aboutsummaryrefslogtreecommitdiff
path: root/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp
diff options
context:
space:
mode:
authorMichalis Spyrou <michalis.spyrou@arm.com>2021-06-07 14:23:57 +0100
committerGeorgios Pinitas <georgios.pinitas@arm.com>2021-06-23 12:25:50 +0000
commit20fca524baf99402f742ce38c538f2fd07d5fff9 (patch)
treeb63d98383d1ba22bb3ca59d393e4ab9d47a9c762 /src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp
parent1d359279e22874121def2ce4bfdb633d94ea5ade (diff)
downloadComputeLibrary-20fca524baf99402f742ce38c538f2fd07d5fff9.tar.gz
Create core library using high priority operators
A smaller core library is created using a subset of the operators. Changed the structure of filelist.json in order to include more information about the kernels and make the selection easier. Resolves: COMPMID-4514 Change-Id: I079ca7d8e64346174eebdd13b834e1dd4dc36ca2 Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5786 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp')
-rw-r--r--src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp31
1 files changed, 18 insertions, 13 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp b/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp
index aea750a475..ac43df979c 100644
--- a/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp
+++ b/src/core/NEON/kernels/arm_conv/depthwise/depthwise_fp32.cpp
@@ -33,7 +33,7 @@
#include "depthwise_implementation_constraints.hpp"
#if defined(__aarch64__)
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ARM_COMPUTE_ENABLE_SVE)
#include "kernels/sve_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst.hpp"
#include "kernels/sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst.hpp"
#include "kernels/sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst.hpp"
@@ -43,7 +43,7 @@
#include "kernels/sve_fp32_packed_to_nhwc_3x3_s2_with_multiplier_output3x3_mla_depthfirst.hpp"
#include "kernels/sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst.hpp"
#include "kernels/sve_fp32_packed_to_nhwc_generic_with_multiplier_output2x8_mla_depthfirst.hpp"
-#endif // defined(__ARM_FEATURE_SVE)
+#endif // defined(ARM_COMPUTE_ENABLE_SVE)
#include "kernels/a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst.hpp"
#include "kernels/a64_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst.hpp"
#include "kernels/a64_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst.hpp"
@@ -85,12 +85,13 @@ namespace
static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
#if defined(__aarch64__)
-#if defined(__ARM_FEATURE_SVE)
+#if defined(ARM_COMPUTE_ENABLE_SVE)
{
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst",
constraint(is_supported<sve_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst>,
- has_no_channel_multiplier),
+ has_no_channel_multiplier,
+ cpu_has_sve),
cycle_estimate<sve_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst>,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirst<sve_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst>(args);
@@ -100,7 +101,8 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst",
constraint(is_supported<sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst>,
- has_no_channel_multiplier),
+ has_no_channel_multiplier,
+ cpu_has_sve),
cycle_estimate<sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst>,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirst<sve_fp32_nhwc_3x3_s1_output3x3_mla_depthfirst>(args);
@@ -110,7 +112,8 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst",
constraint(is_supported<sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst>,
- has_no_channel_multiplier),
+ has_no_channel_multiplier,
+ cpu_has_sve),
cycle_estimate<sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst>,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirst<sve_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst>(args);
@@ -120,7 +123,8 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_3x3_s2_output2x2_mla_depthfirst",
constraint(is_supported<sve_fp32_nhwc_3x3_s2_output2x2_mla_depthfirst>,
- has_no_channel_multiplier),
+ has_no_channel_multiplier,
+ cpu_has_sve),
cycle_estimate<sve_fp32_nhwc_3x3_s2_output2x2_mla_depthfirst>,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirst<sve_fp32_nhwc_3x3_s2_output2x2_mla_depthfirst>(args);
@@ -130,7 +134,8 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_5x5_s1_output2x2_mla_depthfirst",
constraint(is_supported<sve_fp32_nhwc_5x5_s1_output2x2_mla_depthfirst>,
- has_no_channel_multiplier),
+ has_no_channel_multiplier,
+ cpu_has_sve),
cycle_estimate<sve_fp32_nhwc_5x5_s1_output2x2_mla_depthfirst>,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirst<sve_fp32_nhwc_5x5_s1_output2x2_mla_depthfirst>(args);
@@ -139,7 +144,7 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
{
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_generic_output3x3_mla_depthfirst",
- constraint(has_no_channel_multiplier),
+ constraint(has_no_channel_multiplier, cpu_has_sve),
not_preferred,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirstGeneric<sve_fp32_nhwc_generic_output9_mla_depthfirst, 3, 3>(args);
@@ -148,7 +153,7 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
{
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_3x3_s2_with_multiplier_output3x3_mla_depthfirst",
- constraint(is_supported<sve_fp32_packed_to_nhwc_3x3_s2_with_multiplier_output3x3_mla_depthfirst>),
+ constraint(is_supported<sve_fp32_packed_to_nhwc_3x3_s2_with_multiplier_output3x3_mla_depthfirst>, cpu_has_sve),
not_preferred_if_no_multiplier,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirstWithMultiplier<sve_fp32_packed_to_nhwc_3x3_s2_with_multiplier_output3x3_mla_depthfirst>(args);
@@ -157,7 +162,7 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
{
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst",
- constraint(is_supported<sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst>),
+ constraint(is_supported<sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst>, cpu_has_sve),
not_preferred_if_no_multiplier,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirstWithMultiplier<sve_fp32_packed_to_nhwc_5x5_s1_with_multiplier_output2x4_mla_depthfirst>(args);
@@ -166,13 +171,13 @@ static const DepthwiseImplementation<float, float> depthwise_fp32_methods[] = {
{
DepthwiseMethod::DEPTHFIRST,
"sve_fp32_nhwc_generic_with_multiplier_output2x8_mla_depthfirst",
- nullptr,
+ constraint(cpu_has_sve),
not_preferred_if_no_multiplier,
[] (const DepthwiseArgs &args, const Nothing &) -> DepthwiseCommon<float, float, float> * {
return new DepthwiseDepthfirstGenericWithMultiplier<sve_fp32_packed_to_nhwc_generic_with_multiplier_output2x8_mla_depthfirst>(args);
},
},
-#endif // defined(__ARM_FEATURE_SVE)
+#endif // defined(ARM_COMPUTE_ENABLE_SVE)
{
DepthwiseMethod::DEPTHFIRST,
"a64_fp32_nhwc_3x3_s1_output4x4_mla_depthfirst",