base/asm-common.h (x86), and knock-on: Add macros for full-size regs.

[catacomb] / symm / rijndael-arm-crypto.S

diff --git a/symm/rijndael-arm-crypto.S b/symm/rijndael-arm-crypto.S
index 49aa016..1df81d9 100644 (file)
--- a/symm/rijndael-arm-crypto.S
+++ b/symm/rijndael-arm-crypto.S
@@ -25,20 +25,22 @@
 /// MA 02111-1307, USA.
 
 ///--------------------------------------------------------------------------
-/// External definitions.
+/// Preliminaries.
 
 #include "config.h"
 #include "asm-common.h"
 
-       .globl  F(abort)
-       .globl  F(rijndael_rcon)
+       .arch   armv8-a
+       .fpu    crypto-neon-fp-armv8
+
+       .extern F(abort)
+       .extern F(rijndael_rcon)
+
+       .text
 
 ///--------------------------------------------------------------------------
 /// Main code.
 
-       .arch   armv8-a
-       .fpu    crypto-neon-fp-armv8
-
 /// The ARM crypto extension implements a little-endian version of AES
 /// (though the manual doesn't actually spell this out and you have to
 /// experiment), but Catacomb's internal interface presents as big-endian so
@@ -52,7 +54,7 @@
        // Useful constants.
        .equ    maxrounds, 16           // maximum number of rounds
        .equ    maxblksz, 32            // maximum block size, in bytes
-       .equ    kbufsz, maxblksz*(maxrounds + 1) // size of a key-schedule buffer
+       .equ    kbufsz, maxblksz*(maxrounds + 1) // size of key-sched buffer
 
        // Context structure.
        .equ    nr, 0                   // number of rounds
@@ -70,7 +72,7 @@ FUNC(rijndael_setup_arm_crypto)
        //      r2 = pointer to key material
        //      r3 = key size in words
 
-       stmfd   sp!, {r4-r9, r14}
+       pushreg r4-r9, r14
 
        // The initial round key material is taken directly from the input
        // key, so copy it over.  Unfortunately, the key material is not
@@ -104,7 +106,7 @@ FUNC(rijndael_setup_arm_crypto)
        mla     r2, r1, r7, r1          // total key size in words
        leaextq r5, rijndael_rcon       // round constants
        sub     r8, r2, r3              // minus what we've copied already
-       veor    q1, q1                  // all-zero register for the key
+       vmov.i32 q1, #0                 // all-zero register for the key
        add     r8, r9, r8, lsl #2      // limit of the key buffer
        mov     r12, #0                 // position in current cycle
 
@@ -128,7 +130,6 @@ FUNC(rijndael_setup_arm_crypto)
        // First word of the cycle.  Byte substitution, rotation, and round
        // constant.
 1:     ldrb    r14, [r5], #1           // next round constant
-       ldr     r6, [r9, -r3, lsl #2]
        vdup.32 q0, r4
        aese.8  q0, q1                  // effectively, just SubBytes
        vmov.32 r4, d0[0]
@@ -209,11 +210,14 @@ FUNC(rijndael_setup_arm_crypto)
        bl      endswap_block
 
        // All done.
-9:     ldmfd   sp!, {r4-r9, pc}
+9:     popreg  r4-r9, pc
 
-endswap_block:
+ENDFUNC
+
+INTFUNC(endswap_block)
        // End-swap R2 words starting at R1.  R1 is clobbered; R2 is not.
        // It's OK to work in 16-byte chunks.
+
        mov     r4, r2
 0:     vldmia  r1, {d0, d1}
        vrev32.8 q0, q0