SPARCv9 assembler pack: refine CPU detection on Linux, fix for "unaligned

[oweals/openssl.git] / crypto / sha / asm / sha512-armv4.pl

diff --git a/crypto/sha/asm/sha512-armv4.pl b/crypto/sha/asm/sha512-armv4.pl
index 7e7a726b17297bd61c7bfed26a262955ccb9e9bc..4fbb94a914faf8129d5b177cc1c6405e71d1ee43 100644 (file)
--- a/crypto/sha/asm/sha512-armv4.pl
+++ b/crypto/sha/asm/sha512-armv4.pl
@@ -12,12 +12,18 @@
 # This code is ~4.5 (four and a half) times faster than code generated
 # by gcc 3.4 and it spends ~72 clock cycles per byte. 
 
-# This module currently has dependency on byte order, namely *dword*
-# order in ctx->h[0-9]. I have to think of a way to reliably detect
-# "endianness" [and flip below two constants] or arrange given dword
-# order in C.
-$lo=0; # this denotes little-endian platform.
-$hi=4;
+# Byte order [in]dependence. =========================================
+#
+# Caller is expected to maintain specific *dword* order in h[0-7],
+# namely with most significant dword at *lower* address, which is
+# reflected in below two parameters. *Byte* order within these dwords
+# in turn is whatever *native* byte order on current platform.
+$hi=0;
+$lo=4;
+# ====================================================================
+
+$output=shift;
+open STDOUT,">$output";
 
 $ctx="r0";
 $inp="r1";
@@ -384,7 +390,10 @@ $code.=<<___;
        bx      lr                      @ interoperable with Thumb ISA:-)
 .size   sha512_block_data_order,.-sha512_block_data_order
 .asciz  "SHA512 block transform for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
+.align 2
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;
+$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;   # make it possible to compile with -march=armv4
 print $code;
+close STDOUT; # enforce flush