Add/expand compile guards for VAES

Some hardware accelerated implementations (e.g. aegis256x4_avx2) make use of `vaes` instructions without being gated behind `HAVE_VAESINTRIN_H`. This breaks cross compilation using rust's cross [1] build tool, presumably because the MinGW GCC implementation it packages doesn't support `vaes`. The resulting output of `cross build --target=x86_64-pc-windows-gnu` complains about GCC not knowing `vaes` as an attribute and about the respective intrinsics being unknown: ``` cargo:warning=src/c/libaegis/src/aegis128x2/aegis128x2_avx2.c:16:17: error: attribute(target("vaes")) is unknown cargo:warning= # pragma GCC target("vaes,avx2") cargo:warning= ^~~ cargo:warning=src/c/libaegis/src/aegis128x2/aegis128x2_avx2.c: In function 'aegis128x2_update': cargo:warning=src/c/libaegis/src/aegis128x2/aegis128x2_avx2.c:32:39: warning: implicit declaration of function '_mm256_aesenc_epi128'; did you mean '_mm256_bsrli_epi128'? [-Wimplicit-function-declaration] cargo:warning= # define AES_ENC(A, B) _mm256_aesenc_epi128((A), (B)) cargo:warning= ^ cargo:warning=src/c/libaegis/src/aegis128x2/aegis128x2_avx2.c:40:16: note: in expansion of macro 'AES_ENC' cargo:warning= state[7] = AES_ENC(state[6], state[7]); cargo:warning= ^~~~~~~ cargo:warning=src/c/libaegis/src/aegis128x2/aegis128x2_avx2.c:40:14: error: incompatible types when assigning to type 'aes_block_t {aka __vector(4) long long int}' from type 'int' cargo:warning= state[7] = AES_ENC(state[6], state[7]); cargo:warning= ... ``` This change introduces compile-time checks for `HAVE_VAESINTRIN_H` where they aren't present yet and extends some existing checks in some implementation picking functions (which would produce missing linker targets otherwise). I also took the liberty to change the indentation for all nested pecompiler instructions where I added new #ifdef checks in the effort of keeping the overall code style in line with the other files; however, this makes the diff seem larger than it really is. [1]: https://github.com/cross-rs/cross
aegis-aead · Apr 29, 2024 · 3ce51b1 · 3ce51b1
1 parent 3dac3ad
commit 3ce51b1
Show file tree

Hide file tree

Showing 12 changed files with 88 additions and 60 deletions.
diff --git a/src/aegis128x2/aegis128x2.c b/src/aegis128x2/aegis128x2.c
@@ -187,10 +187,12 @@ aegis128x2_pick_best_implementation(void)
 #endif
 
 #if defined(__x86_64__) || defined(_M_AMD64) || defined(__i386__) || defined(_M_IX86)
+#    ifdef HAVE_VAESINTRIN_H
     if (aegis_runtime_has_vaes() && aegis_runtime_has_avx2()) {
         implementation = &aegis128x2_avx2_implementation;
         return 0;
     }
+#    endif
     if (aegis_runtime_has_aesni() && aegis_runtime_has_avx()) {
         implementation = &aegis128x2_aesni_implementation;
         return 0;

diff --git a/src/aegis128x2/aegis128x2_avx2.c b/src/aegis128x2/aegis128x2_avx2.c
@@ -10,26 +10,28 @@
 #    include "aegis128x2.h"
 #    include "aegis128x2_avx2.h"
 
-#    ifdef __clang__
-#        pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
-#    elif defined(__GNUC__)
-#        pragma GCC target("vaes,avx2")
-#    endif
+#    ifdef HAVE_VAESINTRIN_H
+
+#        ifdef __clang__
+#            pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
+#        elif defined(__GNUC__)
+#            pragma GCC target("vaes,avx2")
+#        endif
 
-#    include <immintrin.h>
+#        include <immintrin.h>
 
-#    define AES_BLOCK_LENGTH 32
+#        define AES_BLOCK_LENGTH 32
 
 typedef __m256i aes_block_t;
 
-#    define AES_BLOCK_XOR(A, B) _mm256_xor_si256((A), (B))
-#    define AES_BLOCK_AND(A, B) _mm256_and_si256((A), (B))
-#    define AES_BLOCK_LOAD128_BROADCAST(A) \
-        _mm256_broadcastsi128_si256(_mm_loadu_si128((const void *) (A)))
-#    define AES_BLOCK_LOAD(A)         _mm256_loadu_si256((const aes_block_t *) (const void *) (A))
-#    define AES_BLOCK_LOAD_64x2(A, B) _mm256_broadcastsi128_si256(_mm_set_epi64x((A), (B)))
-#    define AES_BLOCK_STORE(A, B)     _mm256_storeu_si256((aes_block_t *) (void *) (A), (B))
-#    define AES_ENC(A, B)             _mm256_aesenc_epi128((A), (B))
+#        define AES_BLOCK_XOR(A, B) _mm256_xor_si256((A), (B))
+#        define AES_BLOCK_AND(A, B) _mm256_and_si256((A), (B))
+#        define AES_BLOCK_LOAD128_BROADCAST(A) \
+            _mm256_broadcastsi128_si256(_mm_loadu_si128((const void *) (A)))
+#        define AES_BLOCK_LOAD(A)         _mm256_loadu_si256((const aes_block_t *) (const void *) (A))
+#        define AES_BLOCK_LOAD_64x2(A, B) _mm256_broadcastsi128_si256(_mm_set_epi64x((A), (B)))
+#        define AES_BLOCK_STORE(A, B)     _mm256_storeu_si256((aes_block_t *) (void *) (A), (B))
+#        define AES_ENC(A, B)             _mm256_aesenc_epi128((A), (B))
 
 static inline void
 aegis128x2_update(aes_block_t *const state, const aes_block_t d1, const aes_block_t d2)
@@ -50,7 +52,7 @@ aegis128x2_update(aes_block_t *const state, const aes_block_t d1, const aes_bloc
     state[4] = AES_BLOCK_XOR(state[4], d2);
 }
 
-#    include "aegis128x2_common.h"
+#        include "aegis128x2_common.h"
 
 struct aegis128x2_implementation aegis128x2_avx2_implementation = {
     .encrypt_detached              = encrypt_detached,
@@ -66,8 +68,10 @@ struct aegis128x2_implementation aegis128x2_avx2_implementation = {
     .state_decrypt_detached_final  = state_decrypt_detached_final,
 };
 
-#    ifdef __clang__
-#        pragma clang attribute pop
+#         ifdef __clang__
+#             pragma clang attribute pop
+#         endif
+
 #    endif
 
 #endif
diff --git a/src/aegis128x2/aegis128x2_avx2.h b/src/aegis128x2/aegis128x2_avx2.h
@@ -4,6 +4,8 @@
 #include "../common/common.h"
 #include "implementations.h"
 
+#ifdef HAVE_VAESINTRIN_H
 extern struct aegis128x2_implementation aegis128x2_avx2_implementation;
+#endif
 
 #endif
diff --git a/src/aegis128x4/aegis128x4.c b/src/aegis128x4/aegis128x4.c
@@ -193,11 +193,11 @@ aegis128x4_pick_best_implementation(void)
         implementation = &aegis128x4_avx512_implementation;
         return 0;
     }
-#    endif
     if (aegis_runtime_has_vaes() && aegis_runtime_has_avx2()) {
         implementation = &aegis128x4_avx2_implementation;
         return 0;
     }
+#    endif
     if (aegis_runtime_has_aesni() && aegis_runtime_has_avx()) {
         implementation = &aegis128x4_aesni_implementation;
         return 0;

diff --git a/src/aegis128x4/aegis128x4_avx2.c b/src/aegis128x4/aegis128x4_avx2.c
@@ -10,15 +10,17 @@
 #    include "aegis128x4.h"
 #    include "aegis128x4_avx2.h"
 
-#    ifdef __clang__
-#        pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
-#    elif defined(__GNUC__)
-#        pragma GCC target("vaes,avx2")
-#    endif
+#    ifdef HAVE_VAESINTRIN_H
+
+#        ifdef __clang__
+#            pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
+#        elif defined(__GNUC__)
+#            pragma GCC target("vaes,avx2")
+#        endif
 
-#    include <immintrin.h>
+#        include <immintrin.h>
 
-#    define AES_BLOCK_LENGTH 64
+#        define AES_BLOCK_LENGTH 64
 
 typedef struct {
     __m256i b0;
@@ -83,7 +85,7 @@ aegis128x4_update(aes_block_t *const state, const aes_block_t d1, const aes_bloc
     state[4] = AES_BLOCK_XOR(state[4], d2);
 }
 
-#    include "aegis128x4_common.h"
+#        include "aegis128x4_common.h"
 
 struct aegis128x4_implementation aegis128x4_avx2_implementation = {
     .encrypt_detached              = encrypt_detached,
@@ -99,8 +101,10 @@ struct aegis128x4_implementation aegis128x4_avx2_implementation = {
     .state_decrypt_detached_final  = state_decrypt_detached_final,
 };
 
-#    ifdef __clang__
-#        pragma clang attribute pop
+#        ifdef __clang__
+#            pragma clang attribute pop
+#        endif
+
 #    endif
 
-#endif
+#endif
diff --git a/src/aegis128x4/aegis128x4_avx2.h b/src/aegis128x4/aegis128x4_avx2.h
@@ -4,6 +4,8 @@
 #include "../common/common.h"
 #include "implementations.h"
 
+#ifdef HAVE_VAESINTRIN_H
 extern struct aegis128x4_implementation aegis128x4_avx2_implementation;
+#endif
 
 #endif
diff --git a/src/aegis256x2/aegis256x2.c b/src/aegis256x2/aegis256x2.c
@@ -187,10 +187,12 @@ aegis256x2_pick_best_implementation(void)
 #endif
 
 #if defined(__x86_64__) || defined(_M_AMD64) || defined(__i386__) || defined(_M_IX86)
+#    ifdef HAVE_VAESINTRIN_H
     if (aegis_runtime_has_vaes() && aegis_runtime_has_avx2()) {
         implementation = &aegis256x2_avx2_implementation;
         return 0;
     }
+#    endif
     if (aegis_runtime_has_aesni() && aegis_runtime_has_avx()) {
         implementation = &aegis256x2_aesni_implementation;
         return 0;

diff --git a/src/aegis256x2/aegis256x2_avx2.c b/src/aegis256x2/aegis256x2_avx2.c
@@ -10,26 +10,28 @@
 #    include "aegis256x2.h"
 #    include "aegis256x2_avx2.h"
 
-#    ifdef __clang__
-#        pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
-#    elif defined(__GNUC__)
-#        pragma GCC target("vaes,avx2")
-#    endif
+#    ifdef HAVE_VAESINTRIN_H
+
+#        ifdef __clang__
+#            pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
+#        elif defined(__GNUC__)
+#            pragma GCC target("vaes,avx2")
+#        endif
 
-#    include <immintrin.h>
+#        include <immintrin.h>
 
-#    define AES_BLOCK_LENGTH 32
+#        define AES_BLOCK_LENGTH 32
 
 typedef __m256i aes_block_t;
 
-#    define AES_BLOCK_XOR(A, B) _mm256_xor_si256((A), (B))
-#    define AES_BLOCK_AND(A, B) _mm256_and_si256((A), (B))
-#    define AES_BLOCK_LOAD128_BROADCAST(A) \
-        _mm256_broadcastsi128_si256(_mm_loadu_si128((const void *) (A)))
-#    define AES_BLOCK_LOAD(A)         _mm256_loadu_si256((const aes_block_t *) (const void *) (A))
-#    define AES_BLOCK_LOAD_64x2(A, B) _mm256_broadcastsi128_si256(_mm_set_epi64x((A), (B)))
-#    define AES_BLOCK_STORE(A, B)     _mm256_storeu_si256((aes_block_t *) (void *) (A), (B))
-#    define AES_ENC(A, B)             _mm256_aesenc_epi128((A), (B))
+#        define AES_BLOCK_XOR(A, B) _mm256_xor_si256((A), (B))
+#        define AES_BLOCK_AND(A, B) _mm256_and_si256((A), (B))
+#        define AES_BLOCK_LOAD128_BROADCAST(A) \
+            _mm256_broadcastsi128_si256(_mm_loadu_si128((const void *) (A)))
+#        define AES_BLOCK_LOAD(A)         _mm256_loadu_si256((const aes_block_t *) (const void *) (A))
+#        define AES_BLOCK_LOAD_64x2(A, B) _mm256_broadcastsi128_si256(_mm_set_epi64x((A), (B)))
+#        define AES_BLOCK_STORE(A, B)     _mm256_storeu_si256((aes_block_t *) (void *) (A), (B))
+#        define AES_ENC(A, B)             _mm256_aesenc_epi128((A), (B))
 
 static inline void
 aegis256x2_update(aes_block_t *const state, const aes_block_t d)
@@ -45,7 +47,7 @@ aegis256x2_update(aes_block_t *const state, const aes_block_t d)
     state[0] = AES_BLOCK_XOR(AES_ENC(tmp, state[0]), d);
 }
 
-#    include "aegis256x2_common.h"
+#        include "aegis256x2_common.h"
 
 struct aegis256x2_implementation aegis256x2_avx2_implementation = {
     .encrypt_detached              = encrypt_detached,
@@ -61,8 +63,10 @@ struct aegis256x2_implementation aegis256x2_avx2_implementation = {
     .state_decrypt_detached_final  = state_decrypt_detached_final,
 };
 
-#    ifdef __clang__
-#        pragma clang attribute pop
+#        ifdef __clang__
+#            pragma clang attribute pop
+#        endif
+
 #    endif
 
-#endif
+#endif
diff --git a/src/aegis256x2/aegis256x2_avx2.h b/src/aegis256x2/aegis256x2_avx2.h
@@ -4,6 +4,8 @@
 #include "../common/common.h"
 #include "implementations.h"
 
+#ifdef HAVE_VAESINTRIN_H
 extern struct aegis256x2_implementation aegis256x2_avx2_implementation;
+#endif
 
 #endif
diff --git a/src/aegis256x4/aegis256x4.c b/src/aegis256x4/aegis256x4.c
@@ -193,11 +193,11 @@ aegis256x4_pick_best_implementation(void)
         implementation = &aegis256x4_avx512_implementation;
         return 0;
     }
-#    endif
     if (aegis_runtime_has_vaes() && aegis_runtime_has_avx2()) {
         implementation = &aegis256x4_avx2_implementation;
         return 0;
     }
+#    endif
     if (aegis_runtime_has_aesni() && aegis_runtime_has_avx()) {
         implementation = &aegis256x4_aesni_implementation;
         return 0;

diff --git a/src/aegis256x4/aegis256x4_avx2.c b/src/aegis256x4/aegis256x4_avx2.c
@@ -10,15 +10,17 @@
 #    include "aegis256x4.h"
 #    include "aegis256x4_avx2.h"
 
-#    ifdef __clang__
-#        pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
-#    elif defined(__GNUC__)
-#        pragma GCC target("vaes,avx2")
-#    endif
+#    ifdef HAVE_VAESINTRIN_H
+
+#        ifdef __clang__
+#            pragma clang attribute push(__attribute__((target("vaes,avx2"))), apply_to = function)
+#        elif defined(__GNUC__)
+#            pragma GCC target("vaes,avx2")
+#        endif
 
-#    include <immintrin.h>
+#        include <immintrin.h>
 
-#    define AES_BLOCK_LENGTH 64
+#        define AES_BLOCK_LENGTH 64
 
 typedef struct {
     __m256i b0;
@@ -78,7 +80,7 @@ aegis256x4_update(aes_block_t *const state, const aes_block_t d)
     state[0] = AES_BLOCK_XOR(AES_ENC(tmp, state[0]), d);
 }
 
-#    include "aegis256x4_common.h"
+#        include "aegis256x4_common.h"
 
 struct aegis256x4_implementation aegis256x4_avx2_implementation = {
     .encrypt_detached              = encrypt_detached,
@@ -94,8 +96,10 @@ struct aegis256x4_implementation aegis256x4_avx2_implementation = {
     .state_decrypt_detached_final  = state_decrypt_detached_final,
 };
 
-#    ifdef __clang__
-#        pragma clang attribute pop
+#        ifdef __clang__
+#            pragma clang attribute pop
+#        endif
+
 #    endif
 
 #endif
diff --git a/src/aegis256x4/aegis256x4_avx2.h b/src/aegis256x4/aegis256x4_avx2.h
@@ -4,6 +4,8 @@
 #include "../common/common.h"
 #include "implementations.h"
 
+#ifdef HAVE_VAESINTRIN_H
 extern struct aegis256x4_implementation aegis256x4_avx2_implementation;
+#endif
 
 #endif