HP-UX tune-up: new unified configs, HP C compiler bug workaround.
authorAndy Polyakov <appro@openssl.org>
Sat, 12 Feb 2000 23:33:01 +0000 (23:33 +0000)
committerAndy Polyakov <appro@openssl.org>
Sat, 12 Feb 2000 23:33:01 +0000 (23:33 +0000)
CHANGES
Configure
TABLE
config
crypto/md5/md5_dgst.c
crypto/ripemd/rmd_dgst.c
crypto/sha/sha_locl.h

diff --git a/CHANGES b/CHANGES
index 0bb813b..6457df5 100644 (file)
--- a/CHANGES
+++ b/CHANGES
@@ -4,6 +4,9 @@
 
  Changes between 0.9.4 and 0.9.5  [xx XXX 2000]
 
+  *) HP-UX tune-up: new unified configs, HP C compiler bug workaround.
+     [Andy Polyakov]
+
   *) Add -rand argument to smime and pkcs12 applications and read/write
      of seed file.
      [Steve Henson]
index 6ce16f5..c2c0798 100755 (executable)
--- a/Configure
+++ b/Configure
@@ -169,11 +169,43 @@ my %table=(
 "irix64-mips4-gcc","gcc:-mabi=64 -mips4 -mmips-as -O3 -DTERMIOS -DB_ENDIAN -DBN_DIV3W::(unknown)::RC4_CHAR RC4_CHUNK DES_RISC2 DES_UNROLL SIXTY_FOUR_BIT_LONG:asm/mips3.o::",
 "irix64-mips4-cc", "cc:-64 -mips4 -O2 -use_readonly_const -DTERMIOS -DB_ENDIAN -DBN_DIV3W::(unknown)::RC4_CHAR RC4_CHUNK DES_RISC2 DES_UNROLL SIXTY_FOUR_BIT_LONG:asm/mips3.o::",
 
+#### Unified HP-UX ANSI C configs.
+# Special notes:
+# - Originally we were optimizing at +O4 level. It should be noted
+#   that the only difference between +O3 and +O4 is global inter-
+#   procedural analysis. As it has to be performed during the link
+#   stage the compiler leaves behind certain pseudo-code in lib*.a
+#   which might be release or even patch level specific. Generating
+#   the machine code and analyzing the *whole* program appears to be
+#   *extremely* memory demanding while the performance gain is
+#   actually questionable. The situation is intensified by the default
+#   HP-UX data set size limit (infamous 'maxdsiz' tunable) of 64MB
+#   which is way too low for +O4. In other words, doesn't +O3 make
+#   more sense?
+# - Keep in mind that the compiler by default generates code suitable
+#   for execution on the host you're currently compiling at. If you
+#   intend to use it across various PA-RISC processors consider adding
+#   +Dportable.
+# - +DD64 is chosen in favour of +DA2.0W because it's ment to be
+#   compatible with *future* releases.
+# - If you run ./Configure hpux-parisc-[g]cc manually don't forget to
+#   pass -D_REENTRANT on HP-UX 10 and later.
+# - -DMD32_XARRAY triggers workaround for compiler bug we ran into in
+#   32-bit message digests. (For the moment of this writing) HP C
+#   doesn't seem to "digest" too many local variables (they make "him"
+#   chew forever:-). For more details look-up MD32_XARRAY comment in
+#   crypto/sha/sha_lcl.h.
+#                                      <appro@fy.chalmers.se>
+#
+"hpux-parisc-cc","cc:-Ae +O3 +ESlit -z -DB_ENDIAN -DBN_DIV2W -DMD32_XARRAY::::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
+"hpux-parisc-gcc","gcc:-O3 -DB_ENDIAN -DBN_DIV2W::::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
+"hpux64-parisc-cc","cc:-Ae +DD64 +O3 +ESlit -z -DB_ENDIAN -DMD32_XARRAY::-D_REENTRANT::SIXTY_FOUR_BIT_LONG MD2_CHAR RC4_INDEX RC4_CHAR DES_UNROLL DES_RISC1 DES_INT:::",
+
 # HPUX 9.X config.
 # Don't use the bundled cc.  It is broken.  Use HP ANSI C if possible, or
 # egcs.  gcc 2.8.1 is also broken.
 
-"hpux-cc",     "cc:-DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O4 -z::(unknown)::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
+"hpux-cc",     "cc:-DB_ENDIAN -DBN_DIV2W -DMD32_XARRAY -Ae +ESlit +O3 -z::(unknown)::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
 # If hpux-cc fails (e.g. during "make test"), try the next one; otherwise,
 # please report your OS and compiler version to the openssl-bugs@openssl.org
 # mailing list.
@@ -184,7 +216,7 @@ my %table=(
 "hpux-brokengcc",      "gcc:-DB_ENDIAN -DBN_DIV2W -O3::(unknown)::DES_PTR DES_UNROLL DES_RISC1:::",
 
 # HPUX 10.X config.  Supports threads.
-"hpux10-cc",   "cc:-DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O4 -z::-D_REENTRANT::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
+"hpux10-cc",   "cc:-DB_ENDIAN -DBN_DIV2W -DMD32_XARRAY -Ae +ESlit +O3 -z::-D_REENTRANT::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
 # If hpux10-cc fails, try this one (if still fails, try deleting BN_LLONG):
 "hpux10-brokencc",     "cc:-DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O2 -z::-D_REENTRANT::BN_LLONG DES_PTR DES_UNROLL DES_RISC1:::",
 
@@ -194,8 +226,9 @@ my %table=(
 
 # HPUX 11.X from www.globus.org.
 # Only works on PA-RISC 2.0 cpus, and not optimized.  Why?
-"hpux11-32bit-cc","cc:+DA2.0 -DB_ENDIAN -D_HPUX_SOURCE -Aa -Ae +ESlit::-D_REENTRANT::DES_PTR DES_UNROLL DES_RISC1:::",
-"hpux11-64bit-cc","cc:+DA2.0W -g -D_HPUX_SOURCE -Aa -Ae +ESlit::-D_REENTRANT::SIXTY_FOUR_BIT_LONG MD2_CHAR RC4_INDEX RC4_CHAR DES_UNROLL DES_RISC1 DES_INT :::",
+#"hpux11-32bit-cc","cc:+DA2.0 -DB_ENDIAN -D_HPUX_SOURCE -Aa -Ae +ESlit::-D_REENTRANT::DES_PTR DES_UNROLL DES_RISC1:::",
+#"hpux11-64bit-cc","cc:+DA2.0W -g -D_HPUX_SOURCE -Aa -Ae +ESlit::-D_REENTRANT::SIXTY_FOUR_BIT_LONG MD2_CHAR RC4_INDEX RC4_CHAR DES_UNROLL DES_RISC1 DES_INT :::",
+# Use unified settings above instead.
 
 # Dec Alpha, OSF/1 - the alpha164-cc is the flags for a 21164A with
 # the new compiler
diff --git a/TABLE b/TABLE
index 15e71fb..7398fd9 100644 (file)
--- a/TABLE
+++ b/TABLE
@@ -937,7 +937,7 @@ $rc5_obj      =
 
 *** hpux-cc
 $cc           = cc
-$cflags       = -DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O4 -z
+$cflags       = -DB_ENDIAN -DBN_DIV2W -DMD32_XARRAY -Ae +ESlit +O3 -z
 $unistd       = 
 $thread_cflag = (unknown)
 $lflags       = 
@@ -969,11 +969,11 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
-*** hpux10-brokencc
+*** hpux-parisc-cc
 $cc           = cc
-$cflags       = -DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O2 -z
+$cflags       = -Ae +O3 +ESlit -z -DB_ENDIAN -DBN_DIV2W -DMD32_XARRAY
 $unistd       = 
-$thread_cflag = -D_REENTRANT
+$thread_cflag = 
 $lflags       = 
 $bn_ops       = BN_LLONG DES_PTR DES_UNROLL DES_RISC1
 $bn_obj       = 
@@ -986,13 +986,13 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
-*** hpux10-brokengcc
+*** hpux-parisc-gcc
 $cc           = gcc
-$cflags       = -DB_ENDIAN -DBN_DIV2W -O3
+$cflags       = -O3 -DB_ENDIAN -DBN_DIV2W
 $unistd       = 
-$thread_cflag = -D_REENTRANT
+$thread_cflag = 
 $lflags       = 
-$bn_ops       = DES_PTR DES_UNROLL DES_RISC1
+$bn_ops       = BN_LLONG DES_PTR DES_UNROLL DES_RISC1
 $bn_obj       = 
 $des_obj      = 
 $bf_obj       = 
@@ -1003,9 +1003,9 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
-*** hpux10-cc
+*** hpux10-brokencc
 $cc           = cc
-$cflags       = -DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O4 -z
+$cflags       = -DB_ENDIAN -DBN_DIV2W -Ae +ESlit +O2 -z
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $lflags       = 
@@ -1020,13 +1020,13 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
-*** hpux10-gcc
+*** hpux10-brokengcc
 $cc           = gcc
 $cflags       = -DB_ENDIAN -DBN_DIV2W -O3
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $lflags       = 
-$bn_ops       = BN_LLONG DES_PTR DES_UNROLL DES_RISC1
+$bn_ops       = DES_PTR DES_UNROLL DES_RISC1
 $bn_obj       = 
 $des_obj      = 
 $bf_obj       = 
@@ -1037,13 +1037,30 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
-*** hpux11-32bit-cc
+*** hpux10-cc
 $cc           = cc
-$cflags       = +DA2.0 -DB_ENDIAN -D_HPUX_SOURCE -Aa -Ae +ESlit
+$cflags       = -DB_ENDIAN -DBN_DIV2W -DMD32_XARRAY -Ae +ESlit +O3 -z
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $lflags       = 
-$bn_ops       = DES_PTR DES_UNROLL DES_RISC1
+$bn_ops       = BN_LLONG DES_PTR DES_UNROLL DES_RISC1
+$bn_obj       = 
+$des_obj      = 
+$bf_obj       = 
+$md5_obj      = 
+$sha1_obj     = 
+$cast_obj     = 
+$rc4_obj      = 
+$rmd160_obj   = 
+$rc5_obj      = 
+
+*** hpux10-gcc
+$cc           = gcc
+$cflags       = -DB_ENDIAN -DBN_DIV2W -O3
+$unistd       = 
+$thread_cflag = -D_REENTRANT
+$lflags       = 
+$bn_ops       = BN_LLONG DES_PTR DES_UNROLL DES_RISC1
 $bn_obj       = 
 $des_obj      = 
 $bf_obj       = 
@@ -1054,9 +1071,9 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
-*** hpux11-64bit-cc
+*** hpux64-parisc-cc
 $cc           = cc
-$cflags       = +DA2.0W -g -D_HPUX_SOURCE -Aa -Ae +ESlit
+$cflags       = -Ae +DD64 +O3 +ESlit -z -DB_ENDIAN -DMD32_XARRAY
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $lflags       = 
@@ -1428,6 +1445,23 @@ $rc4_obj      =
 $rmd160_obj   = 
 $rc5_obj      = 
 
+*** rhapsody-ppc-cc
+$cc           = cc
+$cflags       = -O3 -DB_ENDIAN
+$unistd       = 
+$thread_cflag = (unknown)
+$lflags       = 
+$bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
+$bn_obj       = 
+$des_obj      = 
+$bf_obj       = 
+$md5_obj      = 
+$sha1_obj     = 
+$cast_obj     = 
+$rc4_obj      = 
+$rmd160_obj   = 
+$rc5_obj      = 
+
 *** sco5-cc
 $cc           = cc
 $cflags       = 
diff --git a/config b/config
index 8752e23..e9ab8be 100755 (executable)
--- a/config
+++ b/config
@@ -449,8 +449,10 @@ case "$GUESSOS" in
   BS2000-siemens-sysv4) OUT="BS2000-OSD" ;;
   RM*-siemens-sysv4) OUT="ReliantUNIX" ;;
   *-siemens-sysv4) OUT="SINIX" ;;
+  *-hpux1*)    OUT="hpux-parisc-$CC"
+               options="$options -D_REENTRANT" ;;
+  *-hpux)      OUT="hpux-parisc-$CC" ;;
   # these are all covered by the catchall below
-  # *-hpux*) OUT="hpux-$CC" ;;
   # *-aix) OUT="aix-$CC" ;;
   # *-dgux) OUT="dgux" ;;
   *) OUT=`echo $GUESSOS | awk -F- '{print $3}'`;;
index 9ff9f93..23d196b 100644 (file)
@@ -186,6 +186,9 @@ void md5_block_host_order (MD5_CTX *c, const void *data, int num)
 #endif
 
 #ifndef md5_block_data_order
+#ifdef X
+#undef X
+#endif
 void md5_block_data_order (MD5_CTX *c, const void *data_, int num)
        {
        const unsigned char *data=data_;
@@ -204,16 +207,15 @@ void md5_block_data_order (MD5_CTX *c, const void *data_, int num)
         *
         *                              <appro@fy.chalmers.se>
         */
-       MD5_LONG X[MD5_LBLOCK];
-       /*
-        * In case you wonder why don't I use c->data for this.
-        * RISCs usually have a handful of registers and if X is
-        * declared as automatic array good optimizing compiler
-        * shall accomodate at least part of it in register bank
-        * instead of memory.
-        *
-        *                              <appro@fy.chalmers.se>
-        */
+#ifndef MD32_XARRAY
+       /* See comment in crypto/sha/sha_locl.h for details. */
+       unsigned long   XX0, XX1, XX2, XX3, XX4, XX5, XX6, XX7,
+                       XX8, XX9,XX10,XX11,XX12,XX13,XX14,XX15;
+# define X(i)  XX##i
+#else
+       MD5_LONG XX[MD5_LBLOCK];
+# define X(i)  XX[i]
+#endif
 
        A=c->A;
        B=c->B;
@@ -222,75 +224,75 @@ void md5_block_data_order (MD5_CTX *c, const void *data_, int num)
 
        for (;num--;)
                {
-       HOST_c2l(data,l); X[ 0]=l;              HOST_c2l(data,l); X[ 1]=l;
+       HOST_c2l(data,l); X( 0)=l;              HOST_c2l(data,l); X( 1)=l;
        /* Round 0 */
-       R0(A,B,C,D,X[ 0], 7,0xd76aa478L);       HOST_c2l(data,l); X[ 2]=l;
-       R0(D,A,B,C,X[ 1],12,0xe8c7b756L);       HOST_c2l(data,l); X[ 3]=l;
-       R0(C,D,A,B,X[ 2],17,0x242070dbL);       HOST_c2l(data,l); X[ 4]=l;
-       R0(B,C,D,A,X[ 3],22,0xc1bdceeeL);       HOST_c2l(data,l); X[ 5]=l;
-       R0(A,B,C,D,X[ 4], 7,0xf57c0fafL);       HOST_c2l(data,l); X[ 6]=l;
-       R0(D,A,B,C,X[ 5],12,0x4787c62aL);       HOST_c2l(data,l); X[ 7]=l;
-       R0(C,D,A,B,X[ 6],17,0xa8304613L);       HOST_c2l(data,l); X[ 8]=l;
-       R0(B,C,D,A,X[ 7],22,0xfd469501L);       HOST_c2l(data,l); X[ 9]=l;
-       R0(A,B,C,D,X[ 8], 7,0x698098d8L);       HOST_c2l(data,l); X[10]=l;
-       R0(D,A,B,C,X[ 9],12,0x8b44f7afL);       HOST_c2l(data,l); X[11]=l;
-       R0(C,D,A,B,X[10],17,0xffff5bb1L);       HOST_c2l(data,l); X[12]=l;
-       R0(B,C,D,A,X[11],22,0x895cd7beL);       HOST_c2l(data,l); X[13]=l;
-       R0(A,B,C,D,X[12], 7,0x6b901122L);       HOST_c2l(data,l); X[14]=l;
-       R0(D,A,B,C,X[13],12,0xfd987193L);       HOST_c2l(data,l); X[15]=l;
-       R0(C,D,A,B,X[14],17,0xa679438eL);
-       R0(B,C,D,A,X[15],22,0x49b40821L);
+       R0(A,B,C,D,X( 0), 7,0xd76aa478L);       HOST_c2l(data,l); X( 2)=l;
+       R0(D,A,B,C,X( 1),12,0xe8c7b756L);       HOST_c2l(data,l); X( 3)=l;
+       R0(C,D,A,B,X( 2),17,0x242070dbL);       HOST_c2l(data,l); X( 4)=l;
+       R0(B,C,D,A,X( 3),22,0xc1bdceeeL);       HOST_c2l(data,l); X( 5)=l;
+       R0(A,B,C,D,X( 4), 7,0xf57c0fafL);       HOST_c2l(data,l); X( 6)=l;
+       R0(D,A,B,C,X( 5),12,0x4787c62aL);       HOST_c2l(data,l); X( 7)=l;
+       R0(C,D,A,B,X( 6),17,0xa8304613L);       HOST_c2l(data,l); X( 8)=l;
+       R0(B,C,D,A,X( 7),22,0xfd469501L);       HOST_c2l(data,l); X( 9)=l;
+       R0(A,B,C,D,X( 8), 7,0x698098d8L);       HOST_c2l(data,l); X(10)=l;
+       R0(D,A,B,C,X( 9),12,0x8b44f7afL);       HOST_c2l(data,l); X(11)=l;
+       R0(C,D,A,B,X(10),17,0xffff5bb1L);       HOST_c2l(data,l); X(12)=l;
+       R0(B,C,D,A,X(11),22,0x895cd7beL);       HOST_c2l(data,l); X(13)=l;
+       R0(A,B,C,D,X(12), 7,0x6b901122L);       HOST_c2l(data,l); X(14)=l;
+       R0(D,A,B,C,X(13),12,0xfd987193L);       HOST_c2l(data,l); X(15)=l;
+       R0(C,D,A,B,X(14),17,0xa679438eL);
+       R0(B,C,D,A,X(15),22,0x49b40821L);
        /* Round 1 */
-       R1(A,B,C,D,X[ 1], 5,0xf61e2562L);
-       R1(D,A,B,C,X[ 6], 9,0xc040b340L);
-       R1(C,D,A,B,X[11],14,0x265e5a51L);
-       R1(B,C,D,A,X[ 0],20,0xe9b6c7aaL);
-       R1(A,B,C,D,X[ 5], 5,0xd62f105dL);
-       R1(D,A,B,C,X[10], 9,0x02441453L);
-       R1(C,D,A,B,X[15],14,0xd8a1e681L);
-       R1(B,C,D,A,X[ 4],20,0xe7d3fbc8L);
-       R1(A,B,C,D,X[ 9], 5,0x21e1cde6L);
-       R1(D,A,B,C,X[14], 9,0xc33707d6L);
-       R1(C,D,A,B,X[ 3],14,0xf4d50d87L);
-       R1(B,C,D,A,X[ 8],20,0x455a14edL);
-       R1(A,B,C,D,X[13], 5,0xa9e3e905L);
-       R1(D,A,B,C,X[ 2], 9,0xfcefa3f8L);
-       R1(C,D,A,B,X[ 7],14,0x676f02d9L);
-       R1(B,C,D,A,X[12],20,0x8d2a4c8aL);
+       R1(A,B,C,D,X( 1), 5,0xf61e2562L);
+       R1(D,A,B,C,X( 6), 9,0xc040b340L);
+       R1(C,D,A,B,X(11),14,0x265e5a51L);
+       R1(B,C,D,A,X( 0),20,0xe9b6c7aaL);
+       R1(A,B,C,D,X( 5), 5,0xd62f105dL);
+       R1(D,A,B,C,X(10), 9,0x02441453L);
+       R1(C,D,A,B,X(15),14,0xd8a1e681L);
+       R1(B,C,D,A,X( 4),20,0xe7d3fbc8L);
+       R1(A,B,C,D,X( 9), 5,0x21e1cde6L);
+       R1(D,A,B,C,X(14), 9,0xc33707d6L);
+       R1(C,D,A,B,X( 3),14,0xf4d50d87L);
+       R1(B,C,D,A,X( 8),20,0x455a14edL);
+       R1(A,B,C,D,X(13), 5,0xa9e3e905L);
+       R1(D,A,B,C,X( 2), 9,0xfcefa3f8L);
+       R1(C,D,A,B,X( 7),14,0x676f02d9L);
+       R1(B,C,D,A,X(12),20,0x8d2a4c8aL);
        /* Round 2 */
-       R2(A,B,C,D,X[ 5], 4,0xfffa3942L);
-       R2(D,A,B,C,X[ 8],11,0x8771f681L);
-       R2(C,D,A,B,X[11],16,0x6d9d6122L);
-       R2(B,C,D,A,X[14],23,0xfde5380cL);
-       R2(A,B,C,D,X[ 1], 4,0xa4beea44L);
-       R2(D,A,B,C,X[ 4],11,0x4bdecfa9L);
-       R2(C,D,A,B,X[ 7],16,0xf6bb4b60L);
-       R2(B,C,D,A,X[10],23,0xbebfbc70L);
-       R2(A,B,C,D,X[13], 4,0x289b7ec6L);
-       R2(D,A,B,C,X[ 0],11,0xeaa127faL);
-       R2(C,D,A,B,X[ 3],16,0xd4ef3085L);
-       R2(B,C,D,A,X[ 6],23,0x04881d05L);
-       R2(A,B,C,D,X[ 9], 4,0xd9d4d039L);
-       R2(D,A,B,C,X[12],11,0xe6db99e5L);
-       R2(C,D,A,B,X[15],16,0x1fa27cf8L);
-       R2(B,C,D,A,X[ 2],23,0xc4ac5665L);
+       R2(A,B,C,D,X( 5), 4,0xfffa3942L);
+       R2(D,A,B,C,X( 8),11,0x8771f681L);
+       R2(C,D,A,B,X(11),16,0x6d9d6122L);
+       R2(B,C,D,A,X(14),23,0xfde5380cL);
+       R2(A,B,C,D,X( 1), 4,0xa4beea44L);
+       R2(D,A,B,C,X( 4),11,0x4bdecfa9L);
+       R2(C,D,A,B,X( 7),16,0xf6bb4b60L);
+       R2(B,C,D,A,X(10),23,0xbebfbc70L);
+       R2(A,B,C,D,X(13), 4,0x289b7ec6L);
+       R2(D,A,B,C,X( 0),11,0xeaa127faL);
+       R2(C,D,A,B,X( 3),16,0xd4ef3085L);
+       R2(B,C,D,A,X( 6),23,0x04881d05L);
+       R2(A,B,C,D,X( 9), 4,0xd9d4d039L);
+       R2(D,A,B,C,X(12),11,0xe6db99e5L);
+       R2(C,D,A,B,X(15),16,0x1fa27cf8L);
+       R2(B,C,D,A,X( 2),23,0xc4ac5665L);
        /* Round 3 */
-       R3(A,B,C,D,X[ 0], 6,0xf4292244L);
-       R3(D,A,B,C,X[ 7],10,0x432aff97L);
-       R3(C,D,A,B,X[14],15,0xab9423a7L);
-       R3(B,C,D,A,X[ 5],21,0xfc93a039L);
-       R3(A,B,C,D,X[12], 6,0x655b59c3L);
-       R3(D,A,B,C,X[ 3],10,0x8f0ccc92L);
-       R3(C,D,A,B,X[10],15,0xffeff47dL);
-       R3(B,C,D,A,X[ 1],21,0x85845dd1L);
-       R3(A,B,C,D,X[ 8], 6,0x6fa87e4fL);
-       R3(D,A,B,C,X[15],10,0xfe2ce6e0L);
-       R3(C,D,A,B,X[ 6],15,0xa3014314L);
-       R3(B,C,D,A,X[13],21,0x4e0811a1L);
-       R3(A,B,C,D,X[ 4], 6,0xf7537e82L);
-       R3(D,A,B,C,X[11],10,0xbd3af235L);
-       R3(C,D,A,B,X[ 2],15,0x2ad7d2bbL);
-       R3(B,C,D,A,X[ 9],21,0xeb86d391L);
+       R3(A,B,C,D,X( 0), 6,0xf4292244L);
+       R3(D,A,B,C,X( 7),10,0x432aff97L);
+       R3(C,D,A,B,X(14),15,0xab9423a7L);
+       R3(B,C,D,A,X( 5),21,0xfc93a039L);
+       R3(A,B,C,D,X(12), 6,0x655b59c3L);
+       R3(D,A,B,C,X( 3),10,0x8f0ccc92L);
+       R3(C,D,A,B,X(10),15,0xffeff47dL);
+       R3(B,C,D,A,X( 1),21,0x85845dd1L);
+       R3(A,B,C,D,X( 8), 6,0x6fa87e4fL);
+       R3(D,A,B,C,X(15),10,0xfe2ce6e0L);
+       R3(C,D,A,B,X( 6),15,0xa3014314L);
+       R3(B,C,D,A,X(13),21,0x4e0811a1L);
+       R3(A,B,C,D,X( 4), 6,0xf7537e82L);
+       R3(D,A,B,C,X(11),10,0xbd3af235L);
+       R3(C,D,A,B,X( 2),15,0x2ad7d2bbL);
+       R3(B,C,D,A,X( 9),21,0xeb86d391L);
 
        A = c->A += A;
        B = c->B += B;
index be3eb22..bdfae27 100644 (file)
@@ -85,14 +85,14 @@ void RIPEMD160_Init(RIPEMD160_CTX *c)
 #ifdef X
 #undef X
 #endif
-#define X(i)   X[(i)]
+#define X(i)   XX[i]
 void ripemd160_block_host_order (RIPEMD160_CTX *ctx, const void *p, int num)
        {
-       const RIPEMD160_LONG *X=p;
+       const RIPEMD160_LONG *XX=p;
        register unsigned long A,B,C,D,E;
        register unsigned long a,b,c,d,e;
 
-       for (;num--;X+=HASH_LBLOCK)
+       for (;num--;XX+=HASH_LBLOCK)
                {
 
        A=ctx->A; B=ctx->B; C=ctx->C; D=ctx->D; E=ctx->E;
@@ -286,21 +286,20 @@ void ripemd160_block_host_order (RIPEMD160_CTX *ctx, const void *p, int num)
 #ifdef X
 #undef X
 #endif
-#define X(i)   X##i
 void ripemd160_block_data_order (RIPEMD160_CTX *ctx, const void *p, int num)
        {
        const unsigned char *data=p;
        register unsigned long A,B,C,D,E;
        unsigned long a,b,c,d,e,l;
-       RIPEMD160_LONG   X0, X1, X2, X3, X4, X5, X6, X7,
-                        X8, X9,X10,X11,X12,X13,X14,X15;
-       /*
-        * Originally the above was declared as RIPEMD160_LONG X[16];
-        * The idea was to make RISC compilers to accomodate at
-        * least part of X in the register bank. Unfortunately not
-        * all compilers get this idea:-(
-        *                                      <appro@fy.chalmers.se>
-        */
+#ifndef MD32_XARRAY
+       /* See comment in crypto/sha/sha_locl.h for details. */
+       unsigned long   XX0, XX1, XX2, XX3, XX4, XX5, XX6, XX7,
+                       XX8, XX9,XX10,XX11,XX12,XX13,XX14,XX15;
+# define X(i)  XX##i
+#else
+       RIPEMD160_LONG  XX[16];
+# define X(i)  XX[i]
+#endif
 
        for (;num--;)
                {
index 0a78b39..3e6f489 100644 (file)
@@ -200,18 +200,39 @@ void HASH_INIT (SHA_CTX *c)
        (f)=xa+(e)+K_60_79+ROTATE((a),5)+F_60_79((b),(c),(d)); \
        (b)=ROTATE((b),30);
 
+#ifdef X
+#undef X
+#endif
+#ifndef MD32_XARRAY
+  /*
+   * Originally X was an array. As it's automatic it's natural
+   * to expect RISC compiler to accomodate at least part of it in
+   * the register bank, isn't it? Unfortunately not all compilers
+   * "find" this expectation reasonable:-( On order to make such
+   * compilers generate better code I replace X[] with a bunch of
+   * X0, X1, etc. See the function body below...
+   *                                   <appro@fy.chalmers.se>
+   */
+# define X(i)  XX##i
+#else
+  /*
+   * However! Some compilers (most notably HP C) get overwhelmed by
+   * that many local variables so that we have to have the way to
+   * fall down to the original behavior.
+   */
+# define X(i)  XX[i]
+#endif
+
 #ifndef DONT_IMPLEMENT_BLOCK_HOST_ORDER
 void HASH_BLOCK_HOST_ORDER (SHA_CTX *c, const void *d, int num)
        {
        const SHA_LONG *W=d;
        register unsigned long A,B,C,D,E,T;
-#ifdef SHA_XARRAY
-       SHA_LONG        X[16];
-# define X(i) X[(i)]
+#ifndef MD32_XARRAY
+       unsigned long   XX0, XX1, XX2, XX3, XX4, XX5, XX6, XX7,
+                       XX8, XX9,XX10,XX11,XX12,XX13,XX14,XX15;
 #else
-       unsigned long    X0, X1, X2, X3, X4, X5, X6, X7,
-                        X8, X9,X10,X11,X12,X13,X14,X15;
-# define X(i) X##i
+       SHA_LONG        XX[16];
 #endif
 
        A=c->h0;
@@ -332,13 +353,11 @@ void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, int num)
        {
        const unsigned char *data=p;
        register unsigned long A,B,C,D,E,T,l;
-#ifdef SHA_XARRAY
-       SHA_LONG        X[16];
-# define X(i) X[(i)]
+#ifndef MD32_XARRAY
+       unsigned long   XX0, XX1, XX2, XX3, XX4, XX5, XX6, XX7,
+                       XX8, XX9,XX10,XX11,XX12,XX13,XX14,XX15;
 #else
-       unsigned long    X0, X1, X2, X3, X4, X5, X6, X7,
-                        X8, X9,X10,X11,X12,X13,X14,X15;
-# define X(i) X##i
+       SHA_LONG        XX[16];
 #endif
 
        A=c->h0;