pq-code-package · mkannwischer · Aug 8, 2025 · Aug 4, 2025
@@ -39,4 +39,5 @@ runs:
       - shell: ${{ env.SHELL }}
         run: |
           make clean
-          tests func --exec-wrapper="valgrind --error-exitcode=1 --track-origins=yes ${{ inputs.valgrind_flags }}" --cflags="-DMLK_CONFIG_CT_TESTING_ENABLED -DNTESTS=50 ${{ inputs.cflags }}"
+          # Disable the AArch64 SHA3 extension as it's not yet supported by valgrind
+          MK_COMPILER_SUPPORTS_SHA3=0 tests func --exec-wrapper="valgrind --error-exitcode=1 --track-origins=yes ${{ inputs.valgrind_flags }}" --cflags="-DMLK_CONFIG_CT_TESTING_ENABLED -DNTESTS=50 ${{ inputs.cflags }}"
@@ -16,7 +16,8 @@
 	build test all \
 	clean quickcheck check-defined-CYCLES \
 	size_512 size_768 size_1024 size \
-	run_size_512 run_size_768 run_size_1024 run_size
+	run_size_512 run_size_768 run_size_1024 run_size \
+	host_info
 
 SHELL := /bin/bash
 .DEFAULT_GOAL := build
@@ -26,6 +27,8 @@ all: build
 W := $(EXEC_WRAPPER)
 
 include test/mk/config.mk
+include test/mk/compiler.mk
+include test/mk/auto.mk
 include test/mk/components.mk
 include test/mk/rules.mk
 
@@ -86,6 +89,15 @@ acvp_1024: $(MLKEM1024_DIR)/bin/acvp_mlkem1024
 	$(Q)echo "  ACVP       ML-KEM-1024:  $^"
 acvp: acvp_512 acvp_768 acvp_1024
 
+ifeq ($(HOST_PLATFORM),Linux-aarch64)
+# valgrind does not work with the AArch64 SHA3 extension
+# Use armv8-a as the target architecture, overwriting a
+# potential earlier addition of armv8.4-a+sha3.
+$(MLKEM512_DIR)/bin/test_stack512:   CFLAGS += -march=armv8-a
+$(MLKEM768_DIR)/bin/test_stack768:   CFLAGS += -march=armv8-a
+$(MLKEM1024_DIR)/bin/test_stack1024: CFLAGS += -march=armv8-a
+endif
+
 stack_512: $(MLKEM512_DIR)/bin/test_stack512
 	$(Q)echo "  STACK      ML-KEM-512:   $^"
 stack_768: $(MLKEM768_DIR)/bin/test_stack768
@@ -176,6 +188,30 @@ run_size: \
 	run_size_768 \
 	run_size_1024
 
+# Display host and compiler feature detection information
+# Shows which architectural features are supported by both the compiler and host CPU
+# Usage: make host_info [AUTO=0|1] [CROSS_PREFIX=...]
+host_info:
+	@echo "=== Host and Compiler Feature Detection ==="
+	@echo "Host Platform: $(HOST_PLATFORM)"
+	@echo "Target Architecture: $(ARCH)"
+	@echo "Compiler: $(CC)"
+	@echo "Cross Prefix: $(if $(CROSS_PREFIX),$(CROSS_PREFIX),<none>)"
+	@echo "AUTO: $(AUTO)"
+	@echo ""
+ifeq ($(ARCH),x86_64)
+	@echo "=== x86_64 Feature Support ==="
+	@echo "AVX2: Host $(if $(filter 1,$(MK_HOST_SUPPORTS_AVX2)),✅,❌) Compiler $(if $(filter 1,$(MK_COMPILER_SUPPORTS_AVX2)),✅,❌)"
+	@echo "SSE2: Host $(if $(filter 1,$(MK_HOST_SUPPORTS_SSE2)),✅,❌) Compiler $(if $(filter 1,$(MK_COMPILER_SUPPORTS_SSE2)),✅,❌)"
+	@echo "BMI2: Host $(if $(filter 1,$(MK_HOST_SUPPORTS_BMI2)),✅,❌) Compiler $(if $(filter 1,$(MK_COMPILER_SUPPORTS_BMI2)),✅,❌)"
+else ifeq ($(ARCH),aarch64)
+	@echo "=== AArch64 Feature Support ==="
+	@echo "SHA3: Host $(if $(filter 1,$(MK_HOST_SUPPORTS_SHA3)),✅,❌) Compiler $(if $(filter 1,$(MK_COMPILER_SUPPORTS_SHA3)),✅,❌)"
+else
+	@echo "=== Architecture Not Supported ==="
+	@echo "No specific feature detection available for $(ARCH)"
+endif
+
 clean:
 	-$(RM) -rf *.gcno *.gcda *.lcov *.o *.so
 	-$(RM) -rf $(BUILD_DIR)

@@ -61,7 +61,7 @@ endif
 # Native compilation
 ifeq ($(CROSS_PREFIX),)
 ifeq ($(HOST_PLATFORM),Linux-x86_64)
-	CFLAGS += -mavx2 -mbmi2 -mpopcnt -maes
+	CFLAGS += -mavx2 -mbmi2
 	CFLAGS += -DMLK_FORCE_X86_64
 else ifeq ($(HOST_PLATFORM),Linux-aarch64)
 	CFLAGS += -DMLK_FORCE_AARCH64
@@ -70,7 +70,7 @@ else ifeq ($(HOST_PLATFORM),Darwin-arm64)
 endif
 # Cross compilation
 else ifneq ($(findstring x86_64, $(CROSS_PREFIX)),)
-	CFLAGS += -mavx2 -mbmi2 -mpopcnt -maes
+	CFLAGS += -mavx2 -mbmi2
 	CFLAGS += -DMLK_FORCE_X86_64
 else ifneq ($(findstring aarch64_be, $(CROSS_PREFIX)),)
 	CFLAGS += -DMLK_FORCE_AARCH64_EB

@@ -28,7 +28,7 @@ endif
 # Native compilation
 ifeq ($(CROSS_PREFIX),)
 ifeq ($(HOST_PLATFORM),Linux-x86_64)
-	CFLAGS += -mavx2 -mbmi2 -mpopcnt -maes
+	CFLAGS += -mavx2 -mbmi2
 	CFLAGS += -DMLK_FORCE_X86_64
 else ifeq ($(HOST_PLATFORM),Linux-aarch64)
 	CFLAGS += -DMLK_FORCE_AARCH64
@@ -37,7 +37,7 @@ else ifeq ($(HOST_PLATFORM),Darwin-arm64)
 endif
 # Cross compilation
 else ifneq ($(findstring x86_64, $(CROSS_PREFIX)),)
-	CFLAGS += -mavx2 -mbmi2 -mpopcnt -maes
+	CFLAGS += -mavx2 -mbmi2
 	CFLAGS += -DMLK_FORCE_X86_64
 else ifneq ($(findstring aarch64_be, $(CROSS_PREFIX)),)
 	CFLAGS += -DMLK_FORCE_AARCH64_EB

@@ -1,29 +1,113 @@
 # SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
 #
-# Automatically detect system architecture and set preprocessor etc accordingly
+# Automatically detect system architecture and set preprocessor flags accordingly
+# This file detects host CPU capabilities and combines them with compiler support
+# to enable optimal compilation flags.
 
-# Native compilation
-ifeq ($(CROSS_PREFIX),)
+ifndef _AUTO_MK
+_AUTO_MK :=
+
+# Helper function to check if host CPU supports a feature
+# Usage: $(call check_host_feature,feature_pattern,source_command)
+define check_host_feature
+$(shell $(2) 2>/dev/null | grep -q "$(1)" && echo 1 || echo 0)
+endef
+
+# x86_64 architecture detection
+ifeq ($(ARCH),x86_64)
+
+# Host CPU feature detection for x86_64
 ifeq ($(HOST_PLATFORM),Linux-x86_64)
-	CFLAGS += -mavx2 -mbmi2 -mpopcnt -maes
-	CFLAGS += -DMLK_FORCE_X86_64
-else ifeq ($(HOST_PLATFORM),Linux-aarch64)
-	CFLAGS += -DMLK_FORCE_AARCH64
+# Linux: Use /proc/cpuinfo
+MK_HOST_SUPPORTS_AVX2 := $(call check_host_feature,avx2,cat /proc/cpuinfo)
+MK_HOST_SUPPORTS_SSE2 := $(call check_host_feature,sse2,cat /proc/cpuinfo)
+MK_HOST_SUPPORTS_BMI2 := $(call check_host_feature,bmi2,cat /proc/cpuinfo)
+else ifeq ($(HOST_PLATFORM),Darwin-x86_64)
+# macOS: Use sysctl
+MK_HOST_SUPPORTS_AVX2 := $(call check_host_feature,AVX2,sysctl -n machdep.cpu.leaf7_features)
+MK_HOST_SUPPORTS_SSE2 := $(call check_host_feature,SSE2,sysctl -n machdep.cpu.features)
+MK_HOST_SUPPORTS_BMI2 := $(call check_host_feature,BMI2,sysctl -n machdep.cpu.leaf7_features)
+else ifneq ($(CROSS_PREFIX),)
+# Cross-compilation: assume all features are supported
+MK_HOST_SUPPORTS_AVX2 := 1
+MK_HOST_SUPPORTS_SSE2 := 1
+MK_HOST_SUPPORTS_BMI2 := 1
+else
+# Other platforms: assume no support
+MK_HOST_SUPPORTS_AVX2 := 0
+MK_HOST_SUPPORTS_SSE2 := 0
+MK_HOST_SUPPORTS_BMI2 := 0
+endif # HOST_PLATFORM x86_64
+
+endif # x86_64
+
+# AArch64 architecture detection
+ifeq ($(ARCH),aarch64)
+
+# Host CPU feature detection for AArch64
+ifeq ($(HOST_PLATFORM),Linux-aarch64)
+# Linux: Use /proc/cpuinfo (look for sha3 in Features line)
+MK_HOST_SUPPORTS_SHA3 := $(call check_host_feature,sha3,cat /proc/cpuinfo)
 else ifeq ($(HOST_PLATFORM),Darwin-arm64)
-	CFLAGS += -DMLK_FORCE_AARCH64
+# macOS: Use sysctl to check for SHA3 support
+MK_HOST_SUPPORTS_SHA3 := $(call check_host_feature,1,sysctl -n hw.optional.armv8_2_sha3)
+else ifneq ($(CROSS_PREFIX),)
+# Cross-compilation: assume all features are supported
+MK_HOST_SUPPORTS_SHA3 := 1
+else
+# Other platforms: assume no support
+MK_HOST_SUPPORTS_SHA3 := 0
+endif # HOST_PLATFORM aarch64
+
+endif # aarch64
+
+# Only apply CFLAGS modifications if AUTO=1
+ifeq ($(AUTO),1)
+
+# x86_64 CFLAGS configuration
+ifeq ($(ARCH),x86_64)
+CFLAGS += -DMLK_FORCE_X86_64
+
+# Add flags only if both compiler and host support the feature
+ifeq ($(MK_COMPILER_SUPPORTS_AVX2)$(MK_HOST_SUPPORTS_AVX2),11)
+CFLAGS += -mavx2
 endif
-# Cross compilation
-else ifneq ($(findstring x86_64, $(CROSS_PREFIX)),)
-	CFLAGS += -mavx2 -mbmi2 -mpopcnt -maes
-	CFLAGS += -DMLK_FORCE_X86_64
-else ifneq ($(findstring aarch64_be, $(CROSS_PREFIX)),)
-	CFLAGS += -DMLK_FORCE_AARCH64_EB
-else ifneq ($(findstring aarch64, $(CROSS_PREFIX)),)
-	CFLAGS += -DMLK_FORCE_AARCH64
-else ifneq ($(findstring riscv64, $(CROSS_PREFIX)),)
-	CFLAGS += -DMLK_FORCE_RISCV64
-else ifneq ($(findstring riscv32, $(CROSS_PREFIX)),)
-	CFLAGS += -DMLK_FORCE_RISCV32
-else ifneq ($(findstring powerpc64le, $(CROSS_PREFIX)),)
-	CFLAGS += -DMLK_FORCE_PPC64LE
+
+ifeq ($(MK_COMPILER_SUPPORTS_BMI2)$(MK_HOST_SUPPORTS_BMI2),11)
+CFLAGS += -mbmi2
 endif
+endif # x86_64
+
+# AArch64 CFLAGS configuration
+ifeq ($(ARCH),aarch64)
+CFLAGS += -DMLK_FORCE_AARCH64
+
+# Add SHA3 flags only if both compiler and host support it
+ifeq ($(MK_COMPILER_SUPPORTS_SHA3)$(MK_HOST_SUPPORTS_SHA3),11)
+CFLAGS += -march=armv8.4-a+sha3
+endif
+endif # aarch64
+
+# AArch64 Big Endian CFLAGS configuration
+ifeq ($(ARCH),aarch64_be)
+CFLAGS += -DMLK_FORCE_AARCH64_EB
+endif # aarch64_be
+
+# RISC-V 64-bit CFLAGS configuration
+ifeq ($(ARCH),riscv64)
+CFLAGS += -DMLK_FORCE_RISCV64
+endif # riscv64
+
+# RISC-V 32-bit CFLAGS configuration
+ifeq ($(ARCH),riscv32)
+CFLAGS += -DMLK_FORCE_RISCV32
+endif # riscv32
+
+# PowerPC 64-bit Little Endian CFLAGS configuration
+ifeq ($(ARCH),powerpc64le)
+CFLAGS += -DMLK_FORCE_PPC64LE
+endif # powerpc64le
+
+endif # AUTO=1
+
+endif # _AUTO_MK
@@ -0,0 +1,65 @@
+# SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
+#
+# Compiler feature detection for mlkem-native
+# This file detects whether the compiler supports various architectural features
+# used by mlkem-native through compile-time tests with C code containing inline assembly.
+#
+# Feature detection can be overridden by setting the corresponding variable on the command line:
+#   make MK_COMPILER_SUPPORTS_SHA3=0    # Disable SHA3 detection
+#   make MK_COMPILER_SUPPORTS_AVX2=0    # Disable AVX2 detection
+#   make MK_COMPILER_SUPPORTS_BMI2=0    # Disable BMI2 detection
+#   make MK_COMPILER_SUPPORTS_SSE2=0    # Disable SSE2 detection
+
+ifndef _COMPILER_MK
+_COMPILER_MK :=
+
+# Normalize architecture names
+ARCH := $(shell uname -m)
+ifeq ($(ARCH),arm64)
+ARCH := aarch64
+endif
+
+# Override ARCH for cross-compilation based on CROSS_PREFIX
+ifneq ($(CROSS_PREFIX),)
+ifneq ($(findstring x86_64, $(CROSS_PREFIX)),)
+ARCH := x86_64
+else ifneq ($(findstring aarch64_be, $(CROSS_PREFIX)),)
+ARCH := aarch64_be
+else ifneq ($(findstring aarch64, $(CROSS_PREFIX)),)
+ARCH := aarch64
+else ifneq ($(findstring riscv64, $(CROSS_PREFIX)),)
+ARCH := riscv64
+else ifneq ($(findstring riscv32, $(CROSS_PREFIX)),)
+ARCH := riscv32
+else ifneq ($(findstring powerpc64le, $(CROSS_PREFIX)),)
+ARCH := powerpc64le
+endif
+endif # CROSS_PREFIX
+
+# x86_64 feature detection
+ifeq ($(ARCH),x86_64)
+
+# Test AVX2 support using C with inline assembly
+# Can be overridden by setting MK_COMPILER_SUPPORTS_AVX2=0/1 on command line
+MK_COMPILER_SUPPORTS_AVX2 ?= $(shell echo 'int main() { __asm__("vpxor %%ymm0, %%ymm1, %%ymm2" ::: "ymm0", "ymm1", "ymm2"); return 0; }' | $(CC) -mavx2 -x c - -c -o /dev/null 2>/dev/null && echo 1 || echo 0)
+
+# Test SSE2 support using C with inline assembly
+# Can be overridden by setting MK_COMPILER_SUPPORTS_SSE2=0/1 on command line
+MK_COMPILER_SUPPORTS_SSE2 ?= $(shell echo 'int main() { __asm__("pxor %%xmm0, %%xmm1" ::: "xmm0", "xmm1"); return 0; }' | $(CC) -msse2 -x c - -c -o /dev/null 2>/dev/null && echo 1 || echo 0)
+
+# Test BMI2 support using C with inline assembly
+# Can be overridden by setting MK_COMPILER_SUPPORTS_BMI2=0/1 on command line
+MK_COMPILER_SUPPORTS_BMI2 ?= $(shell echo 'int main() { __asm__("pdep %%eax, %%ebx, %%ecx" ::: "eax", "ebx", "ecx"); return 0; }' | $(CC) -mbmi2 -x c - -c -o /dev/null 2>/dev/null && echo 1 || echo 0)
+
+endif # x86_64
+
+# AArch64 feature detection
+ifeq ($(ARCH),aarch64)
+
+# Test SHA3 support (Armv8.4-a+SHA3) using C with inline assembly
+# Can be overridden by setting MK_COMPILER_SUPPORTS_SHA3=0/1 on command line
+MK_COMPILER_SUPPORTS_SHA3 ?= $(shell echo 'int main() { __asm__("eor3 v0.16b, v1.16b, v2.16b, v3.16b" ::: "v0", "v1", "v2", "v3"); return 0; }' | $(CC) -march=armv8.4-a+sha3 -x c - -c -o /dev/null 2>/dev/null && echo 1 || echo 0)
+
+endif # aarch64
+
+endif # _COMPILER_MK
@@ -81,10 +81,6 @@ CYCLES ?=
 OPT ?= 1
 RETAINED_VARS := CROSS_PREFIX CYCLES OPT AUTO
 
-ifeq ($(AUTO),1)
-include test/mk/auto.mk
-endif
-
 BUILD_DIR ?= test/build
 
 MAKE_OBJS = $(2:%=$(1)/%.o)