|
| 1 | +// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -Xclang -fdenormal-fp-math-f32="preserve-sign,preserve-sign" %s -o %t.out |
| 2 | +// RUN: %CPU_RUN_PLACEHOLDER %t.out |
| 3 | +// RUN: %GPU_RUN_PLACEHOLDER %t.out |
| 4 | +// RUN: %ACC_RUN_PLACEHOLDER %t.out |
| 5 | + |
| 6 | +#include <cassert> |
| 7 | +#include <sycl/sycl.hpp> |
| 8 | +#include <type_traits> |
| 9 | + |
| 10 | +using namespace sycl; |
| 11 | + |
| 12 | +constexpr float eps = 1e-6; |
| 13 | + |
| 14 | +template <typename T> bool checkClose(T A, T B) { return ((A - B) / A) < eps; } |
| 15 | + |
| 16 | +#define __TEST_FFMATH_UNARY(func) \ |
| 17 | + template <typename T> void test_ffmath_##func(queue &deviceQueue) { \ |
| 18 | + T input[4] = {1.0004f, 1e-4f, 1.4f, 14.0f}; \ |
| 19 | + T res[4] = {-1, -1, -1, -1}; \ |
| 20 | + { \ |
| 21 | + buffer<T> input_buff{input, sycl::range{4}}; \ |
| 22 | + buffer<T> res_buff{res, sycl::range{4}}; \ |
| 23 | + deviceQueue.submit([&](handler &cgh) { \ |
| 24 | + accessor res_acc{res_buff, cgh}; \ |
| 25 | + accessor input_acc{input_buff, cgh}; \ |
| 26 | + cgh.parallel_for(sycl::range{4}, [=](sycl::id<1> idx) { \ |
| 27 | + res_acc[idx] = sycl::func(input_acc[idx]); \ |
| 28 | + }); \ |
| 29 | + }); \ |
| 30 | + } \ |
| 31 | + for (auto i = 0; i < 4; ++i) \ |
| 32 | + assert(checkClose(res[i], sycl::func(input[i]))); \ |
| 33 | + } |
| 34 | + |
| 35 | +#define __TEST_FFMATH_BINARY(func) \ |
| 36 | + template <typename T> void test_ffmath_##func(queue &deviceQueue) { \ |
| 37 | + T input[2][4] = {{1.0004f, 1e-4f, 1.4f, 14.0f}, \ |
| 38 | + {1.0004f, 1e-4f, 1.4f, 14.0f}}; \ |
| 39 | + T res[4] = {-1, -1, -1, -1}; \ |
| 40 | + { \ |
| 41 | + buffer<T> input1_buff{input[0], sycl::range{4}}; \ |
| 42 | + buffer<T> input2_buff{input[1], sycl::range{4}}; \ |
| 43 | + buffer<T> res_buff{res, sycl::range{4}}; \ |
| 44 | + deviceQueue.submit([&](handler &cgh) { \ |
| 45 | + accessor res_acc{res_buff, cgh}; \ |
| 46 | + accessor input1_acc{input1_buff, cgh}; \ |
| 47 | + accessor input2_acc{input2_buff, cgh}; \ |
| 48 | + cgh.parallel_for(sycl::range{4}, [=](sycl::id<1> idx) { \ |
| 49 | + res_acc[idx] = sycl::func(input1_acc[idx], input2_acc[idx]); \ |
| 50 | + }); \ |
| 51 | + }); \ |
| 52 | + } \ |
| 53 | + for (auto i = 0; i < 4; ++i) \ |
| 54 | + assert(checkClose(res[i], sycl::func(input[0][i], input[1][i]))); \ |
| 55 | + } |
| 56 | + |
| 57 | +__TEST_FFMATH_UNARY(cos) |
| 58 | +__TEST_FFMATH_UNARY(exp) |
| 59 | +__TEST_FFMATH_UNARY(exp2) |
| 60 | +__TEST_FFMATH_UNARY(exp10) |
| 61 | +__TEST_FFMATH_UNARY(log) |
| 62 | +__TEST_FFMATH_UNARY(log2) |
| 63 | +__TEST_FFMATH_UNARY(log10) |
| 64 | +__TEST_FFMATH_UNARY(rsqrt) |
| 65 | +__TEST_FFMATH_UNARY(sin) |
| 66 | +__TEST_FFMATH_UNARY(sqrt) |
| 67 | +__TEST_FFMATH_UNARY(tan) |
| 68 | + |
| 69 | +__TEST_FFMATH_BINARY(powr) |
| 70 | + |
| 71 | +int main() { |
| 72 | + |
| 73 | + queue q; |
| 74 | + test_ffmath_cos<float>(q); |
| 75 | + test_ffmath_exp<float>(q); |
| 76 | + test_ffmath_exp2<float>(q); |
| 77 | + test_ffmath_exp10<float>(q); |
| 78 | + test_ffmath_log<float>(q); |
| 79 | + test_ffmath_log2<float>(q); |
| 80 | + test_ffmath_log10<float>(q); |
| 81 | + test_ffmath_powr<float>(q); |
| 82 | + test_ffmath_rsqrt<float>(q); |
| 83 | + test_ffmath_sin<float>(q); |
| 84 | + test_ffmath_sqrt<float>(q); |
| 85 | + test_ffmath_tan<float>(q); |
| 86 | + test_ffmath_powr<float>(q); |
| 87 | + |
| 88 | + return 0; |
| 89 | +} |
0 commit comments