aboutsummaryrefslogtreecommitdiff
path: root/src/core
diff options
context:
space:
mode:
authorOmar Al Khatib <omar.alkhatib@arm.com>2024-04-23 16:26:56 +0100
committerOmar Al Khatib <omar.alkhatib@arm.com>2024-05-08 12:16:10 +0000
commitc1575b2c12a4cee3a60c711fe6521025a814b159 (patch)
treedd88487711f631a83989dec5c7e058d29cc076ec /src/core
parent2fea13593a4753316ae488edf489cb4b00150153 (diff)
downloadComputeLibrary-c1575b2c12a4cee3a60c711fe6521025a814b159.tar.gz
Add SME2 implementation of Softmax for QASYMM8 and QASYMM8_SIGNED.
Resolves: [COMPMID-6917] Change-Id: Id8b96efd29f6c61dd43a371341c6e1fe087953e9 Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11509 Tested-by: Arm Jenkins <bsgcomp@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com> Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core')
-rw-r--r--src/core/CPP/CPPTypes.cpp13
-rw-r--r--src/core/common/Registrars.h22
2 files changed, 32 insertions, 3 deletions
diff --git a/src/core/CPP/CPPTypes.cpp b/src/core/CPP/CPPTypes.cpp
index 9980db42f3..f6761f27b0 100644
--- a/src/core/CPP/CPPTypes.cpp
+++ b/src/core/CPP/CPPTypes.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018-2022 Arm Limited.
+ * Copyright (c) 2018-2022, 2024 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -28,6 +28,7 @@
#include "src/common/cpuinfo/CpuInfo.h"
#include "src/common/cpuinfo/CpuIsaInfo.h"
+#include "src/core/NEON/kernels/arm_gemm/utils.hpp"
namespace arm_compute
{
@@ -135,4 +136,14 @@ unsigned int CPUInfo::get_L2_cache_size() const
{
return _impl->L2_cache_size;
}
+
+unsigned long CPUInfo::get_sme2_vector_length() const
+{
+#ifdef ARM_COMPUTE_ENABLE_SME2
+ return arm_gemm::utils::sme::get_vector_length<int8_t>();
+#else // ARM_COMPUTE_ENABLE_SME2
+ return 0;
+#endif // ARM_COMPUTE_ENABLE_SME2
+}
+
} // namespace arm_compute
diff --git a/src/core/common/Registrars.h b/src/core/common/Registrars.h
index a74316b486..cd849c3666 100644
--- a/src/core/common/Registrars.h
+++ b/src/core/common/Registrars.h
@@ -72,9 +72,13 @@
#endif /* defined(ARM_COMPUTE_ENABLE_SVE2) */
#if defined(ARM_COMPUTE_ENABLE_SME2)
-#define REGISTER_FP32_SME2(func_name) &(func_name)
+#define REGISTER_FP32_SME2(func_name) &(func_name)
+#define REGISTER_QASYMM8_SME2(func_name) &(func_name)
+#define REGISTER_QASYMM8_SIGNED_SME2(func_name) &(func_name)
#else /* !defined(ARM_COMPUTE_ENABLE_SME2) */
-#define REGISTER_FP32_SME2(func_name) nullptr
+#define REGISTER_FP32_SME2(func_name) nullptr
+#define REGISTER_QASYMM8_SME2(func_name) nullptr
+#define REGISTER_QASYMM8_SIGNED_SME2(func_name) nullptr
#endif /* defined(ARM_COMPUTE_ENABLE_SME2) */
#if defined(ARM_COMPUTE_ENABLE_NEON)
@@ -106,10 +110,17 @@
#define REGISTER_QASYMM8_SIGNED_SVE2(func_name) nullptr
#endif /* defined(ARM_COMPUTE_ENABLE_SVE2) */
+#if defined(ARM_COMPUTE_ENABLE_SME2)
+#define REGISTER_QASYMM8_SIGNED_SME2(func_name) &(func_name)
+#else /* !defined(ARM_COMPUTE_ENABLE_SME2) */
+#define REGISTER_QASYMM8_SIGNED_SME2(func_name) nullptr
+#endif /* defined(ARM_COMPUTE_ENABLE_SME2) */
+
#else /* defined(ENABLE_QASYMM8_SIGNED_KERNELS) */
#define REGISTER_QASYMM8_SIGNED_NEON(func_name) nullptr
#define REGISTER_QASYMM8_SIGNED_SVE(func_name) nullptr
#define REGISTER_QASYMM8_SIGNED_SVE2(func_name) nullptr
+#define REGISTER_QASYMM8_SIGNED_SME2(func_name) nullptr
#endif /* defined(ENABLE_QASYMM8_SIGNED_KERNELS) */
#if defined(ENABLE_QASYMM8_KERNELS)
@@ -127,10 +138,17 @@
#define REGISTER_QASYMM8_SVE2(func_name) nullptr
#endif /* defined(ARM_COMPUTE_ENABLE_SVE2) */
+#if defined(ARM_COMPUTE_ENABLE_SME2)
+#define REGISTER_QASYMM8_SME2(func_name) &(func_name)
+#else /* !defined(ARM_COMPUTE_ENABLE_SME2) */
+#define REGISTER_QASYMM8_SME2(func_name) nullptr
+#endif /* defined(ARM_COMPUTE_ENABLE_SME2) */
+
#else /* defined(ENABLE_QASYMM8_KERNELS) */
#define REGISTER_QASYMM8_NEON(func_name) nullptr
#define REGISTER_QASYMM8_SVE(func_name) nullptr
#define REGISTER_QASYMM8_SVE2(func_name) nullptr
+#define REGISTER_QASYMM8_SME2(func_name) nullptr
#endif /* defined(ENABLE_QASYMM8_KERNELS) */
#if defined(ENABLE_QSYMM16_KERNELS)