Enable AMD64 AES implementation for WIN64

* cipher/rijndael-amd64.S: Enable when
HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS defined.
(ELF): New macro to mask lines with ELF specific commands.
* cipher/rijndael-internal.h (USE_AMD64_ASM): Enable when
HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS defined.
(do_encrypt, do_decrypt)
[USE_AMD64_ASM && !HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS]: Use
assembly block to call AMD64 assembly encrypt/decrypt function.
--

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

3 files changed, 48 insertions, 6 deletions

diff --git a/cipher/rijndael-amd64.S b/cipher/rijndael-amd64.S
index 24c555a2..b149e948 100644
--- a/cipher/rijndael-amd64.S
+++ b/cipher/rijndael-amd64.S
@@ -20,7 +20,8 @@
 
 #ifdef __x86_64
 #include <config.h>
-#if defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) && defined(USE_AES)
+#if (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+     defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS)) && defined(USE_AES)
 
 #ifdef __PIC__
 #  define RIP (%rip)
@@ -28,6 +29,12 @@
 #  define RIP
 #endif
 
+#ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
+# define ELF(...) __VA_ARGS__
+#else
+# define ELF(...) /*_*/
+#endif
+
 .text
 
 /* table macros */
@@ -205,7 +212,7 @@
 
 .align 8
 .globl _gcry_aes_amd64_encrypt_block
-.type   _gcry_aes_amd64_encrypt_block,@function;
+ELF(.type   _gcry_aes_amd64_encrypt_block,@function;)
 
 _gcry_aes_amd64_encrypt_block:
 	/* input:
@@ -279,7 +286,7 @@ _gcry_aes_amd64_encrypt_block:
 	lastencround(11);
 
 	jmp .Lenc_done;
-.size _gcry_aes_amd64_encrypt_block,.-_gcry_aes_amd64_encrypt_block;
+ELF(.size _gcry_aes_amd64_encrypt_block,.-_gcry_aes_amd64_encrypt_block;)
 
 #define do_decround(next_r) \
 	do16bit_shr(16, mov, RA, Dsize, D0, RNA, D0, RNB, RT0, RT1); \
@@ -365,7 +372,7 @@ _gcry_aes_amd64_encrypt_block:
 
 .align 8
 .globl _gcry_aes_amd64_decrypt_block
-.type   _gcry_aes_amd64_decrypt_block,@function;
+ELF(.type   _gcry_aes_amd64_decrypt_block,@function;)
 
 _gcry_aes_amd64_decrypt_block:
 	/* input:
@@ -440,7 +447,7 @@ _gcry_aes_amd64_decrypt_block:
 	decround(9);
 
 	jmp .Ldec_tail;
-.size _gcry_aes_amd64_decrypt_block,.-_gcry_aes_amd64_decrypt_block;
+ELF(.size _gcry_aes_amd64_decrypt_block,.-_gcry_aes_amd64_decrypt_block;)
 
 #endif /*USE_AES*/
 #endif /*__x86_64*/
diff --git a/cipher/rijndael-internal.h b/cipher/rijndael-internal.h
index 33ca53f6..6641728c 100644
--- a/cipher/rijndael-internal.h
+++ b/cipher/rijndael-internal.h
@@ -39,7 +39,8 @@
 
 /* USE_AMD64_ASM indicates whether to use AMD64 assembly code. */
 #undef USE_AMD64_ASM
-#if defined(__x86_64__) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS)
+#if defined(__x86_64__) && (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+    defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
 # define USE_AMD64_ASM 1
 #endif
 
diff --git a/cipher/rijndael.c b/cipher/rijndael.c
index ade41c9d..7ebf3293 100644
--- a/cipher/rijndael.c
+++ b/cipher/rijndael.c
@@ -665,8 +665,25 @@ do_encrypt (const RIJNDAEL_context *ctx,
             unsigned char *bx, const unsigned char *ax)
 {
 #ifdef USE_AMD64_ASM
+# ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
   return _gcry_aes_amd64_encrypt_block(ctx->keyschenc, bx, ax, ctx->rounds,
 				       encT);
+# else
+  /* Call SystemV ABI function without storing non-volatile XMM registers,
+   * as target function does not use vector instruction sets. */
+  uintptr_t ret;
+  asm ("movq %[encT], %%r8\n\t"
+       "callq *%[ret]\n\t"
+       : [ret] "=a" (ret)
+       : "0" (_gcry_aes_amd64_encrypt_block),
+         "D" (ctx->keyschenc),
+         "S" (bx),
+         "d" (ax),
+         "c" (ctx->rounds),
+         [encT] "r" (encT)
+       : "cc", "memory", "r8", "r9", "r10", "r11");
+  return ret;
+# endif /* HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS */
 #elif defined(USE_ARM_ASM)
   return _gcry_aes_arm_encrypt_block(ctx->keyschenc, bx, ax, ctx->rounds, encT);
 #else
@@ -1008,8 +1025,25 @@ do_decrypt (const RIJNDAEL_context *ctx, unsigned char *bx,
             const unsigned char *ax)
 {
 #ifdef USE_AMD64_ASM
+# ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
   return _gcry_aes_amd64_decrypt_block(ctx->keyschdec, bx, ax, ctx->rounds,
 				       &dec_tables);
+# else
+  /* Call SystemV ABI function without storing non-volatile XMM registers,
+   * as target function does not use vector instruction sets. */
+  uintptr_t ret;
+  asm ("movq %[dectabs], %%r8\n\t"
+       "callq *%[ret]\n\t"
+       : [ret] "=a" (ret)
+       : "0" (_gcry_aes_amd64_decrypt_block),
+         "D" (ctx->keyschdec),
+         "S" (bx),
+         "d" (ax),
+         "c" (ctx->rounds),
+         [dectabs] "r" (&dec_tables)
+       : "cc", "memory", "r8", "r9", "r10", "r11");
+  return ret;
+# endif /* HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS */
 #elif defined(USE_ARM_ASM)
   return _gcry_aes_arm_decrypt_block(ctx->keyschdec, bx, ax, ctx->rounds,
 				     &dec_tables);