Add quantized CL MatMul kernel for LHS NT, RHS T
Implement a native kernel for batched Matrix Multiplication for the quantized data types QASYMM8 and QASYMM8_SIGNED and with the MatMul attributes `adj_x = false, adj_y = true`.
Resolves: COMPMID-5923
Change-Id: I477b2dd886edfe83beaba9efc7d6b05ed19f5da4
Signed-off-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9467
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/tests/validation/CL/MatMulLowpNativeKernel.cpp b/tests/validation/CL/MatMulLowpNativeKernel.cpp
index a0b2a37..fd7a4cb 100644
--- a/tests/validation/CL/MatMulLowpNativeKernel.cpp
+++ b/tests/validation/CL/MatMulLowpNativeKernel.cpp
@@ -212,7 +212,7 @@
TEST_SUITE(QASYMM8_SIGNED)
FIXTURE_DATA_TEST_CASE(RunTiny, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
framework::dataset::make("TransposeA", { true, false })),
- framework::dataset::make("TransposeB", { false })),
+ framework::dataset::make("TransposeB", { true, false })),
m0_values_precommit),
n0_values_precommit),
k0_values_precommit),
@@ -224,7 +224,7 @@
}
FIXTURE_DATA_TEST_CASE(RunSmall, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
framework::dataset::make("TransposeA", { true, false })),
- framework::dataset::make("TransposeB", { false })),
+ framework::dataset::make("TransposeB", { true, false })),
m0_values_precommit),
n0_values_precommit),
k0_values_precommit),
@@ -234,30 +234,6 @@
// Validate output
validate(CLAccessor(_target), _reference, tolerance_quant);
}
-FIXTURE_DATA_TEST_CASE(RunTiny_T_T, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
- framework::dataset::make("TransposeA", { true })),
- framework::dataset::make("TransposeB", { true })),
- m0_values_precommit),
- n0_values_precommit),
- k0_values_precommit),
- framework::dataset::make("ExportRhsToCLImage", { false })),
- framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
-{
- // Validate output
- validate(CLAccessor(_target), _reference, tolerance_quant);
-}
-FIXTURE_DATA_TEST_CASE(RunSmall_T_T, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
- framework::dataset::make("TransposeA", { true })),
- framework::dataset::make("TransposeB", { true })),
- m0_values_precommit),
- n0_values_precommit),
- k0_values_precommit),
- framework::dataset::make("ExportRhsToCLImage", { false })),
- framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
-{
- // Validate output
- validate(CLAccessor(_target), _reference, tolerance_quant);
-}
FIXTURE_DATA_TEST_CASE(RunLargeNoTranspose, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::NIGHTLY,
combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
framework::dataset::make("TransposeA", { false })),
@@ -271,6 +247,19 @@
// Validate output
validate(CLAccessor(_target), _reference, tolerance_quant);
}
+FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::NIGHTLY,
+ combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+ framework::dataset::make("TransposeA", { false })),
+ framework::dataset::make("TransposeB", { true })),
+ m0_values_nightly_lhs_nt),
+ n0_values_nightly_rhs_t),
+ k0_values_nightly_rhs_t),
+ framework::dataset::make("ExportRhsToCLImage", { false })),
+ framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
+{
+ // Validate output
+ validate(CLAccessor(_target), _reference, tolerance_quant);
+}
FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::NIGHTLY,
combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
framework::dataset::make("TransposeA", { true })),
@@ -302,20 +291,7 @@
FIXTURE_DATA_TEST_CASE(RunHighDimensional, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL,
combine(combine(combine(combine(combine(combine(combine(datasets::HighDimensionalMatMulDataset(),
framework::dataset::make("TransposeA", { true, false })),
- framework::dataset::make("TransposeB", { false })),
- framework::dataset::make("M0", { 2 })),
- framework::dataset::make("N0", { 2 })),
- framework::dataset::make("K0", { 2 })),
- framework::dataset::make("ExportRhsToCLImage", { false })),
- framework::dataset::make("DataType", DataType::QASYMM8_SIGNED)))
-{
- // Validate output
- validate(CLAccessor(_target), _reference, tolerance_quant);
-}
-FIXTURE_DATA_TEST_CASE(RunHighDimensional_T_T, CLMatMulLowpNativeKernelFixture<int8_t>, framework::DatasetMode::ALL,
- combine(combine(combine(combine(combine(combine(combine(datasets::HighDimensionalMatMulDataset(),
- framework::dataset::make("TransposeA", { true })),
- framework::dataset::make("TransposeB", { true })),
+ framework::dataset::make("TransposeB", { true, false })),
framework::dataset::make("M0", { 2 })),
framework::dataset::make("N0", { 2 })),
framework::dataset::make("K0", { 2 })),
@@ -330,7 +306,7 @@
TEST_SUITE(QASYMM8)
FIXTURE_DATA_TEST_CASE(RunTiny, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
framework::dataset::make("TransposeA", { true, false })),
- framework::dataset::make("TransposeB", { false })),
+ framework::dataset::make("TransposeB", { true, false })),
m0_values_precommit),
n0_values_precommit),
k0_values_precommit),
@@ -342,7 +318,7 @@
}
FIXTURE_DATA_TEST_CASE(RunSmall, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
framework::dataset::make("TransposeA", { true, false })),
- framework::dataset::make("TransposeB", { false })),
+ framework::dataset::make("TransposeB", { true, false })),
m0_values_precommit),
n0_values_precommit),
k0_values_precommit),
@@ -352,30 +328,6 @@
// Validate output
validate(CLAccessor(_target), _reference, tolerance_quant);
}
-FIXTURE_DATA_TEST_CASE(RunTiny_T_T, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::TinyMatMulDataset(),
- framework::dataset::make("TransposeA", { true })),
- framework::dataset::make("TransposeB", { true })),
- m0_values_precommit),
- n0_values_precommit),
- k0_values_precommit),
- framework::dataset::make("ExportRhsToCLImage", { false })),
- framework::dataset::make("DataType", DataType::QASYMM8)))
-{
- // Validate output
- validate(CLAccessor(_target), _reference, tolerance_quant);
-}
-FIXTURE_DATA_TEST_CASE(RunSmall_T_T, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::ALL, combine(combine(combine(combine(combine(combine(combine(datasets::SmallMatMulDataset(),
- framework::dataset::make("TransposeA", { true })),
- framework::dataset::make("TransposeB", { true })),
- m0_values_precommit),
- n0_values_precommit),
- k0_values_precommit),
- framework::dataset::make("ExportRhsToCLImage", { false })),
- framework::dataset::make("DataType", DataType::QASYMM8)))
-{
- // Validate output
- validate(CLAccessor(_target), _reference, tolerance_quant);
-}
FIXTURE_DATA_TEST_CASE(RunLargeNoTranspose, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::NIGHTLY,
combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
framework::dataset::make("TransposeA", { false })),
@@ -389,6 +341,19 @@
// Validate output
validate(CLAccessor(_target), _reference, tolerance_quant);
}
+FIXTURE_DATA_TEST_CASE(RunLargeRhsTransposed, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::NIGHTLY,
+ combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
+ framework::dataset::make("TransposeA", { false })),
+ framework::dataset::make("TransposeB", { true })),
+ m0_values_nightly_lhs_nt),
+ n0_values_nightly_rhs_t),
+ k0_values_nightly_rhs_t),
+ framework::dataset::make("ExportRhsToCLImage", { false })),
+ framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+ // Validate output
+ validate(CLAccessor(_target), _reference, tolerance_quant);
+}
FIXTURE_DATA_TEST_CASE(RunLargeLhsTransposed, CLMatMulLowpNativeKernelFixture<uint8_t>, framework::DatasetMode::NIGHTLY,
combine(combine(combine(combine(combine(combine(combine(datasets::LargeMatMulDataset(),
framework::dataset::make("TransposeA", { true })),