Skip to content

GitLab

Commit 72c9f129 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'amd-develop' into amd-master

parents 241c261f ded0d83d
Hide whitespace changes
Inline Side-by-side
Showing with 337 additions and 128 deletions
script/process_perf_data.py
View file @ 72c9f129
......@@ -122,7 +122,7 @@ def parse_logfile(logfile):
#sorted_kernels = [x for _,x in sorted(zip(tests,kernels))]
test_list=list(range(1,len(tests)+1))
#parse conv_fwd and conv_bwd performance tests:
elif 'conv_fwd' in logfile or 'conv_bwd_data' in logfile:
elif 'conv_fwd' in logfile or 'conv_bwd' in logfile:
for line in open(logfile):
if 'tflops:' in line:
lst=line.split()
......@@ -274,14 +274,26 @@ def main():
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_gemm_tflops"
if 'conv_fwd' in filename:
if 'perf_conv_fwd' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_conv_fwd_tflops"
if 'conv_bwd_data' in filename:
if 'perf_conv_bwd_data' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_conv_bwd_data_tflops"
if 'grouped_conv_fwd' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_conv_fwd_tflops"
if 'grouped_conv_bwd_data' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_conv_bwd_data_tflops"
if 'grouped_conv_bwd_weight' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_conv_bwd_weight_tflops"
if 'gemm_bilinear' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
......
script/process_qa_data.sh
View file @ 72c9f129
......@@ -15,8 +15,9 @@ python3 process_perf_data.py perf_resnet50_N256.log
python3 process_perf_data.py perf_resnet50_N4.log
python3 process_perf_data.py perf_batched_gemm.log
python3 process_perf_data.py perf_grouped_gemm.log
python3 process_perf_data.py perf_conv_fwd.log
python3 process_perf_data.py perf_conv_bwd_data.log
python3 process_perf_data.py perf_grouped_conv_fwd.log
python3 process_perf_data.py perf_grouped_conv_bwd_data.log
python3 process_perf_data.py perf_grouped_conv_bwd_weight.log
python3 process_perf_data.py perf_gemm_bilinear.log
python3 process_perf_data.py perf_reduction.log
python3 process_perf_data.py perf_splitK_gemm.log
......
script/profile_conv_fwd.sh → script/profile_grouped_conv_bwd_weight.sh
View file @ 72c9f129
......@@ -12,27 +12,28 @@ INIT=$5
LOG=$6
TIME=$7
N=$8
N=$8
SplitK=$9
# Resnet50
######## op datatype layout verify init log time conv_dim G__ N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3 $SplitK
script/profile_grouped_conv_fwd.sh 0 → 100755
View file @ 72c9f129
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
DRIVER="../build/bin/ckProfiler"
OP=$1
DATATYPE=$2
LAYOUT=$3
INDEXTYPE=$4
VERIFY=$5
INIT=$6
LOG=$7
TIME=$8
N=$9
# Resnet50
######## op datatype indextype layout verify init log time conv_dim G__ N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3
script/run_full_performance_tests.sh
View file @ 72c9f129
......@@ -90,21 +90,27 @@ print_log_header $gemm_bilinear_log $env_type $branch $host_name
./profile_gemm_bilinear.sh gemm_bilinear 1 2 $verify 1 0 1 2>&1 | tee -a $gemm_bilinear_log
./profile_gemm_bilinear.sh gemm_bilinear 1 3 $verify 1 0 1 2>&1 | tee -a $gemm_bilinear_log
#run conv_fwd tests
export conv_fwd_log="perf_conv_fwd.log"
print_log_header $conv_fwd_log $env_type $branch $host_name
./profile_conv_fwd.sh conv_fwd 0 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log
./profile_conv_fwd.sh conv_fwd 1 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log
./profile_conv_fwd.sh conv_fwd 2 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log
./profile_conv_fwd.sh conv_fwd 3 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log
#run grouped_fwd tests
export grouped_conv_fwd_log="perf_grouped_conv_fwd.log"
print_log_header $grouped_conv_fwd_log $env_type $branch $host_name
./profile_grouped_conv_fwd.sh grouped_conv_fwd 0 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
./profile_grouped_conv_fwd.sh grouped_conv_fwd 1 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
./profile_grouped_conv_fwd.sh grouped_conv_fwd 2 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
#run conv_bwd_data tests
export conv_bwd_data_log="perf_conv_bwd_data.log"
print_log_header $conv_bwd_data_log $env_type $branch $host_name
./profile_conv_bwd_data.sh conv_bwd_data 0 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 1 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 2 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 3 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log
#run grouped_bwd_data tests
export grouped_conv_bwd_data_log="perf_grouped_conv_bwd_data.log"
print_log_header $grouped_conv_bwd_data_log $env_type $branch $host_name
./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 0 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 1 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 2 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
#run grouped_bwd_weight tests
export grouped_conv_bwd_weight_log="perf_grouped_conv_bwd_weight.log"
print_log_header $grouped_conv_bwd_weight_log $env_type $branch $host_name
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 0 2 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 1 2 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 2 2 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 1 2 $verify 1 0 1 256 4 2>&1 | tee -a $grouped_conv_bwd_weight_log
#run resnet50 tests
export resnet256_log="perf_resnet50_N256.log"
......
script/run_performance_tests.sh
View file @ 72c9f129
......@@ -51,6 +51,21 @@ print_log_header $gemm_log $env_type $branch $host_name
./profile_gemm.sh gemm 2 3 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 3 $verify 1 0 1 | tee -a $gemm_log
#run grouped_fwd fp16 tests
export grouped_conv_fwd_log="perf_grouped_conv_fwd_fp16.log"
print_log_header $conv_fwd_log $env_type $branch $host_name
./profile_grouped_conv_fwd.sh grouped_conv_fwd 1 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
#run grouped_bwd_data fp16 tests
export grouped_conv_bwd_data_log="perf_grouped_conv_bwd_data_fp16.log"
print_log_header $grouped_conv_bwd_data_log $env_type $branch $host_name
./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 1 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
#run grouped_bwd_weight fp16 tests
export grouped_conv_bwd_weight_log="perf_grouped_conv_bwd_weight_fp16.log"
print_log_header $grouped_conv_bwd_weight_log $env_type $branch $host_name
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 1 1 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
#run resnet50 tests
export resnet256_log="perf_resnet50_N256.log"
print_log_header $resnet256_log $env_type $branch $host_name
......
test/data_type/CMakeLists.txt
View file @ 72c9f129
# temporarily disable flaky test for all architectures
add_definitions(-DCK_SKIP_FLAKY_F8_TEST)
set(CK_SKIP_FLAKY_F8_TEST "ON")
if (USE_BITINT_EXTENSION_INT4)
add_gtest_executable(test_int4 test_int4.cpp)
if(result EQUAL 0)
......
test/data_type/test_bf8.cpp
View file @ 72c9f129
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "ck/utility/data_type.hpp"
#include "ck/utility/type_convert.hpp"
using ck::bf8_t;
using ck::f8_convert_rne;
using ck::f8_convert_sr;
using ck::half_t;
using ck::type_convert;
......@@ -24,33 +25,36 @@ TEST(BF8, ConvertFP32Nearest)
// fix the tolerance value
float abs_tol = 1e-6;
// convert 0 float to bf8 and back, check if holds
ASSERT_NEAR(0.0f, type_convert<float>(type_convert<bf8_t>(0.0f)), abs_tol);
ASSERT_NEAR(0.0f, type_convert<float>(f8_convert_rne<bf8_t>(0.0f)), abs_tol);
// don't run the next test on gfx11 devices
#ifndef CK_SKIP_FLAKY_F8_TEST
// convert minimal float to bf8 and back, check if holds
ASSERT_NEAR(std::numeric_limits<float>::min(),
type_convert<float>(type_convert<bf8_t>(std::numeric_limits<float>::min())),
type_convert<float>(f8_convert_rne<bf8_t>(std::numeric_limits<float>::min())),
abs_tol);
#endif
// convert maximal bf8_t to float and check if equal to 57344.0
ASSERT_NEAR(57344.0f, type_convert<float>(type_convert<bf8_t>(57344.0f)), abs_tol);
ASSERT_NEAR(57344.0f, type_convert<float>(f8_convert_rne<bf8_t>(57344.0f)), abs_tol);
// convert maximal float to bf8 and back, check if clipped to 57344.0
ASSERT_NEAR(57344.0f,
type_convert<float>(type_convert<bf8_t>(std::numeric_limits<float>::max())),
type_convert<float>(f8_convert_rne<bf8_t>(std::numeric_limits<float>::max())),
abs_tol);
// convert inf float to bf8_t and check if it is qNan
ASSERT_NEAR(type_convert<bf8_t>(0x80),
type_convert<bf8_t>(std::numeric_limits<float>::infinity()),
f8_convert_rne<bf8_t>(std::numeric_limits<float>::infinity()),
abs_tol);
// positive norm float value to bf8 and back, check if holds
float pos_float = 0.0000762939f;
ASSERT_NEAR(pos_float, type_convert<float>(type_convert<bf8_t>(pos_float)), abs_tol);
ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_rne<bf8_t>(pos_float)), abs_tol);
// negative norm float value to bf8 and back, check if holds
float neg_float = -0.0000610351f;
ASSERT_NEAR(neg_float, type_convert<float>(type_convert<bf8_t>(neg_float)), abs_tol);
ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_rne<bf8_t>(neg_float)), abs_tol);
// positive subnorm float value to bf8 and back, check if holds
pos_float = 0.0000305175f;
ASSERT_NEAR(pos_float, type_convert<float>(type_convert<bf8_t>(pos_float)), abs_tol);
ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_rne<bf8_t>(pos_float)), abs_tol);
// negative subnorm float value to bf8 and back, check if holds
neg_float = -0.0000152587f;
ASSERT_NEAR(neg_float, type_convert<float>(type_convert<bf8_t>(neg_float)), abs_tol);
ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_rne<bf8_t>(neg_float)), abs_tol);
}
TEST(BF8, ConvertFP32Stochastic)
......@@ -92,34 +96,34 @@ TEST(BF8, ConvertFP16Nearest)
// fix the tolerance value
float abs_tol = 1e-3;
// convert 0 fp16 to bf8 and back, check if holds
ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(type_convert<bf8_t>(half_t{0.0})), abs_tol);
ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(f8_convert_rne<bf8_t>(half_t{0.0})), abs_tol);
// convert minimal fp16 to bf8 and back, check if holds
ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
type_convert<half_t>(type_convert<bf8_t>(ck::NumericLimits<half_t>::Min())),
type_convert<half_t>(f8_convert_rne<bf8_t>(ck::NumericLimits<half_t>::Min())),
abs_tol);
// convert maximal bf8_t to fp16 and check if equal to 57344.0
ASSERT_NEAR(
half_t{57344.0}, type_convert<half_t>(type_convert<bf8_t>(half_t{57344.0})), abs_tol);
half_t{57344.0}, type_convert<half_t>(f8_convert_rne<bf8_t>(half_t{57344.0})), abs_tol);
// convert maximal fp16 to bf8 and back, check if clipped to 57344.0
ASSERT_NEAR(half_t{57344.0},
type_convert<half_t>(type_convert<bf8_t>(ck::NumericLimits<half_t>::Max())),
type_convert<half_t>(f8_convert_rne<bf8_t>(ck::NumericLimits<half_t>::Max())),
abs_tol);
// convert QuietNaN fp16 to bf8_t and check if it is QuietNaN
ASSERT_NEAR(type_convert<bf8_t>(0x80),
type_convert<bf8_t>(ck::NumericLimits<half_t>::QuietNaN()),
f8_convert_rne<bf8_t>(ck::NumericLimits<half_t>::QuietNaN()),
abs_tol);
// positive norm fp16 value to bf8 and back, check if holds
half_t pos_half = half_t{0.0000762939};
ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<bf8_t>(pos_half)), abs_tol);
ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_rne<bf8_t>(pos_half)), abs_tol);
// negative norm fp16 value to bf8 and back, check if holds
half_t neg_half = half_t{-0.0000610351};
ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<bf8_t>(neg_half)), abs_tol);
ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_rne<bf8_t>(neg_half)), abs_tol);
// positive subnorm fp16 value to bf8 and back, check if holds
pos_half = half_t{0.0000305175};
ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<bf8_t>(pos_half)), abs_tol);
ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_rne<bf8_t>(pos_half)), abs_tol);
// negative subnorm fp16 value to bf8 and back, check if holds
neg_half = half_t{-0.0000152587};
ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<bf8_t>(neg_half)), abs_tol);
ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_rne<bf8_t>(neg_half)), abs_tol);
}
TEST(BF8, ConvertFP16Stochastic)
......
test/data_type/test_fp8.cpp
View file @ 72c9f129
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "ck/utility/data_type.hpp"
#include "ck/utility/type_convert.hpp"
using ck::f8_convert_rne;
using ck::f8_convert_sr;
using ck::f8_t;
using ck::half_t;
......@@ -24,33 +25,36 @@ TEST(FP8, ConvertFP32Nearest)
// fix the tolerance value
float abs_tol = 1e-6;
// convert 0 float to fp8 and back, check if holds
ASSERT_NEAR(0.0f, type_convert<float>(type_convert<f8_t>(0.0f)), abs_tol);
ASSERT_NEAR(0.0f, type_convert<float>(f8_convert_rne<f8_t>(0.0f)), abs_tol);
// don't run the next test on gfx11 devices
#ifndef CK_SKIP_FLAKY_F8_TEST
// convert minimal float to fp8 and back, check if holds
ASSERT_NEAR(std::numeric_limits<float>::min(),
type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::min())),
type_convert<float>(f8_convert_rne<f8_t>(std::numeric_limits<float>::min())),
abs_tol);
#endif
// convert maximal f8_t to float and check if equal to 240.0
ASSERT_NEAR(240.0f, type_convert<float>(type_convert<f8_t>(240.0f)), abs_tol);
ASSERT_NEAR(240.0f, type_convert<float>(f8_convert_rne<f8_t>(240.0f)), abs_tol);
// convert maximal float to fp8 and back, check if clipped to 240.0
ASSERT_NEAR(240.0f,
type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::max())),
type_convert<float>(f8_convert_rne<f8_t>(std::numeric_limits<float>::max())),
abs_tol);
// convert inf float to f8_t and check if it is qNan
ASSERT_NEAR(type_convert<f8_t>(0x80),
type_convert<f8_t>(std::numeric_limits<float>::infinity()),
f8_convert_rne<f8_t>(std::numeric_limits<float>::infinity()),
abs_tol);
// positive norm float value to fp8 and back, check if holds
float pos_float = 0.017578125f;
ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_rne<f8_t>(pos_float)), abs_tol);
// negative norm float value to fp8 and back, check if holds
float neg_float = -0.015625f;
ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_rne<f8_t>(neg_float)), abs_tol);
// positive subnorm float value to fp8 and back, check if holds
pos_float = 0.00390625f;
ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_rne<f8_t>(pos_float)), abs_tol);
// negative subnorm float value to fp8 and back, check if holds
neg_float = -0.001953125f;
ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_rne<f8_t>(neg_float)), abs_tol);
}
TEST(FP8, ConvertFP32Stochastic)
......@@ -92,33 +96,33 @@ TEST(FP8, ConvertFP16Nearest)
// fix the tolerance value
float abs_tol = 1e-3;
// convert 0 fp16 to fp8 and back, check if holds
ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(type_convert<f8_t>(half_t{0.0})), abs_tol);
ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(f8_convert_rne<f8_t>(half_t{0.0})), abs_tol);
// convert minimal fp16 to fp8 and back, check if holds
ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Min())),
type_convert<half_t>(f8_convert_rne<f8_t>(ck::NumericLimits<half_t>::Min())),
abs_tol);
// convert maximal f8_t to fp16 and check if equal to 240.0
ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(type_convert<f8_t>(half_t{240.0})), abs_tol);
ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(f8_convert_rne<f8_t>(half_t{240.0})), abs_tol);
// convert maximal fp16 to fp8 and back, check if clipped to 240.0
ASSERT_NEAR(half_t{240.0},
type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Max())),
type_convert<half_t>(f8_convert_rne<f8_t>(ck::NumericLimits<half_t>::Max())),
abs_tol);
// convert QuietNaN fp16 to f8_t and check if it is QuietNaN
ASSERT_NEAR(type_convert<f8_t>(0x80),
type_convert<f8_t>(ck::NumericLimits<half_t>::QuietNaN()),
f8_convert_rne<f8_t>(ck::NumericLimits<half_t>::QuietNaN()),
abs_tol);
// positive norm fp16 value to fp8 and back, check if holds
half_t pos_half = half_t{0.017578125};
ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_rne<f8_t>(pos_half)), abs_tol);
// negative norm fp16 value to fp8 and back, check if holds
half_t neg_half = half_t{-0.015625};
ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_rne<f8_t>(neg_half)), abs_tol);
// positive subnorm fp16 value to fp8 and back, check if holds
pos_half = half_t{0.00390625};
ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_rne<f8_t>(pos_half)), abs_tol);
// negative subnorm fp16 value to fp8 and back, check if holds
neg_half = half_t{-0.001953125};
ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_rne<f8_t>(neg_half)), abs_tol);
}
TEST(FP8, ConvertFP16Stochastic)
......
test/gemm_universal/test_gemm_universal_xdl.cpp
View file @ 72c9f129
......@@ -44,17 +44,22 @@ class TestGemmUniversal_MK_NK
using KernelTypes_MK_KN = ::testing::Types<
// ADataType, BDataType, ComputeDataType, CDataType
std::tuple< F16, F16, F16, F16>,
#if (defined CK_ENABLE_FP8)
std::tuple< F16, F8, F16, F16>,
std::tuple< F8, F16, F16, F16>,
std::tuple< F8, F8, F8, BF16>,
#endif
std::tuple< BF16, BF16, BF16, BF16>
>;
using KernelTypes_MK_NK = ::testing::Types<
// ADataType, BDataType, ComputeDataType, CDataType
std::tuple< F16, F16, F16, F16>,
#if (defined CK_ENABLE_FP8)
std::tuple< F16, F8, F16, F16>,
std::tuple< F8, F16, F16, F16>,
std::tuple< BF16, BF16, BF16, BF16>,
std::tuple< F8, F8, F8, BF16>
std::tuple< F8, F8, F8, BF16>,
#endif
std::tuple< BF16, BF16, BF16, BF16>
>;
// clang-format on
......
test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
View file @ 72c9f129
......@@ -66,6 +66,12 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
{
return true;
}
// Skip due to the lack of kernels for NGCDHW
if constexpr(std::is_same_v<InLayout, NGCW> || std::is_same_v<InLayout, NGCHW> ||
std::is_same_v<InLayout, NGCDHW>)
{
return true;
}
}
else
{
......@@ -139,7 +145,8 @@ using KernelTypes2d = ::testing::Types<
std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
std::tuple<float, float, float, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
std::tuple<ck::bhalf_t, float, ck::bhalf_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>>;
std::tuple<ck::bhalf_t, float, ck::bhalf_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, NGCHW, GKYXC, NGKHW, ck::Number<2>>>;
using KernelTypes3d = ::testing::Types<
std::tuple<float, float, float, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
......@@ -148,7 +155,8 @@ using KernelTypes3d = ::testing::Types<
std::tuple<float, float, float, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
std::tuple<ck::bhalf_t, float, ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
std::tuple<int8_t, int8_t, int8_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>>;
std::tuple<int8_t, int8_t, int8_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
std::tuple<ck::half_t, ck::half_t, ck::half_t, NGCDHW, GKZYXC, NGKDHW, ck::Number<3>>>;
TYPED_TEST_SUITE(TestGroupedConvndBwdWeight1d, KernelTypes1d);
TYPED_TEST_SUITE(TestGroupedConvndBwdWeight2d, KernelTypes2d);
......
test/grouped_convnd_fwd/CMakeLists.txt
View file @ 72c9f129
......@@ -7,6 +7,12 @@ if(GPU_TARGETS MATCHES "gfx9" OR GPU_TARGETS MATCHES "gfx11")
endif()
endif()
if(GPU_TARGETS MATCHES "gfx9")
add_executable(test_grouped_convnd_fwd_large_cases_xdl test_grouped_convnd_fwd_large_cases_xdl.cpp)
target_compile_options(test_grouped_convnd_fwd_large_cases_xdl PRIVATE -Wno-global-constructors -Wno-undef)
target_link_libraries(test_grouped_convnd_fwd_large_cases_xdl PRIVATE gtest_main getopt::getopt utility device_grouped_conv1d_fwd_instance device_grouped_conv2d_fwd_instance device_grouped_conv3d_fwd_instance)
endif()
add_gtest_executable(test_grouped_convnd_fwd_multi_ab_interface test_grouped_convnd_fwd_multi_ab_interface.cpp)
if(result EQUAL 0)
target_link_libraries(test_grouped_convnd_fwd_multi_ab_interface PRIVATE utility)
......
test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp
View file @ 72c9f129
......@@ -17,7 +17,7 @@ class TestGroupedConvndFwd : public ::testing::Test
using InLayout = std::tuple_element_t<1, Tuple>;
using WeiLayout = std::tuple_element_t<2, Tuple>;
using OutLayout = std::tuple_element_t<3, Tuple>;
using IndexType = std::tuple_element_t<4, Tuple>;
using IndexType = ck::index_t;
std::vector<ck::utils::conv::ConvParam> conv_params;
......@@ -50,31 +50,28 @@ class TestGroupedConvndFwd : public ::testing::Test
using namespace ck::tensor_layout::convolution;
using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK, ck::index_t>,
std::tuple<ck::half_t, GNWC, GKXC, GNWK, ck::index_t>,
std::tuple<ck::bhalf_t, GNWC, GKXC, GNWK, ck::index_t>,
std::tuple<int8_t, GNWC, GKXC, GNWK, ck::index_t>>;
using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK, ck::index_t>,
std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK, ck::index_t>,
std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK, ck::index_t>,
std::tuple<int8_t, GNHWC, GKYXC, GNHWK, ck::index_t>,
std::tuple<float, NHWGC, GKYXC, NHWGK, ck::index_t>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK, ck::index_t>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK, ck::index_t>,
std::tuple<int8_t, NHWGC, GKYXC, NHWGK, ck::index_t>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK, ck::index_t>,
std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::index_t>,
std::tuple<ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::index_t>,
std::tuple<int8_t, GNDHWC, GKZYXC, GNDHWK, ck::index_t>,
std::tuple<float, NDHWGC, GKZYXC, NDHWGK, ck::index_t>,
std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK, ck::index_t>,
std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK, ck::index_t>,
std::tuple<int8_t, NDHWGC, GKZYXC, NDHWGK, ck::index_t>>;
using KernelTypes2dLargeCases =
::testing::Types<std::tuple<float, NHWGC, GKYXC, NHWGK, ck::long_index_t>>;
using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK>,
std::tuple<ck::half_t, GNWC, GKXC, GNWK>,
std::tuple<ck::bhalf_t, GNWC, GKXC, GNWK>,
std::tuple<int8_t, GNWC, GKXC, GNWK>>;
using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK>,
std::tuple<int8_t, GNHWC, GKYXC, GNHWK>,
std::tuple<float, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>,
std::tuple<int8_t, NHWGC, GKYXC, NHWGK>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<int8_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<float, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<int8_t, NDHWGC, GKZYXC, NDHWGK>>;
template <typename Tuple>
class TestGroupedConvndFwd1d : public TestGroupedConvndFwd<Tuple>
......@@ -91,15 +88,9 @@ class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple>
{
};
template <typename Tuple>
class TestGroupedConvndFwd2dLargeCases : public TestGroupedConvndFwd<Tuple>
{
};
TYPED_TEST_SUITE(TestGroupedConvndFwd1d, KernelTypes1d);
TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d);
TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d);
TYPED_TEST_SUITE(TestGroupedConvndFwd2dLargeCases, KernelTypes2dLargeCases);
TYPED_TEST(TestGroupedConvndFwd1d, Test1D)
{
......@@ -149,17 +140,3 @@ TYPED_TEST(TestGroupedConvndFwd3d, Test3D)
{3, 96, 1, 1, 1, {3, 3, 3}, {4, 30, 160}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->template Run<3>();
}
TYPED_TEST(TestGroupedConvndFwd2dLargeCases, Test2DLargeCases)
{
// Case larger than 2GB
this->conv_params.push_back(
{2, 1, 64, 4, 192, {2, 2}, {224, 224}, {224, 224}, {1, 1}, {0, 0}, {0, 0}});
// With supported NumGroupsToMerge > 1
this->conv_params.push_back(
{2, 32, 64, 1, 1, {2, 2}, {672, 672}, {672, 672}, {1, 1}, {0, 0}, {0, 0}});
// When image is larger than 2GB
this->conv_params.push_back(
{2, 1, 1, 256, 256, {3, 3}, {4096, 2048}, {1024, 1024}, {3, 3}, {1, 1}, {1, 1}});
this->template Run<2>();
}
test/grouped_convnd_fwd/test_grouped_convnd_fwd_large_cases_xdl.cpp 0 → 100644
View file @ 72c9f129
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <initializer_list>
#include <vector>
#include <gtest/gtest.h>
#include "profiler/profile_grouped_conv_fwd_impl.hpp"
template <typename Tuple>
class TestGroupedConvndFwd : public ::testing::Test
{
protected:
using DataType = std::tuple_element_t<0, Tuple>;
using InLayout = std::tuple_element_t<1, Tuple>;
using WeiLayout = std::tuple_element_t<2, Tuple>;
using OutLayout = std::tuple_element_t<3, Tuple>;
using IndexType = ck::long_index_t;
std::vector<ck::utils::conv::ConvParam> conv_params;
template <ck::index_t NDimSpatial>
void Run()
{
EXPECT_FALSE(conv_params.empty());
bool pass = true;
for(auto& param : conv_params)
{
pass = pass && ck::profiler::profile_grouped_conv_fwd_impl<NDimSpatial,
InLayout,
WeiLayout,
OutLayout,
DataType,
DataType,
DataType,
DataType,
DataType,
IndexType>(
true, // do_verification
1, // init_method: integer value
false, // do_log
false, // time_kernel
param);
}
EXPECT_TRUE(pass);
}
};
using namespace ck::tensor_layout::convolution;
using KernelTypes2d = ::testing::Types<std::tuple<float, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK>>;
template <typename Tuple>
class TestGroupedConvndFwd2d : public TestGroupedConvndFwd<Tuple>
{
};
template <typename Tuple>
class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple>
{
};
TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d);
TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d);
TYPED_TEST(TestGroupedConvndFwd2d, Test2D)
{
// Case larger than 2GB
this->conv_params.push_back(
{2, 1, 128, 4, 192, {2, 2}, {224, 224}, {224, 224}, {1, 1}, {0, 0}, {0, 0}});
// With supported NumGroupsToMerge > 1
this->conv_params.push_back(
{2, 32, 64, 1, 1, {2, 2}, {672, 672}, {672, 672}, {1, 1}, {0, 0}, {0, 0}});
// When image is larger than 2GB
this->conv_params.push_back(
{2, 2, 2, 128, 128, {3, 3}, {4096, 2048}, {300, 300}, {3, 3}, {1, 1}, {1, 1}});
this->template Run<2>();
}
TYPED_TEST(TestGroupedConvndFwd3d, Test3D)
{
// Case larger than 2GB
this->conv_params.push_back({3,
1,
128,
4,
192,
{2, 2, 2},
{2, 224, 224},
{1, 224, 224},
{1, 1, 1},
{0, 0, 0},
{0, 0, 0}});
// With supported NumGroupsToMerge > 1
this->conv_params.push_back({3,
32,
64,
1,
1,
{2, 2, 2},
{360, 2, 672},
{360, 2, 672},
{1, 1, 1},
{0, 0, 0},
{0, 0, 0}});
// When image is larger than 2GB
this->conv_params.push_back({3,
1,
2,
128,
128,
{3, 1, 3},
{900, 2, 2048},
{300, 1, 300},
{3, 2, 3},
{1, 1, 1},
{1, 1, 1}});
this->template Run<3>();
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment