aboutsummaryrefslogtreecommitdiff
path: root/src/gpu/cl/kernels/gemm/reshaped
diff options
context:
space:
mode:
authorFelix Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com>2023-09-27 17:46:17 +0100
committerfelixjohnny.thomasmathibalan <felixjohnny.thomasmathibalan@arm.com>2023-09-28 12:08:05 +0000
commitafd38f0c617d6f89b2b4532c6c44f116617e2b6f (patch)
tree03bc7d5a762099989b16a656fa8d397b490ed70e /src/gpu/cl/kernels/gemm/reshaped
parentbdcb4c148ee2fdeaaddf4cf1e57bbb0de02bb894 (diff)
downloadComputeLibrary-afd38f0c617d6f89b2b4532c6c44f116617e2b6f.tar.gz
Apply clang-format on repository
Code is formatted as per a revised clang format configuration file(not part of this delivery). Version 14.0.6 is used. Exclusion List: - files with .cl extension - files that are not strictly C/C++ (e.g. Android.bp, Sconscript ...) And the following directories - compute_kernel_writer/validation/ - tests/ - include/ - src/core/NEON/kernels/convolution/ - src/core/NEON/kernels/arm_gemm/ - src/core/NEON/kernels/arm_conv/ - data/ There will be a follow up for formatting of .cl files and the files under tests/ and compute_kernel_writer/validation/. Signed-off-by: Felix Thomasmathibalan <felixjohnny.thomasmathibalan@arm.com> Change-Id: Ib7eb1fcf4e7537b9feaefcfc15098a804a3fde0a Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10391 Benchmark: Arm Jenkins <bsgcomp@arm.com> Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Diffstat (limited to 'src/gpu/cl/kernels/gemm/reshaped')
-rw-r--r--src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.cpp163
-rw-r--r--src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.h27
-rw-r--r--src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.cpp168
-rw-r--r--src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.h18
-rw-r--r--src/gpu/cl/kernels/gemm/reshaped/ClGemmReshapedKernelConfig.h2
5 files changed, 206 insertions, 172 deletions
diff --git a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.cpp b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.cpp
index 657018eb53..c956c347ef 100644
--- a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.cpp
+++ b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.cpp
@@ -29,6 +29,7 @@
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/TensorShape.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
#include "src/gpu/cl/kernels/gemm/ClGemmHelpers.h"
#include <utility>
@@ -43,30 +44,31 @@ namespace gemm
{
using namespace arm_compute::misc::shape_calculator;
-ClGemmDefaultConfigReshapedBifrost::ClGemmDefaultConfigReshapedBifrost(GPUTarget gpu)
- : IClGemmKernelConfig(gpu)
+ClGemmDefaultConfigReshapedBifrost::ClGemmDefaultConfigReshapedBifrost(GPUTarget gpu) : IClGemmKernelConfig(gpu)
{
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure(
+ unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (ClGemmDefaultConfigReshapedBifrost::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (
+ ClGemmDefaultConfigReshapedBifrost::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G7x(&ClGemmDefaultConfigReshapedBifrost::configure_G7x_f32,
- &ClGemmDefaultConfigReshapedBifrost::configure_G7x_f16,
- &ClGemmDefaultConfigReshapedBifrost::configure_G7x_u8);
+ CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G7x(
+ &ClGemmDefaultConfigReshapedBifrost::configure_G7x_f32, &ClGemmDefaultConfigReshapedBifrost::configure_G7x_f16,
+ &ClGemmDefaultConfigReshapedBifrost::configure_G7x_u8);
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G52(&ClGemmDefaultConfigReshapedBifrost::configure_G52_f32,
- &ClGemmDefaultConfigReshapedBifrost::configure_G52_f16,
- &ClGemmDefaultConfigReshapedBifrost::configure_G7x_u8);
+ CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G52(
+ &ClGemmDefaultConfigReshapedBifrost::configure_G52_f32, &ClGemmDefaultConfigReshapedBifrost::configure_G52_f16,
+ &ClGemmDefaultConfigReshapedBifrost::configure_G7x_u8);
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G76(&ClGemmDefaultConfigReshapedBifrost::configure_G76_f32,
- &ClGemmDefaultConfigReshapedBifrost::configure_G76_f16,
- &ClGemmDefaultConfigReshapedBifrost::configure_G76_u8);
+ CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G76(
+ &ClGemmDefaultConfigReshapedBifrost::configure_G76_f32, &ClGemmDefaultConfigReshapedBifrost::configure_G76_f16,
+ &ClGemmDefaultConfigReshapedBifrost::configure_G76_u8);
ConfigurationFunctionExecutorPtr func = nullptr;
- switch(_target)
+ switch (_target)
{
case GPUTarget::G76:
func = configs_G76.get_function(data_type);
@@ -83,12 +85,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
return (this->*func)(m, n, k, b);
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 8, 16, 16, true, false, false, true);
}
@@ -98,12 +101,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 8, 8, 2, true, true, true, false);
}
@@ -113,14 +117,15 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
- if(dot8_supported(CLKernelLibrary::get().get_device()))
+ if (dot8_supported(CLKernelLibrary::get().get_device()))
{
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 16, 2, 2, true, false, false, true);
}
@@ -131,7 +136,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
else
{
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 8, 2, 2, true, false, false, true);
}
@@ -142,7 +147,8 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
const float r_mn = static_cast<float>(m) / static_cast<float>(n);
const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
@@ -154,100 +160,108 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
GEMMLHSMatrixInfo lhs_info_img;
GEMMRHSMatrixInfo rhs_info_img;
- if(workload <= 274.4000f)
+ if (workload <= 274.4000f)
{
- if(r_nk <= 0.7461f)
+ if (r_nk <= 0.7461f)
{
- if(r_mn <= 21.1667f)
+ if (r_mn <= 21.1667f)
{
return configure_lhs_rhs_info(m, n, 4, 2, 4, 4, 4, false, true, true, false, false);
}
else
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
}
else
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
}
else
{
- if(r_mk <= 17.3926f)
+ if (r_mk <= 17.3926f)
{
- if(workload <= 542.4000f)
+ if (workload <= 542.4000f)
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
else
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
}
else
{
- if(r_nk <= 0.5463f)
+ if (r_nk <= 0.5463f)
{
- if(workload <= 11767.6001f)
+ if (workload <= 11767.6001f)
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
else
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
}
else
{
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
+ std::tie(lhs_info_buf, rhs_info_buf) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F32);
}
}
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- if(workload <= 323.4000f)
+ if (workload <= 323.4000f)
{
return configure_lhs_rhs_info(m, n, 2, 2, 8, 4, 8, false, false, false, true, false);
}
@@ -257,7 +271,8 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
@@ -268,7 +283,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
GEMMRHSMatrixInfo rhs_info_img;
// Get lhs_info/rhs_info in case of OpenCL buffer
- if(n <= 4)
+ if (n <= 4)
{
std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 2, 8, 16, 16, true, false, false, true);
}
@@ -279,15 +294,17 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
// Get lhs_info/rhs_info in case of OpenCL image
// Condition on the GPU workload
- if((m / 4) * (n / 4) >= 2560)
+ if ((m / 4) * (n / 4) >= 2560)
{
// Big workload
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 8, true, true, true, false, true);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 8, true, true, true, false, true);
}
else
{
// Small workload
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 1, true, true, true, false, true);
+ std::tie(lhs_info_img, rhs_info_img) =
+ configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 1, true, true, true, false, true);
}
const TensorInfo tensor_rhs_info(TensorShape(n, k, b), 1, DataType::F32);
@@ -297,7 +314,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
// In case of vector by matrix with few work-items, we use the OpenCL buffer rather than the OpenCL image2d
const bool use_cl_image2d = (n <= 4) ? false : true;
- if(bool(validate_image2d_support_on_rhs(tensor_reshaped_info, rhs_info_img)) && use_cl_image2d)
+ if (bool(validate_image2d_support_on_rhs(tensor_reshaped_info, rhs_info_img)) && use_cl_image2d)
{
return std::make_pair(lhs_info_img, rhs_info_img);
}
@@ -307,16 +324,17 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- if(workload <= 1595.2000f)
+ if (workload <= 1595.2000f)
{
- if(r_mk <= 2.1044f)
+ if (r_mk <= 2.1044f)
{
- if(workload <= 870.4000f)
+ if (workload <= 870.4000f)
{
return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 2, true, false, true, false, false);
}
@@ -336,12 +354,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifro
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedBifrost::configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedBifrost::configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 16, 4, 1, false, false, false, true);
}
diff --git a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.h b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.h
index d86d1ba0a7..9227ec2551 100644
--- a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.h
+++ b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedBifrost.h
@@ -45,17 +45,26 @@ public:
ClGemmDefaultConfigReshapedBifrost(GPUTarget gpu);
// Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
};
} // namespace gemm
} // namespace kernels
diff --git a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.cpp b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.cpp
index 58d0873b86..70b324eb5a 100644
--- a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.cpp
+++ b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.cpp
@@ -26,6 +26,7 @@
#include "arm_compute/core/CL/CLHelpers.h"
#include "arm_compute/core/CL/CLKernelLibrary.h"
#include "arm_compute/core/GPUTarget.h"
+
#include "src/gpu/cl/kernels/gemm/ClGemmHelpers.h"
#include <utility>
@@ -38,26 +39,27 @@ namespace kernels
{
namespace gemm
{
-ClGemmDefaultConfigReshapedValhall::ClGemmDefaultConfigReshapedValhall(GPUTarget gpu)
- : IClGemmKernelConfig(gpu)
+ClGemmDefaultConfigReshapedValhall::ClGemmDefaultConfigReshapedValhall(GPUTarget gpu) : IClGemmKernelConfig(gpu)
{
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure(
+ unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (ClGemmDefaultConfigReshapedValhall::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (
+ ClGemmDefaultConfigReshapedValhall::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G77(&ClGemmDefaultConfigReshapedValhall::configure_G77_f32,
- &ClGemmDefaultConfigReshapedValhall::configure_G77_f16,
- &ClGemmDefaultConfigReshapedValhall::configure_G77_u8);
+ CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G77(
+ &ClGemmDefaultConfigReshapedValhall::configure_G77_f32, &ClGemmDefaultConfigReshapedValhall::configure_G77_f16,
+ &ClGemmDefaultConfigReshapedValhall::configure_G77_u8);
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G78(&ClGemmDefaultConfigReshapedValhall::configure_G78_f32,
- &ClGemmDefaultConfigReshapedValhall::configure_G78_f16,
- &ClGemmDefaultConfigReshapedValhall::configure_G77_u8);
+ CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G78(
+ &ClGemmDefaultConfigReshapedValhall::configure_G78_f32, &ClGemmDefaultConfigReshapedValhall::configure_G78_f16,
+ &ClGemmDefaultConfigReshapedValhall::configure_G77_u8);
ConfigurationFunctionExecutorPtr func = nullptr;
- switch(_target)
+ switch (_target)
{
case GPUTarget::G78:
func = configs_G78.get_function(data_type);
@@ -72,12 +74,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
return (this->*func)(m, n, k, b);
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedValhall::configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 8, 16, 16, 1, 0, 0, 1);
}
@@ -87,7 +90,8 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedValhall::configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
@@ -104,17 +108,17 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 0);
- if(r_mk <= 0.11824845522642136)
+ if (r_mk <= 0.11824845522642136)
{
- if(workload <= 880.0)
+ if (workload <= 880.0)
{
return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, 0, 0, 1, 0, 0);
}
else
{
- if(r_nk <= 0.42521367967128754)
+ if (r_nk <= 0.42521367967128754)
{
- if(workload <= 1726.4000244140625)
+ if (workload <= 1726.4000244140625)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 0);
}
@@ -123,13 +127,12 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
}
else
{
- if(workload <= 1241.6000366210938)
+ if (workload <= 1241.6000366210938)
{
return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, 0, 0, 1, 0, 0);
}
@@ -142,17 +145,16 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 11404.7998046875)
+ if (workload <= 11404.7998046875)
{
- if(r_mk <= 1.0126488208770752)
+ if (r_mk <= 1.0126488208770752)
{
- if(r_mn <= 2.545312523841858)
+ if (r_mn <= 2.545312523841858)
{
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
else
{
@@ -161,43 +163,39 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 2881.199951171875)
+ if (workload <= 2881.199951171875)
{
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, 0, 0, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
else
{
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
}
}
else
{
- if(r_nk <= 0.5765306055545807)
+ if (r_nk <= 0.5765306055545807)
{
- if(r_mn <= 6.010416746139526)
+ if (r_mn <= 6.010416746139526)
{
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
else
{
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 1, 0, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
}
else
@@ -205,27 +203,27 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 1, 0, 1, 0, 1);
return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
+ std::make_pair(lhs_info_buf, rhs_info_buf), n, k, b, DataType::F16);
}
}
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedValhall::configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
const float r_mn = static_cast<float>(m) / static_cast<float>(n);
const float r_mk = static_cast<float>(m) / static_cast<float>(k);
const float r_nk = static_cast<float>(n) / static_cast<float>(k);
const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- if(workload <= 1288.0000f)
+ if (workload <= 1288.0000f)
{
- if(workload <= 505.6000f)
+ if (workload <= 505.6000f)
{
- if(r_mn <= 0.4466f)
+ if (r_mn <= 0.4466f)
{
- if(r_nk <= 0.2384f)
+ if (r_nk <= 0.2384f)
{
return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
}
@@ -241,9 +239,9 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(r_mn <= 0.2250f)
+ if (r_mn <= 0.2250f)
{
- if(r_mn <= 0.1599f)
+ if (r_mn <= 0.1599f)
{
return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
}
@@ -254,11 +252,11 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(r_mk <= 0.7609f)
+ if (r_mk <= 0.7609f)
{
- if(r_mn <= 2.5453f)
+ if (r_mn <= 2.5453f)
{
- if(workload <= 1089.6000f)
+ if (workload <= 1089.6000f)
{
return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
}
@@ -281,29 +279,29 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 5434.4001f)
+ if (workload <= 5434.4001f)
{
- if(workload <= 1603.2000f)
+ if (workload <= 1603.2000f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
else
{
- if(r_nk <= 0.6192f)
+ if (r_nk <= 0.6192f)
{
- if(r_mn <= 16.1016f)
+ if (r_mn <= 16.1016f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
else
{
- if(workload <= 2750.0000f)
+ if (workload <= 2750.0000f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
else
{
- if(r_mk <= 6.3151f)
+ if (r_mk <= 6.3151f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 0, 1, 1);
}
@@ -316,15 +314,15 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(r_mk <= 0.0387f)
+ if (r_mk <= 0.0387f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
}
else
{
- if(r_mk <= 2.5859f)
+ if (r_mk <= 2.5859f)
{
- if(r_mk <= 0.2734f)
+ if (r_mk <= 0.2734f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
}
@@ -343,13 +341,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(r_mk <= 25.7500f)
+ if (r_mk <= 25.7500f)
{
- if(r_mk <= 0.3615f)
+ if (r_mk <= 0.3615f)
{
- if(r_mn <= 0.0913f)
+ if (r_mn <= 0.0913f)
{
- if(r_mk <= 0.0683f)
+ if (r_mk <= 0.0683f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 4, 2, 0, 0, 1, 0, 1);
}
@@ -365,15 +363,15 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 11174.3999f)
+ if (workload <= 11174.3999f)
{
- if(r_mk <= 0.8047f)
+ if (r_mk <= 0.8047f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
else
{
- if(workload <= 7185.5999f)
+ if (workload <= 7185.5999f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
}
@@ -385,9 +383,9 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 17917.5000f)
+ if (workload <= 17917.5000f)
{
- if(r_mk <= 1.5078f)
+ if (r_mk <= 1.5078f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
@@ -398,7 +396,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 34449.6016f)
+ if (workload <= 34449.6016f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
@@ -412,11 +410,11 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(r_mk <= 331.1111f)
+ if (r_mk <= 331.1111f)
{
- if(workload <= 53397.5996f)
+ if (workload <= 53397.5996f)
{
- if(r_mn <= 57.8063f)
+ if (r_mn <= 57.8063f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
@@ -427,7 +425,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(r_nk <= 0.9211f)
+ if (r_nk <= 0.9211f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 4, 2, 0, 0, 1, 0, 1);
}
@@ -439,7 +437,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 38070.4004f)
+ if (workload <= 38070.4004f)
{
return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 0, 1, 1);
}
@@ -453,27 +451,28 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedValhall::configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
const float r_mn = static_cast<float>(m) / static_cast<float>(n);
const float r_nk = static_cast<float>(n) / static_cast<float>(k);
const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- if(workload <= 801.6000f)
+ if (workload <= 801.6000f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
}
else
{
- if(r_mn <= 0.1211f)
+ if (r_mn <= 0.1211f)
{
- if(workload <= 3296.0000f)
+ if (workload <= 3296.0000f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
else
{
- if(r_nk <= 1.0625f)
+ if (r_nk <= 1.0625f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
@@ -485,15 +484,15 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 5068.8000f)
+ if (workload <= 5068.8000f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
}
else
{
- if(r_nk <= 0.2361f)
+ if (r_nk <= 0.2361f)
{
- if(workload <= 12630.0000f)
+ if (workload <= 12630.0000f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
}
@@ -504,7 +503,7 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
else
{
- if(workload <= 178790.3984f)
+ if (workload <= 178790.3984f)
{
return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
}
@@ -518,12 +517,13 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValha
}
}
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> ClGemmDefaultConfigReshapedValhall::configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ClGemmDefaultConfigReshapedValhall::configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
{
ARM_COMPUTE_UNUSED(k);
ARM_COMPUTE_UNUSED(b);
- if(n <= 4)
+ if (n <= 4)
{
return configure_lhs_rhs_info(m, n, 4, 2, 16, 4, 1, 0, 0, 0, 1);
}
diff --git a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.h b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.h
index 466eda00a6..5f62efb59e 100644
--- a/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.h
+++ b/src/gpu/cl/kernels/gemm/reshaped/ClGemmDefaultConfigReshapedValhall.h
@@ -45,14 +45,20 @@ public:
ClGemmDefaultConfigReshapedValhall(GPUTarget gpu);
// Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo>
+ configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
};
} // namespace gemm
} // namespace kernels
diff --git a/src/gpu/cl/kernels/gemm/reshaped/ClGemmReshapedKernelConfig.h b/src/gpu/cl/kernels/gemm/reshaped/ClGemmReshapedKernelConfig.h
index 1c32f1358b..83928b3f4f 100644
--- a/src/gpu/cl/kernels/gemm/reshaped/ClGemmReshapedKernelConfig.h
+++ b/src/gpu/cl/kernels/gemm/reshaped/ClGemmReshapedKernelConfig.h
@@ -50,7 +50,7 @@ public:
*/
static std::unique_ptr<IClGemmKernelConfig> create(GPUTarget gpu)
{
- switch(get_arch_from_target(gpu))
+ switch (get_arch_from_target(gpu))
{
case GPUTarget::MIDGARD:
case GPUTarget::BIFROST: