mirror of
https://github.com/KhronosGroup/OpenCL-CTS.git
synced 2026-03-19 06:09:01 +00:00
Complementation and modernization of commonfns tests (#1694)
* Unified common functions tests due to preparation for adding cl_khr_fp16 support * Renamed base structure, few cosmetic corrections * Added corrections due to code review * Removed comment separators * Added review related corrections
This commit is contained in:
Marcin Hajder
@@ -1,6 +1,6 @@
|
||||
//
|
||||
// Copyright (c) 2017 The Khronos Group Inc.
|
||||
//
|
||||
// Copyright (c) 2023 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
|
||||
@@ -13,524 +13,252 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
//
|
||||
#include "harness/compat.h"
|
||||
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
#include <sys/types.h>
|
||||
#include <sys/stat.h>
|
||||
|
||||
#include "procs.h"
|
||||
|
||||
static int
|
||||
test_step_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems);
|
||||
#include "test_base.h"
|
||||
|
||||
|
||||
const char *step_kernel_code =
|
||||
"__kernel void test_step(__global float *srcA, __global float *srcB, __global float *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
const char *step_fn_code_pattern = "%s\n" /* optional pragma */
|
||||
"__kernel void test_fn(__global %s%s *edge, "
|
||||
"__global %s%s *x, __global %s%s *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
" dst[tid] = step(edge[tid], x[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step2_kernel_code =
|
||||
"__kernel void test_step2(__global float2 *srcA, __global float2 *srcB, __global float2 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
const char *step_fn_code_pattern_v3 =
|
||||
"%s\n" /* optional pragma */
|
||||
"__kernel void test_fn(__global %s *edge, __global %s *x, __global %s "
|
||||
"*dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
" vstore3(step(vload3(tid,edge), vload3(tid,x)), tid, dst);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step4_kernel_code =
|
||||
"__kernel void test_step4(__global float4 *srcA, __global float4 *srcB, __global float4 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step8_kernel_code =
|
||||
"__kernel void test_step8(__global float8 *srcA, __global float8 *srcB, __global float8 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step16_kernel_code =
|
||||
"__kernel void test_step16(__global float16 *srcA, __global float16 *srcB, __global float16 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step3_kernel_code =
|
||||
"__kernel void test_step3(__global float *srcA, __global float *srcB, __global float *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" vstore3(step(vload3(tid,srcA), vload3(tid,srcB)),tid,dst);\n"
|
||||
"}\n";
|
||||
const char *step_fn_code_pattern_v3_scalar =
|
||||
"%s\n" /* optional pragma */
|
||||
"__kernel void test_fn(__global %s *edge, __global %s *x, __global %s "
|
||||
"*dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
" vstore3(step(edge[tid], vload3(tid,x)), tid, dst);\n"
|
||||
"}\n";
|
||||
|
||||
|
||||
int
|
||||
verify_step(float *inptrA, float *inptrB, float *outptr, int n)
|
||||
namespace {
|
||||
|
||||
template <typename T>
|
||||
int verify_step(const T *const inptrA, const T *const inptrB,
|
||||
const T *const outptr, const int n, const int veclen,
|
||||
const bool vecParam)
|
||||
{
|
||||
float r;
|
||||
int i;
|
||||
T r;
|
||||
|
||||
for (i=0; i<n; i++)
|
||||
if (vecParam)
|
||||
{
|
||||
r = (inptrB[i] < inptrA[i]) ? 0.0f : 1.0f;
|
||||
if (r != outptr[i])
|
||||
return -1;
|
||||
for (int i = 0; i < n * veclen; i++)
|
||||
{
|
||||
r = (inptrB[i] < inptrA[i]) ? 0.0 : 1.0;
|
||||
if (r != outptr[i]) return -1;
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
for (int i = 0; i < n;)
|
||||
{
|
||||
int ii = i / veclen;
|
||||
for (int j = 0; j < veclen && i < n; ++j, ++i)
|
||||
{
|
||||
r = (inptrB[i] < inptrA[ii]) ? 0.0f : 1.0f;
|
||||
if (r != outptr[i])
|
||||
{
|
||||
log_error("Failure @ {%d, element %d}: step(%a,%a) -> *%a "
|
||||
"vs %a\n",
|
||||
ii, j, inptrA[ii], inptrB[i], r, outptr[i]);
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
int
|
||||
test_step(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
|
||||
}
|
||||
|
||||
|
||||
template <typename T>
|
||||
int test_step_fn(cl_device_id device, cl_context context,
|
||||
cl_command_queue queue, int n_elems, bool vecParam)
|
||||
{
|
||||
cl_mem streams[3];
|
||||
cl_float *input_ptr[2], *output_ptr, *p;
|
||||
cl_program program[kTotalVecCount];
|
||||
cl_kernel kernel[kTotalVecCount];
|
||||
size_t threads[1];
|
||||
int num_elements;
|
||||
int err;
|
||||
int i;
|
||||
MTdata d;
|
||||
num_elements = n_elems * 16;
|
||||
clMemWrapper streams[3];
|
||||
std::vector<T> input_ptr[2], output_ptr;
|
||||
|
||||
input_ptr[0] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
|
||||
input_ptr[1] = (cl_float*)malloc(sizeof(cl_float) * num_elements);
|
||||
output_ptr = (cl_float*)malloc(sizeof(cl_float) * num_elements);
|
||||
streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(cl_float) * num_elements, NULL, NULL);
|
||||
if (!streams[0])
|
||||
std::vector<clProgramWrapper> programs;
|
||||
std::vector<clKernelWrapper> kernels;
|
||||
|
||||
int err, i;
|
||||
MTdataHolder d = MTdataHolder(gRandomSeed);
|
||||
|
||||
assert(BaseFunctionTest::type2name.find(sizeof(T))
|
||||
!= BaseFunctionTest::type2name.end());
|
||||
auto tname = BaseFunctionTest::type2name[sizeof(T)];
|
||||
int num_elements = n_elems * (1 << (kTotalVecCount - 1));
|
||||
|
||||
programs.resize(kTotalVecCount);
|
||||
kernels.resize(kTotalVecCount);
|
||||
|
||||
for (i = 0; i < 2; i++) input_ptr[i].resize(num_elements);
|
||||
output_ptr.resize(num_elements);
|
||||
|
||||
for (i = 0; i < 3; i++)
|
||||
{
|
||||
log_error("clCreateBuffer failed\n");
|
||||
return -1;
|
||||
}
|
||||
streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(cl_float) * num_elements, NULL, NULL);
|
||||
if (!streams[1])
|
||||
{
|
||||
log_error("clCreateBuffer failed\n");
|
||||
return -1;
|
||||
}
|
||||
streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(cl_float) * num_elements, NULL, NULL);
|
||||
if (!streams[2])
|
||||
{
|
||||
log_error("clCreateBuffer failed\n");
|
||||
return -1;
|
||||
streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(T) * num_elements, NULL, &err);
|
||||
test_error(err, "clCreateBuffer failed");
|
||||
}
|
||||
|
||||
p = input_ptr[0];
|
||||
d = init_genrand( gRandomSeed );
|
||||
for (i=0; i<num_elements; i++)
|
||||
std::string pragma_str;
|
||||
if (std::is_same<T, float>::value)
|
||||
{
|
||||
p[i] = get_random_float(-0x40000000, 0x40000000, d);
|
||||
}
|
||||
p = input_ptr[1];
|
||||
for (i=0; i<num_elements; i++)
|
||||
{
|
||||
p[i] = get_random_float(-0x40000000, 0x40000000, d);
|
||||
}
|
||||
free_mtdata(d); d = NULL;
|
||||
|
||||
err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clWriteArray failed\n");
|
||||
return -1;
|
||||
}
|
||||
err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clWriteArray failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &step_kernel_code, "test_step" );
|
||||
if (err) return -1;
|
||||
err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &step2_kernel_code, "test_step2" );
|
||||
if (err) return -1;
|
||||
err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &step4_kernel_code, "test_step4" );
|
||||
if (err) return -1;
|
||||
err = create_single_kernel_helper(context, &program[3], &kernel[3], 1,
|
||||
&step8_kernel_code, "test_step8");
|
||||
if (err) return -1;
|
||||
err = create_single_kernel_helper(context, &program[4], &kernel[4], 1,
|
||||
&step16_kernel_code, "test_step16");
|
||||
if (err) return -1;
|
||||
err = create_single_kernel_helper(context, &program[5], &kernel[5], 1,
|
||||
&step3_kernel_code, "test_step3");
|
||||
if (err) return -1;
|
||||
|
||||
for (i=0; i <kTotalVecCount; i++)
|
||||
{
|
||||
err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
|
||||
err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
|
||||
err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
|
||||
if (err != CL_SUCCESS)
|
||||
for (i = 0; i < num_elements; i++)
|
||||
{
|
||||
log_error("clSetKernelArgs failed\n");
|
||||
return -1;
|
||||
input_ptr[0][i] = get_random_float(-0x40000000, 0x40000000, d);
|
||||
input_ptr[1][i] = get_random_float(-0x40000000, 0x40000000, d);
|
||||
}
|
||||
}
|
||||
else if (std::is_same<T, double>::value)
|
||||
{
|
||||
pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
|
||||
for (i = 0; i < num_elements; i++)
|
||||
{
|
||||
input_ptr[0][i] = get_random_double(-0x40000000, 0x40000000, d);
|
||||
input_ptr[1][i] = get_random_double(-0x40000000, 0x40000000, d);
|
||||
}
|
||||
}
|
||||
|
||||
threads[0] = (size_t)n_elems;
|
||||
for (i=0; i<kTotalVecCount; i++)
|
||||
for (i = 0; i < 2; i++)
|
||||
{
|
||||
err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
err = clEnqueueWriteBuffer(queue, streams[i], CL_TRUE, 0,
|
||||
sizeof(T) * num_elements,
|
||||
&input_ptr[i].front(), 0, NULL, NULL);
|
||||
test_error(err, "Unable to write input buffer");
|
||||
}
|
||||
|
||||
char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" };
|
||||
|
||||
for (i = 0; i < kTotalVecCount; i++)
|
||||
{
|
||||
std::string kernelSource;
|
||||
if (i >= kVectorSizeCount)
|
||||
{
|
||||
log_error("clEnqueueNDRangeKernel failed\n");
|
||||
return -1;
|
||||
if (vecParam)
|
||||
{
|
||||
std::string str = step_fn_code_pattern_v3;
|
||||
kernelSource =
|
||||
string_format(str, pragma_str.c_str(), tname.c_str(),
|
||||
tname.c_str(), tname.c_str());
|
||||
}
|
||||
else
|
||||
{
|
||||
std::string str = step_fn_code_pattern_v3_scalar;
|
||||
kernelSource =
|
||||
string_format(str, pragma_str.c_str(), tname.c_str(),
|
||||
tname.c_str(), tname.c_str());
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
// regular path
|
||||
std::string str = step_fn_code_pattern;
|
||||
kernelSource =
|
||||
string_format(str, pragma_str.c_str(), tname.c_str(),
|
||||
vecParam ? vecSizeNames[i] : "", tname.c_str(),
|
||||
vecSizeNames[i], tname.c_str(), vecSizeNames[i]);
|
||||
}
|
||||
const char *programPtr = kernelSource.c_str();
|
||||
err =
|
||||
create_single_kernel_helper(context, &programs[i], &kernels[i], 1,
|
||||
(const char **)&programPtr, "test_fn");
|
||||
test_error(err, "Unable to create kernel");
|
||||
|
||||
for (int j = 0; j < 3; j++)
|
||||
{
|
||||
err =
|
||||
clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]);
|
||||
test_error(err, "Unable to set kernel argument");
|
||||
}
|
||||
|
||||
err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
size_t threads = (size_t)n_elems;
|
||||
|
||||
err = clEnqueueNDRangeKernel(queue, kernels[i], 1, NULL, &threads, NULL,
|
||||
0, NULL, NULL);
|
||||
test_error(err, "Unable to execute kernel");
|
||||
|
||||
err = clEnqueueReadBuffer(queue, streams[2], true, 0,
|
||||
sizeof(T) * num_elements, &output_ptr[0], 0,
|
||||
NULL, NULL);
|
||||
test_error(err, "Unable to read results");
|
||||
|
||||
err = verify_step(&input_ptr[0].front(), &input_ptr[1].front(),
|
||||
&output_ptr.front(), n_elems, g_arrVecSizes[i],
|
||||
vecParam);
|
||||
if (err)
|
||||
{
|
||||
log_error("clEnqueueReadBuffer failed\n");
|
||||
return -1;
|
||||
log_error("step %s%d%s test failed\n", tname.c_str(),
|
||||
((g_arrVecSizes[i])),
|
||||
vecParam ? "" : std::string(", " + tname).c_str());
|
||||
err = -1;
|
||||
}
|
||||
|
||||
switch (i)
|
||||
else
|
||||
{
|
||||
case 0:
|
||||
err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems);
|
||||
if (err)
|
||||
log_error("STEP float test failed\n");
|
||||
else
|
||||
log_info("STEP float test passed\n");
|
||||
break;
|
||||
|
||||
case 1:
|
||||
err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*2);
|
||||
if (err)
|
||||
log_error("STEP float2 test failed\n");
|
||||
else
|
||||
log_info("STEP float2 test passed\n");
|
||||
break;
|
||||
|
||||
case 2:
|
||||
err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*4);
|
||||
if (err)
|
||||
log_error("STEP float4 test failed\n");
|
||||
else
|
||||
log_info("STEP float4 test passed\n");
|
||||
break;
|
||||
|
||||
case 3:
|
||||
err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*8);
|
||||
if (err)
|
||||
log_error("STEP float8 test failed\n");
|
||||
else
|
||||
log_info("STEP float8 test passed\n");
|
||||
break;
|
||||
|
||||
case 4:
|
||||
err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*16);
|
||||
if (err)
|
||||
log_error("STEP float16 test failed\n");
|
||||
else
|
||||
log_info("STEP float16 test passed\n");
|
||||
break;
|
||||
|
||||
case 5:
|
||||
err = verify_step(input_ptr[0], input_ptr[1], output_ptr, n_elems*3);
|
||||
if (err)
|
||||
log_error("STEP float3 test failed\n");
|
||||
else
|
||||
log_info("STEP float3 test passed\n");
|
||||
break;
|
||||
log_info("step %s%d%s test passed\n", tname.c_str(),
|
||||
((g_arrVecSizes[i])),
|
||||
vecParam ? "" : std::string(", " + tname).c_str());
|
||||
err = 0;
|
||||
}
|
||||
|
||||
if (err)
|
||||
break;
|
||||
}
|
||||
|
||||
clReleaseMemObject(streams[0]);
|
||||
clReleaseMemObject(streams[1]);
|
||||
clReleaseMemObject(streams[2]);
|
||||
for (i=0; i<kTotalVecCount; i++)
|
||||
{
|
||||
clReleaseKernel(kernel[i]);
|
||||
clReleaseProgram(program[i]);
|
||||
}
|
||||
free(input_ptr[0]);
|
||||
free(input_ptr[1]);
|
||||
free(output_ptr);
|
||||
|
||||
if( err )
|
||||
return err;
|
||||
|
||||
if( ! is_extension_available( device, "cl_khr_fp64" ))
|
||||
return 0;
|
||||
|
||||
return test_step_double( device, context, queue, n_elems);
|
||||
}
|
||||
|
||||
|
||||
#pragma mark -
|
||||
|
||||
const char *step_kernel_code_double =
|
||||
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
||||
"__kernel void test_step_double(__global double *srcA, __global double *srcB, __global double *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step2_kernel_code_double =
|
||||
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
||||
"__kernel void test_step2_double(__global double2 *srcA, __global double2 *srcB, __global double2 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step4_kernel_code_double =
|
||||
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
||||
"__kernel void test_step4_double(__global double4 *srcA, __global double4 *srcB, __global double4 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step8_kernel_code_double =
|
||||
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
||||
"__kernel void test_step8_double(__global double8 *srcA, __global double8 *srcB, __global double8 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step16_kernel_code_double =
|
||||
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
||||
"__kernel void test_step16_double(__global double16 *srcA, __global double16 *srcB, __global double16 *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" dst[tid] = step(srcA[tid], srcB[tid]);\n"
|
||||
"}\n";
|
||||
|
||||
const char *step3_kernel_code_double =
|
||||
"#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n"
|
||||
"__kernel void test_step3_double(__global double *srcA, __global double *srcB, __global double *dst)\n"
|
||||
"{\n"
|
||||
" int tid = get_global_id(0);\n"
|
||||
"\n"
|
||||
" vstore3(step(vload3(tid,srcA), vload3(tid,srcB)),tid,dst);\n"
|
||||
"}\n";
|
||||
|
||||
|
||||
int
|
||||
verify_step_double(double *inptrA, double *inptrB, double *outptr, int n)
|
||||
{
|
||||
double r;
|
||||
int i;
|
||||
|
||||
for (i=0; i<n; i++)
|
||||
{
|
||||
r = (inptrB[i] < inptrA[i]) ? 0.0 : 1.0;
|
||||
if (r != outptr[i])
|
||||
return -1;
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
test_step_double(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems)
|
||||
{
|
||||
cl_mem streams[3];
|
||||
cl_double *input_ptr[2], *output_ptr, *p;
|
||||
cl_program program[kTotalVecCount];
|
||||
cl_kernel kernel[kTotalVecCount];
|
||||
size_t threads[1];
|
||||
int num_elements;
|
||||
int err;
|
||||
int i;
|
||||
MTdata d;
|
||||
num_elements = n_elems * 16;
|
||||
|
||||
input_ptr[0] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
|
||||
input_ptr[1] = (cl_double*)malloc(sizeof(cl_double) * num_elements);
|
||||
output_ptr = (cl_double*)malloc(sizeof(cl_double) * num_elements);
|
||||
streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(cl_double) * num_elements, NULL, NULL);
|
||||
if (!streams[0])
|
||||
{
|
||||
log_error("clCreateBuffer failed\n");
|
||||
return -1;
|
||||
}
|
||||
streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(cl_double) * num_elements, NULL, NULL);
|
||||
if (!streams[1])
|
||||
{
|
||||
log_error("clCreateBuffer failed\n");
|
||||
return -1;
|
||||
}
|
||||
streams[2] = clCreateBuffer(context, CL_MEM_READ_WRITE,
|
||||
sizeof(cl_double) * num_elements, NULL, NULL);
|
||||
if (!streams[2])
|
||||
{
|
||||
log_error("clCreateBuffer failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
p = input_ptr[0];
|
||||
d = init_genrand( gRandomSeed );
|
||||
for (i=0; i<num_elements; i++)
|
||||
{
|
||||
p[i] = get_random_double(-0x40000000, 0x40000000, d);
|
||||
}
|
||||
p = input_ptr[1];
|
||||
for (i=0; i<num_elements; i++)
|
||||
{
|
||||
p[i] = get_random_double(-0x40000000, 0x40000000, d);
|
||||
}
|
||||
free_mtdata(d); d = NULL;
|
||||
|
||||
err = clEnqueueWriteBuffer( queue, streams[0], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[0], 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clWriteArray failed\n");
|
||||
return -1;
|
||||
}
|
||||
err = clEnqueueWriteBuffer( queue, streams[1], true, 0, sizeof(cl_double)*num_elements, (void *)input_ptr[1], 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clWriteArray failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
err = create_single_kernel_helper( context, &program[0], &kernel[0], 1, &step_kernel_code_double, "test_step_double" );
|
||||
if (err)
|
||||
return -1;
|
||||
err = create_single_kernel_helper( context, &program[1], &kernel[1], 1, &step2_kernel_code_double, "test_step2_double" );
|
||||
if (err)
|
||||
return -1;
|
||||
err = create_single_kernel_helper( context, &program[2], &kernel[2], 1, &step4_kernel_code_double, "test_step4_double" );
|
||||
if (err)
|
||||
return -1;
|
||||
err = create_single_kernel_helper( context, &program[3], &kernel[3], 1, &step8_kernel_code_double, "test_step8_double" );
|
||||
if (err)
|
||||
return -1;
|
||||
err = create_single_kernel_helper( context, &program[4], &kernel[4], 1, &step16_kernel_code_double, "test_step16_double" );
|
||||
if (err)
|
||||
return -1;
|
||||
err = create_single_kernel_helper( context, &program[5], &kernel[5], 1, &step3_kernel_code_double, "test_step3_double" );
|
||||
if (err)
|
||||
return -1;
|
||||
|
||||
for (i=0; i < kTotalVecCount; i++)
|
||||
{
|
||||
err = clSetKernelArg(kernel[i], 0, sizeof streams[0], &streams[0] );
|
||||
err |= clSetKernelArg(kernel[i], 1, sizeof streams[1], &streams[1] );
|
||||
err |= clSetKernelArg(kernel[i], 2, sizeof streams[2], &streams[2] );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clSetKernelArgs failed\n");
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
|
||||
threads[0] = (size_t)n_elems;
|
||||
for (i=0; i<kTotalVecCount; i++)
|
||||
{
|
||||
err = clEnqueueNDRangeKernel( queue, kernel[i], 1, NULL, threads, NULL, 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clEnqueueNDRangeKernel failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
err = clEnqueueReadBuffer( queue, streams[2], true, 0, sizeof(cl_double)*num_elements, (void *)output_ptr, 0, NULL, NULL );
|
||||
if (err != CL_SUCCESS)
|
||||
{
|
||||
log_error("clEnqueueReadBuffer failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
switch (i)
|
||||
{
|
||||
case 0:
|
||||
err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems);
|
||||
if (err)
|
||||
log_error("STEP double test failed\n");
|
||||
else
|
||||
log_info("STEP double test passed\n");
|
||||
break;
|
||||
|
||||
case 1:
|
||||
err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*2);
|
||||
if (err)
|
||||
log_error("STEP double2 test failed\n");
|
||||
else
|
||||
log_info("STEP double2 test passed\n");
|
||||
break;
|
||||
|
||||
case 2:
|
||||
err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*4);
|
||||
if (err)
|
||||
log_error("STEP double4 test failed\n");
|
||||
else
|
||||
log_info("STEP double4 test passed\n");
|
||||
break;
|
||||
|
||||
case 3:
|
||||
err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*8);
|
||||
if (err)
|
||||
log_error("STEP double8 test failed\n");
|
||||
else
|
||||
log_info("STEP double8 test passed\n");
|
||||
break;
|
||||
|
||||
case 4:
|
||||
err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*16);
|
||||
if (err)
|
||||
log_error("STEP double16 test failed\n");
|
||||
else
|
||||
log_info("STEP double16 test passed\n");
|
||||
break;
|
||||
|
||||
case 5:
|
||||
err = verify_step_double(input_ptr[0], input_ptr[1], output_ptr, n_elems*3);
|
||||
if (err)
|
||||
log_error("STEP double3 test failed\n");
|
||||
else
|
||||
log_info("STEP double3 test passed\n");
|
||||
break;
|
||||
}
|
||||
|
||||
if (err)
|
||||
break;
|
||||
}
|
||||
|
||||
clReleaseMemObject(streams[0]);
|
||||
clReleaseMemObject(streams[1]);
|
||||
clReleaseMemObject(streams[2]);
|
||||
for (i=0; i<kTotalVecCount; i++)
|
||||
{
|
||||
clReleaseKernel(kernel[i]);
|
||||
clReleaseProgram(program[i]);
|
||||
}
|
||||
free(input_ptr[0]);
|
||||
free(input_ptr[1]);
|
||||
free(output_ptr);
|
||||
|
||||
return err;
|
||||
}
|
||||
|
||||
|
||||
cl_int StepTest::Run()
|
||||
{
|
||||
cl_int error = CL_SUCCESS;
|
||||
|
||||
error = test_step_fn<float>(device, context, queue, num_elems, vecParam);
|
||||
test_error(error, "StepTest::Run<float> failed");
|
||||
|
||||
if (is_extension_available(device, "cl_khr_fp64"))
|
||||
{
|
||||
error =
|
||||
test_step_fn<double>(device, context, queue, num_elems, vecParam);
|
||||
test_error(error, "StepTest::Run<double> failed");
|
||||
}
|
||||
|
||||
return error;
|
||||
}
|
||||
|
||||
|
||||
int test_step(cl_device_id device, cl_context context, cl_command_queue queue,
|
||||
int n_elems)
|
||||
{
|
||||
return MakeAndRunTest<StepTest>(device, context, queue, n_elems, "step",
|
||||
true);
|
||||
}
|
||||
|
||||
|
||||
int test_stepf(cl_device_id device, cl_context context, cl_command_queue queue,
|
||||
int n_elems)
|
||||
{
|
||||
return MakeAndRunTest<StepTest>(device, context, queue, n_elems, "step",
|
||||
false);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user