42 cub3D

XPM load

XPM tex files can be loaded either via

t_img *img = mlx_xpm_file_to_image(app->mlx, (char *)filename, &width, &height);

or, if statically, linked via

t_img *img = mlx_int_parse_xpm(app->mlx, (char *)xpm_data, 0, mlx_int_static_line);

to be able to refer to the static data we shall make sure that the xpm structs have unique identifiers it can be done via running this little sed script in the loop

for file in *.xpm; do
  name=$(basename "$file" .xpm)
  sed -E -i 's|^static (const )?char \*\w+\[]|static \1char *'"${name}"'_xpm[]|' "$file"
done

then we just need to compile them into object files

$(BUILD_DIR)/%.xpm.o: %.xpm
		@$(CC) -Dstatic= -x c -c $^ -o $@

Info

clang doesn't let us compiling unused static variables into object files due to Dead Code Elimination. Therefore we apply a simple trick: removing the static modifier via preprocessor statement -Dstatic= which is similar to

#define static 

We can double check the result via

$ nm build/textures/arm_cannon_big.xpm.o
0000000000000000 D arm_cannon_big_xpm

where capital D means it is global (external) and lies in the initialized data section

then load it at runtime like this

void calc_idxs_avx2(void)
{
	t_tex			*tex;
	extern const char	*title_card_xpm[];

	tex->data = img_to_tex_static_row_major(app, title_card_xpm, &tex->x, &tex->y);
}

Bug

Original mlx_int_static_line() has a bug preventing it from working properly https://github.com/42paris/minilibx-linux/pull/32/files

thus we are patching the library with overriding the symbol strlcpy_is_not_posix with our own one via globilizing the strlcpy_is_not_posix and weakening the link

## mlx
$(LIBX): $(LIBX_DIR)/Makefile.gen
		+$(MAKE) -C $(LIBX_DIR) -f Makefile.gen all
		@if [ ! -d $(BUILD_DIR) ]; then mkdir -p $(BUILD_DIR); fi
		@ar x $(LIBX) mlx_xpm.o --output $(BUILD_DIR)
		@ar d $(LIBX) $(BUILD_DIR)/mlx_xpm.o
		@objcopy --globalize-symbol=strlcpy_is_not_posix $(BUILD_DIR)/mlx_xpm.o
		@objcopy --weaken-symbol=strlcpy_is_not_posix $(BUILD_DIR)/mlx_xpm.o
		@ar rcs $(LIBX) $(BUILD_DIR)/mlx_xpm.o
		@$(RM) -v $(BUILD_DIR)/mlx_xpm.o
		@echo "LIBX BUILD COMPLETE!"

Build

• To build project with CMmake, run:

cmake -S . -B build
cmake --build build

or using Ninja:

cmake -S . -B build -G Ninja
cmake --build build

Optimisations

LLVM IR-level struct decomposition: There is an optimization called Scalar Replacement of Aggregates (SROA).

Clang/LLVM is SROA, which effectively decomposes structs into Static Single Assignment values (SSA) during optimization. There's no direct source-level feature for this, but we can help the compiler by:

• Keeping structs small (fewer fields, all PODs). // POD - Plain Old Data
• Avoiding pointers to structs or taking addresses of struct fields.
• Declaring structs locally in the function if possible.

Intrinsics

#include <mmintrin.h>  // MMX
#include <xmmintrin.h> // SSE
#include <emmintrin.h> // SSE2
#include <pmmintrin.h> // SSE3
#include <tmmintrin.h> // SSSE3
#include <smmintrin.h> // SSE4.1
#include <nmmintrin.h> // SSE4.2
#include <ammintrin.h> // SSE4A
#include <wmmintrin.h> // AES
#include <immintrin.h> // AVX, AVX2, FMA
#include <zmmintrin.h> // AVX512

Including one of these pulls in all previous ones

Sanitizers

export ASAN_OPTIONS=color=always:print_summary=1:verbosity=0:symbolize=1:detect_stack_use_after_return=true:malloc_context_size=20:detect_leaks=1:handle_segv=2:abort_on_error=1:fast_unwind_on_malloc=0

With Clang:

export ASAN_SYMBOLIZER_PATH="$(command -v llvm-symbolizer)"

If you need object lists and more chatter:

export LSAN_OPTIONS=verbosity=1:report_objects=1 ASAN_OPTIONS=fast_unwind_on_malloc=0"

export UBSAN_OPTIONS=print_stacktrace=1

External libraries

SDL_mixer 2.0.4:

• curl https://raw.githubusercontent.com/libsdl-org/SDL_mixer/refs/tags/release-2.0.4/SDL_mixer.h -O

FreeType2 6.18.1:

• /lib/x86_64-linux-gnu/libfreetype.so.6