Function createTranspose1

examples/transpose/run.cpp:31–41

)";

inline KernelCode createTranspose1(const char *shaderTemplate,
				   const size_t M, const size_t N,
                                   const Shape &workgroupSize = {256, 1, 1},
                                   NumType precision = kf32) {
  std::string codeString(shaderTemplate);
  replaceAll(codeString, {{"{{workgroupSize}}", toString(workgroupSize)},
                          {"{{precision}}", toString(precision)},
                          {"{{M}}", toString(M)},
                          {"{{N}}", toString(N)}});
  return {codeString, workgroupSize};
}

// Shared memory cache-blocking
static const char *kShaderTranspose2 = R"(