395 lines
16 KiB
C++
395 lines
16 KiB
C++
//
|
|
// 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 <string.h>
|
|
#include "cl_utils.h"
|
|
#include "tests.h"
|
|
#include "harness/testHarness.h"
|
|
|
|
int test_roundTrip( cl_device_id device, cl_context context, cl_command_queue queue, int num_elements )
|
|
{
|
|
int vectorSize, error;
|
|
uint64_t i, j;
|
|
cl_program programs[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
cl_kernel kernels[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
cl_program doublePrograms[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
cl_kernel doubleKernels[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
uint64_t time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
uint64_t min_time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
uint64_t doubleTime[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
uint64_t min_double_time[kVectorSizeCount+kStrangeVectorSizeCount] = {0};
|
|
memset( min_time, -1, sizeof( min_time ) );
|
|
memset( min_double_time, -1, sizeof( min_double_time ) );
|
|
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
{
|
|
const char *source[] = {
|
|
"__kernel void test( const __global half *in, __global half *out )\n"
|
|
"{\n"
|
|
" size_t i = get_global_id(0);\n"
|
|
" vstore_half",vector_size_name_extensions[vectorSize],"( vload_half",vector_size_name_extensions[vectorSize],"(i, in), i, out);\n"
|
|
"}\n"
|
|
};
|
|
|
|
const char *doubleSource[] = {
|
|
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
|
"__kernel void test( const __global half *in, __global half *out )\n"
|
|
"{\n"
|
|
" size_t i = get_global_id(0);\n"
|
|
" vstore_half",vector_size_name_extensions[vectorSize],"( convert_double", vector_size_name_extensions[vectorSize], "( vload_half",vector_size_name_extensions[vectorSize],"(i, in)), i, out);\n"
|
|
"}\n"
|
|
};
|
|
|
|
const char *sourceV3[] = {
|
|
"__kernel void test( const __global half *in, __global half *out,"
|
|
" uint extra_last_thread )\n"
|
|
"{\n"
|
|
" size_t i = get_global_id(0);\n"
|
|
" size_t last_i = get_global_size(0)-1;\n"
|
|
" size_t adjust = 0;\n"
|
|
" if(i == last_i && extra_last_thread != 0) { \n"
|
|
" adjust = 3-extra_last_thread;\n"
|
|
" }\n"
|
|
" vstore_half3( vload_half3(i, in-adjust), i, out-adjust);\n"
|
|
"}\n"
|
|
};
|
|
|
|
const char *doubleSourceV3[] = {
|
|
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
|
"__kernel void test( const __global half *in, __global half *out,"
|
|
" uint extra_last_thread )\n"
|
|
"{\n"
|
|
" size_t i = get_global_id(0);\n"
|
|
" size_t last_i = get_global_size(0)-1;\n"
|
|
" size_t adjust = 0;\n"
|
|
" if(i == last_i && extra_last_thread != 0) { \n"
|
|
" adjust = 3-extra_last_thread;\n"
|
|
" }\n"
|
|
" vstore_half3( vload_half3(i, in-adjust), i, out-adjust);\n"
|
|
"}\n"
|
|
};
|
|
|
|
/*
|
|
const char *sourceV3aligned[] = {
|
|
"__kernel void test( const __global half *in, __global half *out )\n"
|
|
"{\n"
|
|
" size_t i = get_global_id(0);\n"
|
|
" vstorea_half3( vloada_half3(i, in), i, out);\n"
|
|
" vstore_half(vload_half(4*i+3, in), 4*i+3, out);\n"
|
|
"}\n"
|
|
};
|
|
|
|
const char *doubleSourceV3aligned[] = {
|
|
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
|
"__kernel void test( const __global half *in, __global half *out )\n"
|
|
"{\n"
|
|
" size_t i = get_global_id(0);\n"
|
|
" vstorea_half3( vloada_half3(i, in), i, out);\n"
|
|
" vstore_half(vload_half(4*i+3, in), 4*i+3, out);\n"
|
|
"}\n"
|
|
};
|
|
*/
|
|
|
|
if(g_arrVecSizes[vectorSize] == 3) {
|
|
programs[vectorSize] = MakeProgram( device, sourceV3, sizeof( sourceV3) / sizeof( sourceV3[0]) );
|
|
if( NULL == programs[ vectorSize ] )
|
|
{
|
|
gFailCount++;
|
|
|
|
return -1;
|
|
}
|
|
} else {
|
|
programs[vectorSize] = MakeProgram( device, source, sizeof( source) / sizeof( source[0]) );
|
|
if( NULL == programs[ vectorSize ] )
|
|
{
|
|
gFailCount++;
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
kernels[ vectorSize ] = clCreateKernel( programs[ vectorSize ], "test", &error );
|
|
if( NULL == kernels[vectorSize] )
|
|
{
|
|
gFailCount++;
|
|
vlog_error( "\t\tFAILED -- Failed to create kernel. (%d)\n", error );
|
|
return error;
|
|
}
|
|
|
|
if( gTestDouble )
|
|
{
|
|
if(g_arrVecSizes[vectorSize] == 3) {
|
|
doublePrograms[vectorSize] = MakeProgram( device, doubleSourceV3, sizeof( doubleSourceV3) / sizeof( doubleSourceV3[0]) );
|
|
if( NULL == doublePrograms[ vectorSize ] )
|
|
{
|
|
gFailCount++;
|
|
return -1;
|
|
}
|
|
} else {
|
|
doublePrograms[vectorSize] = MakeProgram( device, doubleSource, sizeof( doubleSource) / sizeof( doubleSource[0]) );
|
|
if( NULL == doublePrograms[ vectorSize ] )
|
|
{
|
|
gFailCount++;
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
doubleKernels[ vectorSize ] = clCreateKernel( doublePrograms[ vectorSize ], "test", &error );
|
|
if( NULL == doubleKernels[vectorSize] )
|
|
{
|
|
gFailCount++;
|
|
vlog_error( "\t\tFAILED -- Failed to create kernel. (%d)\n", error );
|
|
return error;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Figure out how many elements are in a work block
|
|
size_t elementSize = MAX( sizeof(cl_half), sizeof(cl_float));
|
|
size_t blockCount = (size_t)getBufferSize(device) / elementSize; //elementSize is a power of two
|
|
uint64_t lastCase = 1ULL << (8*sizeof(cl_half)); // number of cl_half
|
|
size_t stride = blockCount;
|
|
|
|
error = 0;
|
|
uint64_t printMask = (lastCase >> 4) - 1;
|
|
uint32_t count;
|
|
size_t loopCount;
|
|
|
|
for( i = 0; i < (uint64_t)lastCase; i += stride )
|
|
{
|
|
count = (uint32_t) MIN( blockCount, lastCase - i );
|
|
|
|
//Init the input stream
|
|
uint16_t *p = (uint16_t *)gIn_half;
|
|
for( j = 0; j < count; j++ )
|
|
p[j] = j + i;
|
|
|
|
if( (error = clEnqueueWriteBuffer(gQueue, gInBuffer_half, CL_TRUE, 0, count * sizeof( cl_half ), gIn_half, 0, NULL, NULL)) )
|
|
{
|
|
vlog_error( "Failure in clWriteArray\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
//Check the vector lengths
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
{ // here we loop through vector sizes -- 3 is last.
|
|
uint32_t pattern = 0xdeaddead;
|
|
memset_pattern4( gOut_half, &pattern, (size_t)getBufferSize(device)/2);
|
|
|
|
if( (error = clEnqueueWriteBuffer(gQueue, gOutBuffer_half, CL_TRUE, 0, count * sizeof(cl_half), gOut_half, 0, NULL, NULL)) )
|
|
{
|
|
vlog_error( "Failure in clWriteArray\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
|
|
// here is where "3" starts to cause problems.
|
|
error = RunKernel(device, kernels[vectorSize], gInBuffer_half, gOutBuffer_half, numVecs(count, vectorSize, false) ,
|
|
runsOverBy(count, vectorSize, false) );
|
|
if(error)
|
|
{
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
if( (error = clEnqueueReadBuffer(gQueue, gOutBuffer_half, CL_TRUE, 0, count * sizeof(cl_half), gOut_half, 0, NULL, NULL)) )
|
|
{
|
|
vlog_error( "Failure in clReadArray\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
if( (memcmp( gOut_half, gIn_half, count * sizeof(cl_half))) )
|
|
{
|
|
uint16_t *u1 = (uint16_t *)gOut_half;
|
|
uint16_t *u2 = (uint16_t *)gIn_half;
|
|
for( j = 0; j < count; j++ )
|
|
{
|
|
if( u1[j] != u2[j] )
|
|
{
|
|
uint16_t abs1 = u1[j] & 0x7fff;
|
|
uint16_t abs2 = u2[j] & 0x7fff;
|
|
if( abs1 > 0x7c00 && abs2 > 0x7c00 )
|
|
continue; //any NaN is okay if NaN is input
|
|
|
|
// if reference result is sub normal, test if the output is flushed to zero
|
|
if( IsHalfSubnormal(u2[j]) && ( (u1[j] == 0) || (u1[j] == 0x8000) ) )
|
|
continue;
|
|
|
|
vlog_error( "%lld) (of %lld) Failure at 0x%4.4x: 0x%4.4x vector_size = %d \n", j, (uint64_t)count, u2[j], u1[j], (g_arrVecSizes[vectorSize]) );
|
|
gFailCount++;
|
|
error = -1;
|
|
goto exit;
|
|
}
|
|
}
|
|
}
|
|
|
|
if( gTestDouble )
|
|
{
|
|
memset_pattern4( gOut_half, &pattern, (size_t)getBufferSize(device)/2);
|
|
if( (error = clEnqueueWriteBuffer(gQueue, gOutBuffer_half, CL_TRUE, 0, count * sizeof(cl_half), gOut_half, 0, NULL, NULL)) )
|
|
{
|
|
vlog_error( "Failure in clWriteArray\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
|
|
if( (error = RunKernel(device, doubleKernels[vectorSize], gInBuffer_half, gOutBuffer_half, numVecs(count, vectorSize, false) ,
|
|
runsOverBy(count, vectorSize, false) ) ) )
|
|
{
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
if( (error = clEnqueueReadBuffer(gQueue, gOutBuffer_half, CL_TRUE, 0, count * sizeof(cl_half), gOut_half, 0, NULL, NULL)) )
|
|
{
|
|
vlog_error( "Failure in clReadArray\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
if( (memcmp( gOut_half, gIn_half, count * sizeof(cl_half))) )
|
|
{
|
|
uint16_t *u1 = (uint16_t *)gOut_half;
|
|
uint16_t *u2 = (uint16_t *)gIn_half;
|
|
for( j = 0; j < count; j++ )
|
|
{
|
|
if( u1[j] != u2[j] )
|
|
{
|
|
uint16_t abs1 = u1[j] & 0x7fff;
|
|
uint16_t abs2 = u2[j] & 0x7fff;
|
|
if( abs1 > 0x7c00 && abs2 > 0x7c00 )
|
|
continue; //any NaN is okay if NaN is input
|
|
|
|
// if reference result is sub normal, test if the output is flushed to zero
|
|
if( IsHalfSubnormal(u2[j]) && ( (u1[j] == 0) || (u1[j] == 0x8000) ) )
|
|
continue;
|
|
|
|
vlog_error( "%lld) Failure at 0x%4.4x: 0x%4.4x vector_size = %d (double precsion)\n", j, u2[j], u1[j], (g_arrVecSizes[vectorSize]) );
|
|
gFailCount++;
|
|
error = -1;
|
|
goto exit;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if( ((i+blockCount) & ~printMask) == (i+blockCount) )
|
|
{
|
|
vlog( "." );
|
|
fflush( stdout );
|
|
}
|
|
}
|
|
|
|
vlog( "\n" );
|
|
|
|
loopCount = 100;
|
|
if( gReportTimes )
|
|
{
|
|
//Run again for timing
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
{
|
|
uint64_t bestTime = -1ULL;
|
|
|
|
for( j = 0; j < loopCount; j++ )
|
|
{
|
|
uint64_t startTime = ReadTime();
|
|
if( (error = RunKernel(device, kernels[vectorSize], gInBuffer_half, gOutBuffer_half,numVecs(count, vectorSize, false) ,
|
|
runsOverBy(count, vectorSize, false)) ) )
|
|
{
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
if( (error = clFinish(gQueue)) )
|
|
{
|
|
vlog_error( "Failure in clFinish\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
uint64_t currentTime = ReadTime() - startTime;
|
|
if( currentTime < bestTime )
|
|
bestTime = currentTime;
|
|
time[ vectorSize ] += currentTime;
|
|
}
|
|
if( bestTime < min_time[ vectorSize ] )
|
|
min_time[ vectorSize ] = bestTime;
|
|
|
|
if( gTestDouble )
|
|
{
|
|
bestTime = -1ULL;
|
|
for( j = 0; j < loopCount; j++ )
|
|
{
|
|
uint64_t startTime = ReadTime();
|
|
if( (error = RunKernel(device, doubleKernels[vectorSize], gInBuffer_half, gOutBuffer_half, numVecs(count, vectorSize, false) ,
|
|
runsOverBy(count, vectorSize, false)) ) )
|
|
{
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
|
|
if( (error = clFinish(gQueue)) )
|
|
{
|
|
vlog_error( "Failure in clFinish\n" );
|
|
gFailCount++;
|
|
goto exit;
|
|
}
|
|
uint64_t currentTime = ReadTime() - startTime;
|
|
if( currentTime < bestTime )
|
|
bestTime = currentTime;
|
|
doubleTime[ vectorSize ] += currentTime;
|
|
}
|
|
if( bestTime < min_double_time[ vectorSize ] )
|
|
min_double_time[ vectorSize ] = bestTime;
|
|
}
|
|
}
|
|
}
|
|
|
|
if( gReportTimes )
|
|
{
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
vlog_perf( SubtractTime( time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) (count * loopCount), 0, "average us/elem", "roundTrip avg. (vector size: %d)", (g_arrVecSizes[vectorSize]) );
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
vlog_perf( SubtractTime( min_time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) count, 0, "best us/elem", "roundTrip best (vector size: %d)", (g_arrVecSizes[vectorSize]) );
|
|
if( gTestDouble )
|
|
{
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
vlog_perf( SubtractTime( doubleTime[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) (count * loopCount), 0, "average us/elem (double)", "roundTrip avg. d (vector size: %d)", (g_arrVecSizes[vectorSize]) );
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
vlog_perf( SubtractTime( min_double_time[ vectorSize ], 0 ) * 1e6 * gDeviceFrequency * gComputeDevices / (double) count, 0, "best us/elem (double)", "roundTrip best d (vector size: %d)", (g_arrVecSizes[vectorSize]) );
|
|
}
|
|
}
|
|
|
|
exit:
|
|
//clean up
|
|
for( vectorSize = kMinVectorSize; vectorSize < kLastVectorSizeToTest; vectorSize++)
|
|
{
|
|
clReleaseKernel( kernels[ vectorSize ] );
|
|
clReleaseProgram( programs[ vectorSize ] );
|
|
if( gTestDouble )
|
|
{
|
|
clReleaseKernel( doubleKernels[ vectorSize ] );
|
|
clReleaseProgram( doublePrograms[ vectorSize ] );
|
|
}
|
|
}
|
|
|
|
return error;
|
|
}
|
|
|
|
|