| // |
| // 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 <string.h> |
| #include <sys/types.h> |
| #include <sys/stat.h> |
| |
| #include "procs.h" |
| |
| |
| const char *wg_broadcast_1D_kernel_code = |
| "__kernel void test_wg_broadcast_1D(global float *input, global float *output)\n" |
| "{\n" |
| " int tid = get_global_id(0);\n" |
| "\n" |
| " float result = work_group_broadcast(input[tid], get_group_id(0) % get_local_size(0));\n" |
| " output[tid] = result;\n" |
| "}\n"; |
| |
| const char *wg_broadcast_2D_kernel_code = |
| "__kernel void test_wg_broadcast_2D(global float *input, global float *output)\n" |
| "{\n" |
| " size_t tid_x = get_global_id(0);\n" |
| " size_t tid_y = get_global_id(1);\n" |
| " size_t x = get_group_id(0) % get_local_size(0);\n" |
| " size_t y = get_group_id(1) % get_local_size(1);\n" |
| "\n" |
| " size_t indx = (tid_y * get_global_size(0)) + tid_x;\n" |
| " float result = work_group_broadcast(input[indx], x, y);\n" |
| " output[indx] = result;\n" |
| "}\n"; |
| |
| const char *wg_broadcast_3D_kernel_code = |
| "__kernel void test_wg_broadcast_3D(global float *input, global float *output)\n" |
| "{\n" |
| " size_t tid_x = get_global_id(0);\n" |
| " size_t tid_y = get_global_id(1);\n" |
| " size_t tid_z = get_global_id(2);\n" |
| " size_t x = get_group_id(0) % get_local_size(0);\n" |
| " size_t y = get_group_id(1) % get_local_size(1);\n" |
| " size_t z = get_group_id(2) % get_local_size(2);\n" |
| "\n" |
| " size_t indx = (tid_z * get_global_size(1) * get_global_size(0)) + (tid_y * get_global_size(0)) + tid_x;\n" |
| " float result = work_group_broadcast(input[indx], x, y, z);\n" |
| " output[indx] = result;\n" |
| "}\n"; |
| |
| static int |
| verify_wg_broadcast_1D(float *inptr, float *outptr, size_t n, size_t wg_size) |
| { |
| size_t i, j; |
| size_t group_id; |
| |
| for (i=0,group_id=0; i<n; i+=wg_size,group_id++) |
| { |
| int local_size = (n-i) > wg_size ? wg_size : (n-i); |
| float broadcast_result = inptr[i + (group_id % local_size)]; |
| for (j=0; j<local_size; j++) |
| { |
| if ( broadcast_result != outptr[i+j] ) |
| { |
| log_info("work_group_broadcast: Error at %u: expected = %f, got = %f\n", i+j, broadcast_result, outptr[i+j]); |
| return -1; |
| } |
| } |
| } |
| |
| return 0; |
| } |
| |
| static int |
| verify_wg_broadcast_2D(float *inptr, float *outptr, size_t nx, size_t ny, size_t wg_size_x, size_t wg_size_y) |
| { |
| size_t i, j, _i, _j; |
| size_t group_id_x, group_id_y; |
| |
| for (i=0,group_id_y=0; i<ny; i+=wg_size_y,group_id_y++) |
| { |
| size_t y = group_id_y % wg_size_y; |
| size_t local_size_y = (ny-i) > wg_size_y ? wg_size_y : (ny-i); |
| for (_i=0; _i < local_size_y; _i++) |
| { |
| for (j=0,group_id_x=0; j<nx; j+=wg_size_x,group_id_x++) |
| { |
| size_t x = group_id_x % wg_size_x; |
| size_t local_size_x = (nx-j) > wg_size_x ? wg_size_x : (nx-j); |
| float broadcast_result = inptr[(i + y) * nx + (j + x)]; |
| for (_j=0; _j < local_size_x; _j++) |
| { |
| size_t indx = (i + _i) * nx + (j + _j); |
| if ( broadcast_result != outptr[indx] ) |
| { |
| log_info("work_group_broadcast: Error at (%u, %u): expected = %f, got = %f\n", j+_j, i+_i, broadcast_result, outptr[indx]); |
| return -1; |
| } |
| } |
| } |
| } |
| } |
| |
| return 0; |
| } |
| |
| static int |
| verify_wg_broadcast_3D(float *inptr, float *outptr, size_t nx, size_t ny, size_t nz, size_t wg_size_x, size_t wg_size_y, size_t wg_size_z) |
| { |
| size_t i, j, k, _i, _j, _k; |
| size_t group_id_x, group_id_y, group_id_z; |
| |
| for (i=0,group_id_z=0; i<nz; i+=wg_size_z,group_id_z++) |
| { |
| size_t z = group_id_z % wg_size_z; |
| size_t local_size_z = (nz-i) > wg_size_z ? wg_size_z : (nz-i); |
| for (_i=0; _i < local_size_z; _i++) |
| { |
| for (j=0,group_id_y=0; j<ny; j+=wg_size_y,group_id_y++) |
| { |
| size_t y = group_id_y % wg_size_y; |
| size_t local_size_y = (ny-j) > wg_size_y ? wg_size_y : (ny-j); |
| for (_j=0; _j < local_size_y; _j++) |
| { |
| for (k=0,group_id_x=0; k<nx; k+=wg_size_x,group_id_x++) |
| { |
| size_t x = group_id_x % wg_size_x; |
| size_t local_size_x = (nx-k) > wg_size_x ? wg_size_x : (nx-k); |
| float broadcast_result = inptr[(i + z) * ny * nz + (j + y) * nx + (k + x)]; |
| for (_k=0; _k < local_size_x; _k++) |
| { |
| size_t indx = (i + _i) * ny * nx + (j + _j) * nx + (k + _k); |
| if ( broadcast_result != outptr[indx] ) |
| { |
| log_info("work_group_broadcast: Error at (%u, %u, %u): expected = %f, got = %f\n", k+_k, j+_j, i+_i, broadcast_result, outptr[indx]); |
| return -1; |
| } |
| } |
| } |
| } |
| } |
| } |
| } |
| |
| return 0; |
| } |
| |
| |
| int |
| test_work_group_broadcast_1D(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) |
| { |
| cl_mem streams[2]; |
| cl_float *input_ptr[1], *p; |
| cl_float *output_ptr; |
| cl_program program; |
| cl_kernel kernel; |
| void *values[2]; |
| size_t globalsize[1]; |
| size_t wg_size[1]; |
| size_t num_elements; |
| int err; |
| int i; |
| MTdata d; |
| |
| err = create_single_kernel_helper(context, &program, &kernel, 1, |
| &wg_broadcast_1D_kernel_code, |
| "test_wg_broadcast_1D"); |
| if (err) |
| return -1; |
| |
| // "wg_size" is limited to that of the first dimension as only a 1DRange is executed. |
| err = get_max_allowed_1d_work_group_size_on_device(device, kernel, wg_size); |
| test_error(err, "get_max_allowed_1d_work_group_size_on_device failed"); |
| |
| num_elements = n_elems; |
| |
| input_ptr[0] = (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]) |
| { |
| 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; |
| } |
| |
| p = input_ptr[0]; |
| d = init_genrand( gRandomSeed ); |
| for (i=0; i<num_elements; i++) |
| { |
| p[i] = get_random_float((float)(-100000.f * M_PI), (float)(100000.f * M_PI) ,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; |
| } |
| |
| values[0] = streams[0]; |
| values[1] = streams[1]; |
| err = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0] ); |
| err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1] ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clSetKernelArgs failed\n"); |
| return -1; |
| } |
| |
| // Line below is troublesome... |
| globalsize[0] = (size_t)n_elems; |
| err = clEnqueueNDRangeKernel( queue, kernel, 1, NULL, globalsize, wg_size, 0, NULL, NULL ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clEnqueueNDRangeKernel failed\n"); |
| return -1; |
| } |
| |
| cl_uint dead = 0xdeaddead; |
| memset_pattern4(output_ptr, &dead, sizeof(cl_float)*num_elements); |
| err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clEnqueueReadBuffer failed\n"); |
| return -1; |
| } |
| |
| if (verify_wg_broadcast_1D(input_ptr[0], output_ptr, num_elements, wg_size[0])) |
| { |
| log_error("work_group_broadcast_1D test failed\n"); |
| return -1; |
| } |
| log_info("work_group_broadcast_1D test passed\n"); |
| |
| clReleaseMemObject(streams[0]); |
| clReleaseMemObject(streams[1]); |
| clReleaseKernel(kernel); |
| clReleaseProgram(program); |
| free(input_ptr[0]); |
| free(output_ptr); |
| |
| return err; |
| } |
| |
| |
| int |
| test_work_group_broadcast_2D(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) |
| { |
| cl_mem streams[2]; |
| cl_float *input_ptr[1], *p; |
| cl_float *output_ptr; |
| cl_program program; |
| cl_kernel kernel; |
| void *values[2]; |
| size_t globalsize[2]; |
| size_t localsize[2]; |
| size_t wg_size[1]; |
| size_t num_workgroups; |
| size_t num_elements; |
| int err; |
| int i; |
| MTdata d; |
| |
| err = create_single_kernel_helper(context, &program, &kernel, 1, |
| &wg_broadcast_2D_kernel_code, |
| "test_wg_broadcast_2D"); |
| if (err) |
| return -1; |
| |
| err = clGetKernelWorkGroupInfo( kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), wg_size, NULL); |
| if (err) |
| return -1; |
| |
| if (wg_size[0] >= 256) |
| { |
| localsize[0] = localsize[1] = 16; |
| } |
| else if (wg_size[0] >=64) |
| { |
| localsize[0] = localsize[1] = 8; |
| } |
| else if (wg_size[0] >= 16) |
| { |
| localsize[0] = localsize[1] = 4; |
| } |
| else |
| { |
| localsize[0] = localsize[1] = 1; |
| } |
| |
| num_workgroups = MAX(n_elems/wg_size[0], 16); |
| globalsize[0] = num_workgroups * localsize[0]; |
| globalsize[1] = num_workgroups * localsize[1]; |
| num_elements = globalsize[0] * globalsize[1]; |
| |
| input_ptr[0] = (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]) |
| { |
| 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; |
| } |
| |
| p = input_ptr[0]; |
| d = init_genrand( gRandomSeed ); |
| for (i=0; i<num_elements; i++) |
| { |
| p[i] = get_random_float((float)(-100000.f * M_PI), (float)(100000.f * M_PI) ,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; |
| } |
| |
| values[0] = streams[0]; |
| values[1] = streams[1]; |
| err = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0] ); |
| err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1] ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clSetKernelArgs failed\n"); |
| return -1; |
| } |
| |
| err = clEnqueueNDRangeKernel( queue, kernel, 2, NULL, globalsize, localsize, 0, NULL, NULL ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clEnqueueNDRangeKernel failed\n"); |
| return -1; |
| } |
| |
| cl_uint dead = 0xdeaddead; |
| memset_pattern4(output_ptr, &dead, sizeof(cl_float)*num_elements); |
| err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clEnqueueReadBuffer failed\n"); |
| return -1; |
| } |
| |
| if (verify_wg_broadcast_2D(input_ptr[0], output_ptr, globalsize[0], globalsize[1], localsize[0], localsize[1])) |
| { |
| log_error("work_group_broadcast_2D test failed\n"); |
| return -1; |
| } |
| log_info("work_group_broadcast_2D test passed\n"); |
| |
| clReleaseMemObject(streams[0]); |
| clReleaseMemObject(streams[1]); |
| clReleaseKernel(kernel); |
| clReleaseProgram(program); |
| free(input_ptr[0]); |
| free(output_ptr); |
| |
| return err; |
| } |
| |
| |
| int |
| test_work_group_broadcast_3D(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) |
| { |
| cl_mem streams[2]; |
| cl_float *input_ptr[1], *p; |
| cl_float *output_ptr; |
| cl_program program; |
| cl_kernel kernel; |
| void *values[2]; |
| size_t globalsize[3]; |
| size_t localsize[3]; |
| size_t wg_size[1]; |
| size_t num_workgroups; |
| size_t num_elements; |
| int err; |
| int i; |
| MTdata d; |
| |
| err = create_single_kernel_helper(context, &program, &kernel, 1, |
| &wg_broadcast_3D_kernel_code, |
| "test_wg_broadcast_3D"); |
| if (err) |
| return -1; |
| |
| err = clGetKernelWorkGroupInfo( kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), wg_size, NULL); |
| if (err) |
| return -1; |
| |
| if (wg_size[0] >=512) |
| { |
| localsize[0] = localsize[1] = localsize[2] = 8; |
| } |
| else if (wg_size[0] >= 64) |
| { |
| localsize[0] = localsize[1] = localsize[2] = 4; |
| } |
| else if (wg_size[0] >= 8) |
| { |
| localsize[0] = localsize[1] = localsize[2] = 2; |
| } |
| else |
| { |
| localsize[0] = localsize[1] = localsize[2] = 1; |
| } |
| |
| num_workgroups = MAX(n_elems/wg_size[0], 8); |
| globalsize[0] = num_workgroups * localsize[0]; |
| globalsize[1] = num_workgroups * localsize[1]; |
| globalsize[2] = num_workgroups * localsize[2]; |
| num_elements = globalsize[0] * globalsize[1] * globalsize[2]; |
| |
| input_ptr[0] = (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]) |
| { |
| 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; |
| } |
| |
| p = input_ptr[0]; |
| d = init_genrand( gRandomSeed ); |
| for (i=0; i<num_elements; i++) |
| { |
| p[i] = get_random_float((float)(-100000.f * M_PI), (float)(100000.f * M_PI) ,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; |
| } |
| |
| values[0] = streams[0]; |
| values[1] = streams[1]; |
| err = clSetKernelArg(kernel, 0, sizeof streams[0], &streams[0] ); |
| err |= clSetKernelArg(kernel, 1, sizeof streams[1], &streams[1] ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clSetKernelArgs failed\n"); |
| return -1; |
| } |
| |
| err = clEnqueueNDRangeKernel( queue, kernel, 3, NULL, globalsize, localsize, 0, NULL, NULL ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clEnqueueNDRangeKernel failed\n"); |
| return -1; |
| } |
| |
| cl_uint dead = 0xdeaddead; |
| memset_pattern4(output_ptr, &dead, sizeof(cl_float)*num_elements); |
| err = clEnqueueReadBuffer( queue, streams[1], true, 0, sizeof(cl_float)*num_elements, (void *)output_ptr, 0, NULL, NULL ); |
| if (err != CL_SUCCESS) |
| { |
| log_error("clEnqueueReadBuffer failed\n"); |
| return -1; |
| } |
| |
| if (verify_wg_broadcast_3D(input_ptr[0], output_ptr, globalsize[0], globalsize[1], globalsize[2], localsize[0], localsize[1], localsize[2])) |
| { |
| log_error("work_group_broadcast_3D test failed\n"); |
| return -1; |
| } |
| log_info("work_group_broadcast_3D test passed\n"); |
| |
| clReleaseMemObject(streams[0]); |
| clReleaseMemObject(streams[1]); |
| clReleaseKernel(kernel); |
| clReleaseProgram(program); |
| free(input_ptr[0]); |
| free(output_ptr); |
| |
| return err; |
| } |
| |
| |
| int |
| test_work_group_broadcast(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) |
| { |
| int err; |
| |
| err = test_work_group_broadcast_1D(device, context, queue, n_elems); |
| if (err) return err; |
| err = test_work_group_broadcast_2D(device, context, queue, n_elems); |
| if (err) return err; |
| return err; |
| } |
| |
| |
