat16 expected, in ExpectValuesNear()
658 bfloat16 actual) { in ExpectValuesNear()
670 void NearComparator::UpdateAndLogMiscompares<bfloat16>( in UpdateAndLogMiscompares()
671 const bfloat16 expected, const bfloat16 actual, const Shape& shape, in UpdateAndLogMiscompares()
|
| D | convolution_test.cc | 749 Array4D<bfloat16> input_data(1, 1, 1, 2); in TEST_F()
750 input_data.FillWithYX(Array2D<bfloat16>({ in TEST_F()
751 {bfloat16(1), bfloat16(2)}, in TEST_F()
753 Array4D<bfloat16> filter_data(1, 1, 1, 2); in TEST_F()
754 filter_data.FillWithYX(Array2D<bfloat16>({ in TEST_F()
755 {bfloat16(5), bfloat16(6)}, in TEST_F()
|
| /external/tensorflow/tensorflow/python/lib/core/ |
| D | bfloat16_test.py | 32 bfloat16 = pywrap_tensorflow.TF_bfloat16_type() variable
53 self._assertFloatIdentical(v, float(bfloat16(v)))
57 self.assertEqual(v, int(bfloat16(v)))
60 self.assertEqual("0", str(bfloat16(0.0)))
61 self.assertEqual("1", str(bfloat16(1.0)))
62 self.assertEqual("-3.5", str(bfloat16(-3.5)))
63 self.assertEqual("0.0078125", str(bfloat16(float.fromhex("1.0p-7"))))
64 self.assertEqual("inf", str(bfloat16(float("inf"))))
65 self.assertEqual("-inf", str(bfloat16(float("-inf"))))
66 self.assertEqual("nan", str(bfloat16(float("nan"))))
[all …]
|
| D | bfloat16.cc | 81 bfloat16 value;
91 bfloat16 PyBfloat16_Bfloat16(PyObject* object) { in PyBfloat16_Bfloat16()
96 Safe_PyObjectPtr PyBfloat16_FromBfloat16(bfloat16 x) { in PyBfloat16_FromBfloat16()
108 bool AsBfloat16(PyObject* arg, bfloat16* output) { in AsBfloat16()
119 *output = bfloat16(d); in AsBfloat16()
128 *output = bfloat16(static_cast<float>(l)); in AsBfloat16()
134 *output = bfloat16(f); in AsBfloat16()
144 bfloat16 x = PyBfloat16_Bfloat16(self); in PyBfloat16_Float()
150 bfloat16 x = PyBfloat16_Bfloat16(self); in PyBfloat16_Int()
157 bfloat16 x = PyBfloat16_Bfloat16(self); in PyBfloat16_Negative()
[all …]
|
| /external/tensorflow/tensorflow/core/lib/bfloat16/ |
| D | bfloat16.h | 42 struct bfloat16 { struct
43 B16_DEVICE_FUNC bfloat16() {} in bfloat16() argument
45 B16_DEVICE_FUNC explicit bfloat16(const float v) { in bfloat16() argument
58 B16_DEVICE_FUNC explicit bfloat16(const double val) in bfloat16() argument
59 : bfloat16(static_cast<float>(val)) {} in bfloat16()
62 B16_DEVICE_FUNC explicit bfloat16(const complex64& val) in bfloat16() function
63 : bfloat16(val.real()) {} in bfloat16()
65 B16_DEVICE_FUNC explicit bfloat16(const complex128& val) in bfloat16() argument
66 : bfloat16(static_cast<float>(val.real())) {} in bfloat16()
68 B16_DEVICE_FUNC explicit bfloat16(const unsigned short val) in bfloat16() function
[all …]
|
| /external/tensorflow/tensorflow/core/framework/ |
| D | numeric_types.h | 48 static inline tensorflow::bfloat16 FloatToBFloat16(float float_val) { in FloatToBFloat16()
50 return *reinterpret_cast<tensorflow::bfloat16*>( in FloatToBFloat16()
53 return *reinterpret_cast<tensorflow::bfloat16*>( in FloatToBFloat16()
62 struct NumTraits<tensorflow::bfloat16>
63 : GenericNumTraits<tensorflow::bfloat16> {
69 static EIGEN_STRONG_INLINE tensorflow::bfloat16 highest() {
73 static EIGEN_STRONG_INLINE tensorflow::bfloat16 lowest() {
77 static EIGEN_STRONG_INLINE tensorflow::bfloat16 infinity() {
81 static EIGEN_STRONG_INLINE tensorflow::bfloat16 quiet_NaN() {
93 EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE tensorflow::bfloat16 log(
[all …]
|
| D | bfloat16_test.cc | 27 bfloat16 a(12); in TEST()
47 bfloat16 a(GetParam().input); in TEST_P()
97 bfloat16 b[100]; in TEST()
109 EXPECT_LT(1.0f, static_cast<float>(bfloat16::epsilon() + bfloat16(1.0f))); in TEST()
110 EXPECT_EQ(1.0f, static_cast<float>((bfloat16::epsilon() / bfloat16(2.0f)) + in TEST()
111 bfloat16(1.0f))); in TEST()
115 EXPECT_EQ(-3.0f, static_cast<float>(-bfloat16(3.0f))); in TEST()
116 EXPECT_EQ(4.5f, static_cast<float>(-bfloat16(-4.5f))); in TEST()
124 testing::BytesProcessed(tot * (sizeof(float) + sizeof(bfloat16))); in BM_FloatToBFloat16()
127 bfloat16* out = new bfloat16[N]; in BM_FloatToBFloat16()
[all …]
|
| D | bfloat16.h | 57 void FloatToBFloat16(const float* src, bfloat16* dst, int64 size);
58 void BFloat16ToFloat(const bfloat16* src, float* dst, int64 size);
|
| D | bfloat16.cc | 20 void FloatToBFloat16(const float* src, bfloat16* dst, int64 size) { in FloatToBFloat16()
34 void BFloat16ToFloat(const bfloat16* src, float* dst, int64 size) { in BFloat16ToFloat()
|
| /external/tensorflow/tensorflow/core/kernels/ |
| D | concat_op_test.cc | 85 ConcatHelper<bfloat16>(iters, 1, dim2); in BM_ConcatDim1bfloat16()
124 ConcatManyHelper<bfloat16>(iters, 1, dim2); in BM_ConcatManyDim1bfloat16()
159 typedef Eigen::TensorMap<Eigen::Tensor<bfloat16, 1, Eigen::RowMajor>,
168 bfloat16* data = new bfloat16[size]; in MemcpyManyAlternative1()
173 kNumCopies * sizeof(bfloat16)); in MemcpyManyAlternative1()
176 std::vector<bfloat16*> inputs(kNumCopies); in MemcpyManyAlternative1()
180 bfloat16* result = new bfloat16[size]; in MemcpyManyAlternative1()
182 bfloat16* output = &result[j * dim2]; in MemcpyManyAlternative1()
187 memcpy(output, inputs[j], dim2 * sizeof(bfloat16)); in MemcpyManyAlternative1()
203 bfloat16* data = new bfloat16[size]; in MemcpyManyAlternative2()
[all …]
|
| D | cast_op.h | 123 struct scalar_cast_op<::tensorflow::bfloat16, float> {
127 const ::tensorflow::bfloat16& a) const {
144 struct functor_traits<scalar_cast_op<::tensorflow::bfloat16, float>> {
149 struct scalar_cast_op<float, ::tensorflow::bfloat16> {
151 typedef ::tensorflow::bfloat16 result_type;
152 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ::tensorflow::bfloat16 operator()(
154 return ::tensorflow::bfloat16(a);
159 struct functor_traits<scalar_cast_op<float, ::tensorflow::bfloat16>> {
|
| D | sparse_matmul_op.cc | 277 ALWAYS_INLINE float ConvertBfloat16ToFloat(const bfloat16* src) { in ConvertBfloat16ToFloat()
279 auto tmp = reinterpret_cast<bfloat16*>(&out); in ConvertBfloat16ToFloat()
288 ALWAYS_INLINE Packet ConvertFourBfloat16ToFloat(const bfloat16* src) { in ConvertFourBfloat16ToFloat()
293 ALWAYS_INLINE Packet ConvertTwoBfloat16ToFloat(const bfloat16* src) { in ConvertTwoBfloat16ToFloat()
304 ALWAYS_INLINE void ScalarMulAdd(const float a, const bfloat16** inp, in ScalarMulAdd()
312 const float a3, const bfloat16** inp1, in ScalarMulAdd3Way()
313 const bfloat16** inp2, in ScalarMulAdd3Way()
314 const bfloat16** inp3, float** out) { in ScalarMulAdd3Way()
336 ALWAYS_INLINE void LoadSingleScalar(const bfloat16** data, Packet* l) { in LoadSingleScalar()
342 ALWAYS_INLINE void LoadTwoScalars(const bfloat16** data, Packet* l1, in LoadTwoScalars()
[all …]
|
| D | cast_op_impl_bfloat.cc | 32 BFloat16ToFloat(inp.flat<bfloat16>().data() + start, in GetCpuCastFromBfloat()
46 functor::CastFunctor<GPUDevice, float, bfloat16> func; in GetGpuCastFromBfloat()
48 inp.flat<bfloat16>()); in GetGpuCastFromBfloat()
|
| D | cast_op_test.cc | 143 (sizeof(float) + sizeof(bfloat16))); in BM_cpu_float_bfloat16()
145 test::Benchmark("cpu", Cast<float, bfloat16>(num)).Run(iters); in BM_cpu_float_bfloat16()
152 (sizeof(float) + sizeof(bfloat16))); in BM_cpu_bfloat16_float()
154 test::Benchmark("cpu", Cast<bfloat16, float>(num)).Run(iters); in BM_cpu_bfloat16_float()
|
| D | concat_lib_gpu_impl.cu.cc | 205 REGISTER_GPUCONCAT32(bfloat16);
212 REGISTER_GPUCONCAT64(bfloat16);
219 REGISTER_GPU32(bfloat16);
226 REGISTER_GPU64(bfloat16);
|
| D | list_kernels.cu.cc | 46 REGISTER_TENSOR_LIST_STACK_GPU(bfloat16);
62 REGISTER_TENSOR_LIST_FROM_TENSOR_GPU(bfloat16);
|
| /external/tensorflow/tensorflow/python/kernel_tests/ |
| D | sparse_matmul_op_test.py | 77 for x_dtype in (dtypes.float32, dtypes.bfloat16):
78 for y_dtype in (dtypes.float32, dtypes.bfloat16):
84 for x_dtype in (dtypes.float32, dtypes.bfloat16):
85 for y_dtype in (dtypes.float32, dtypes.bfloat16):
91 for x_dtype in (dtypes.float32, dtypes.bfloat16):
92 for y_dtype in (dtypes.float32, dtypes.bfloat16):
101 for x_dtype in (dtypes.float32, dtypes.bfloat16):
102 for y_dtype in (dtypes.float32, dtypes.bfloat16):
113 for x_dtype in (dtypes.float32, dtypes.bfloat16):
114 for y_dtype in (dtypes.float32, dtypes.bfloat16):
[all …]
|
| /external/tensorflow/tensorflow/compiler/xla/ |
| D | literal_util_test.cc | 115 auto bf16_lit = Literal::CreateR0<bfloat16>(static_cast<bfloat16>(0.5f)); in TEST_F()
120 Literal::CreateR0<bfloat16>(static_cast<bfloat16>(3.14f)); in TEST_F()
124 Literal::CreateR0<bfloat16>(static_cast<bfloat16>(9.001f)); in TEST_F()
440 bfloat16 b8(8.0f); in TEST_F()
441 bfloat16 b9(9.0f); in TEST_F()
443 EXPECT_TRUE(Literal::CreateR2<bfloat16>({{b8}, {b8}})->IsAll(8)); in TEST_F()
444 EXPECT_FALSE(Literal::CreateR2<bfloat16>({{b8}, {b9}})->IsAll(8)); in TEST_F()
445 EXPECT_FALSE(Literal::CreateR2<bfloat16>({{b9}, {b8}})->IsAll(8)); in TEST_F()
448 bfloat16 b91(9.001f); in TEST_F()
449 bfloat16 b90(9.00f); in TEST_F()
[all …]
|
| D | literal_util.cc | 338 COPY_ELEMENTS(BF16, bfloat16); in CopyFrom()
477 return CopySliceFromInternal<bfloat16>(src_literal, src_base, dest_base, in CopySliceFrom()
515 *Literal::CreateR0<bfloat16>(static_cast<bfloat16>(0.0f))); in Zero()
554 *Literal::CreateR0<bfloat16>(static_cast<bfloat16>(1.0f))); in One()
612 return std::move(*Literal::CreateR0<bfloat16>( in MinValue()
613 static_cast<bfloat16>(-std::numeric_limits<float>::infinity()))); in MinValue()
658 return std::move(*Literal::CreateR0<bfloat16>( in MaxValue()
659 static_cast<bfloat16>(std::numeric_limits<float>::infinity()))); in MaxValue()
911 static_cast<float>(Get<bfloat16>(multi_index, shape_index))); in GetAsString()
962 GetSparseElement<bfloat16>(sparse_element_number, shape_index))); in GetSparseElementAsString()
[all …]
|
| /external/tensorflow/tensorflow/core/util/ |
| D | cuda_kernel_helper.h | 32 __host__ __device__ inline tensorflow::bfloat16 CudaLdg( in CudaLdg()
33 const tensorflow::bfloat16* address) { in CudaLdg()
34 tensorflow::bfloat16 return_value; in CudaLdg()
|
| /external/tensorflow/tensorflow/compiler/tf2xla/lib/ |
| D | util.cc | 44 return builder->ConstantR0<bfloat16>(static_cast<bfloat16>(value)); in FloatLiteral()
99 *xla::Literal::CreateR0<bfloat16>(static_cast<bfloat16>(value))); in IntegerLiteral()
|