diff options
author | rander <rander.wang@intel.com> | 2017-06-19 15:22:25 +0800 |
---|---|---|
committer | Yang Rong <rong.r.yang@intel.com> | 2017-07-06 17:53:15 +0800 |
commit | 01389c558d5e966a9052e34c540b48180d02a436 (patch) | |
tree | 0ae239ec031d736b238a2606528b40e95cc74c17 /utests/compiler_fdiv2rcp.cpp | |
parent | 49b63a260c66fc9438ad8b29344d8e1a4ba0285f (diff) | |
download | beignet-01389c558d5e966a9052e34c540b48180d02a436.tar.gz |
utests: add utest for fdiv to rcp
for this case 1.0f/src, 2.0f/src can be converted,
but 3.0f/src and i/src cant
Signed-off-by: rander.wang <rander.wang@intel.com>
Reviewed-by: Yang Rong <rong.r.yang@intel.com>
Diffstat (limited to 'utests/compiler_fdiv2rcp.cpp')
-rw-r--r-- | utests/compiler_fdiv2rcp.cpp | 61 |
1 files changed, 61 insertions, 0 deletions
diff --git a/utests/compiler_fdiv2rcp.cpp b/utests/compiler_fdiv2rcp.cpp new file mode 100644 index 00000000..7a983920 --- /dev/null +++ b/utests/compiler_fdiv2rcp.cpp @@ -0,0 +1,61 @@ +#include "utest_helper.hpp" +#include <cmath> + +void compiler_fdiv2rcp(void) { + const int n = 1024; + float src[n]; + + // Setup kernel and buffers + OCL_CREATE_KERNEL("compiler_fdiv2rcp"); + OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(float), NULL); + OCL_CREATE_BUFFER(buf[1], 0, n * 4 * 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 = 1.0f / (src[i]); + float gpu = dst[4 * i]; + if (fabsf(cpu - gpu) >= 1e-3) { + printf("%f %f %f", src[i], cpu, gpu); + OCL_ASSERT(0); + } + + cpu = i / (src[i]); + gpu = dst[4 * i + 1]; + if (fabsf(cpu - gpu) >= 1e-3) { + printf("%f %f %f", src[i], cpu, gpu); + OCL_ASSERT(0); + } + + cpu = 2.0f / (src[i]); + gpu = dst[4 * i + 2]; + if (fabsf(cpu - gpu) >= 1e-3) { + printf("%f %f %f", src[i], cpu, gpu); + OCL_ASSERT(0); + } + + cpu = 3.0f / (src[i]); + gpu = dst[4 * i + 3]; + 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_fdiv2rcp); |