AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

EXSWHTEC-106 - Reimplement tests for hipOccupancyMaxActiveBlocksPerMultiprocessor and hipOccupancyMaxPotentialBlockSize APIs (#46)

Browse Source 
- Reimplement tests for hipOccupancyMaxActiveBlocksPerMultiprocessor and hipOccupancyMaxPotentialBlockSize APIs
- Add helper file occupancy_common.hh with parameterized templates
- Expand all positive and negative tests to use the templates
- Change section disable macros
- Disable AMD specific test due to defect
- Disable negative test using json file
- Fix disabled files
This commit is contained in:
nives-vukovic
2023-06-28 09:19:28 +02:00
committed by GitHub
parent 329a350ec0
commit dd9b9b027f
8 changed files with 396 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files
catch/hipTestMain/config/config_amd_linux_common.json
+1
View File
@@ -13,6 +13,7 @@
"Unit_hipInit_Negative",
"Unit_hipDeviceReset_Positive_Basic",
"Unit_hipDeviceReset_Positive_Threaded",
"Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Negative_Parameters",
"Unit_hipGraphMemcpyNodeSetParamsToSymbol_Positive_Basic",
"Unit_hipGraphMemcpyNodeSetParamsFromSymbol_Positive_Basic",
"Unit_hipKernelNameRef_Negative_Parameters",
catch/hipTestMain/config/config_amd_windows_common.json
+1
View File
@@ -95,6 +95,7 @@
"Unit_hipStreamSynchronize_NullStreamAndStreamPerThread",
"Note: intermittent Seg fault failure ",
"Unit_hipGraphAddEventRecordNode_Functional_WithoutFlags",
"Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Negative_Parameters",
"Unit_hipGraphMemcpyNodeSetParamsToSymbol_Positive_Basic",
"Unit_hipGraphExecMemcpyNodeSetParamsToSymbol_Positive_Basic",
"Unit_hipGraphMemcpyNodeSetParamsFromSymbol_Positive_Basic",
catch/unit/occupancy/CMakeLists.txt
+2
View File
@@ -1,7 +1,9 @@
# Common Tests - Test independent of all platforms
set(TEST_SRC
hipOccupancyMaxActiveBlocksPerMultiprocessor.cc
hipOccupancyMaxActiveBlocksPerMultiprocessor_old.cc
hipOccupancyMaxPotentialBlockSize.cc
hipOccupancyMaxPotentialBlockSize_old.cc
hipOccupancyMaxPotentialBlockSizeVariableSMemWithFlags.cc
)
catch/unit/occupancy/hipOccupancyMaxActiveBlocksPerMultiprocessor.cc
+85 -58
View File
@@ -1,5 +1,5 @@
/*
Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
@@ -16,76 +16,103 @@ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include <hip_test_common.hh>
/*
Testcase Scenarios :
Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Positive_RangeValidation - Test correct execution
of hipOccupancyMaxActiveBlocksPerMultiprocessor for diffrent parameter values
Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Positive_TemplateInvocation - Test correct
execution of hipOccupancyMaxActiveBlocksPerMultiprocessor template for diffrent parameter values
Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Negative_Parameters - Test unsuccessful execution
of hipOccupancyMaxActiveBlocksPerMultiprocessor api when parameters are invalid
*/
#include "occupancy_common.hh"
static __global__ void f1(float *a) { *a = 1.0; }
static __global__ void f1(float* a) { *a = 1.0; }
template <typename T>
static __global__ void f2(T *a) { *a = 1; }
template <typename T> static __global__ void f2(T* a) { *a = 1; }
/**
* Defines
*/
#define OccupancyDisableCachingOverride 0x01
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Negative") {
hipError_t ret;
int numBlock = 0, blockSize = 0;
int gridSize = 0, defBlkSize = 32;
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
// Validate each argument
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(NULL, f1, blockSize, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, NULL, blockSize, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, 0, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, 0,
std::numeric_limits<std::size_t>::max());
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(&numBlock, f1,
defBlkSize, 0, OccupancyDisableCachingOverride);
REQUIRE(ret == hipSuccess);
}
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_rangeValidation") {
hipDeviceProp_t devProp;
int numBlock = 0, blockSize = 0;
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Negative_Parameters") {
int numBlocks = 0;
int blockSize = 0;
int gridSize = 0;
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
// Common negative tests
MaxActiveBlocksPerMultiprocessorNegative(
[](int* numBlocks, int blockSize, size_t dynSharedMemPerBlk) {
return hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f1, blockSize,
dynSharedMemPerBlk);
},
blockSize);
SECTION("Kernel function is NULL") {
HIP_CHECK_ERROR(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlocks, NULL, blockSize, 0),
hipErrorInvalidDeviceFunction);
}
}
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Positive_RangeValidation") {
hipDeviceProp_t devProp;
int blockSize = 0;
int gridSize = 0;
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, blockSize, 0));
SECTION("dynSharedMemPerBlk = 0") {
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
// Check if numBlocks and blockSize are within limits
REQUIRE(numBlock > 0);
REQUIRE((numBlock * blockSize) <= devProp.maxThreadsPerMultiProcessor);
MaxActiveBlocksPerMultiprocessor(
[blockSize](int* numBlocks) {
return hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f1, blockSize, 0);
},
blockSize, devProp.maxThreadsPerMultiProcessor);
}
SECTION("dynSharedMemPerBlk = sharedMemPerBlock") {
// Get potential blocksize
HIP_CHECK(
hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, devProp.sharedMemPerBlock, 0));
// Validate numBlock after passing dynSharedMemPerBlk
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, blockSize,
devProp.sharedMemPerBlock));
// Check if numBlocks and blockSize are within limits
REQUIRE(numBlock > 0);
REQUIRE((numBlock * blockSize) <= devProp.maxThreadsPerMultiProcessor);
MaxActiveBlocksPerMultiprocessor(
[blockSize, devProp](int* numBlocks) {
return hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f1, blockSize,
devProp.sharedMemPerBlock);
},
blockSize, devProp.maxThreadsPerMultiProcessor);
}
}
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_templateInvocation") {
int blockSize = 32;
int numBlock = 0;
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Positive_TemplateInvocation") {
hipDeviceProp_t devProp;
int blockSize = 0;
int gridSize = 0;
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor<void(*)(int *)>
(&numBlock, f2, blockSize, 0));
REQUIRE(numBlock > 0);
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
SECTION("dynSharedMemPerBlk = 0") {
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize<void (*)(int*)>(&gridSize, &blockSize, f2, 0, 0));
MaxActiveBlocksPerMultiprocessor(
[blockSize](int* numBlocks) {
return hipOccupancyMaxActiveBlocksPerMultiprocessor<void (*)(int*)>(numBlocks, f2,
blockSize, 0);
},
blockSize, devProp.maxThreadsPerMultiProcessor);
}
SECTION("dynSharedMemPerBlk = sharedMemPerBlock") {
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize<void (*)(int*)>(&gridSize, &blockSize, f2,
devProp.sharedMemPerBlock, 0));
MaxActiveBlocksPerMultiprocessor(
[blockSize, devProp](int* numBlocks) {
return hipOccupancyMaxActiveBlocksPerMultiprocessor<void (*)(int*)>(
numBlocks, f2, blockSize, devProp.sharedMemPerBlock);
},
blockSize, devProp.maxThreadsPerMultiProcessor);
}
}
catch/unit/occupancy/hipOccupancyMaxActiveBlocksPerMultiprocessor_old.cc
+91
View File
@@ -0,0 +1,91 @@
/*
Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include <hip_test_common.hh>
static __global__ void f1(float *a) { *a = 1.0; }
template <typename T>
static __global__ void f2(T *a) { *a = 1; }
/**
* Defines
*/
#define OccupancyDisableCachingOverride 0x01
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_Negative") {
hipError_t ret;
int numBlock = 0, blockSize = 0;
int gridSize = 0, defBlkSize = 32;
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
// Validate each argument
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(NULL, f1, blockSize, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, NULL, blockSize, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, 0, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, 0,
std::numeric_limits<std::size_t>::max());
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(&numBlock, f1,
defBlkSize, 0, OccupancyDisableCachingOverride);
REQUIRE(ret == hipSuccess);
}
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_rangeValidation") {
hipDeviceProp_t devProp;
int numBlock = 0, blockSize = 0;
int gridSize = 0;
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, blockSize, 0));
// Check if numBlocks and blockSize are within limits
REQUIRE(numBlock > 0);
REQUIRE((numBlock * blockSize) <= devProp.maxThreadsPerMultiProcessor);
// Validate numBlock after passing dynSharedMemPerBlk
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, blockSize,
devProp.sharedMemPerBlock));
// Check if numBlocks and blockSize are within limits
REQUIRE(numBlock > 0);
REQUIRE((numBlock * blockSize) <= devProp.maxThreadsPerMultiProcessor);
}
TEST_CASE("Unit_hipOccupancyMaxActiveBlocksPerMultiprocessor_templateInvocation") {
int blockSize = 32;
int numBlock = 0;
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor<void(*)(int *)>
(&numBlock, f2, blockSize, 0));
REQUIRE(numBlock > 0);
}
catch/unit/occupancy/hipOccupancyMaxPotentialBlockSize.cc
+64 -47
View File
@@ -1,5 +1,5 @@
/*
Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
@@ -16,65 +16,82 @@ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include <hip_test_common.hh>
/*
Testcase Scenarios :
Unit_hipOccupancyMaxPotentialBlockSize_Positive_RangeValidation - Test correct execution of
hipOccupancyMaxPotentialBlockSize for diffrent parameter values
Unit_hipOccupancyMaxPotentialBlockSize_Positive_TemplateInvocation - Test correct execution of
hipOccupancyMaxPotentialBlockSize template for diffrent parameter values
Unit_hipOccupancyMaxPotentialBlockSize_Negative_Parameters - Test unsuccessful execution of
hipOccupancyMaxPotentialBlockSize api when parameters are invalid
*/
#include "occupancy_common.hh"
static __global__ void f1(float *a) { *a = 1.0; }
static __global__ void f1(float* a) { *a = 1.0; }
template <typename T>
static __global__ void f2(T *a) { *a = 1; }
template <typename T> static __global__ void f2(T* a) { *a = 1; }
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_Negative") {
hipError_t ret;
int blockSize = 0;
int gridSize = 0;
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_Negative_Parameters") {
// Common negative tests
MaxPotentialBlockSizeNegative([](int* gridSize, int* blockSize) {
return hipOccupancyMaxPotentialBlockSize(gridSize, blockSize, f1, 0, 0);
});
// Validate each argument
ret = hipOccupancyMaxPotentialBlockSize(NULL, &blockSize, f1, 0, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxPotentialBlockSize(&gridSize, NULL, f1, 0, 0);
REQUIRE(ret != hipSuccess);
#ifndef __HIP_PLATFORM_NVIDIA__
// nvcc doesnt support kernelfunc(NULL) for api
ret = hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, NULL, 0, 0);
REQUIRE(ret != hipSuccess);
#if HT_AMD
#if 0 // EXSWHTEC-219
SECTION("Kernel function is NULL") {
int blockSize = 0;
int gridSize = 0;
// nvcc doesnt support kernelfunc(NULL) for api
HIP_CHECK_ERROR(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, NULL, 0, 0),
hipErrorInvalidDeviceFunction);
}
#endif
#endif
}
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_rangeValidation") {
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_Positive_RangeValidation") {
hipDeviceProp_t devProp;
int blockSize = 0;
int gridSize = 0;
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
// Check if blockSize doen't exceed maxThreadsPerBlock
REQUIRE(gridSize > 0); REQUIRE(blockSize > 0);
REQUIRE(blockSize <= devProp.maxThreadsPerBlock);
// Pass dynSharedMemPerBlk, blockSizeLimit and check out param
blockSize = 0;
gridSize = 0;
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1,
devProp.sharedMemPerBlock, devProp.maxThreadsPerBlock));
// Check if blockSize doen't exceed maxThreadsPerBlock
REQUIRE(gridSize > 0); REQUIRE(blockSize > 0);
REQUIRE(blockSize <= devProp.maxThreadsPerBlock);
SECTION("dynSharedMemPerBlk = 0, blockSizeLimit = 0") {
MaxPotentialBlockSize(
[](int* gridSize, int* blockSize) {
return hipOccupancyMaxPotentialBlockSize(gridSize, blockSize, f1, 0, 0);
},
devProp.maxThreadsPerBlock);
}
SECTION("dynSharedMemPerBlk = sharedMemPerBlock, blockSizeLimit = maxThreadsPerBlock") {
MaxPotentialBlockSize(
[devProp](int* gridSize, int* blockSize) {
return hipOccupancyMaxPotentialBlockSize(
gridSize, blockSize, f1, devProp.sharedMemPerBlock, devProp.maxThreadsPerBlock);
},
devProp.maxThreadsPerBlock);
}
}
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_templateInvocation") {
int gridSize = 0, blockSize = 0;
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_Positive_TemplateInvocation") {
hipDeviceProp_t devProp;
HIP_CHECK(hipOccupancyMaxPotentialBlockSize<void(*)(int *)>(&gridSize,
&blockSize, f2, 0, 0));
REQUIRE(gridSize > 0);
REQUIRE(blockSize > 0);
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
SECTION("dynSharedMemPerBlk = 0, blockSizeLimit = 0") {
MaxPotentialBlockSize(
[](int* gridSize, int* blockSize) {
return hipOccupancyMaxPotentialBlockSize<void (*)(int*)>(gridSize, blockSize, f2, 0, 0);
},
devProp.maxThreadsPerBlock);
}
SECTION("dynSharedMemPerBlk = sharedMemPerBlock, blockSizeLimit = maxThreadsPerBlock") {
MaxPotentialBlockSize(
[devProp](int* gridSize, int* blockSize) {
return hipOccupancyMaxPotentialBlockSize<void (*)(int*)>(
gridSize, blockSize, f2, devProp.sharedMemPerBlock, devProp.maxThreadsPerBlock);
},
devProp.maxThreadsPerBlock);
}
}
catch/unit/occupancy/hipOccupancyMaxPotentialBlockSize_old.cc
+80
View File
@@ -0,0 +1,80 @@
/*
Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include <hip_test_common.hh>
static __global__ void f1(float *a) { *a = 1.0; }
template <typename T>
static __global__ void f2(T *a) { *a = 1; }
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_Negative") {
hipError_t ret;
int blockSize = 0;
int gridSize = 0;
// Validate each argument
ret = hipOccupancyMaxPotentialBlockSize(NULL, &blockSize, f1, 0, 0);
REQUIRE(ret != hipSuccess);
ret = hipOccupancyMaxPotentialBlockSize(&gridSize, NULL, f1, 0, 0);
REQUIRE(ret != hipSuccess);
#ifndef __HIP_PLATFORM_NVIDIA__
// nvcc doesnt support kernelfunc(NULL) for api
ret = hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, NULL, 0, 0);
REQUIRE(ret != hipSuccess);
#endif
}
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_rangeValidation") {
hipDeviceProp_t devProp;
int blockSize = 0;
int gridSize = 0;
// Get potential blocksize
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0));
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
// Check if blockSize doen't exceed maxThreadsPerBlock
REQUIRE(gridSize > 0); REQUIRE(blockSize > 0);
REQUIRE(blockSize <= devProp.maxThreadsPerBlock);
// Pass dynSharedMemPerBlk, blockSizeLimit and check out param
blockSize = 0;
gridSize = 0;
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1,
devProp.sharedMemPerBlock, devProp.maxThreadsPerBlock));
// Check if blockSize doen't exceed maxThreadsPerBlock
REQUIRE(gridSize > 0); REQUIRE(blockSize > 0);
REQUIRE(blockSize <= devProp.maxThreadsPerBlock);
}
TEST_CASE("Unit_hipOccupancyMaxPotentialBlockSize_templateInvocation") {
int gridSize = 0, blockSize = 0;
HIP_CHECK(hipOccupancyMaxPotentialBlockSize<void(*)(int *)>(&gridSize,
&blockSize, f2, 0, 0));
REQUIRE(gridSize > 0);
REQUIRE(blockSize > 0);
}
catch/unit/occupancy/occupancy_common.hh
+72
View File
@@ -0,0 +1,72 @@
/*
Copyright (c) 2022 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include <hip_test_common.hh>
template <typename F> void MaxPotentialBlockSize(F func, int maxThreadsPerBlock) {
int gridSize = 0;
int blockSize = 0;
// Get potential blocksize
HIP_CHECK(func(&gridSize, &blockSize));
// Check if blockSize doesn't exceed maxThreadsPerBlock
REQUIRE(gridSize > 0);
REQUIRE(blockSize > 0);
REQUIRE(blockSize <= maxThreadsPerBlock);
REQUIRE(gridSize * blockSize < static_cast<int64_t>(std::pow(2, 32)));
}
template <typename F> void MaxPotentialBlockSizeNegative(F func) {
int blockSize = 0;
int gridSize = 0;
// Validate common arguments
SECTION("gridSize is nullptr") {
HIP_CHECK_ERROR(func(nullptr, &blockSize), hipErrorInvalidValue);
}
SECTION("blockSize is nullptr") {
HIP_CHECK_ERROR(func(&gridSize, nullptr), hipErrorInvalidValue);
}
}
template <typename F>
void MaxActiveBlocksPerMultiprocessor(F func, int blockSize, int maxThreadsPerMultiProcessor) {
int numBlocks = 0;
// Validate maximum active block pre multiprocessor
HIP_CHECK(func(&numBlocks));
// Check if numBlocks and blockSize are within limits
REQUIRE(numBlocks > 0);
REQUIRE((numBlocks * blockSize) <= maxThreadsPerMultiProcessor);
}
template <typename F> void MaxActiveBlocksPerMultiprocessorNegative(F func, int blockSize) {
int numBlocks = 0;
// Validate common arguments
SECTION("numBlocks is nullptr") {
HIP_CHECK_ERROR(func(nullptr, blockSize, 0), hipErrorInvalidValue);
}
SECTION("Block size is 0") {
HIP_CHECK_ERROR(func(&numBlocks, 0, 0), hipErrorInvalidValue);
}
}