aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp')
-rw-r--r--src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp126
1 files changed, 95 insertions, 31 deletions
diff --git a/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp b/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
index 094c6aa301..a7f3dd3a93 100644
--- a/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
+++ b/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2021 Arm Limited.
+ * Copyright (c) 2021-2022 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -33,12 +33,18 @@
#include "kernels/cpp_nhwc_1x1_stride_any_depthfirst.hpp"
#if defined(__aarch64__)
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+#include "kernels/sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
+#include "kernels/sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp"
+#include "kernels/sme_fp16_nhwc_avg_generic_depthfirst.hpp"
+#include "kernels/sme_fp16_nhwc_max_generic_depthfirst.hpp"
+#endif // defined(ARM_COMPUTE_ENABLE_SME)
+#if defined(ARM_COMPUTE_ENABLE_SVE)
#include "kernels/sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
#include "kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp"
#include "kernels/sve_fp16_nhwc_avg_generic_depthfirst.hpp"
#include "kernels/sve_fp16_nhwc_max_generic_depthfirst.hpp"
-#endif // defined(__ARM_FEATURE_SVE)
+#endif // defined(ARM_COMPUTE_ENABLE_SVE)
#include "kernels/a64_fp16_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
#include "kernels/a64_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp"
#include "kernels/a64_fp16_nhwc_avg_generic_depthfirst.hpp"
@@ -48,19 +54,6 @@
namespace arm_conv {
namespace pooling {
-namespace
-{
- template <class Strategy>
- bool is_supported(const PoolingArgs &args, const Nothing &)
- {
- return ((args.pool_type == Strategy::pooling_type()) &&
- (args.pool_window.rows == Strategy::pool_rows()) &&
- (args.pool_window.cols == Strategy::pool_cols()) &&
- (args.pool_stride.rows == Strategy::stride_rows()) &&
- (args.pool_stride.cols == Strategy::stride_cols()));
- }
-}
-
static const PoolingImplementation<__fp16, __fp16> pooling_fp16_methods[] = {
{
PoolingMethod::DEPTHFIRST,
@@ -70,48 +63,115 @@ static const PoolingImplementation<__fp16, __fp16> pooling_fp16_methods[] = {
},
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirstGeneric<cpp_nhwc_1x1_stride_any_depthfirst<__fp16>>(args);
+ auto strat = new cpp_nhwc_1x1_stride_any_depthfirst<__fp16>(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
},
},
#if defined(__aarch64__)
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+ {
+ PoolingMethod::DEPTHFIRST,
+ "sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst",
+ [] (const PoolingArgs &args, const Nothing &os) -> bool {
+ return args.cpu_info->has_sme() &&
+ is_supported<sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>(args, os);
+ },
+ nullptr,
+ [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+ auto strat = new sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst(args.cpu_info);
+ return new PoolingDepthfirst<__fp16>(strat, args);
+ },
+ },
+ {
+ PoolingMethod::DEPTHFIRST,
+ "sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst",
+ [] (const PoolingArgs &args, const Nothing &os) -> bool {
+ return args.cpu_info->has_sme() &&
+ is_supported<sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>(args, os);
+ },
+ nullptr,
+ [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+ auto strat = new sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst(args.cpu_info);
+ return new PoolingDepthfirst<__fp16>(strat, args);
+ },
+ },
+ {
+ PoolingMethod::DEPTHFIRST,
+ "sme_fp16_nhwc_avg_generic_depthfirst",
+ [] (const PoolingArgs &args, const Nothing &) -> bool {
+ return args.cpu_info->has_sme() && args.pool_type == PoolingType::AVERAGE;
+ },
+ nullptr,
+ [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+ auto strat = new sme_fp16_nhwc_avg_generic_depthfirst(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
+ },
+ },
+ {
+ PoolingMethod::DEPTHFIRST,
+ "sme_fp16_nhwc_max_generic_depthfirst",
+ [] (const PoolingArgs &args, const Nothing &) -> bool {
+ return args.cpu_info->has_sme() && args.pool_type == PoolingType::MAX;
+ },
+ nullptr,
+ [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+ auto strat = new sme_fp16_nhwc_max_generic_depthfirst(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
+ },
+ },
+#endif // defined(ARM_COMPUTE_ENABLE_SME)
+#if defined(ARM_COMPUTE_ENABLE_SVE)
{
PoolingMethod::DEPTHFIRST,
"sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst",
- is_supported<sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>,
+ [] (const PoolingArgs &args, const Nothing &os) -> bool {
+ return args.cpu_info->has_sve() &&
+ is_supported<sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>(args, os);
+ },
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirst<sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>(args);
+ auto strat = new sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst(args.cpu_info);
+ return new PoolingDepthfirst<__fp16>(strat, args);
},
},
{
PoolingMethod::DEPTHFIRST,
"sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst",
- is_supported<sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>,
+ [] (const PoolingArgs &args, const Nothing &os) -> bool {
+ return args.cpu_info->has_sve() &&
+ is_supported<sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>(args, os);
+ },
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirst<sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>(args);
+ auto strat = new sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst(args.cpu_info);
+ return new PoolingDepthfirst<__fp16>(strat, args);
},
},
{
PoolingMethod::DEPTHFIRST,
"sve_fp16_nhwc_avg_generic_depthfirst",
- [] (const PoolingArgs &args, const Nothing &) -> bool { return args.pool_type == PoolingType::AVERAGE; },
+ [] (const PoolingArgs &args, const Nothing &) -> bool {
+ return args.cpu_info->has_sve() && args.pool_type == PoolingType::AVERAGE;
+ },
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirstGeneric<sve_fp16_nhwc_avg_generic_depthfirst>(args);
+ auto strat = new sve_fp16_nhwc_avg_generic_depthfirst(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
},
},
{
PoolingMethod::DEPTHFIRST,
"sve_fp16_nhwc_max_generic_depthfirst",
- [] (const PoolingArgs &args, const Nothing &) -> bool { return args.pool_type == PoolingType::MAX; },
+ [] (const PoolingArgs &args, const Nothing &) -> bool {
+ return args.cpu_info->has_sve() && args.pool_type == PoolingType::MAX;
+ },
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirstGeneric<sve_fp16_nhwc_max_generic_depthfirst>(args);
+ auto strat = new sve_fp16_nhwc_max_generic_depthfirst(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
},
},
-#endif // defined(__ARM_FEATURE_SVE)
+#endif // defined(ARM_COMPUTE_ENABLE_SVE)
#if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
{
PoolingMethod::DEPTHFIRST,
@@ -119,7 +179,8 @@ static const PoolingImplementation<__fp16, __fp16> pooling_fp16_methods[] = {
is_supported<a64_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>,
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirst<a64_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>(args);
+ auto strat = new a64_fp16_nhwc_max_2x2_s1_output2x2_depthfirst(args.cpu_info);
+ return new PoolingDepthfirst<__fp16>(strat, args);
},
},
{
@@ -128,7 +189,8 @@ static const PoolingImplementation<__fp16, __fp16> pooling_fp16_methods[] = {
is_supported<a64_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>,
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirst<a64_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>(args);
+ auto strat = new a64_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst(args.cpu_info);
+ return new PoolingDepthfirst<__fp16>(strat, args);
},
},
{
@@ -137,7 +199,8 @@ static const PoolingImplementation<__fp16, __fp16> pooling_fp16_methods[] = {
[] (const PoolingArgs &args, const Nothing &) -> bool { return args.pool_type == PoolingType::AVERAGE; },
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirstGeneric<a64_fp16_nhwc_avg_generic_depthfirst>(args);
+ auto strat = new a64_fp16_nhwc_avg_generic_depthfirst(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
},
},
{
@@ -146,7 +209,8 @@ static const PoolingImplementation<__fp16, __fp16> pooling_fp16_methods[] = {
[] (const PoolingArgs &args, const Nothing &) -> bool { return args.pool_type == PoolingType::MAX; },
nullptr,
[] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
- return new PoolingDepthfirstGeneric<a64_fp16_nhwc_max_generic_depthfirst>(args);
+ auto strat = new a64_fp16_nhwc_max_generic_depthfirst(args.cpu_info);
+ return new PoolingDepthfirstGeneric<__fp16>(strat, args);
},
},
#endif // defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-03 18:21:08 +0000