base/asm-common.h, *.S: Use consistent little-endian notation for SIMD regs.

[catacomb] / symm / rijndael-x86ish-aesni.S

diff --git a/symm/rijndael-x86ish-aesni.S b/symm/rijndael-x86ish-aesni.S
index e556aa5..ad9236a 100644 (file)
--- a/symm/rijndael-x86ish-aesni.S
+++ b/symm/rijndael-x86ish-aesni.S
@@ -25,20 +25,21 @@
 /// MA 02111-1307, USA.
 
 ///--------------------------------------------------------------------------
-/// External definitions.
+/// Preliminaries.
 
 #include "config.h"
 #include "asm-common.h"
 
+       .arch   .aes
+
        .extern F(abort)
        .extern F(rijndael_rcon)
 
+       .text
+
 ///--------------------------------------------------------------------------
 /// Main code.
 
-       .arch   .aes
-       .text
-
 /// The AESNI instructions implement a little-endian version of AES, but
 /// Catacomb's internal interface presents as big-endian so as to work better
 /// with things like GCM.  We therefore maintain the round keys in
@@ -61,17 +62,20 @@
 ///--------------------------------------------------------------------------
 /// Key setup.
 
-FUNC(rijndael_setup_x86ish_aesni)
+FUNC(rijndael_setup_x86ish_aesni_avx)
+       vzeroupper                    // avoid penalty on `legacy' XMM access
+  endprologue
+       // and drop through...
+ENDFUNC
 
-#define SI WHOLE(si)
-#define DI WHOLE(di)
+FUNC(rijndael_setup_x86ish_aesni)
 
 #if CPUFAM_X86
        // Arguments are on the stack.  We'll need to stack the caller's
        // register veriables, but we'll manage.
 
-#  define CTX ebp                      // context pointer
-#  define BLKSZ [esp + 24]             // block size
+#  define CTX BP                       // context pointer
+#  define BLKSZ [SP + 24]              // block size
 
 #  define KSZ ebx                      // key size
 #  define NKW edx                      // total number of key words
@@ -85,15 +89,15 @@ FUNC(rijndael_setup_x86ish_aesni)
 #  define BLKOFF edx                   // block size in bytes
 
        // Stack the caller's registers.
-       pushreg ebp
+       pushreg BP
        pushreg ebx
        pushreg esi
        pushreg edi
 
        // Set up our own variables.
-       mov     CTX, [esp + 20]         // context base pointer
-       mov     SI, [esp + 28]          // key material
-       mov     KSZ, [esp + 32]         // key size, in words
+       mov     CTX, [SP + 20]          // context base pointer
+       mov     SI, [SP + 28]           // key material
+       mov     KSZ, [SP + 32]          // key size, in words
 #endif
 
 #if CPUFAM_AMD64 && ABI_SYSV
@@ -323,7 +327,7 @@ FUNC(rijndael_setup_x86ish_aesni)
        popreg  edi
        popreg  esi
        popreg  ebx
-       popreg  ebp
+       popreg  BP
 #endif
 #if CPUFAM_AMD64 && ABI_WIN
        popreg  rdi
@@ -365,6 +369,12 @@ ENDFUNC
 /// Encrypting and decrypting blocks.
 
 .macro encdec  op, aes, koff
+  FUNC(rijndael_\op\()_x86ish_aesni_avx)
+       vzeroupper                      // avoid XMM penalties
+  endprologue
+       // and drop through...
+  ENDFUNC
+
   FUNC(rijndael_\op\()_x86ish_aesni)
 
 #if CPUFAM_X86
@@ -376,8 +386,8 @@ ENDFUNC
 #  define DST edx
 #  define NR ecx
 
-       mov     K, [esp + 4]
-       mov     SRC, [esp + 8]
+       mov     K, [SP + 4]
+       mov     SRC, [SP + 8]
 #endif
 
 #if CPUFAM_AMD64 && ABI_SYSV
@@ -415,7 +425,7 @@ ENDFUNC
        add     K, 16
        pxor    xmm0, xmm1
 #if CPUFAM_X86
-       mov     DST, [esp + 12]
+       mov     DST, [SP + 12]
 #endif
 
        // Dispatch to the correct code.