1 files changed, 23 insertions, 21 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/interleave_indirect.cpp b/src/core/NEON/kernels/arm_gemm/interleave_indirect.cpp
index bb86d9e41d..59591935cd 100644
--- a/src/core/NEON/kernels/arm_gemm/interleave_indirect.cpp
+++ b/src/core/NEON/kernels/arm_gemm/interleave_indirect.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,7 +28,9 @@
 #include "interleave_indirect.hpp"
 #include "bfloat.hpp"
 
+#if !defined(_WIN64) && !defined(__OpenBSD__)
 #include <alloca.h>
+#endif /* !defined(_WIN64) && !defined(__OpenBSD__) */
 
 #include <algorithm>
 #include <cstddef>
@@ -292,21 +294,21 @@ void Interleave(TOut *out, const TIn *in, size_t in_stride, const unsigned int y
 /* AArch32 */
 #ifdef __arm__
 /* FP32 */
-/* Neon implementation (height 6) */
+/* Arm® Neon™ implementation (height 6) */
 template void IndirectInterleave<6, 1, VLType::None>(float *, const float * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<6, 1, VLType::None>(float *, const float *, size_t, const convolver<float> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<6, 1, VLType::None>(float *, const float *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
 /* FP16 */
 #if __ARM_FP16_ARGS
-/* Neon implementation using FP32 kernel (height 6) */
+/* Arm® Neon™ implementation using FP32 kernel (height 6) */
 template void IndirectInterleave<6, 1, VLType::None>(float *, const __fp16 * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<6, 1, VLType::None>(float *, const __fp16 *, size_t, const convolver<__fp16> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<6, 1, VLType::None>(float *, const __fp16 *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 #endif /* __ARM_FP16_ARGS */
 
 /* BF16 */
-/* Neon implementation using FP32 kernel */
+/* Arm® Neon™ implementation using FP32 kernel */
 template void IndirectInterleave<6, 1, VLType::None>(float *, const bfloat16 * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<6, 1, VLType::None>(float *, const bfloat16 *, size_t, const convolver<bfloat16> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<6, 1, VLType::None>(float *, const bfloat16 *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
@@ -315,17 +317,17 @@ template void Interleave<6, 1, VLType::None>(float *, const bfloat16 *, size_t,
 /* AArch64 */
 #ifdef __aarch64__
 /* FP32 */
-/* Neon/SVE implementation (height 8) */
+/* Arm® Neon™/SVE implementation (height 8) */
 template void IndirectInterleave<8, 1, VLType::None>(float *, const float * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<8, 1, VLType::None>(float *, const float *, size_t, const convolver<float> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 1, VLType::None>(float *, const float *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
-#if defined(__ARM_FEATURE_SVE) && defined(MMLA_FP32)
+#if defined(ARM_COMPUTE_ENABLE_SVE) && defined(ARM_COMPUTE_ENABLE_SVEF32MM)
 /* FMMLA */
 template void IndirectInterleave<8, 2, VLType::None>(float *, const float * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<8, 2, VLType::None>(float *, const float *, size_t, const convolver<float> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 2, VLType::None>(float *, const float *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
-#endif // SVE && MMLA_FP32
+#endif // ARM_COMPUTE_ENABLE_SVE && ARM_COMPUTE_ENABLE_SVEF32MM
 
 /* FP16 */
 #if defined(FP16_KERNELS) || defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
@@ -339,8 +341,8 @@ template void ConvolutionInterleave<8, 1, VLType::None>(float *, const __fp16 *,
 template void Interleave<8, 1, VLType::None>(float *, const __fp16 *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
 /* BF16 */
-/* Neon/SVE BFDOT */
-#ifdef V8P6_BF
+/* Arm® Neon™/SVE BFDOT */
+#ifdef ARM_COMPUTE_ENABLE_BF16
 template void IndirectInterleave<8, 2, VLType::None>(bfloat16 *, const bfloat16 * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<8, 2, VLType::None>(bfloat16 *, const bfloat16 *, size_t, const convolver<bfloat16> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 2, VLType::None>(bfloat16 *, const bfloat16 *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
@@ -348,9 +350,13 @@ template void Interleave<8, 2, VLType::None>(bfloat16 *, const bfloat16 *, size_
 template void IndirectInterleave<8, 4, VLType::None>(bfloat16 *, const bfloat16 * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<8, 4, VLType::None>(bfloat16 *, const bfloat16 *, size_t, const convolver<bfloat16> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 4, VLType::None>(bfloat16 *, const bfloat16 *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
-#endif // V8P6_BF
 
-/* Neon/SVE using FP32 kernel */
+template void IndirectInterleave<8, 4, VLType::None>(bfloat16 *, const float * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
+template void ConvolutionInterleave<8, 4, VLType::None>(bfloat16 *, const float *, size_t, const convolver<float> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
+template void Interleave<8, 4, VLType::None>(bfloat16 *, const float *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
+#endif // ARM_COMPUTE_ENABLE_BF16
+
+/* Arm® Neon™/SVE using FP32 kernel */
 template void IndirectInterleave<8, 1, VLType::None>(float *, const bfloat16 * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<8, 1, VLType::None>(float *, const bfloat16 *, size_t, const convolver<bfloat16> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 1, VLType::None>(float *, const bfloat16 *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
@@ -365,46 +371,42 @@ template void ConvolutionInterleave<8, 1, VLType::None>(uint16_t *, const uint16
 template void Interleave<8, 1, VLType::None>(uint16_t *, const uint16_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
 /* INT8 */
-/* Neon SMLA/SMLAL (height 4, block 16) */
+/* Arm® Neon™ SMLA/SMLAL (height 4, block 16) */
 template void IndirectInterleave<4, 16, VLType::None>(int8_t *, const int8_t * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<4, 16, VLType::None>(int8_t *, const int8_t *, size_t, const convolver<int8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<4, 16, VLType::None>(int8_t *, const int8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
-/* Neon SDOT (height 8, block 4) */
+/* Arm® Neon™ SDOT (height 8, block 4) */
 template void IndirectInterleave<8, 4, VLType::None>(int8_t *, const int8_t * const * const *, unsigned int, unsigned int, unsigned int y0, unsigned int ymax, unsigned int k0, unsigned int kmax, bool, int32_t);
 template void ConvolutionInterleave<8, 4, VLType::None>(int8_t *, const int8_t *, size_t, const convolver<int8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 4, VLType::None>(int8_t *, const int8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
-#ifdef MMLA_INT8
 /* MMLA SMMLA (height 8, block 8) */
 template void IndirectInterleave<8, 8, VLType::None>(int8_t *, const int8_t * const * const *, unsigned int, unsigned int, unsigned int y0, unsigned int ymax, unsigned int k0, unsigned int kmax, bool, int32_t);
 template void ConvolutionInterleave<8, 8, VLType::None>(int8_t *, const int8_t *, size_t, const convolver<int8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 8, VLType::None>(int8_t *, const int8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
-#endif // MMLA_INT8
 
-/* Neon SDOT (height 8, block 1) */
+/* Arm® Neon™ SDOT (height 8, block 1) */
 template void IndirectInterleave<8, 1, VLType::None>(int16_t *, const int8_t * const * const *, unsigned int, unsigned int, unsigned int y0, unsigned int ymax, unsigned int k0, unsigned int kmax, bool, int32_t);
 template void ConvolutionInterleave<8, 1, VLType::None>(int16_t *, const int8_t *, size_t, const convolver<int8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 1, VLType::None>(int16_t *, const int8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
-/* Neon SMLA/SMLAL (height 4, block 16) */
+/* Arm® Neon™ SMLA/SMLAL (height 4, block 16) */
 template void IndirectInterleave<4, 16, VLType::None>(uint8_t *, const uint8_t * const * const *, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void ConvolutionInterleave<4, 16, VLType::None>(uint8_t *, const uint8_t *, size_t, const convolver<uint8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<4, 16, VLType::None>(uint8_t *, const uint8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
-/* Neon SDOT (height 8, block 4) */
+/* Arm® Neon™ SDOT (height 8, block 4) */
 template void IndirectInterleave<8, 4, VLType::None>(uint8_t *, const uint8_t * const * const *, unsigned int, unsigned int, unsigned int y0, unsigned int ymax, unsigned int k0, unsigned int kmax, bool, int32_t);
 template void ConvolutionInterleave<8, 4, VLType::None>(uint8_t *, const uint8_t *, size_t, const convolver<uint8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 4, VLType::None>(uint8_t *, const uint8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 
-#ifdef MMLA_INT8
 /* MMLA SMMLA (height 8, block 8) */
 template void IndirectInterleave<8, 8, VLType::None>(uint8_t *, const uint8_t * const * const *, unsigned int, unsigned int, unsigned int y0, unsigned int ymax, unsigned int k0, unsigned int kmax, bool, int32_t);
 template void ConvolutionInterleave<8, 8, VLType::None>(uint8_t *, const uint8_t *, size_t, const convolver<uint8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 8, VLType::None>(uint8_t *, const uint8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
-#endif // MMLA_INT8
 
-/* Neon 16-bit (height 8, block 1) */
+/* Arm® Neon™ 16-bit (height 8, block 1) */
 template void IndirectInterleave<8, 1, VLType::None>(uint16_t *, const uint8_t * const * const *, unsigned int, unsigned int, unsigned int y0, unsigned int ymax, unsigned int k0, unsigned int kmax, bool, int32_t);
 template void ConvolutionInterleave<8, 1, VLType::None>(uint16_t *, const uint8_t *, size_t, const convolver<uint8_t> &, unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);
 template void Interleave<8, 1, VLType::None>(uint16_t *, const uint8_t *, size_t, unsigned int, unsigned int, unsigned int, unsigned int, bool, int32_t);