blob: 74f6e407153d465d4e9a881dc85a528e10472cdf [file] [log] [blame]
//
// Copyright (c) 2017 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
#include "../../test_common/harness/compat.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include "../../test_common/harness/conversions.h"
#include "procs.h"
static const char *import_after_export_aliased_local_kernel =
"#pragma OPENCL EXTENSION cl_khr_async_work_group_copy_fence : enable\n"
"%s\n" // optional pragma string
"__kernel void test_fn( const __global %s *exportSrc, __global %s "
"*exportDst,\n"
" const __global %s *importSrc, __global %s "
"*importDst,\n"
" __local %s *localBuffer, /* there isn't another "
"__local %s local buffer since export src and import dst are aliased*/\n"
" int exportSrcLocalSize, int "
"exportCopiesPerWorkItem,\n"
" int importSrcLocalSize, int "
"importCopiesPerWorkItem )\n"
"{\n"
" int i;\n"
" int localImportOffset = exportSrcLocalSize - importSrcLocalSize;\n"
// Zero the local storage first
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" localBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] = "
"(%s)(%s)0;\n"
" }\n"
" // no need to set another local buffer values to (%s)(%s)0 since "
"export src and import dst are aliased (use the same buffer)\n"
// Do this to verify all kernels are done zeroing the local buffer before we
// try the export and import
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" localBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] = "
"exportSrc[ get_global_id( 0 )*exportCopiesPerWorkItem+i ];\n"
" }\n"
// Do this to verify all kernels are done copying to the local buffer before
// we try the export and import
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" event_t events;\n"
" events = async_work_group_copy((__global "
"%s*)(exportDst+exportSrcLocalSize*get_group_id(0)), (__local const "
"%s*)localBuffer, (size_t)exportSrcLocalSize, 0 );\n"
" async_work_group_copy_fence( CLK_LOCAL_MEM_FENCE );\n"
" events = async_work_group_copy( (__local "
"%s*)(localBuffer+localImportOffset), (__global const "
"%s*)(importSrc+importSrcLocalSize*get_group_id(0)), "
"(size_t)importSrcLocalSize, events );\n"
// Wait for the export and import to complete, then verify by manually
// copying to the dest
" wait_group_events( 2, &events );\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importDst[ get_global_id( 0 )*importCopiesPerWorkItem+i ] = "
"(localBuffer+localImportOffset)[ get_local_id( 0 "
")*importCopiesPerWorkItem+i ];\n"
" }\n"
"}\n";
static const char *import_after_export_aliased_global_kernel =
"#pragma OPENCL EXTENSION cl_khr_async_work_group_copy_fence : enable\n"
"%s\n" // optional pragma string
"__kernel void test_fn( const __global %s *exportSrc, __global %s "
"*exportDstImportSrc,\n"
" __global %s *importDst, /* there isn't a dedicated "
"__global %s buffer for import src since export dst and import src are "
"aliased*/\n"
" __local %s *exportLocalBuffer, __local %s "
"*importLocalBuffer,\n"
" int exportSrcLocalSize, int "
"exportCopiesPerWorkItem,\n"
" int importSrcLocalSize, int "
"importCopiesPerWorkItem )\n"
"{\n"
" int i;\n"
// Zero the local storage first
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" exportLocalBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] "
"= (%s)(%s)0;\n"
" }\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importLocalBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ] "
"= (%s)(%s)0;\n"
" }\n"
// Do this to verify all kernels are done zeroing the local buffer before we
// try the export and import
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" exportLocalBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] "
"= exportSrc[ get_global_id( 0 )*exportCopiesPerWorkItem+i ];\n"
" }\n"
// Do this to verify all kernels are done copying to the local buffer before
// we try the export and import
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" event_t events;\n"
" events = async_work_group_copy((__global "
"%s*)(exportDstImportSrc+exportSrcLocalSize*get_group_id(0)), (__local "
"const %s*)exportLocalBuffer, (size_t)exportSrcLocalSize, 0 );\n"
" async_work_group_copy_fence( CLK_GLOBAL_MEM_FENCE );\n"
" events = async_work_group_copy( (__local %s*)importLocalBuffer, "
"(__global const "
"%s*)(exportDstImportSrc+exportSrcLocalSize*get_group_id(0) + "
"(exportSrcLocalSize - importSrcLocalSize)), (size_t)importSrcLocalSize, "
"events );\n"
// Wait for the export and import to complete, then verify by manually
// copying to the dest
" wait_group_events( 2, &events );\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importDst[ get_global_id( 0 )*importCopiesPerWorkItem+i ] = "
"importLocalBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ];\n"
" }\n"
"}\n";
static const char *import_after_export_aliased_global_and_local_kernel =
"#pragma OPENCL EXTENSION cl_khr_async_work_group_copy_fence : enable\n"
"%s\n" // optional pragma string
"__kernel void test_fn( const __global %s *exportSrc, __global %s "
"*exportDstImportSrc,\n"
" __global %s *importDst, /* there isn't a dedicated "
"__global %s buffer for import src since export dst and import src are "
"aliased*/\n"
" __local %s *localBuffer, /* there isn't another "
"__local %s local buffer since export src and import dst are aliased*/\n"
" int exportSrcLocalSize, int "
"exportCopiesPerWorkItem,\n"
" int importSrcLocalSize, int "
"importCopiesPerWorkItem )\n"
"{\n"
" int i;\n"
" int localImportOffset = exportSrcLocalSize - importSrcLocalSize;\n"
// Zero the local storage first
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" localBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] = "
"(%s)(%s)0;\n"
" }\n"
" // no need to set another local buffer values to (%s)(%s)0 since "
"export src and import dst are aliased (use the same buffer)\n"
// Do this to verify all kernels are done zeroing the local buffer before we
// try the export and import
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" localBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] = "
"exportSrc[ get_global_id( 0 )*exportCopiesPerWorkItem+i ];\n"
" }\n"
// Do this to verify all kernels are done copying to the local buffer before
// we try the export and import
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" event_t events;\n"
" events = async_work_group_copy((__global "
"%s*)(exportDstImportSrc+exportSrcLocalSize*get_group_id(0)), (__local "
"const %s*)localBuffer, (size_t)exportSrcLocalSize, 0 );\n"
" async_work_group_copy_fence( CLK_GLOBAL_MEM_FENCE | "
"CLK_LOCAL_MEM_FENCE );\n"
" events = async_work_group_copy( (__local "
"%s*)(localBuffer+localImportOffset), (__global const "
"%s*)(exportDstImportSrc+exportSrcLocalSize*get_group_id(0) + "
"(exportSrcLocalSize - importSrcLocalSize)), (size_t)importSrcLocalSize, "
"events );\n"
// Wait for the export and import to complete, then verify by manually
// copying to the dest
" wait_group_events( 2, &events );\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importDst[ get_global_id( 0 )*importCopiesPerWorkItem+i ] = "
"(localBuffer+localImportOffset)[ get_local_id( 0 "
")*importCopiesPerWorkItem+i ];\n"
" }\n"
"}\n";
static const char *export_after_import_aliased_local_kernel =
"#pragma OPENCL EXTENSION cl_khr_async_work_group_copy_fence : enable\n"
"%s\n" // optional pragma string
"__kernel void test_fn( const __global %s *importSrc, __global %s "
"*importDst,\n"
" const __global %s *exportDst, /* there isn't a "
"dedicated __global %s buffer for export src since the local memory is "
"aliased, so the export src is taken from it */\n"
" __local %s *localBuffer, /* there isn't another "
"__local %s local buffer since import dst and export src are aliased*/\n"
" int importSrcLocalSize, int "
"importCopiesPerWorkItem,\n"
" int exportSrcLocalSize, int "
"exportCopiesPerWorkItem )\n"
"{\n"
" int i;\n"
// Zero the local storage first
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" localBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ] = "
"(%s)(%s)0;\n"
" }\n"
" // no need to set another local buffer values to (%s)(%s)0 since "
"import dst and export src are aliased (use the same buffer)\n"
// Do this to verify all kernels are done zeroing the local buffer before we
// try the import and export
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" event_t events;\n"
" events = async_work_group_copy( (__local %s*)localBuffer, (__global "
"const %s*)(importSrc+importSrcLocalSize*get_group_id(0)), "
"(size_t)importSrcLocalSize, events );\n"
" async_work_group_copy_fence( CLK_LOCAL_MEM_FENCE );\n"
" events = async_work_group_copy((__global "
"%s*)(exportDst+exportSrcLocalSize*get_group_id(0)), (__local const "
"%s*)(localBuffer + (importSrcLocalSize - exportSrcLocalSize)), "
"(size_t)exportSrcLocalSize, 0 );\n"
// Wait for the import and export to complete, then verify by manually
// copying to the dest
" wait_group_events( 2, &events );\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importDst[ get_global_id( 0 )*importCopiesPerWorkItem+i ] = "
"localBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ];\n"
" }\n"
"}\n";
static const char *export_after_import_aliased_global_kernel =
"#pragma OPENCL EXTENSION cl_khr_async_work_group_copy_fence : enable\n"
"%s\n" // optional pragma string
"__kernel void test_fn( const __global %s *importSrcExportDst, __global %s "
"*importDst,\n"
" const __global %s *exportSrc,\n"
" /* there isn't a dedicated __global %s buffer for "
"export dst since import src and export dst are aliased */\n"
" __local %s *importLocalBuffer, __local %s "
"*exportLocalBuffer,\n"
" int importSrcLocalSize, int "
"importCopiesPerWorkItem,\n"
" int exportSrcLocalSize, int "
"exportCopiesPerWorkItem )\n"
"{\n"
" int i;\n"
// Zero the local storage first
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importLocalBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ] "
"= (%s)(%s)0;\n"
" }\n"
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" exportLocalBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] "
"= (%s)(%s)0;\n"
" }\n"
// Do this to verify all kernels are done zeroing the local buffer before we
// try the import and export
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" for(i=0; i<exportCopiesPerWorkItem; i++) {\n"
" exportLocalBuffer[ get_local_id( 0 )*exportCopiesPerWorkItem+i ] "
"= exportSrc[ get_global_id( 0 )*exportCopiesPerWorkItem+i ];\n"
" }\n"
// Do this to verify all kernels are done copying to the local buffer before
// we try the import and export
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" event_t events;\n"
" events = async_work_group_copy( (__local %s*)importLocalBuffer, "
"(__global const "
"%s*)(importSrcExportDst+importSrcLocalSize*get_group_id(0)), "
"(size_t)importSrcLocalSize, 0 );\n"
" async_work_group_copy_fence( CLK_GLOBAL_MEM_FENCE );\n"
" events = async_work_group_copy((__global "
"%s*)(importSrcExportDst+importSrcLocalSize*get_group_id(0) + "
"(importSrcLocalSize - exportSrcLocalSize)), (__local const "
"%s*)exportLocalBuffer, (size_t)exportSrcLocalSize, events );\n"
// Wait for the import and export to complete, then verify by manually
// copying to the dest
" wait_group_events( 2, &events );\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importDst[ get_global_id( 0 )*importCopiesPerWorkItem+i ] = "
"importLocalBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ];\n"
" }\n"
"}\n";
static const char *export_after_import_aliased_global_and_local_kernel =
"#pragma OPENCL EXTENSION cl_khr_async_work_group_copy_fence : enable\n"
"%s\n" // optional pragma string
"__kernel void test_fn( const __global %s *importSrcExportDst, __global %s "
"*importDst,\n"
" /* there isn't a dedicated __global %s buffer for "
"export src since the local memory is aliased, so the export src is taken "
"from it */\n"
" /* there isn't a dedicated __global %s buffer for "
"export dst since import src and export dst are aliased */\n"
" __local %s *localBuffer, /* there isn't another "
"__local %s local buffer since import dst and export src are aliased*/\n"
" int importSrcLocalSize, int "
"importCopiesPerWorkItem,\n"
" int exportSrcLocalSize, int "
"exportCopiesPerWorkItem )\n"
"{\n"
" int i;\n"
// Zero the local storage first
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" localBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ] = "
"(%s)(%s)0;\n"
" }\n"
" // no need to set another local buffer values to (%s)(%s)0 since "
"import dst and export src are aliased (use the same buffer)\n"
// Do this to verify all kernels are done zeroing the local buffer before we
// try the import and export
" barrier( CLK_LOCAL_MEM_FENCE );\n"
" event_t events;\n"
" events = async_work_group_copy( (__local %s*)localBuffer, (__global "
"const %s*)(importSrcExportDst+importSrcLocalSize*get_group_id(0)), "
"(size_t)importSrcLocalSize, 0 );\n"
" async_work_group_copy_fence( CLK_GLOBAL_MEM_FENCE | "
"CLK_LOCAL_MEM_FENCE );\n"
" events = async_work_group_copy((__global "
"%s*)(importSrcExportDst+importSrcLocalSize*get_group_id(0) + "
"(importSrcLocalSize - exportSrcLocalSize)), (__local const "
"%s*)(localBuffer + (importSrcLocalSize - exportSrcLocalSize)), "
"(size_t)exportSrcLocalSize, events );\n"
// Wait for the import and export to complete, then verify by manually
// copying to the dest
" wait_group_events( 2, &events );\n"
" for(i=0; i<importCopiesPerWorkItem; i++) {\n"
" importDst[ get_global_id( 0 )*importCopiesPerWorkItem+i ] = "
"localBuffer[ get_local_id( 0 )*importCopiesPerWorkItem+i ];\n"
" }\n"
"}\n";
int test_copy_fence(cl_device_id deviceID, cl_context context,
cl_command_queue queue, const char *kernelCode,
ExplicitType vecType, int vecSize, bool export_after_import,
bool aliased_local_mem, bool aliased_global_mem)
{
int error;
clProgramWrapper program;
clKernelWrapper kernel;
clMemWrapper streams[4];
size_t threads[1], localThreads[1];
void *transaction1InBuffer, *transaction1OutBuffer, *transaction2InBuffer,
*transaction2OutBuffer;
MTdata d;
bool transaction1DstIsTransaction2Src =
(aliased_global_mem && !export_after_import)
|| (aliased_local_mem && export_after_import);
bool transaction1SrcIsTransaction2Dst =
aliased_global_mem && export_after_import;
char vecNameString[64];
vecNameString[0] = 0;
if (vecSize == 1)
sprintf(vecNameString, "%s", get_explicit_type_name(vecType));
else
sprintf(vecNameString, "%s%d", get_explicit_type_name(vecType),
vecSize);
size_t elementSize = get_explicit_type_size(vecType) * vecSize;
log_info("Testing %s\n", vecNameString);
if (!is_extension_available(deviceID, "cl_khr_async_work_group_copy_fence"))
{
log_info(
"Device does not support extended async copies fence. Skipping "
"test.\n");
return 0;
}
cl_long max_local_mem_size;
error =
clGetDeviceInfo(deviceID, CL_DEVICE_LOCAL_MEM_SIZE,
sizeof(max_local_mem_size), &max_local_mem_size, NULL);
test_error(error, "clGetDeviceInfo for CL_DEVICE_LOCAL_MEM_SIZE failed.");
unsigned int num_of_compute_devices;
error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_COMPUTE_UNITS,
sizeof(num_of_compute_devices),
&num_of_compute_devices, NULL);
test_error(error,
"clGetDeviceInfo for CL_DEVICE_MAX_COMPUTE_UNITS failed.");
char programSource[4096];
programSource[0] = 0;
char *programPtr;
sprintf(programSource, kernelCode,
vecType == kDouble ? "#pragma OPENCL EXTENSION cl_khr_fp64 : enable"
: "",
vecNameString, vecNameString, vecNameString, vecNameString,
vecNameString, vecNameString, vecNameString,
get_explicit_type_name(vecType), vecNameString,
get_explicit_type_name(vecType), vecNameString, vecNameString,
vecNameString, vecNameString);
// log_info("program: %s\n", programSource);
programPtr = programSource;
error = create_single_kernel_helper(context, &program, &kernel, 1,
(const char **)&programPtr, "test_fn");
test_error(error, "Unable to create testing kernel");
size_t max_workgroup_size;
error = clGetKernelWorkGroupInfo(
kernel, deviceID, CL_KERNEL_WORK_GROUP_SIZE, sizeof(max_workgroup_size),
&max_workgroup_size, NULL);
test_error(
error,
"clGetKernelWorkGroupInfo failed for CL_KERNEL_WORK_GROUP_SIZE.");
size_t max_local_workgroup_size[3];
error = clGetDeviceInfo(deviceID, CL_DEVICE_MAX_WORK_ITEM_SIZES,
sizeof(max_local_workgroup_size),
max_local_workgroup_size, NULL);
test_error(error,
"clGetDeviceInfo failed for CL_DEVICE_MAX_WORK_ITEM_SIZES");
// Pick the minimum of the device and the kernel
if (max_workgroup_size > max_local_workgroup_size[0])
max_workgroup_size = max_local_workgroup_size[0];
size_t transaction1NumberOfCopiesPerWorkitem = 13;
size_t transaction2NumberOfCopiesPerWorkitem = 2;
elementSize =
get_explicit_type_size(vecType) * ((vecSize == 3) ? 4 : vecSize);
size_t localStorageSpacePerWorkitem =
transaction1NumberOfCopiesPerWorkitem * elementSize
+ (aliased_local_mem
? 0
: transaction2NumberOfCopiesPerWorkitem * elementSize);
size_t maxLocalWorkgroupSize =
(((int)max_local_mem_size / 2) / localStorageSpacePerWorkitem);
// Calculation can return 0 on embedded devices due to 1KB local mem limit
if (maxLocalWorkgroupSize == 0)
{
maxLocalWorkgroupSize = 1;
}
size_t localWorkgroupSize = maxLocalWorkgroupSize;
if (maxLocalWorkgroupSize > max_workgroup_size)
localWorkgroupSize = max_workgroup_size;
size_t transaction1LocalBufferSize = localWorkgroupSize * elementSize
* transaction1NumberOfCopiesPerWorkitem;
size_t transaction2LocalBufferSize = localWorkgroupSize * elementSize
* transaction2NumberOfCopiesPerWorkitem; // irrelevant if
// aliased_local_mem
size_t numberOfLocalWorkgroups = 1111;
size_t transaction1GlobalBufferSize =
numberOfLocalWorkgroups * transaction1LocalBufferSize;
size_t transaction2GlobalBufferSize =
numberOfLocalWorkgroups * transaction2LocalBufferSize;
size_t globalWorkgroupSize = numberOfLocalWorkgroups * localWorkgroupSize;
transaction1InBuffer = (void *)malloc(transaction1GlobalBufferSize);
transaction1OutBuffer = (void *)malloc(transaction1GlobalBufferSize);
transaction2InBuffer = (void *)malloc(transaction2GlobalBufferSize);
transaction2OutBuffer = (void *)malloc(transaction2GlobalBufferSize);
memset(transaction1OutBuffer, 0, transaction1GlobalBufferSize);
memset(transaction2OutBuffer, 0, transaction2GlobalBufferSize);
cl_int transaction1CopiesPerWorkitemInt, transaction1CopiesPerWorkgroup,
transaction2CopiesPerWorkitemInt, transaction2CopiesPerWorkgroup;
transaction1CopiesPerWorkitemInt =
(int)transaction1NumberOfCopiesPerWorkitem;
transaction1CopiesPerWorkgroup =
(int)(transaction1NumberOfCopiesPerWorkitem * localWorkgroupSize);
transaction2CopiesPerWorkitemInt =
(int)transaction2NumberOfCopiesPerWorkitem;
transaction2CopiesPerWorkgroup =
(int)(transaction2NumberOfCopiesPerWorkitem * localWorkgroupSize);
log_info(
"Global: %d, local %d. 1st Transaction: local buffer %db, global "
"buffer %db, each work group will copy %d elements and each work "
"item item will copy %d elements. 2nd Transaction: local buffer "
"%db, global buffer %db, each work group will copy %d elements and "
"each work item will copy %d elements\n",
(int)globalWorkgroupSize, (int)localWorkgroupSize,
(int)transaction1LocalBufferSize, (int)transaction1GlobalBufferSize,
transaction1CopiesPerWorkgroup, transaction1CopiesPerWorkitemInt,
(int)transaction2LocalBufferSize, (int)transaction2GlobalBufferSize,
transaction2CopiesPerWorkgroup, transaction2CopiesPerWorkitemInt);
threads[0] = globalWorkgroupSize;
localThreads[0] = localWorkgroupSize;
d = init_genrand(gRandomSeed);
generate_random_data(
vecType, transaction1GlobalBufferSize / get_explicit_type_size(vecType),
d, transaction1InBuffer);
if (!transaction1DstIsTransaction2Src)
{
generate_random_data(vecType,
transaction2GlobalBufferSize
/ get_explicit_type_size(vecType),
d, transaction2InBuffer);
}
free_mtdata(d);
d = NULL;
streams[0] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
transaction1GlobalBufferSize,
transaction1InBuffer, &error);
test_error(error, "Unable to create input buffer");
streams[1] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
transaction1GlobalBufferSize,
transaction1OutBuffer, &error);
test_error(error, "Unable to create output buffer");
if (!transaction1DstIsTransaction2Src)
{
streams[2] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
transaction2GlobalBufferSize,
transaction2InBuffer, &error);
test_error(error, "Unable to create input buffer");
}
if (!transaction1SrcIsTransaction2Dst)
{
streams[3] = clCreateBuffer(context, CL_MEM_COPY_HOST_PTR,
transaction2GlobalBufferSize,
transaction2OutBuffer, &error);
test_error(error, "Unable to create output buffer");
}
cl_uint argIndex = 0;
error = clSetKernelArg(kernel, argIndex, sizeof(streams[0]), &streams[0]);
test_error(error, "Unable to set kernel argument");
++argIndex;
error = clSetKernelArg(kernel, argIndex, sizeof(streams[1]), &streams[1]);
test_error(error, "Unable to set kernel argument");
++argIndex;
if (!transaction1DstIsTransaction2Src)
{
error =
clSetKernelArg(kernel, argIndex, sizeof(streams[2]), &streams[2]);
test_error(error, "Unable to set kernel argument");
++argIndex;
}
if (!transaction1SrcIsTransaction2Dst)
{
error =
clSetKernelArg(kernel, argIndex, sizeof(streams[3]), &streams[3]);
test_error(error, "Unable to set kernel argument");
++argIndex;
}
error = clSetKernelArg(kernel, argIndex, transaction1LocalBufferSize, NULL);
test_error(error, "Unable to set kernel argument");
++argIndex;
if (!aliased_local_mem)
{
error =
clSetKernelArg(kernel, argIndex, transaction2LocalBufferSize, NULL);
test_error(error, "Unable to set kernel argument");
++argIndex;
}
error =
clSetKernelArg(kernel, argIndex, sizeof(transaction1CopiesPerWorkgroup),
&transaction1CopiesPerWorkgroup);
test_error(error, "Unable to set kernel argument");
++argIndex;
error = clSetKernelArg(kernel, argIndex,
sizeof(transaction1CopiesPerWorkitemInt),
&transaction1CopiesPerWorkitemInt);
test_error(error, "Unable to set kernel argument");
++argIndex;
error =
clSetKernelArg(kernel, argIndex, sizeof(transaction2CopiesPerWorkgroup),
&transaction2CopiesPerWorkgroup);
test_error(error, "Unable to set kernel argument");
++argIndex;
error = clSetKernelArg(kernel, argIndex,
sizeof(transaction2CopiesPerWorkitemInt),
&transaction2CopiesPerWorkitemInt);
test_error(error, "Unable to set kernel argument");
// Enqueue
error = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, threads,
localThreads, 0, NULL, NULL);
test_error(error, "Unable to queue kernel");
// Read
error = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0,
transaction1GlobalBufferSize,
transaction1OutBuffer, 0, NULL, NULL);
test_error(error, "Unable to read results");
if (transaction1DstIsTransaction2Src)
{
for (size_t idx = 0; idx < numberOfLocalWorkgroups; idx++)
{
memcpy(
(void *)((unsigned char *)transaction2InBuffer
+ idx * transaction2CopiesPerWorkgroup * elementSize),
(const void *)((unsigned char *)transaction1OutBuffer
+ (idx * transaction1CopiesPerWorkgroup
+ (transaction1CopiesPerWorkgroup
- transaction2CopiesPerWorkgroup))
* elementSize),
(size_t)transaction2CopiesPerWorkgroup * elementSize);
}
}
if (transaction1SrcIsTransaction2Dst)
{
void *transaction1SrcBuffer =
(void *)malloc(transaction1GlobalBufferSize);
error = clEnqueueReadBuffer(queue, streams[0], CL_TRUE, 0,
transaction1GlobalBufferSize,
transaction1SrcBuffer, 0, NULL, NULL);
test_error(error, "Unable to read results");
for (size_t idx = 0; idx < numberOfLocalWorkgroups; idx++)
{
memcpy(
(void *)((unsigned char *)transaction2OutBuffer
+ idx * transaction2CopiesPerWorkgroup * elementSize),
(const void *)((unsigned char *)transaction1SrcBuffer
+ (idx * transaction1CopiesPerWorkgroup
+ (transaction1CopiesPerWorkgroup
- transaction2CopiesPerWorkgroup))
* elementSize),
(size_t)transaction2CopiesPerWorkgroup * elementSize);
}
free(transaction1SrcBuffer);
}
else
{
error = clEnqueueReadBuffer(queue, streams[3], CL_TRUE, 0,
transaction2GlobalBufferSize,
transaction2OutBuffer, 0, NULL, NULL);
test_error(error, "Unable to read results");
}
// Verify
int failuresPrinted = 0;
if (memcmp(transaction1InBuffer, transaction1OutBuffer,
transaction1GlobalBufferSize)
!= 0)
{
size_t typeSize = get_explicit_type_size(vecType) * vecSize;
unsigned char *inchar = (unsigned char *)transaction1InBuffer;
unsigned char *outchar = (unsigned char *)transaction1OutBuffer;
for (int i = 0; i < (int)transaction1GlobalBufferSize;
i += (int)elementSize)
{
if (memcmp(((char *)inchar) + i, ((char *)outchar) + i, typeSize)
!= 0)
{
char values[4096];
values[0] = 0;
if (failuresPrinted == 0)
{
// Print first failure message
log_error("ERROR: Results of 1st transaction did not "
"validate!\n");
}
sprintf(values + strlen(values), "%d -> [", i);
for (int j = 0; j < (int)elementSize; j++)
sprintf(values + strlen(values), "%2x ", inchar[i + j]);
sprintf(values + strlen(values), "] != [");
for (int j = 0; j < (int)elementSize; j++)
sprintf(values + strlen(values), "%2x ", outchar[i + j]);
sprintf(values + strlen(values), "]");
log_error("%s\n", values);
failuresPrinted++;
}
if (failuresPrinted > 5)
{
log_error("Not printing further failures...\n");
break;
}
}
}
if (memcmp(transaction2InBuffer, transaction2OutBuffer,
transaction2GlobalBufferSize)
!= 0)
{
size_t typeSize = get_explicit_type_size(vecType) * vecSize;
unsigned char *inchar = (unsigned char *)transaction2InBuffer;
unsigned char *outchar = (unsigned char *)transaction2OutBuffer;
for (int i = 0; i < (int)transaction2GlobalBufferSize;
i += (int)elementSize)
{
if (memcmp(((char *)inchar) + i, ((char *)outchar) + i, typeSize)
!= 0)
{
char values[4096];
values[0] = 0;
if (failuresPrinted == 0)
{
// Print first failure message
log_error("ERROR: Results of 2nd transaction did not "
"validate!\n");
}
sprintf(values + strlen(values), "%d -> [", i);
for (int j = 0; j < (int)elementSize; j++)
sprintf(values + strlen(values), "%2x ", inchar[i + j]);
sprintf(values + strlen(values), "] != [");
for (int j = 0; j < (int)elementSize; j++)
sprintf(values + strlen(values), "%2x ", outchar[i + j]);
sprintf(values + strlen(values), "]");
log_error("%s\n", values);
failuresPrinted++;
}
if (failuresPrinted > 5)
{
log_error("Not printing further failures...\n");
break;
}
}
}
free(transaction1InBuffer);
free(transaction1OutBuffer);
free(transaction2InBuffer);
free(transaction2OutBuffer);
return failuresPrinted ? -1 : 0;
}
int test_copy_fence_all_types(cl_device_id deviceID, cl_context context,
cl_command_queue queue, const char *kernelCode,
bool export_after_import, bool aliased_local_mem,
bool aliased_global_mem)
{
ExplicitType vecType[] = {
kChar, kUChar, kShort, kUShort, kInt, kUInt, kLong,
kULong, kFloat, kDouble, kNumExplicitTypes
};
unsigned int vecSizes[] = { 1, 2, 3, 4, 8, 16, 0 };
unsigned int size, typeIndex;
int errors = 0;
for (typeIndex = 0; vecType[typeIndex] != kNumExplicitTypes; typeIndex++)
{
if (vecType[typeIndex] == kDouble
&& !is_extension_available(deviceID, "cl_khr_fp64"))
continue;
if ((vecType[typeIndex] == kLong || vecType[typeIndex] == kULong)
&& !gHasLong)
continue;
for (size = 0; vecSizes[size] != 0; size++)
{
if (test_copy_fence(deviceID, context, queue, kernelCode,
vecType[typeIndex], vecSizes[size],
export_after_import, aliased_local_mem,
aliased_global_mem))
{
errors++;
}
}
}
if (errors) return -1;
return 0;
}
int test_async_work_group_copy_fence_import_after_export_aliased_local(
cl_device_id deviceID, cl_context context, cl_command_queue queue,
int num_elements)
{
return test_copy_fence_all_types(deviceID, context, queue,
import_after_export_aliased_local_kernel,
false, true, false);
}
int test_async_work_group_copy_fence_import_after_export_aliased_global(
cl_device_id deviceID, cl_context context, cl_command_queue queue,
int num_elements)
{
return test_copy_fence_all_types(deviceID, context, queue,
import_after_export_aliased_global_kernel,
false, false, true);
}
int test_async_work_group_copy_fence_import_after_export_aliased_global_and_local(
cl_device_id deviceID, cl_context context, cl_command_queue queue,
int num_elements)
{
return test_copy_fence_all_types(
deviceID, context, queue,
import_after_export_aliased_global_and_local_kernel, false, true, true);
}
int test_async_work_group_copy_fence_export_after_import_aliased_local(
cl_device_id deviceID, cl_context context, cl_command_queue queue,
int num_elements)
{
return test_copy_fence_all_types(deviceID, context, queue,
export_after_import_aliased_local_kernel,
true, true, false);
}
int test_async_work_group_copy_fence_export_after_import_aliased_global(
cl_device_id deviceID, cl_context context, cl_command_queue queue,
int num_elements)
{
return test_copy_fence_all_types(deviceID, context, queue,
export_after_import_aliased_global_kernel,
true, false, true);
}
int test_async_work_group_copy_fence_export_after_import_aliased_global_and_local(
cl_device_id deviceID, cl_context context, cl_command_queue queue,
int num_elements)
{
return test_copy_fence_all_types(
deviceID, context, queue,
export_after_import_aliased_global_and_local_kernel, true, true, true);
}