the negtive Add is like: exp -a llvm transfer it to: add x -a, 0 exp x
Signed-off-by: rander.wang <rander.w...@intel.com> --- kernels/compiler_remove_negative_add.cl | 4 ++++ utests/CMakeLists.txt | 3 ++- utests/compiler_remove_negative_add.cpp | 40 +++++++++++++++++++++++++++++++++++ 3 files changed, 46 insertions(+), 1 deletion(-) create mode 100644 kernels/compiler_remove_negative_add.cl create mode 100644 utests/compiler_remove_negative_add.cpp diff --git a/kernels/compiler_remove_negative_add.cl b/kernels/compiler_remove_negative_add.cl new file mode 100644 index 0000000..92799db --- /dev/null +++ b/kernels/compiler_remove_negative_add.cl @@ -0,0 +1,4 @@ +kernel void compiler_remove_negative_add(global float *src, global float *dst) { + int i = get_global_id(0); + dst[i] = exp2(-src[i]); +}; diff --git a/utests/CMakeLists.txt b/utests/CMakeLists.txt index ae9e2bd..8967389 100644 --- a/utests/CMakeLists.txt +++ b/utests/CMakeLists.txt @@ -301,7 +301,8 @@ set (utests_sources runtime_pipe_query.cpp compiler_pipe_builtin.cpp compiler_device_enqueue.cpp - compiler_sqrtDiv.cpp) + compiler_sqrtDiv.cpp + compiler_remove_negative_add.cpp) if (LLVM_VERSION_NODOT VERSION_GREATER 34) SET(utests_sources diff --git a/utests/compiler_remove_negative_add.cpp b/utests/compiler_remove_negative_add.cpp new file mode 100644 index 0000000..335ba43 --- /dev/null +++ b/utests/compiler_remove_negative_add.cpp @@ -0,0 +1,40 @@ +#include <cmath> +#include "utest_helper.hpp" + +void compiler_remove_negative_add(void) { + const int n = 1024; + float src[n]; + + // Setup kernel and buffers + OCL_CREATE_KERNEL("compiler_remove_negative_add"); + OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(float), NULL); + OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(float), NULL); + OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]); + OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]); + globals[0] = n; + locals[0] = 16; + + for (int j = 0; j < 1024; j++) { + OCL_MAP_BUFFER(0); + for (int i = 0; i < n; ++i) { + src[i] = ((float*) buf_data[0])[i] = (j * n + i + 1) * 0.001f; + } + OCL_UNMAP_BUFFER(0); + + OCL_NDRANGE(1); + + OCL_MAP_BUFFER(1); + float *dst = (float*) buf_data[1]; + for (int i = 0; i < n; ++i) { + float cpu = exp2(-src[i]); + float gpu = dst[i]; + if (fabsf(cpu - gpu) >= 1e-3) { + printf("%f %f %f", src[i], cpu, gpu); + OCL_ASSERT(0); + } + } + OCL_UNMAP_BUFFER(1); + } +} + +MAKE_UTEST_FROM_FUNCTION (compiler_remove_negative_add); -- 2.7.4 _______________________________________________ Beignet mailing list Beignet@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/beignet