✅ Many changes to benchmarks

- Support and test lpc4078
- Add benchmarks that uses empty strings (way worse than returning data)

Author: kammce

benchmark/CMakeLists.txt

@@ -110,6 +110,49 @@ foreach(test_path ${GENERATED_TEST})
   endif()
 endforeach()
 
+
+file(GLOB WRITTEN_TEST "${CMAKE_CURRENT_SOURCE_DIR}/written_tests/*.cpp")
+
+foreach(test_path ${WRITTEN_TEST})
+  get_filename_component(test_file_name ${test_path} NAME)
+  set(elf ${test_file_name}.elf)
+
+  message(STATUS "Generating written Demo for \"${elf}\"")
+  add_executable(${elf} ${test_path})
+  target_include_directories(${elf} PRIVATE .)
+  target_compile_options(${elf} PRIVATE ${BENCHMARK_COMPILE_OPTIONS})
+
+  # Check if the test file name starts with "result_"
+  string(REGEX MATCH "^result_" is_result_file ${test_file_name})
+  if(is_result_file)
+    target_compile_options(${elf} PRIVATE
+      $<$<COMPILE_LANGUAGE:CXX>:-fno-exceptions>)
+  else()
+    target_compile_options(${elf} PRIVATE
+      $<$<COMPILE_LANGUAGE:CXX>:-fexceptions>)
+  endif()
+
+  target_link_libraries(${elf} PRIVATE
+    startup_code
+    libhal::$ENV{LIBHAL_PLATFORM_LIBRARY}
+  )
+  target_link_options(${elf} PRIVATE -fno-threadsafe-statics
+    -L${CMAKE_SOURCE_DIR}/
+    -Wl,-Map=${CMAKE_BINARY_DIR}/${elf}.map
+  )
+  if(prebuilt-picolibc_FOUND)
+    target_link_libraries(${elf} PRIVATE picolibc)
+  endif()
+
+  if(${CMAKE_CROSSCOMPILING})
+    # Convert elf into .bin, .hex and other formats needed for programming
+    # devices.
+    libhal_post_build(${elf})
+    libhal_disassemble(${elf})
+  endif()
+endforeach()
+
+
 # Build nearpoint binary if if available
 if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/nearpoint.cpp")
   message(STATUS "nearpoint.cpp exists!")

benchmark/README.md

@@ -78,11 +78,33 @@ To execute the benchmark you can either supply the python script with a binary
 file to flash OR provide it a newline delimitated list of executable paths.
 
 ```bash
-python3 halbord_benchmark.py -o all_tests.csv -f all_tests.list
+python3 halbord_benchmark.py -n 10000000 -f test_order.list
 ```
 
 We currently provide the following test lists within the directory:
 
-1. `all_tests.list`
-2. `except_tests.list`
-3. `result_tests.list`
+1. `test_order.csv`
+2. `test_order_lpc4078.csv`
+3. `test_nearpoint.csv`
+
+## About `fast_gcc_except`
+
+This test can only fully reach its maximum performance if you make the
+`search_EIT_Table()` function visible and weakened so the implementation in
+this project can take over.
+
+To do this you must run the following command on ARM GCC:
+
+```bash
+arm-none-eabi-objcopy \
+  /path/to/arm-none-eabi-gcc/lib/gcc/arm-none-eabi/14.2.1/thumb/v7-m/nofp/libgcc.a \
+  --globalize-symbol=search_EIT_table \
+  --weaken-symbol=search_EIT_table \
+  /path/to/arm-none-eabi-gcc/lib/gcc/arm-none-eabi/14.2.1/thumb/v7-m/nofp/libgcc.a
+```
+
+You need to change the `/path/to/` to the actual path of your
+`arm-none-eabi-gcc` toolchain.
+
+Once you've done this, delete the `build` directory if it exists and rebuilt
+the project.

benchmark/fast_gcc_unwind.cpp

@@ -1,3 +1,4 @@
+#include <atomic>
 #include <unwind.h>
 
 #include <algorithm>
@@ -6,6 +7,8 @@
 #include <cstdlib>
 #include <span>
 
+#include <platform.hpp>
+
 /* Misc constants.  */
 #define R_IP 12
 #define R_SP 13
@@ -76,48 +79,18 @@ typedef struct __EIT_entry
   std::uint32_t content;
 } __EIT_entry;
 
-std::uint32_t selfrel_offset31(std::uint32_t const* p)
+std::uintptr_t selfrel_offset31(std::uint32_t const* p)
 {
-  std::uint32_t offset;
+  auto offset = static_cast<std::int32_t>(*p);
 
-  offset = *p;
   /* Sign extend to 32 bits.  */
-  if (offset & (1 << 30))
-    offset |= 1u << 31;
-  else
-    offset &= ~(1u << 31);
+  offset <<= 1;
+  offset >>= 1;
 
-  return offset + (std::uint32_t)p;
+  return std::bit_cast<std::uintptr_t>(offset +
+                                       std::bit_cast<std::intptr_t>(p));
 }
 
-struct eit_entry_less_than
-{
-  [[gnu::always_inline]] static std::uint32_t to_absolute(
-    __EIT_entry const& entry)
-  {
-    auto entry_addr = reinterpret_cast<std::uint32_t>(&entry.content);
-    // Sign extend :D
-    entry_addr <<= 1;
-    entry_addr >>= 1;
-    return entry_addr;
-  }
-
-  bool operator()(__EIT_entry const& left, __EIT_entry const& right)
-  {
-    return left.fnoffset < right.fnoffset;
-  }
-  bool operator()(__EIT_entry const& left, std::uint32_t right)
-  {
-    std::uint32_t absolute_left = to_absolute(left);
-    return absolute_left < right;
-  }
-  bool operator()(std::uint32_t left, __EIT_entry const& right)
-  {
-    std::uint32_t absolute_right = to_absolute(right);
-    return left < absolute_right;
-  }
-};
-
 /* Return the next byte of unwinding information, or CODE_FINISH if there is
    no data remaining.  */
 [[gnu::always_inline]] _uw8 next_unwind_byte(__gnu_unwind_state* uws)
@@ -411,18 +384,76 @@ extern "C"
     return __wrap___gnu_unwind_execute(context, &uws);
   }
 
-  __EIT_entry const* search_EIT_table(__EIT_entry const* table,
-                                      int nrec,  // NOLINT
+  struct eit_entry_less_than
+  {
+    bool operator()(__EIT_entry const& left, __EIT_entry const& right)
+    {
+      return selfrel_offset31(&left.fnoffset) <
+             selfrel_offset31(&right.fnoffset);
+    }
+
+    bool operator()(__EIT_entry const& left, std::uint32_t right)
+    {
+      return selfrel_offset31(&left.fnoffset) < right;
+    }
+    bool operator()(std::uint32_t left, __EIT_entry const& right)
+    {
+      return left < selfrel_offset31(&right.fnoffset);
+    }
+    bool operator()(std::uint32_t left, std::uint32_t right)
+    {
+      return left < right;
+    }
+  };
+
+#if 1
+  __EIT_entry const* search_EIT_table(__EIT_entry const* table,  // NOLINT
+                                      int nrec,                  // NOLINT
                                       std::uint32_t return_address)
   {
     if (nrec == 0) {
       return nullptr;
     }
-    std::span<__EIT_entry const> table_span(table, nrec);
-    auto const& entry = std::upper_bound(table_span.begin(),
-                                         table_span.end(),
-                                         return_address,
-                                         eit_entry_less_than{});
-    return entry.base();
+
+    auto const& next_entry = std::ranges::upper_bound(
+      std::span(table, nrec), return_address, eit_entry_less_than{});
+    auto const& actual_entry = *(next_entry - 1);
+
+    return &actual_entry;
+  }
+#else
+  __EIT_entry const* search_EIT_table(__EIT_entry const* table,  // NOLINT
+                                      int nrec,                  // NOLINT
+                                      _uw return_address)
+  {
+    _uw next_fn;
+    _uw this_fn;
+    int n, left, right;
+    if (nrec == 0)
+      return (__EIT_entry*)0;
+
+    left = 0;
+    right = nrec - 1;
+
+    while (1) {
+      n = (left + right) / 2;
+      this_fn = selfrel_offset31(&table[n].fnoffset);
+      if (n != nrec - 1) {
+        next_fn = selfrel_offset31(&table[n + 1].fnoffset) - 1;
+      } else {
+        next_fn = (_uw)0 - 1;
+      }
+      if (return_address < this_fn) {
+        if (n == left) {
+          return (__EIT_entry*)0;
+        }
+        right = n - 1;
+      } else if (return_address <= next_fn) {
+        return &table[n];
+      } else {
+        left = n + 1;
+      }
+    }
   }
+#endif
 }

benchmark/generated_tests/estell_except.cpp.csv

@@ -1,61 +1,61 @@
 depth,error_size,destructor_percentage,pulse_us
-50,4,0,341.33
-35,4,0,244.79
-15,4,0,117.21
-5,4,0,52.12
-1,4,0,26.54
-50,4,25,429.54
-35,4,25,306.83
-15,4,25,144.96
-5,4,25,66.79
-1,4,25,34.25
-50,4,50,510.33
-35,4,50,366.88
-15,4,50,171.33
-5,4,50,73.96
-1,4,50,34.25
-50,4,100,677.79
-35,4,100,480.96
-15,4,100,217.92
-5,4,100,86.96
-1,4,100,34.25
-50,16,0,341.33
-35,16,0,244.83
-15,16,0,117.12
-5,16,0,52.17
-1,16,0,26.58
-50,16,25,429.5
-35,16,25,307.21
-15,16,25,144.96
-5,16,25,66.42
-1,16,25,34.29
-50,16,50,510.33
-35,16,50,366.88
-15,16,50,171.38
-5,16,50,74.0
-1,16,50,34.25
-50,16,100,677.75
-35,16,100,481.0
-15,16,100,217.92
-5,16,100,87.04
-1,16,100,34.25
-50,65,0,357.0
-35,65,0,260.42
-15,65,0,132.75
-5,65,0,67.75
-1,65,0,42.17
-50,65,25,445.5
-35,65,25,322.75
-15,65,25,160.12
-5,65,25,81.96
-1,65,25,49.75
-50,65,50,525.88
-35,65,50,382.38
-15,65,50,186.88
-5,65,50,89.5
-1,65,50,49.75
-50,65,100,693.33
-35,65,100,496.54
-15,65,100,233.46
-5,65,100,102.5
-1,65,100,49.79
+50,4,0,335.04
+35,4,0,240.29
+15,4,0,115.17
+5,4,0,51.33
+1,4,0,26.29
+50,4,25,423.21
+35,4,25,302.25
+15,4,25,142.96
+5,4,25,65.96
+1,4,25,33.96
+50,4,50,504.0
+35,4,50,362.29
+15,4,50,169.29
+5,4,50,73.12
+1,4,50,33.92
+50,4,100,671.38
+35,4,100,476.38
+15,4,100,215.92
+5,4,100,86.17
+1,4,100,33.96
+50,16,0,335.0
+35,16,0,240.25
+15,16,0,115.12
+5,16,0,51.33
+1,16,0,26.25
+50,16,25,423.21
+35,16,25,302.67
+15,16,25,142.96
+5,16,25,65.62
+1,16,25,34.0
+50,16,50,504.04
+35,16,50,362.38
+15,16,50,169.38
+5,16,50,73.12
+1,16,50,33.96
+50,16,100,671.42
+35,16,100,476.42
+15,16,100,215.92
+5,16,100,86.17
+1,16,100,33.96
+50,65,0,345.54
+35,65,0,250.75
+15,65,0,125.62
+5,65,0,61.83
+1,65,0,36.75
+50,65,25,434.0
+35,65,25,313.04
+15,65,25,152.96
+5,65,25,76.04
+1,65,25,44.38
+50,65,50,514.42
+35,65,50,372.71
+15,65,50,179.75
+5,65,50,83.58
+1,65,50,44.38
+50,65,100,681.83
+35,65,100,486.83
+15,65,100,226.33
+5,65,100,96.58
+1,65,100,44.38

benchmark/generated_tests/except.cpp

@@ -1,6 +1,7 @@
 #include <cstdint>
 
 #include <array>
+#include <string>
 
 #include <platform.hpp>