makefile

CC      = gcc
CFLAGS  = -O3 -Wall -Wextra -std=c11 -I./include -march=native \
          -D_POSIX_C_SOURCE=199309L
LDFLAGS = -lm

SIMDFLAGS = -mavx2 -mfma

SRC = src/main.c src/matrix.c src/multiply_naive.c src/multiply_opt.c \
      src/multiply_simd.c src/timer.c src/verify.c

ifeq ($(OS),Windows_NT)
    OUT = matrix_app.exe
    RUN = .\$(OUT)
    RM  = del /Q
else
    OUT = matrix_app
    RUN = ./$(OUT)
    RM  = rm -f
endif

# ── Targets ───────────────────────────────────────────────────────────────────

.PHONY: all run clean cachegrind perf help

## Default: build with AVX2 SIMD
all:
	$(CC) $(CFLAGS) $(SIMDFLAGS) $(SRC) -o $(OUT) $(LDFLAGS)
	@echo "Build OK → $(OUT)"

## Build without AVX2 (fallback scalar path in multiply_simd.c)
nosimd:
	$(CC) $(CFLAGS) $(SRC) -o $(OUT) $(LDFLAGS)
	@echo "Build OK (no SIMD) → $(OUT)"

run: all
	$(RUN)

## Run Cachegrind cache profiling (only Linux )
## Output: cachegrind.out.<pid>  →  parsed to cachegrind_report.txt
cachegrind: all
	valgrind --tool=cachegrind \
	         --cachegrind-out-file=cachegrind.out \
	         --D1=32768,8,64 \
	         --LL=8388608,16,64 \
	         $(RUN)
	cg_annotate cachegrind.out > cachegrind_report.txt
	@echo "Cache profile written to cachegrind_report.txt"

## Run perf hardware-counter profiling (Linux only, requires perf installed)
## Output: perf_report.txt
perf: all
	perf stat -e cycles,instructions,cache-misses,cache-references,\
L1-dcache-loads,L1-dcache-load-misses,LLC-loads,LLC-load-misses \
	$(RUN) 2> perf_report.txt
	@echo "Perf counters written to perf_report.txt"

## Remove build artifacts
clean:
	-$(RM) $(OUT) benchmark_results.csv cachegrind.out cachegrind_report.txt perf_report.txt

help:
	@echo "Targets:"
	@echo "  make          — build with AVX2"
	@echo "  make nosimd   — build without AVX2"
	@echo "  make run      — build and run benchmark"
	@echo "  make cachegrind — Valgrind cache profiling (Linux)"
	@echo "  make perf     — perf hardware counters (Linux)"
	@echo "  make clean    — remove artifacts"