// // 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 // // 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 #include #include #include #include "harness/stringHelpers.h" #include "procs.h" #include "test_base.h" const char *smoothstep_fn_code_pattern = "%s\n" /* optional pragma */ "__kernel void test_fn(__global %s%s *e0, __global %s%s *e1, __global %s%s " "*x, __global %s%s *dst)\n" "{\n" " int tid = get_global_id(0);\n" "\n" " dst[tid] = smoothstep(e0[tid], e1[tid], x[tid]);\n" "}\n"; const char *smoothstep_fn_code_pattern_v3 = "%s\n" /* optional pragma */ "__kernel void test_fn(__global %s *e0, __global %s *e1, __global %s *x, " "__global %s *dst)\n" "{\n" " int tid = get_global_id(0);\n" "\n" " vstore3(smoothstep(vload3(tid,e0), vload3(tid,e1), vload3(tid,x)), " "tid, dst);\n" "}\n"; const char *smoothstep_fn_code_pattern_v3_scalar = "%s\n" /* optional pragma */ "__kernel void test_fn(__global %s *e0, __global %s *e1, __global %s *x, " "__global %s *dst)\n" "{\n" " int tid = get_global_id(0);\n" "\n" " vstore3(smoothstep(e0[tid], e1[tid], vload3(tid,x)), tid, dst);\n" "}\n"; #define MAX_ERR (1e-5f) namespace { template int verify_smoothstep(const T *const edge0, const T *const edge1, const T *const x, const T *const outptr, const int n, const int veclen, const bool vecParam) { double r, t; float delta = 0, max_delta = 0; if (vecParam) { for (int i = 0; i < n * veclen; i++) { t = (conv_to_dbl(x[i]) - conv_to_dbl(edge0[i])) / (conv_to_dbl(edge1[i]) - conv_to_dbl(edge0[i])); if (t < 0.0) t = 0.0; else if (t > 1.0) t = 1.0; r = t * t * (3.0 - 2.0 * t); delta = (float)fabs(r - conv_to_dbl(outptr[i])); if (!std::is_same::value) { if (delta > MAX_ERR) { log_error( "%d) verification error: smoothstep(%a, %a, %a) = " "*%a vs. %a\n", i, x[i], edge0[i], edge1[i], r, outptr[i]); return -1; } } else max_delta = std::max(max_delta, delta); } } else { for (int i = 0; i < n; ++i) { int ii = i / veclen; int vi = i * veclen; for (int j = 0; j < veclen; ++j, ++vi) { t = (conv_to_dbl(x[vi]) - conv_to_dbl(edge0[i])) / (conv_to_dbl(edge1[i]) - conv_to_dbl(edge0[i])); if (t < 0.0) t = 0.0; else if (t > 1.0) t = 1.0; r = t * t * (3.0 - 2.0 * t); delta = (float)fabs(r - conv_to_dbl(outptr[vi])); if (!std::is_same::value) { if (delta > MAX_ERR) { log_error("{%d, element %d}) verification error: " "smoothstep(%a, %a, %a) = *%a vs. %a\n", ii, j, x[vi], edge0[i], edge1[i], r, outptr[vi]); return -1; } } else max_delta = std::max(max_delta, delta); } } } // due to the fact that accuracy of smoothstep for cl_khr_fp16 is // implementation defined this test only reports maximum error without // testing maximum error threshold if (std::is_same::value) log_error("smoothstep half verification result, max delta: %a\n", max_delta); return 0; } } template int test_smoothstep_fn(cl_device_id device, cl_context context, cl_command_queue queue, const int n_elems, const bool vecParam) { clMemWrapper streams[4]; std::vector input_ptr[3], output_ptr; std::vector programs; std::vector kernels; int err, i; MTdataHolder d = MTdataHolder(gRandomSeed); assert(BaseFunctionTest::type2name.find(sizeof(T)) != BaseFunctionTest::type2name.end()); auto tname = BaseFunctionTest::type2name[sizeof(T)]; programs.resize(kTotalVecCount); kernels.resize(kTotalVecCount); int num_elements = n_elems * (1 << (kTotalVecCount - 1)); for (i = 0; i < 3; i++) input_ptr[i].resize(num_elements); output_ptr.resize(num_elements); for (i = 0; i < 4; i++) { streams[i] = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(T) * num_elements, NULL, &err); test_error(err, "clCreateBuffer failed"); } std::string pragma_str; if (std::is_same::value) { for (i = 0; i < num_elements; i++) { input_ptr[0][i] = get_random_float(-0x00200000, 0x00010000, d); input_ptr[1][i] = get_random_float(input_ptr[0][i], 0x00200000, d); input_ptr[2][i] = get_random_float(-0x20000000, 0x20000000, d); } } else if (std::is_same::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(-0x00200000, 0x00010000, d); input_ptr[1][i] = get_random_double(input_ptr[0][i], 0x00200000, d); input_ptr[2][i] = get_random_double(-0x20000000, 0x20000000, d); } } else if (std::is_same::value) { pragma_str = "#pragma OPENCL EXTENSION cl_khr_fp16 : enable\n"; for (i = 0; i < num_elements; i++) { input_ptr[0][i] = conv_to_half(get_random_float(-65503, 65503, d)); input_ptr[1][i] = conv_to_half( get_random_float(conv_to_flt(input_ptr[0][i]), 65503, d)); input_ptr[2][i] = conv_to_half(get_random_float(-65503, 65503, d)); } } for (i = 0; i < 3; i++) { 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"); } const char vecSizeNames[][3] = { "", "2", "4", "8", "16", "3" }; for (i = 0; i < kTotalVecCount; i++) { std::string kernelSource; if (i >= kVectorSizeCount) { if (vecParam) { std::string str = smoothstep_fn_code_pattern_v3; kernelSource = str_sprintf(str, pragma_str.c_str(), tname.c_str(), tname.c_str(), tname.c_str(), tname.c_str()); } else { std::string str = smoothstep_fn_code_pattern_v3_scalar; kernelSource = str_sprintf(str, pragma_str.c_str(), tname.c_str(), tname.c_str(), tname.c_str(), tname.c_str()); } } else { // regular path std::string str = smoothstep_fn_code_pattern; kernelSource = str_sprintf(str, pragma_str.c_str(), tname.c_str(), vecParam ? vecSizeNames[i] : "", 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 < 4; j++) { err = clSetKernelArg(kernels[i], j, sizeof(streams[j]), &streams[j]); test_error(err, "Unable to set kernel argument"); } 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[3], true, 0, sizeof(T) * num_elements, &output_ptr[0], 0, NULL, NULL); test_error(err, "Unable to read results"); if (verify_smoothstep((T *)&input_ptr[0].front(), (T *)&input_ptr[1].front(), (T *)&input_ptr[2].front(), &output_ptr[0], n_elems, g_arrVecSizes[i], vecParam)) { log_error("smoothstep %s%d%s test failed\n", tname.c_str(), ((g_arrVecSizes[i])), vecParam ? "" : std::string(", " + tname).c_str()); err = -1; } else { log_info("smoothstep %s%d%s test passed\n", tname.c_str(), ((g_arrVecSizes[i])), vecParam ? "" : std::string(", " + tname).c_str()); err = 0; } if (err) break; } return err; } cl_int SmoothstepTest::Run() { cl_int error = CL_SUCCESS; if (is_extension_available(device, "cl_khr_fp16")) { error = test_smoothstep_fn(device, context, queue, num_elems, vecParam); test_error(error, "SmoothstepTest::Run failed"); } error = test_smoothstep_fn(device, context, queue, num_elems, vecParam); test_error(error, "SmoothstepTest::Run failed"); if (is_extension_available(device, "cl_khr_fp64")) { error = test_smoothstep_fn(device, context, queue, num_elems, vecParam); test_error(error, "SmoothstepTest::Run failed"); } return error; } int test_smoothstep(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) { return MakeAndRunTest(device, context, queue, n_elems, "smoothstep", true); } int test_smoothstepf(cl_device_id device, cl_context context, cl_command_queue queue, int n_elems) { return MakeAndRunTest(device, context, queue, n_elems, "smoothstep", false); }