369 lines
11 KiB
C++
369 lines
11 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 "harness/compat.h"
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <sys/types.h>
|
|
#include <sys/stat.h>
|
|
|
|
|
|
#include "procs.h"
|
|
|
|
const char *barrier_with_localmem_kernel_code[] = {
|
|
"__kernel void compute_sum_with_localmem(__global int *a, int n, __local int *tmp_sum, __global int *sum)\n"
|
|
"{\n"
|
|
" int tid = get_local_id(0);\n"
|
|
" int lsize = get_local_size(0);\n"
|
|
" int i;\n"
|
|
"\n"
|
|
" tmp_sum[tid] = 0;\n"
|
|
" for (i=tid; i<n; i+=lsize)\n"
|
|
" tmp_sum[tid] += a[i];\n"
|
|
"\n"
|
|
" if( lsize == 1 )\n"
|
|
" {\n"
|
|
" if( tid == 0 )\n"
|
|
" *sum = tmp_sum[0];\n"
|
|
" return;\n"
|
|
" }\n"
|
|
"\n"
|
|
" do\n"
|
|
" {\n"
|
|
" barrier(CLK_LOCAL_MEM_FENCE);\n"
|
|
" if (tid < lsize/2)\n"
|
|
" {\n"
|
|
" int sum = tmp_sum[tid];\n"
|
|
" if( (lsize & 1) && tid == 0 )\n"
|
|
" sum += tmp_sum[tid + lsize - 1];\n"
|
|
" tmp_sum[tid] = sum + tmp_sum[tid + lsize/2];\n"
|
|
" }\n"
|
|
" lsize = lsize/2; \n"
|
|
" }while( lsize );\n"
|
|
"\n"
|
|
" if( tid == 0 )\n"
|
|
" *sum = tmp_sum[0];\n"
|
|
"}\n",
|
|
"__kernel void compute_sum_with_localmem(__global int *a, int n, __global int *sum)\n"
|
|
"{\n"
|
|
" __local int tmp_sum[%d];\n"
|
|
" int tid = get_local_id(0);\n"
|
|
" int lsize = get_local_size(0);\n"
|
|
" int i;\n"
|
|
"\n"
|
|
" tmp_sum[tid] = 0;\n"
|
|
" for (i=tid; i<n; i+=lsize)\n"
|
|
" tmp_sum[tid] += a[i];\n"
|
|
"\n"
|
|
" if( lsize == 1 )\n"
|
|
" {\n"
|
|
" if( tid == 0 )\n"
|
|
" *sum = tmp_sum[0];\n"
|
|
" return;\n"
|
|
" }\n"
|
|
"\n"
|
|
" do\n"
|
|
" {\n"
|
|
" barrier(CLK_LOCAL_MEM_FENCE);\n"
|
|
" if (tid < lsize/2)\n"
|
|
" {\n"
|
|
" int sum = tmp_sum[tid];\n"
|
|
" if( (lsize & 1) && tid == 0 )\n"
|
|
" sum += tmp_sum[tid + lsize - 1];\n"
|
|
" tmp_sum[tid] = sum + tmp_sum[tid + lsize/2];\n"
|
|
" }\n"
|
|
" lsize = lsize/2; \n"
|
|
" }while( lsize );\n"
|
|
"\n"
|
|
" if( tid == 0 )\n"
|
|
" *sum = tmp_sum[0];\n"
|
|
"}\n"
|
|
};
|
|
|
|
static int
|
|
verify_sum(int *inptr, int *outptr, int n)
|
|
{
|
|
int r = 0;
|
|
int i;
|
|
|
|
for (i=0; i<n; i++)
|
|
{
|
|
r += inptr[i];
|
|
}
|
|
|
|
if (r != outptr[0])
|
|
{
|
|
log_error("LOCAL test failed: *%d vs %d\n", r, outptr[0] );
|
|
return -1;
|
|
}
|
|
|
|
log_info("LOCAL test passed\n");
|
|
return 0;
|
|
}
|
|
|
|
int test_local_arg_def(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements)
|
|
{
|
|
cl_mem streams[2];
|
|
cl_program program;
|
|
cl_kernel kernel;
|
|
|
|
cl_int *input_ptr, *output_ptr;
|
|
size_t global_threads[1], local_threads[1];
|
|
size_t wgsize, kwgsize;
|
|
size_t max_local_workgroup_size[3];
|
|
int err, i;
|
|
MTdata d = init_genrand( gRandomSeed );
|
|
|
|
err = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof wgsize, &wgsize, NULL);
|
|
if (err) {
|
|
log_error("clGetDeviceInfo failed, %d\n\n", err);
|
|
return -1;
|
|
}
|
|
wgsize/=2;
|
|
if (wgsize < 1)
|
|
wgsize = 1;
|
|
|
|
size_t in_length = sizeof(cl_int) * num_elements;
|
|
size_t out_length = sizeof(cl_int) * wgsize;
|
|
|
|
input_ptr = (cl_int *)malloc(in_length);
|
|
output_ptr = (cl_int *)malloc(out_length);
|
|
|
|
streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE, in_length, NULL, NULL);
|
|
if (!streams[0])
|
|
{
|
|
log_error("clCreateBuffer failed\n");
|
|
return -1;
|
|
}
|
|
streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE, out_length, NULL, NULL);
|
|
if (!streams[1])
|
|
{
|
|
log_error("clCreateBuffer failed\n");
|
|
return -1;
|
|
}
|
|
|
|
for (i=0; i<num_elements; i++)
|
|
input_ptr[i] = (int)genrand_int32(d);
|
|
|
|
free_mtdata(d); d = NULL;
|
|
|
|
err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0, in_length, input_ptr, 0, NULL, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clEnqueueWriteBuffer failed\n");
|
|
return -1;
|
|
}
|
|
|
|
err = create_single_kernel_helper(context, &program, &kernel, 1, &barrier_with_localmem_kernel_code[0], "compute_sum_with_localmem" );
|
|
if (err)
|
|
return -1;
|
|
|
|
err = clGetKernelWorkGroupInfo(kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof kwgsize, &kwgsize, NULL);
|
|
test_error(err, "clGetKernelWorkGroupInfo failed for CL_KERNEL_WORK_GROUP_SIZE");
|
|
|
|
err = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(max_local_workgroup_size), max_local_workgroup_size, NULL);
|
|
test_error(err, "clGetDeviceInfo failed for CL_DEVICE_MAX_WORK_ITEM_SIZES");
|
|
|
|
// Pick the minimum of the device and the kernel
|
|
if (kwgsize > max_local_workgroup_size[0])
|
|
kwgsize = max_local_workgroup_size[0];
|
|
|
|
// err = clSetKernelArgs(context, kernel, 4, NULL, values, sizes);
|
|
err = clSetKernelArg(kernel, 0, sizeof(streams[0]), &streams[0]);
|
|
err |= clSetKernelArg(kernel, 1, sizeof num_elements, &num_elements);
|
|
err |= clSetKernelArg(kernel, 2, wgsize * sizeof(cl_int), NULL);
|
|
err |= clSetKernelArg(kernel, 3, sizeof streams[1], &streams[1]);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clSetKernelArgs failed\n");
|
|
return -1;
|
|
}
|
|
|
|
global_threads[0] = wgsize;
|
|
local_threads[0] = wgsize;
|
|
|
|
// Adjust the local thread size to fit and be a nice multiple.
|
|
if (kwgsize < wgsize) {
|
|
log_info("Adjusting wgsize down from %lu to %lu.\n", wgsize, kwgsize);
|
|
local_threads[0] = kwgsize;
|
|
}
|
|
while (global_threads[0] % local_threads[0] != 0)
|
|
local_threads[0]--;
|
|
|
|
err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, global_threads, local_threads, 0, NULL, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clEnqueueNDRangeKernel failed\n");
|
|
return -1;
|
|
}
|
|
|
|
err = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, out_length, output_ptr, 0, NULL, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clEnqueueReadBuffer failed\n");
|
|
return -1;
|
|
}
|
|
|
|
err = verify_sum(input_ptr, output_ptr, num_elements);
|
|
|
|
// cleanup
|
|
clReleaseMemObject(streams[0]);
|
|
clReleaseMemObject(streams[1]);
|
|
clReleaseKernel(kernel);
|
|
clReleaseProgram(program);
|
|
free(input_ptr);
|
|
free(output_ptr);
|
|
|
|
return err;
|
|
}
|
|
|
|
int test_local_kernel_def(cl_device_id device, cl_context context, cl_command_queue queue, int num_elements)
|
|
{
|
|
cl_mem streams[2];
|
|
cl_program program;
|
|
cl_kernel kernel;
|
|
|
|
cl_int *input_ptr, *output_ptr;
|
|
size_t global_threads[1], local_threads[1];
|
|
size_t wgsize, kwgsize;
|
|
int err, i;
|
|
char *program_source = (char*)malloc(sizeof(char)*2048);
|
|
MTdata d = init_genrand( gRandomSeed );
|
|
size_t max_local_workgroup_size[3];
|
|
memset(program_source, 0, 2048);
|
|
|
|
err = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof wgsize, &wgsize, NULL);
|
|
if (err) {
|
|
log_error("clGetDeviceInfo failed, %d\n\n", err);
|
|
return -1;
|
|
}
|
|
wgsize/=2;
|
|
if (wgsize < 1)
|
|
wgsize = 1;
|
|
|
|
size_t in_length = sizeof(cl_int) * num_elements;
|
|
size_t out_length = sizeof(cl_int) * wgsize;
|
|
|
|
input_ptr = (cl_int *)malloc(in_length);
|
|
output_ptr = (cl_int *)malloc(out_length);
|
|
|
|
streams[0] = clCreateBuffer(context, CL_MEM_READ_WRITE, in_length, NULL, NULL);
|
|
if (!streams[0])
|
|
{
|
|
log_error("clCreateBuffer failed\n");
|
|
return -1;
|
|
}
|
|
streams[1] = clCreateBuffer(context, CL_MEM_READ_WRITE, out_length, NULL, NULL);
|
|
if (!streams[1])
|
|
{
|
|
log_error("clCreateBuffer failed\n");
|
|
return -1;
|
|
}
|
|
|
|
for (i=0; i<num_elements; i++)
|
|
input_ptr[i] = (cl_int) genrand_int32(d);
|
|
|
|
free_mtdata(d); d = NULL;
|
|
|
|
err = clEnqueueWriteBuffer(queue, streams[0], CL_TRUE, 0, in_length, input_ptr, 0, NULL, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clEnqueueWriteBuffer failed\n");
|
|
return -1;
|
|
}
|
|
|
|
// Validate that created kernel doesn't violate local memory size allowed by the device
|
|
cl_ulong localMemSize = 0;
|
|
err = clGetDeviceInfo(device, CL_DEVICE_LOCAL_MEM_SIZE, sizeof(localMemSize), &localMemSize, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clGetDeviceInfo failed\n");
|
|
return -1;
|
|
}
|
|
if ( wgsize > (localMemSize / (sizeof(cl_int)*sizeof(cl_int))) )
|
|
{
|
|
wgsize = localMemSize / (sizeof(cl_int)*sizeof(cl_int));
|
|
}
|
|
|
|
sprintf(program_source, barrier_with_localmem_kernel_code[1], (int)(wgsize * sizeof(cl_int)));
|
|
|
|
err = create_single_kernel_helper(context, &program, &kernel, 1, (const char**)&program_source, "compute_sum_with_localmem" );
|
|
free(program_source);
|
|
if (err)
|
|
return -1;
|
|
|
|
err = clGetKernelWorkGroupInfo(kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof kwgsize, &kwgsize, NULL);
|
|
test_error(err, "clGetKernelWorkGroupInfo failed for CL_KERNEL_WORK_GROUP_SIZE");
|
|
|
|
err = clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(max_local_workgroup_size), max_local_workgroup_size, NULL);
|
|
test_error(err, "clGetDeviceInfo failed for CL_DEVICE_MAX_WORK_ITEM_SIZES");
|
|
|
|
// Pick the minimum of the device and the kernel
|
|
if (kwgsize > max_local_workgroup_size[0])
|
|
kwgsize = max_local_workgroup_size[0];
|
|
|
|
// err = clSetKernelArgs(context, kernel, 4, NULL, values, sizes);
|
|
err = clSetKernelArg(kernel, 0, sizeof(streams[0]), &streams[0]);
|
|
err |= clSetKernelArg(kernel, 1, sizeof num_elements, &num_elements);
|
|
err |= clSetKernelArg(kernel, 2, sizeof streams[1], &streams[1]);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clSetKernelArgs failed\n");
|
|
return -1;
|
|
}
|
|
|
|
global_threads[0] = wgsize;
|
|
local_threads[0] = wgsize;
|
|
|
|
// Adjust the local thread size to fit and be a nice multiple.
|
|
if (kwgsize < wgsize) {
|
|
log_info("Adjusting wgsize down from %lu to %lu.\n", wgsize, kwgsize);
|
|
local_threads[0] = kwgsize;
|
|
}
|
|
while (global_threads[0] % local_threads[0] != 0)
|
|
local_threads[0]--;
|
|
|
|
err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, global_threads, local_threads, 0, NULL, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clEnqueueNDRangeKernel failed\n");
|
|
return -1;
|
|
}
|
|
|
|
err = clEnqueueReadBuffer(queue, streams[1], CL_TRUE, 0, out_length, output_ptr, 0, NULL, NULL);
|
|
if (err != CL_SUCCESS)
|
|
{
|
|
log_error("clEnqueueReadBuffer failed\n");
|
|
return -1;
|
|
}
|
|
|
|
err = verify_sum(input_ptr, output_ptr, num_elements);
|
|
|
|
// cleanup
|
|
clReleaseMemObject(streams[0]);
|
|
clReleaseMemObject(streams[1]);
|
|
clReleaseKernel(kernel);
|
|
clReleaseProgram(program);
|
|
free(input_ptr);
|
|
free(output_ptr);
|
|
|
|
return err;
|
|
}
|
|
|
|
|
|
|