tint/resolver: Test f16 types in materialize unittests
This patch add f16 types materialize unittests. Bug: tint:1473, tint:1502 Change-Id: Ifc4e991cf19e58a9fbfb4f032462d548694c2a5c Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/97820 Reviewed-by: Ben Clayton <bclayton@google.com> Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
This commit is contained in:
parent
8c590acda0
commit
0ea0d4b16e
|
@ -35,18 +35,20 @@ using f16V = builder::vec<3, f16>;
|
||||||
using i32V = builder::vec<3, i32>;
|
using i32V = builder::vec<3, i32>;
|
||||||
using u32V = builder::vec<3, u32>;
|
using u32V = builder::vec<3, u32>;
|
||||||
using f32M = builder::mat<3, 2, f32>;
|
using f32M = builder::mat<3, 2, f32>;
|
||||||
|
using f16M = builder::mat<3, 2, f16>;
|
||||||
using f32A = builder::array<3, f32>;
|
using f32A = builder::array<3, f32>;
|
||||||
|
using f16A = builder::array<3, f16>;
|
||||||
using i32A = builder::array<3, i32>;
|
using i32A = builder::array<3, i32>;
|
||||||
using u32A = builder::array<3, u32>;
|
using u32A = builder::array<3, u32>;
|
||||||
|
|
||||||
constexpr double kTooBigF32 = static_cast<double>(3.5e+38);
|
constexpr double kTooBigF32 = static_cast<double>(3.5e+38);
|
||||||
// constexpr double kTooBigF16 = static_cast<double>(6.6e+4);
|
constexpr double kTooBigF16 = static_cast<double>(6.6e+4);
|
||||||
constexpr double kPiF64 = 3.141592653589793;
|
constexpr double kPiF64 = 3.141592653589793;
|
||||||
constexpr double kPiF32 = 3.1415927410125732; // kPiF64 quantized to f32
|
constexpr double kPiF32 = 3.1415927410125732; // kPiF64 quantized to f32
|
||||||
// constexpr double kPiF16 = 3.140625; // kPiF64 quantized to f16
|
constexpr double kPiF16 = 3.140625; // kPiF64 quantized to f16
|
||||||
|
|
||||||
constexpr double kSubnormalF32 = 0x1.0p-128;
|
constexpr double kSubnormalF32 = 0x1.0p-128;
|
||||||
// constexpr double kSubnormalF16 = 0x1.0p-16;
|
constexpr double kSubnormalF16 = 0x1.0p-16;
|
||||||
|
|
||||||
enum class Expectation {
|
enum class Expectation {
|
||||||
kMaterialize,
|
kMaterialize,
|
||||||
|
@ -308,8 +310,7 @@ using MaterializeAbstractNumericToConcreteType =
|
||||||
MaterializeTest<std::tuple<Expectation, Method, Data>>;
|
MaterializeTest<std::tuple<Expectation, Method, Data>>;
|
||||||
|
|
||||||
TEST_P(MaterializeAbstractNumericToConcreteType, Test) {
|
TEST_P(MaterializeAbstractNumericToConcreteType, Test) {
|
||||||
// Once built-in and ops using f16 is properly supported, we'll need to enable this:
|
Enable(ast::Extension::kF16);
|
||||||
// Enable(ast::Extension::kF16);
|
|
||||||
|
|
||||||
const auto& param = GetParam();
|
const auto& param = GetParam();
|
||||||
const auto& expectation = std::get<0>(param);
|
const auto& expectation = std::get<0>(param);
|
||||||
|
@ -489,13 +490,13 @@ INSTANTIATE_TEST_SUITE_P(
|
||||||
Types<f32, AFloat>(AFloat(kPiF32), kPiF64), //
|
Types<f32, AFloat>(AFloat(kPiF32), kPiF64), //
|
||||||
Types<f32, AFloat>(AFloat(kSubnormalF32), kSubnormalF32), //
|
Types<f32, AFloat>(AFloat(kSubnormalF32), kSubnormalF32), //
|
||||||
Types<f32, AFloat>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
Types<f32, AFloat>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
||||||
/* Types<f16, AFloat>(0.0_a, 0.0), */ //
|
Types<f16, AFloat>(0.0_a, 0.0), //
|
||||||
/* Types<f16, AFloat>(1.0_a, 1.0), */ //
|
Types<f16, AFloat>(1.0_a, 1.0), //
|
||||||
/* Types<f16, AFloat>(AFloat(kHighestF16), kHighestF16), */ //
|
Types<f16, AFloat>(AFloat(f16::Highest()), static_cast<double>(f16::Highest())), //
|
||||||
/* Types<f16, AFloat>(AFloat(kLowestF16), kLowestF16), */ //
|
Types<f16, AFloat>(AFloat(f16::Lowest()), static_cast<double>(f16::Lowest())), //
|
||||||
/* Types<f16, AFloat>(AFloat(kPiF16), kPiF64), */ //
|
Types<f16, AFloat>(AFloat(kPiF16), kPiF64), //
|
||||||
/* Types<f16, AFloat>(AFloat(kSubnormalF16), kSubnormalF16), */ //
|
Types<f16, AFloat>(AFloat(kSubnormalF16), kSubnormalF16), //
|
||||||
/* Types<f16, AFloat>(AFloat(-kSubnormalF16), -kSubnormalF16), */ //
|
Types<f16, AFloat>(AFloat(-kSubnormalF16), -kSubnormalF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
INSTANTIATE_TEST_SUITE_P(
|
INSTANTIATE_TEST_SUITE_P(
|
||||||
|
@ -522,14 +523,14 @@ INSTANTIATE_TEST_SUITE_P(
|
||||||
Types<f32V, AFloatV>(AFloat(kPiF32), kPiF64), //
|
Types<f32V, AFloatV>(AFloat(kPiF32), kPiF64), //
|
||||||
Types<f32V, AFloatV>(AFloat(kSubnormalF32), kSubnormalF32), //
|
Types<f32V, AFloatV>(AFloat(kSubnormalF32), kSubnormalF32), //
|
||||||
Types<f32V, AFloatV>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
Types<f32V, AFloatV>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
||||||
/* Types<f16V, AFloatV>(0.0_a, 0.0), */ //
|
Types<f16V, AFloatV>(0.0_a, 0.0), //
|
||||||
/* Types<f16V, AFloatV>(1.0_a, 1.0), */ //
|
Types<f16V, AFloatV>(1.0_a, 1.0), //
|
||||||
/* Types<f16V, AFloatV>(-1.0_a, -1.0), */ //
|
Types<f16V, AFloatV>(-1.0_a, -1.0), //
|
||||||
/* Types<f16V, AFloatV>(AFloat(kHighestF16), kHighestF16), */ //
|
Types<f16V, AFloatV>(AFloat(f16::Highest()), static_cast<double>(f16::Highest())), //
|
||||||
/* Types<f16V, AFloatV>(AFloat(kLowestF16), kLowestF16), */ //
|
Types<f16V, AFloatV>(AFloat(f16::Lowest()), static_cast<double>(f16::Lowest())), //
|
||||||
/* Types<f16V, AFloatV>(AFloat(kPiF16), kPiF64), */ //
|
Types<f16V, AFloatV>(AFloat(kPiF16), kPiF64), //
|
||||||
/* Types<f16V, AFloatV>(AFloat(kSubnormalF16), kSubnormalF16), */ //
|
Types<f16V, AFloatV>(AFloat(kSubnormalF16), kSubnormalF16), //
|
||||||
/* Types<f16V, AFloatV>(AFloat(-kSubnormalF16), -kSubnormalF16), */ //
|
Types<f16V, AFloatV>(AFloat(-kSubnormalF16), -kSubnormalF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
INSTANTIATE_TEST_SUITE_P(
|
INSTANTIATE_TEST_SUITE_P(
|
||||||
|
@ -569,14 +570,14 @@ INSTANTIATE_TEST_SUITE_P(
|
||||||
Types<f32M, AFloatM>(AFloat(kPiF32), kPiF64), //
|
Types<f32M, AFloatM>(AFloat(kPiF32), kPiF64), //
|
||||||
Types<f32M, AFloatM>(AFloat(kSubnormalF32), kSubnormalF32), //
|
Types<f32M, AFloatM>(AFloat(kSubnormalF32), kSubnormalF32), //
|
||||||
Types<f32M, AFloatM>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
Types<f32M, AFloatM>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
||||||
/* Types<f16M, AFloatM>(0.0_a, 0.0), */ //
|
Types<f16M, AFloatM>(0.0_a, 0.0), //
|
||||||
/* Types<f16M, AFloatM>(1.0_a, 1.0), */ //
|
Types<f16M, AFloatM>(1.0_a, 1.0), //
|
||||||
/* Types<f16M, AFloatM>(-1.0_a, -1.0), */ //
|
Types<f16M, AFloatM>(-1.0_a, -1.0), //
|
||||||
/* Types<f16M, AFloatM>(AFloat(kHighestF16), kHighestF16), */ //
|
Types<f16M, AFloatM>(AFloat(f16::Highest()), static_cast<double>(f16::Highest())), //
|
||||||
/* Types<f16M, AFloatM>(AFloat(kLowestF16), kLowestF16), */ //
|
Types<f16M, AFloatM>(AFloat(f16::Lowest()), static_cast<double>(f16::Lowest())), //
|
||||||
/* Types<f16M, AFloatM>(AFloat(kPiF16), kPiF64), */ //
|
Types<f16M, AFloatM>(AFloat(kPiF16), kPiF64), //
|
||||||
/* Types<f16M, AFloatM>(AFloat(kSubnormalF16), kSubnormalF16), */ //
|
Types<f16M, AFloatM>(AFloat(kSubnormalF16), kSubnormalF16), //
|
||||||
/* Types<f16M, AFloatM>(AFloat(-kSubnormalF16), -kSubnormalF16), */ //
|
Types<f16M, AFloatM>(AFloat(-kSubnormalF16), -kSubnormalF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
INSTANTIATE_TEST_SUITE_P(
|
INSTANTIATE_TEST_SUITE_P(
|
||||||
|
@ -637,14 +638,14 @@ INSTANTIATE_TEST_SUITE_P(
|
||||||
Types<f32A, AFloatA>(AFloat(kPiF32), kPiF64), //
|
Types<f32A, AFloatA>(AFloat(kPiF32), kPiF64), //
|
||||||
Types<f32A, AFloatA>(AFloat(kSubnormalF32), kSubnormalF32), //
|
Types<f32A, AFloatA>(AFloat(kSubnormalF32), kSubnormalF32), //
|
||||||
Types<f32A, AFloatA>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
Types<f32A, AFloatA>(AFloat(-kSubnormalF32), -kSubnormalF32), //
|
||||||
/* Types<f16A, AFloatA>(0.0_a, 0.0), */ //
|
Types<f16A, AFloatA>(0.0_a, 0.0), //
|
||||||
/* Types<f16A, AFloatA>(1.0_a, 1.0), */ //
|
Types<f16A, AFloatA>(1.0_a, 1.0), //
|
||||||
/* Types<f16A, AFloatA>(-1.0_a, -1.0), */ //
|
Types<f16A, AFloatA>(-1.0_a, -1.0), //
|
||||||
/* Types<f16A, AFloatA>(AFloat(kHighestF16), kHighestF16), */ //
|
Types<f16A, AFloatA>(AFloat(f16::Highest()), static_cast<double>(f16::Highest())), //
|
||||||
/* Types<f16A, AFloatA>(AFloat(kLowestF16), kLowestF16), */ //
|
Types<f16A, AFloatA>(AFloat(f16::Lowest()), static_cast<double>(f16::Lowest())), //
|
||||||
/* Types<f16A, AFloatA>(AFloat(kPiF16), kPiF64), */ //
|
Types<f16A, AFloatA>(AFloat(kPiF16), kPiF64), //
|
||||||
/* Types<f16A, AFloatA>(AFloat(kSubnormalF16), kSubnormalF16), */ //
|
Types<f16A, AFloatA>(AFloat(kSubnormalF16), kSubnormalF16), //
|
||||||
/* Types<f16A, AFloatA>(AFloat(-kSubnormalF16), -kSubnormalF16), */ //
|
Types<f16A, AFloatA>(AFloat(-kSubnormalF16), -kSubnormalF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
INSTANTIATE_TEST_SUITE_P(
|
INSTANTIATE_TEST_SUITE_P(
|
||||||
|
@ -716,8 +717,8 @@ INSTANTIATE_TEST_SUITE_P(
|
||||||
Types<u32, AInt>(0_a, static_cast<double>(u32::kLowestValue) - 1), //
|
Types<u32, AInt>(0_a, static_cast<double>(u32::kLowestValue) - 1), //
|
||||||
Types<f32, AFloat>(0.0_a, kTooBigF32), //
|
Types<f32, AFloat>(0.0_a, kTooBigF32), //
|
||||||
Types<f32, AFloat>(0.0_a, -kTooBigF32), //
|
Types<f32, AFloat>(0.0_a, -kTooBigF32), //
|
||||||
/* Types<f16, AFloat>(0.0_a, kTooBigF16), */ //
|
Types<f16, AFloat>(0.0_a, kTooBigF16), //
|
||||||
/* Types<f16, AFloat>(0.0_a, -kTooBigF16), */ //
|
Types<f16, AFloat>(0.0_a, -kTooBigF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
INSTANTIATE_TEST_SUITE_P(
|
INSTANTIATE_TEST_SUITE_P(
|
||||||
|
@ -732,8 +733,8 @@ INSTANTIATE_TEST_SUITE_P(
|
||||||
Types<u32V, AIntV>(0_a, static_cast<double>(u32::kLowestValue) - 1), //
|
Types<u32V, AIntV>(0_a, static_cast<double>(u32::kLowestValue) - 1), //
|
||||||
Types<f32V, AFloatV>(0.0_a, kTooBigF32), //
|
Types<f32V, AFloatV>(0.0_a, kTooBigF32), //
|
||||||
Types<f32V, AFloatV>(0.0_a, -kTooBigF32), //
|
Types<f32V, AFloatV>(0.0_a, -kTooBigF32), //
|
||||||
/* Types<f16V, AFloatV>(0.0_a, kTooBigF16), */ //
|
Types<f16V, AFloatV>(0.0_a, kTooBigF16), //
|
||||||
/* Types<f16V, AFloatV>(0.0_a, -kTooBigF16), */ //
|
Types<f16V, AFloatV>(0.0_a, -kTooBigF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
INSTANTIATE_TEST_SUITE_P(MatrixValueCannotBeRepresented,
|
INSTANTIATE_TEST_SUITE_P(MatrixValueCannotBeRepresented,
|
||||||
|
@ -743,8 +744,8 @@ INSTANTIATE_TEST_SUITE_P(MatrixValueCannotBeRepresented,
|
||||||
testing::ValuesIn(std::vector<Data>{
|
testing::ValuesIn(std::vector<Data>{
|
||||||
Types<f32M, AFloatM>(0.0_a, kTooBigF32), //
|
Types<f32M, AFloatM>(0.0_a, kTooBigF32), //
|
||||||
Types<f32M, AFloatM>(0.0_a, -kTooBigF32), //
|
Types<f32M, AFloatM>(0.0_a, -kTooBigF32), //
|
||||||
/* Types<f16M, AFloatM>(0.0_a, kTooBigF16), */ //
|
Types<f16M, AFloatM>(0.0_a, kTooBigF16), //
|
||||||
/* Types<f16M, AFloatM>(0.0_a, -kTooBigF16), */ //
|
Types<f16M, AFloatM>(0.0_a, -kTooBigF16), //
|
||||||
})));
|
})));
|
||||||
|
|
||||||
} // namespace materialize_abstract_numeric_to_concrete_type
|
} // namespace materialize_abstract_numeric_to_concrete_type
|
||||||
|
|
Loading…
Reference in New Issue