From bdcdc39d89b6a6556f5c0483af5379f75eae0c55 Mon Sep 17 00:00:00 2001
From: Georgios Pinitas <georgios.pinitas@arm.com>
Date: Thu, 22 Apr 2021 16:42:03 +0100
Subject: Enable fat binary support

Changes our build system to allow building both Neon(TM) and SVE
kernels and package them in the same binary. This will allow
runtime selection of the underlying architecture.

Adds new build option, fat_binary, for enabling this feature.

Change-Id: I8e8386149773ce28e071a2fb7ddd8c8ae0f28a4a
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5704
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp  | 7 ++++---
 src/core/NEON/kernels/batchnormalization/impl/SVE/fp16.cpp | 4 ++--
 src/core/NEON/kernels/batchnormalization/impl/SVE/fp32.cpp | 4 ++--
 3 files changed, 8 insertions(+), 7 deletions(-)

(limited to 'src/core/NEON/kernels')

diff --git a/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp b/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp
index 1691943b07..92000bb2f6 100644
--- a/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEBatchNormalizationLayerKernel.cpp
@@ -63,7 +63,7 @@ struct BatchNormalizationKernel
 
 static const BatchNormalizationKernel available_kernels[] =
 {
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ENABLE_SVE)
     {
         "fp16_sve_batch_normalization",
         [](const BatchNormalizationSelectorData & data) { return data.dt == DataType::F16; },
@@ -74,7 +74,8 @@ static const BatchNormalizationKernel available_kernels[] =
         [](const BatchNormalizationSelectorData & data) { return data.dt == DataType::F32; },
         REGISTER_FP32_SVE(arm_compute::cpu::fp32_sve_batch_normalization)
     },
-#else /* !defined(__ARM_FEATURE_SVE) */
+#endif /* !defined(ENABLE_SVE) */
+#if defined(ENABLE_NEON)
 #if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
     {
         "fp16_neon_batch_normalization",
@@ -87,7 +88,7 @@ static const BatchNormalizationKernel available_kernels[] =
         [](const BatchNormalizationSelectorData & data) { return data.dt == DataType::F32; },
         REGISTER_FP32_NEON(arm_compute::cpu::fp32_neon_batch_normalization)
     },
-#endif /* !defined(__ARM_FEATURE_SVE) */
+#endif /* !defined(ENABLE_NEON) */
 };
 
 const BatchNormalizationKernel *get_implementation(const BatchNormalizationSelectorData &data)
diff --git a/src/core/NEON/kernels/batchnormalization/impl/SVE/fp16.cpp b/src/core/NEON/kernels/batchnormalization/impl/SVE/fp16.cpp
index 3e3e81d044..a715b9d3ee 100644
--- a/src/core/NEON/kernels/batchnormalization/impl/SVE/fp16.cpp
+++ b/src/core/NEON/kernels/batchnormalization/impl/SVE/fp16.cpp
@@ -29,7 +29,7 @@
 #include <cmath>
 #include <cstddef>
 
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ENABLE_SVE)
 #include <arm_sve.h>
 
 namespace arm_compute
@@ -114,4 +114,4 @@ void fp16_sve_batch_normalization(ITensor *src, ITensor *dst, const ITensor *mea
 }
 } // namespace cpu
 } // namespace arm_compute
-#endif // __ARM_FEATURE_SVE
+#endif // ENABLE_SVE
diff --git a/src/core/NEON/kernels/batchnormalization/impl/SVE/fp32.cpp b/src/core/NEON/kernels/batchnormalization/impl/SVE/fp32.cpp
index b0d4cbb684..7cc570d8aa 100644
--- a/src/core/NEON/kernels/batchnormalization/impl/SVE/fp32.cpp
+++ b/src/core/NEON/kernels/batchnormalization/impl/SVE/fp32.cpp
@@ -29,7 +29,7 @@
 #include <cmath>
 #include <cstddef>
 
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ENABLE_SVE)
 #include <arm_sve.h>
 
 namespace arm_compute
@@ -114,4 +114,4 @@ void fp32_sve_batch_normalization(ITensor *src, ITensor *dst, const ITensor *mea
 }
 } // namespace cpu
 } // namespace arm_compute
-#endif // __ARM_FEATURE_SVE
+#endif // ENABLE_SVE
-- 
cgit v1.2.1