diff options
Diffstat (limited to 'test_conformance/math_brute_force/mad_float.cpp')
| -rw-r--r-- | test_conformance/math_brute_force/mad_float.cpp | 126 |
1 files changed, 25 insertions, 101 deletions
diff --git a/test_conformance/math_brute_force/mad_float.cpp b/test_conformance/math_brute_force/mad_float.cpp index 095a22ff..04ac5aa6 100644 --- a/test_conformance/math_brute_force/mad_float.cpp +++ b/test_conformance/math_brute_force/mad_float.cpp @@ -14,127 +14,52 @@ // limitations under the License. // +#include "common.h" #include "function_list.h" #include "test_functions.h" #include "utility.h" #include <cstring> -static int BuildKernel(const char *name, int vectorSize, cl_kernel *k, - cl_program *p, bool relaxedMode) -{ - const char *c[] = { "__kernel void math_kernel", - sizeNames[vectorSize], - "( __global float", - sizeNames[vectorSize], - "* out, __global float", - sizeNames[vectorSize], - "* in1, __global float", - sizeNames[vectorSize], - "* in2, __global float", - sizeNames[vectorSize], - "* in3 )\n" - "{\n" - " size_t i = get_global_id(0);\n" - " out[i] = ", - name, - "( in1[i], in2[i], in3[i] );\n" - "}\n" }; - - const char *c3[] = { - "__kernel void math_kernel", - sizeNames[vectorSize], - "( __global float* out, __global float* in, __global float* in2, " - "__global float* in3)\n" - "{\n" - " size_t i = get_global_id(0);\n" - " if( i + 1 < get_global_size(0) )\n" - " {\n" - " float3 f0 = vload3( 0, in + 3 * i );\n" - " float3 f1 = vload3( 0, in2 + 3 * i );\n" - " float3 f2 = vload3( 0, in3 + 3 * i );\n" - " f0 = ", - name, - "( f0, f1, f2 );\n" - " vstore3( f0, 0, out + 3*i );\n" - " }\n" - " else\n" - " {\n" - " size_t parity = i & 1; // Figure out how many elements are " - "left over after BUFFER_SIZE % (3*sizeof(float)). Assume power of two " - "buffer size \n" - " float3 f0;\n" - " float3 f1;\n" - " float3 f2;\n" - " switch( parity )\n" - " {\n" - " case 1:\n" - " f0 = (float3)( in[3*i], NAN, NAN ); \n" - " f1 = (float3)( in2[3*i], NAN, NAN ); \n" - " f2 = (float3)( in3[3*i], NAN, NAN ); \n" - " break;\n" - " case 0:\n" - " f0 = (float3)( in[3*i], in[3*i+1], NAN ); \n" - " f1 = (float3)( in2[3*i], in2[3*i+1], NAN ); \n" - " f2 = (float3)( in3[3*i], in3[3*i+1], NAN ); \n" - " break;\n" - " }\n" - " f0 = ", - name, - "( f0, f1, f2 );\n" - " switch( parity )\n" - " {\n" - " case 0:\n" - " out[3*i+1] = f0.y; \n" - " // fall through\n" - " case 1:\n" - " out[3*i] = f0.x; \n" - " break;\n" - " }\n" - " }\n" - "}\n" - }; - - const char **kern = c; - size_t kernSize = sizeof(c) / sizeof(c[0]); +namespace { - if (sizeValues[vectorSize] == 3) - { - kern = c3; - kernSize = sizeof(c3) / sizeof(c3[0]); - } - - char testName[32]; - snprintf(testName, sizeof(testName) - 1, "math_kernel%s", - sizeNames[vectorSize]); - - return MakeKernel(kern, (cl_uint)kernSize, testName, k, p, relaxedMode); +int BuildKernel(const char *name, int vectorSize, cl_kernel *k, cl_program *p, + bool relaxedMode) +{ + auto kernel_name = GetKernelName(vectorSize); + auto source = GetTernaryKernel(kernel_name, name, ParameterType::Float, + ParameterType::Float, ParameterType::Float, + ParameterType::Float, vectorSize); + std::array<const char *, 1> sources{ source.c_str() }; + return MakeKernel(sources.data(), sources.size(), kernel_name.c_str(), k, p, + relaxedMode); } -typedef struct BuildKernelInfo +struct BuildKernelInfo2 { - cl_uint offset; // the first vector size to build cl_kernel *kernels; - cl_program *programs; + Programs &programs; const char *nameInCode; bool relaxedMode; // Whether to build with -cl-fast-relaxed-math. -} BuildKernelInfo; +}; -static cl_int BuildKernelFn(cl_uint job_id, cl_uint thread_id UNUSED, void *p) +cl_int BuildKernelFn(cl_uint job_id, cl_uint thread_id UNUSED, void *p) { - BuildKernelInfo *info = (BuildKernelInfo *)p; - cl_uint i = info->offset + job_id; - return BuildKernel(info->nameInCode, i, info->kernels + i, - info->programs + i, info->relaxedMode); + BuildKernelInfo2 *info = (BuildKernelInfo2 *)p; + cl_uint vectorSize = gMinVectorSizeIndex + job_id; + return BuildKernel(info->nameInCode, vectorSize, info->kernels + vectorSize, + &(info->programs[vectorSize]), info->relaxedMode); } +} // anonymous namespace + int TestFunc_mad_Float(const Func *f, MTdata d, bool relaxedMode) { int error; logFunctionInfo(f->name, sizeof(cl_float), relaxedMode); - cl_program programs[VECTOR_SIZE_COUNT]; + Programs programs; cl_kernel kernels[VECTOR_SIZE_COUNT]; float maxError = 0.0f; float maxErrorVal = 0.0f; @@ -144,8 +69,8 @@ int TestFunc_mad_Float(const Func *f, MTdata d, bool relaxedMode) // Init the kernels { - BuildKernelInfo build_info = { gMinVectorSizeIndex, kernels, programs, - f->nameInCode, relaxedMode }; + BuildKernelInfo2 build_info{ kernels, programs, f->nameInCode, + relaxedMode }; if ((error = ThreadPool_Do(BuildKernelFn, gMaxVectorSizeIndex - gMinVectorSizeIndex, &build_info))) @@ -293,7 +218,6 @@ exit: for (auto k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++) { clReleaseKernel(kernels[k]); - clReleaseProgram(programs[k]); } return error; |