Merge pull request #66 from db-tu-dresden/mwe_rtl_fpga

Improved validation and cmake target selection
db-tu-dresden · Oct 6, 2023 · d864d70 · d864d70
2 parents 9a1ead5 + eac44e3
commit d864d70
Show file tree

Hide file tree

Showing 2 changed files with 32 additions and 12 deletions.
diff --git a/examples/oneAPIfpga/CMakeLists.txt b/examples/oneAPIfpga/CMakeLists.txt
@@ -72,10 +72,11 @@ function(create_fpga_target targetName mainFile)
   target_link_options(${exec_target_name} PRIVATE -qactypes -fsycl -fintelfpga -Xshardware -Xsboard=${fpga_board} -reuse-exe=${CMAKE_CURRENT_BINARY_DIR}/${exec_target_name}) 
 endfunction()
 
-if(NOT DEFINED ${TARGET})
+if(NOT DEFINED TARGET)
   message(STATUS "No target specified. Assuming emulator")
   set(TARGET EMULATOR)
 endif()
+message(STATUS "BUILDING for ${TARGET}")
 if(${TARGET} STREQUAL "EMULATOR")
   create_fpga_emulator_target(clz_rtl clz_rtl_example.cpp)
 elseif(${TARGET} STREQUAL "FPGA_HARDWARE")

diff --git a/examples/oneAPIfpga/clz_rtl_example.cpp b/examples/oneAPIfpga/clz_rtl_example.cpp
@@ -15,6 +15,8 @@ struct count_leading_zero_kernel {
 int main(void) {
   // so far, only 32-bit unsigned integers are supported as RTL code
 
+  bool all_ok = true;
+
   using namespace tsl;
   executor<runtime::cpu> cpu_executor;
   using cpu_simd = simd<uint32_t, avx512>;
@@ -24,6 +26,7 @@ int main(void) {
     };
   using fpga_simd = simd<uint32_t, oneAPIfpgaRTL, 512>;
 
+  auto expected_result     = cpu_executor.allocate<uint32_t>(128);
   // allocate memory on host
   auto host_mem_data       = cpu_executor.allocate<uint32_t>(128);
   auto host_mem_result     = cpu_executor.allocate<uint32_t>(128);
@@ -36,9 +39,13 @@ int main(void) {
   auto usm_dev_mem_result  = fpga_executor.allocate<uint32_t>(128, oneAPI::MEMORY_ON_DEVICE{});
 
   // initialize input data
-  for (size_t i = 0; i < 128; i++) {
+  host_mem_data[0] = 0;
+  usm_host_mem_data[0] = 0;
+  expected_result[0] = 32;
+  for (uint32_t i = 1; i < 128; i++) {
     host_mem_data[i] = i;
     usm_host_mem_data[i] = i;
+    expected_result[i] = __builtin_clz(i);
   }
   // copy input data to FPGA device
   fpga_executor.copy(usm_dev_mem_data, usm_host_mem_data, 128);
@@ -54,15 +61,21 @@ int main(void) {
 
   // run kernel on CPU using avx512
   cpu_executor.submit<cpu_simd, count_leading_zero_kernel>(host_mem_result, host_mem_data, (size_t)128);
+  // check results
+  for (size_t i = 0; i < 128; i++) {
+    if (host_mem_result[i] != expected_result[i]) {
+      std::cerr << "ERROR: host_mem_result[" << i << "] = " << host_mem_result[i] << " != expected_result[" << i << "] = " << expected_result[i] << std::endl;
+      all_ok = false;
+    }
+  }
 
   // run kernel on FPGA using oneAPIfpgaRTL (RTL code is built seperately). Use USM-Host memory
   fpga_executor.submit<fpga_simd, count_leading_zero_kernel>(usm_host_mem_result, usm_host_mem_data, (size_t)128);
-
   // check results
   for (size_t i = 0; i < 128; i++) {
-    if (host_mem_result[i] != usm_host_mem_result[i]) {
-      std::cerr << "ERROR: host_mem_result[" << i << "] = " << host_mem_result[i] << " != usm_host_mem_result[" << i << "] = " << usm_host_mem_result[i] << std::endl;
-      std::terminate();
+    if (host_mem_result[i] != expected_result[i]) {
+      std::cerr << "ERROR: usm_host_mem_result[" << i << "] = " << usm_host_mem_result[i] << " != expected_result[" << i << "] = " << expected_result[i] << std::endl;
+      all_ok = false;
     }
   }
 
@@ -71,12 +84,11 @@ int main(void) {
 
   // copy output to host
   fpga_executor.copy(usm_host_mem_result, usm_dev_mem_result, 128);
-
   // check results
   for (size_t i = 0; i < 128; i++) {
-    if (host_mem_result[i] != usm_host_mem_result[i]) {
-      std::cerr << "ERROR: host_mem_result[" << i << "] = " << host_mem_result[i] << " != usm_host_mem_result[" << i << "] = " << usm_host_mem_result[i] << std::endl;
-      std::terminate();
+    if (host_mem_result[i] != expected_result[i]) {
+      std::cerr << "ERROR: usm_host_mem_result[" << i << "] = " << usm_host_mem_result[i] << " != expected_result[" << i << "] = " << expected_result[i] << std::endl;
+      all_ok = false;
     }
   }
 
@@ -87,8 +99,15 @@ int main(void) {
   fpga_executor.deallocate(usm_host_mem_data);
   cpu_executor.deallocate(host_mem_result);
   cpu_executor.deallocate(host_mem_data);
+  cpu_executor.deallocate(expected_result);
 
   // done
-  std::cout << "Everything worked fine!" << std::endl;
-  return 0;
+  if (all_ok) {
+    std::cout << "Everything worked fine!" << std::endl;
+    return 0;
+  } else {
+    std::cout << "There were errors!" << std::endl;
+    return 1;
+  }
+
 }