Makefile Optimization Guide

Overview

This guide covers techniques for optimizing Makefile performance, including parallel builds, dependency tracking, incremental builds, caching strategies, and performance profiling.

Parallel Builds

Enabling Parallel Execution

# Run with 4 parallel jobs
make -j4

# Use all CPU cores
make -j$(nproc)

# Unlimited parallel jobs (careful!)
make -j

Making Makefiles Parallel-Safe

Problem: Shared resources

# WRONG: Multiple rules write to same file
target1:
	echo "data1" >> shared.log

target2:
	echo "data2" >> shared.log

# With -j2, file corruption likely!

Solution: Proper dependencies

# RIGHT: Serialize access with dependencies
target2: target1

# Or use separate files
target1:
	echo "data1" > target1.log

target2:
	echo "data2" > target2.log

Controlling Parallelism

# Disable parallel builds for this Makefile
.NOTPARALLEL:

# Disable parallelism for specific targets
.NOTPARALLEL: install clean

# Serialize specific targets
install: build
	# Install runs after build completes

GNU Make 4.4+ Parallel Control Features

GNU Make 4.4 (released October 2022) introduced new features for fine-grained parallel control. These are becoming part of the upcoming POSIX standard.

.WAIT Special Target

The .WAIT target provides explicit ordering without creating artificial dependencies:

# .WAIT ensures prerequisites to its left complete
# before starting prerequisites to its right
all: compile .WAIT link .WAIT package

# Equivalent behavior without .WAIT would require:
# link: compile
# package: link
# But .WAIT is cleaner when targets are independent conceptually

Use Cases for .WAIT:

# Build phases with explicit ordering
build: setup .WAIT compile .WAIT test .WAIT package
	@echo "Build complete"

# Parallel within phases, serial between phases
ci: lint fmt .WAIT test.unit test.integration .WAIT build
# lint and fmt run in parallel
# then unit and integration tests run in parallel
# finally build runs

# Database migrations before tests
test: migrate .WAIT run-tests
	@echo "Tests complete"

Important Notes:

.WAIT only affects parallel builds (make -j)
With sequential execution, .WAIT has no effect
.WAIT doesn't create actual dependencies, just ordering
Available in GNU Make 4.4+ (check with make --version)

.NOTPARALLEL with Prerequisites (Enhanced)

In Make 4.4+, .NOTPARALLEL can take specific targets as prerequisites:

# Traditional: Disable ALL parallel execution
.NOTPARALLEL:

# NEW in 4.4: Serialize only specific targets
.NOTPARALLEL: install deploy cleanup

# This implicitly adds .WAIT between each prerequisite
# of the listed targets

When to Use .NOTPARALLEL with Prerequisites:

# Deployment must be serial (avoid race conditions)
.NOTPARALLEL: deploy

deploy: deploy-database deploy-backend deploy-frontend
	@echo "Deployment complete"
# deploy-database -> deploy-backend -> deploy-frontend (serial)

# But compilation can still be parallel
build: $(OBJECTS)
	$(CC) $^ -o $(TARGET)
# Object files compile in parallel (unaffected by .NOTPARALLEL: deploy)

Version Checking for Make 4.4+ Features

Check Make version before using 4.4+ features:

# Check Make version (4.4 = 4.4, need >= 4.4)
MAKE_VERSION_MAJOR := $(word 1,$(subst ., ,$(MAKE_VERSION)))
MAKE_VERSION_MINOR := $(word 2,$(subst ., ,$(MAKE_VERSION)))

# Simple version check
ifeq ($(shell expr $(MAKE_VERSION_MAJOR) \>= 4),1)
ifeq ($(shell expr $(MAKE_VERSION_MINOR) \>= 4),1)
    HAVE_WAIT := 1
endif
endif

# Alternative: Graceful degradation
ifdef HAVE_WAIT
    # Use .WAIT for modern Make
    all: compile .WAIT link
else
    # Fall back to dependencies for older Make
    link: compile
    all: link
endif

Practical Version Check Pattern:

# At the top of your Makefile
MIN_MAKE_VERSION := 4.4
CURRENT_MAKE_VERSION := $(MAKE_VERSION)

# Check and warn if using older Make
ifeq ($(shell printf '%s\n' "$(MIN_MAKE_VERSION)" "$(CURRENT_MAKE_VERSION)" | sort -V | head -n1),$(MIN_MAKE_VERSION))
    # Make version is sufficient
else
    $(warning GNU Make $(MIN_MAKE_VERSION)+ recommended. You have $(CURRENT_MAKE_VERSION))
    $(warning Some parallel control features may not work)
endif

Comparison: .WAIT vs Dependencies vs .NOTPARALLEL

Feature	Use Case	Make Version
Dependencies (`b: a`)	Actual dependency relationship	All
`.WAIT`	Ordering without dependency	4.4+
`.NOTPARALLEL:` (global)	Disable all parallel	All
`.NOTPARALLEL: target`	Serialize specific target's prereqs	4.4+

Example Comparison:

# Using dependencies (works in all Make versions)
# Problem: Creates false dependency relationship
link: compile
package: link
all: package

# Using .WAIT (Make 4.4+)
# Cleaner: Explicit ordering, no false dependencies
all: compile .WAIT link .WAIT package

# Using .NOTPARALLEL with targets (Make 4.4+)
# Best for: Targets that must never run in parallel
.NOTPARALLEL: deploy
deploy: step1 step2 step3

Optimal Parallel Structure

# Good parallel structure
SOURCES := src1.c src2.c src3.c src4.c
OBJECTS := $(SOURCES:.c=.o)

# All .o files can build in parallel
program: $(OBJECTS)
	$(CC) $^ -o $@

%.o: %.c
	$(CC) -c $< -o $@

Parallel execution:

make -j4
# Compiles 4 .c files simultaneously
# Then links when all are done

Dependency Tracking

Accurate Dependencies

Problem: Incorrect dependencies

# WRONG: Missing header dependencies
main.o: main.c
	$(CC) -c $< -o $@

# If common.h changes, main.o won't rebuild!

Solution: Automatic dependency generation

# Generate dependencies during compilation
%.o: %.c
	$(CC) $(CFLAGS) -MMD -MP -c $< -o $@

# Include generated .d files
-include $(OBJECTS:.o=.d)

Generated dependency file (main.d):

main.o: main.c common.h utils.h
common.h:
utils.h:

Dependency Flags

# -MMD: Generate dependency file (.d)
# -MP: Add phony targets for headers
# -MF file: Specify dependency file name

DEPFLAGS = -MMD -MP -MF $(@:.o=.d)

%.o: %.c
	$(CC) $(CFLAGS) $(DEPFLAGS) -c $< -o $@

Why -MP is Important

Without -MP:

# Generated main.d:
main.o: main.c utils.h

# If utils.h is deleted:
make: *** No rule to make target 'utils.h'. Stop.

With -MP:

# Generated main.d:
main.o: main.c utils.h
utils.h:

# If utils.h is deleted, make continues
# (assumes you also removed #include "utils.h")

Incremental Builds

Timestamp-Based Builds

Make rebuilds targets when prerequisites are newer:

# program rebuilt if any .o is newer
program: $(OBJECTS)
	$(CC) $^ -o $@

# main.o rebuilt if main.c or headers are newer
main.o: main.c common.h
	$(CC) -c main.c -o main.o

Optimizing Dependency Chains

Inefficient:

# Every source depends on config.h
# Changing config.h rebuilds EVERYTHING
main.o: main.c config.h
utils.o: utils.c config.h
helper.o: helper.c config.h

Better: Only include where needed

# Only main.c actually uses config.h
main.o: main.c config.h
utils.o: utils.c
helper.o: helper.c

Best: Use automatic dependencies

%.o: %.c
	$(CC) $(CFLAGS) -MMD -MP -c $< -o $@

-include $(DEPENDS)
# Automatically tracks which headers each file uses

Intermediate File Management

# Mark intermediate files
.INTERMEDIATE: $(OBJECTS)
# Deleted after use

# Keep important intermediate files
.SECONDARY: important.o
# Not deleted

# Never delete these files
.PRECIOUS: %.o %.d
# Protected from deletion

Avoiding Unnecessary Rebuilds

Problem: Timestamp updates without changes

# WRONG: Always updates config.h
config.h: config.h.in
	sed 's/@VERSION@/$(VERSION)/g' $< > $@
	# Updates timestamp even if content unchanged!

Solution: Conditional update

# RIGHT: Only update if different
config.h: config.h.in
	sed 's/@VERSION@/$(VERSION)/g' $< > $@.tmp
	cmp -s $@.tmp $@ || mv $@.tmp $@
	rm -f $@.tmp

Build Caching

Compiler Cache (ccache)

# Use ccache for faster recompilation
CC := ccache gcc
CXX := ccache g++

# Or conditionally:
ifeq ($(shell command -v ccache 2>/dev/null),)
CC ?= gcc
else
CC ?= ccache gcc
endif

Benefits:

Caches compilation results
Speeds up clean rebuilds
Useful for CI/CD and switching branches

Distcc for Distributed Compilation

# Distributed compilation across network
CC := distcc gcc
CXX := distcc g++

# Set number of jobs based on available hosts
DISTCC_HOSTS := localhost/2 build1/4 build2/4
JOBS := 10

Build Directory Caching

# Keep build artifacts between clean builds
.PHONY: clean distclean

clean:
	$(RM) $(TARGET)
	# Keep .o and .d files for faster rebuild

distclean: clean
	$(RM) -r $(BUILDDIR)
	# Complete clean

Performance Optimization Techniques

1. Use := Instead of =

# SLOW: Recursive expansion (evaluated every use)
SOURCES = $(wildcard src/*.c)
OBJECTS = $(SOURCES:.c=.o)
# $(OBJECTS) re-runs wildcard every time!

# FAST: Simple expansion (evaluated once)
SOURCES := $(wildcard src/*.c)
OBJECTS := $(SOURCES:.c=.o)
# Evaluated once when defined

2. Minimize Shell Invocations

# SLOW: Multiple shell calls
FILES = $(shell ls *.c)
COUNT = $(shell ls *.c | wc -l)

# FAST: Single shell call
FILES := $(wildcard *.c)
COUNT := $(words $(FILES))

3. Use Static Pattern Rules

OBJECTS := main.o utils.o helper.o

# FASTER: Static pattern rule (make knows exact files)
$(OBJECTS): %.o: %.c
	$(CC) -c $< -o $@

# SLOWER: Pattern rule (make searches for matches)
%.o: %.c
	$(CC) -c $< -o $@

4. Reduce Makefile Parsing Time

# SLOW: Complex shell commands in variable assignment
VERSION = $(shell git describe --tags --always --dirty)

# FAST: Use := to evaluate once
VERSION := $(shell git describe --tags --always --dirty)

# FASTER: Cache in file
VERSION := $(file < VERSION.txt)

5. Avoid Recursive Make

Inefficient: Recursive Make

# Top-level Makefile
SUBDIRS := lib1 lib2 app

all:
	for dir in $(SUBDIRS); do $(MAKE) -C $$dir; done

Problems:

Multiple make invocations (slow)
Incorrect dependency tracking
Parallel builds broken

Efficient: Non-Recursive Make

# Single Makefile
LIB1_SRC := $(wildcard lib1/*.c)
LIB2_SRC := $(wildcard lib2/*.c)
APP_SRC := $(wildcard app/*.c)

ALL_SRC := $(LIB1_SRC) $(LIB2_SRC) $(APP_SRC)
OBJECTS := $(ALL_SRC:.c=.o)

# Single dependency tree
# Accurate parallel builds

Reference: "Recursive Make Considered Harmful" by Peter Miller

Performance Profiling

Timing Individual Targets

# Time recipe execution
%.o: %.c
	@echo "Compiling $<..."
	@time $(CC) $(CFLAGS) -c $< -o $@

Build Time Measurement

# Time entire build
time make -j4

# Output:
# real    0m12.345s
# user    0m45.678s
# sys     0m3.456s

Debug Output for Performance Analysis

# Show what make is doing
make -d

# Show only remake decisions
make -d --debug=basic

# Show implicit rule search
make -d --debug=implicit

# Profile make itself
make --profile=profile.log

Finding Bottlenecks

# Add timing to critical paths
$(TARGET): $(OBJECTS)
	@echo "==> Linking $(TARGET)"
	@time $(CC) $(LDFLAGS) $^ $(LDLIBS) -o $@

%.o: %.c
	@echo "==> Compiling $<"
	@time $(CC) $(CFLAGS) -c $< -o $@

Optimization Best Practices

1. Structure for Parallelism

# Good: Independent compilation
OBJECTS := a.o b.o c.o d.o

program: $(OBJECTS)
	$(CC) $^ -o $@

%.o: %.c
	$(CC) -c $< -o $@

# make -j4 compiles 4 files at once

2. Accurate Dependencies

# Use automatic dependency generation
CFLAGS += -MMD -MP
-include $(OBJECTS:.o=.d)

# Not manual maintenance

3. Minimal Clean

# Keep intermediate files by default
clean:
	$(RM) $(TARGET)

# Full clean only when needed
distclean: clean
	$(RM) $(OBJECTS) $(DEPENDS)

4. Efficient Variable Usage

# Use := for computed values
SOURCES := $(wildcard src/*.c)
OBJECTS := $(SOURCES:.c=.o)

# Use ?= for user overrides
CC ?= gcc
CFLAGS ?= -O2

5. Avoid Unnecessary Work

# Don't rebuild if nothing changed
config.h: config.h.in Makefile
	@sed 's/@VERSION@/$(VERSION)/g' $< > $@.tmp
	@if ! cmp -s $@ $@.tmp; then \
		echo "  GEN     $@"; \
		mv $@.tmp $@; \
	else \
		rm -f $@.tmp; \
	fi

Complete Optimized Example

# Optimized Makefile for C project
.DELETE_ON_ERROR:
.SUFFIXES:

PROJECT := optimized
VERSION := 1.0.0

# User-overridable (use ?=)
CC ?= gcc
CFLAGS ?= -Wall -Wextra -O2
PREFIX ?= /usr/local

# Computed once (use :=)
SRCDIR := src
BUILDDIR := build
OBJDIR := $(BUILDDIR)/obj

SOURCES := $(wildcard $(SRCDIR)/*.c)
OBJECTS := $(SOURCES:$(SRCDIR)/%.c=$(OBJDIR)/%.o)
DEPENDS := $(OBJECTS:.o=.d)
TARGET := $(BUILDDIR)/$(PROJECT)

# Check for ccache
ifneq ($(shell command -v ccache 2>/dev/null),)
    CC := ccache $(CC)
endif

# Optimization flags for dependencies
DEPFLAGS = -MMD -MP

.PHONY: all clean distclean profile

all: $(TARGET)

# Link (serial)
$(TARGET): $(OBJECTS)
	@mkdir -p $(@D)
	@echo "  LD      $@"
	$(CC) $(LDFLAGS) $^ $(LDLIBS) -o $@

# Compile (parallel-safe)
$(OBJDIR)/%.o: $(SRCDIR)/%.c
	@mkdir -p $(@D)
	@echo "  CC      $<"
	$(CC) $(CPPFLAGS) $(CFLAGS) $(DEPFLAGS) -c $< -o $@

# Include auto-generated dependencies
-include $(DEPENDS)

# Minimal clean (keeps .o for faster rebuild)
clean:
	$(RM) $(TARGET)

# Full clean
distclean:
	$(RM) -r $(BUILDDIR)

# Profile build
profile:
	time $(MAKE) clean
	time $(MAKE) -j$(shell nproc) all

Benchmarking Results

Example project: 100 C files

Configuration	Build Time	Rebuild Time
Sequential (make)	45s	12s
Parallel -j2	25s	7s
Parallel -j4	15s	4s
Parallel -j8	12s	3s
Parallel + ccache (cold)	14s	3s
Parallel + ccache (warm)	3s	1s

Key takeaways:

Parallel builds: 3-4x speedup
ccache (warm): 10x speedup on clean builds
Accurate dependencies: Only rebuild what changed

Advanced Optimization

Precompiled Headers

# Generate precompiled header
$(OBJDIR)/common.h.gch: $(SRCDIR)/common.h
	@mkdir -p $(@D)
	$(CC) $(CPPFLAGS) $(CFLAGS) -x c-header $< -o $@

# Use precompiled header
$(OBJDIR)/%.o: $(SRCDIR)/%.c $(OBJDIR)/common.h.gch
	$(CC) $(CPPFLAGS) $(CFLAGS) -include $(OBJDIR)/common.h -c $< -o $@

Link-Time Optimization (LTO)

# Enable LTO for release builds
release: CFLAGS += -flto -O3
release: LDFLAGS += -flto -O3
release: $(TARGET)

Unity Builds

# Combine all sources into one compilation unit
unity.c: $(SOURCES)
	@echo "Generating unity build..."
	@for src in $(SOURCES); do \
		echo "#include \"$$src\"" >> $@; \
	done

unity.o: unity.c
	$(CC) $(CFLAGS) -c $< -o $@

# Fast single-file compilation
# Trade-off: No parallel compilation

Profiling Tools

# Make's built-in profiling
make --profile=profile.log
# Analyze profile.log

# Time individual targets
make -d 2>&1 | grep -E "Considering|Must remake"

# strace for system call analysis
strace -c make 2>&1 | tail -20

# Remake (make debugger)
remake --debug

/makefile-generator