PPC assembler pack: adhere closer to ABI specs, add PowerOpen traceback data.

crypto/aes/asm/aes-ppc.pl

@@ -18,7 +18,7 @@
 
 # February 2010
 #
-# Rescheduling instructions to favour Power6 pipeline gives 10%
+# Rescheduling instructions to favour Power6 pipeline gave 10%
 # performance improvement on the platfrom in question (and marginal
 # improvement even on others). It should be noted that Power6 fails
 # to process byte in 18 cycles, only in 23, because it fails to issue
@@ -33,11 +33,13 @@ $flavour = shift;
 
 if ($flavour =~ /64/) {
 	$SIZE_T	=8;
+	$LRSAVE	=2*$SIZE_T;
 	$STU	="stdu";
 	$POP	="ld";
 	$PUSH	="std";
 } elsif ($flavour =~ /32/) {
 	$SIZE_T	=4;
+	$LRSAVE	=$SIZE_T;
 	$STU	="stwu";
 	$POP	="lwz";
 	$PUSH	="stw";
@@ -116,15 +118,19 @@ LAES_Te:
 	addi	$Tbl0,$Tbl0,`128-8`
 	mtlr	r0
 	blr
-	.space	`32-24`
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+	.space	`64-9*4`
 LAES_Td:
 	mflr	r0
 	bcl	20,31,\$+4
 	mflr	$Tbl0	;    vvvvvvvv "distance" between . and 1st data entry
-	addi	$Tbl0,$Tbl0,`128-8-32+2048+256`
+	addi	$Tbl0,$Tbl0,`128-64-8+2048+256`
 	mtlr	r0
 	blr
-	.space	`128-32-24`
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+	.space	`128-64-9*4`
 ___
 &_data_word(
 	0xc66363a5, 0xf87c7c84, 0xee777799, 0xf67b7b8d,
@@ -328,10 +334,9 @@ $code.=<<___;
 .globl	.AES_encrypt
 .align	7
 .AES_encrypt:
-	mflr	r0
 	$STU	$sp,-$FRAME($sp)
+	mflr	r0
 
-	$PUSH	r0,`$FRAME-$SIZE_T*21`($sp)
 	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
 	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
 	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
@@ -352,6 +357,7 @@ $code.=<<___;
 	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
 	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
 	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
 
 	lwz	$s0,0($inp)
 	lwz	$s1,4($inp)
@@ -364,7 +370,7 @@ $code.=<<___;
 	stw	$s2,8($out)
 	stw	$s3,12($out)
 
-	$POP	r0,`$FRAME-$SIZE_T*21`($sp)
+	$POP	r0,`$FRAME+$LRSAVE`($sp)
 	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
 	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
 	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
@@ -388,6 +394,9 @@ $code.=<<___;
 	mtlr	r0
 	addi	$sp,$sp,$FRAME
 	blr
+	.long	0
+	.byte	0,12,4,1,0x80,18,3,0
+	.long	0
 
 .align	5
 Lppc_AES_encrypt:
@@ -530,6 +539,8 @@ Lenc_loop:
 	xor	$s2,$s2,$t2
 	xor	$s3,$s3,$t3
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .align	4
 Lppc_AES_encrypt_compact:
@@ -673,14 +684,15 @@ Lenc_compact_done:
 	xor	$s2,$s2,$t2
 	xor	$s3,$s3,$t3
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .globl	.AES_decrypt
 .align	7
 .AES_decrypt:
-	mflr	r0
 	$STU	$sp,-$FRAME($sp)
+	mflr	r0
 
-	$PUSH	r0,`$FRAME-$SIZE_T*21`($sp)
 	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
 	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
 	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
@@ -701,6 +713,7 @@ Lenc_compact_done:
 	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
 	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
 	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
 
 	lwz	$s0,0($inp)
 	lwz	$s1,4($inp)
@@ -713,7 +726,7 @@ Lenc_compact_done:
 	stw	$s2,8($out)
 	stw	$s3,12($out)
 
-	$POP	r0,`$FRAME-$SIZE_T*21`($sp)
+	$POP	r0,`$FRAME+$LRSAVE`($sp)
 	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
 	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
 	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
@@ -737,6 +750,9 @@ Lenc_compact_done:
 	mtlr	r0
 	addi	$sp,$sp,$FRAME
 	blr
+	.long	0
+	.byte	0,12,4,1,0x80,18,3,0
+	.long	0
 
 .align	5
 Lppc_AES_decrypt:
@@ -879,6 +895,8 @@ Ldec_loop:
 	xor	$s2,$s2,$t2
 	xor	$s3,$s3,$t3
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .align	4
 Lppc_AES_decrypt_compact:
@@ -1179,7 +1197,9 @@ Ldec_compact_done:
 	xor	$s2,$s2,$t2
 	xor	$s3,$s3,$t3
 	blr
-.long	0
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+
 .asciz	"AES for PPC, CRYPTOGAMS by <appro\@openssl.org>"
 .align	7
 ___

crypto/bn/asm/ppc-mont.pl

@@ -31,7 +31,6 @@ if ($flavour =~ /32/) {
 	$BNSZ=	$BITS/8;
 	$SIZE_T=4;
 	$RZONE=	224;
-	$FRAME=	$SIZE_T*16;
 
 	$LD=	"lwz";		# load
 	$LDU=	"lwzu";		# load and update
@@ -51,7 +50,6 @@ if ($flavour =~ /32/) {
 	$BNSZ=	$BITS/8;
 	$SIZE_T=8;
 	$RZONE=	288;
-	$FRAME=	$SIZE_T*16;
 
 	# same as above, but 64-bit mnemonics...
 	$LD=	"ld";		# load
@@ -69,6 +67,9 @@ if ($flavour =~ /32/) {
 	$POP=	$LD;
 } else { die "nonsense $flavour"; }
 
+$FRAME=8*$SIZE_T+$RZONE;
+$LOCALS=8*$SIZE_T;
+
 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
@@ -89,18 +90,18 @@ $aj="r10";
 $nj="r11";
 $tj="r12";
 # non-volatile registers
-$i="r14";
-$j="r15";
-$tp="r16";
-$m0="r17";
-$m1="r18";
-$lo0="r19";
-$hi0="r20";
-$lo1="r21";
-$hi1="r22";
-$alo="r23";
-$ahi="r24";
-$nlo="r25";
+$i="r20";
+$j="r21";
+$tp="r22";
+$m0="r23";
+$m1="r24";
+$lo0="r25";
+$hi0="r26";
+$lo1="r27";
+$hi1="r28";
+$alo="r29";
+$ahi="r30";
+$nlo="r31";
 #
 $nhi="r0";
 
@@ -123,32 +124,33 @@ ___
 $code.=<<___;
 	slwi	$num,$num,`log($BNSZ)/log(2)`
 	li	$tj,-4096
-	addi	$ovf,$num,`$FRAME+$RZONE`
+	addi	$ovf,$num,$FRAME
 	subf	$ovf,$ovf,$sp	; $sp-$ovf
 	and	$ovf,$ovf,$tj	; minimize TLB usage
 	subf	$ovf,$sp,$ovf	; $ovf-$sp
+	mr	$tj,$sp
 	srwi	$num,$num,`log($BNSZ)/log(2)`
 	$STUX	$sp,$sp,$ovf
 
-	$PUSH	r14,`4*$SIZE_T`($sp)
-	$PUSH	r15,`5*$SIZE_T`($sp)
-	$PUSH	r16,`6*$SIZE_T`($sp)
-	$PUSH	r17,`7*$SIZE_T`($sp)
-	$PUSH	r18,`8*$SIZE_T`($sp)
-	$PUSH	r19,`9*$SIZE_T`($sp)
-	$PUSH	r20,`10*$SIZE_T`($sp)
-	$PUSH	r21,`11*$SIZE_T`($sp)
-	$PUSH	r22,`12*$SIZE_T`($sp)
-	$PUSH	r23,`13*$SIZE_T`($sp)
-	$PUSH	r24,`14*$SIZE_T`($sp)
-	$PUSH	r25,`15*$SIZE_T`($sp)
+	$PUSH	r20,`-12*$SIZE_T`($tj)
+	$PUSH	r21,`-11*$SIZE_T`($tj)
+	$PUSH	r22,`-10*$SIZE_T`($tj)
+	$PUSH	r23,`-9*$SIZE_T`($tj)
+	$PUSH	r24,`-8*$SIZE_T`($tj)
+	$PUSH	r25,`-7*$SIZE_T`($tj)
+	$PUSH	r26,`-6*$SIZE_T`($tj)
+	$PUSH	r27,`-5*$SIZE_T`($tj)
+	$PUSH	r28,`-4*$SIZE_T`($tj)
+	$PUSH	r29,`-3*$SIZE_T`($tj)
+	$PUSH	r30,`-2*$SIZE_T`($tj)
+	$PUSH	r31,`-1*$SIZE_T`($tj)
 
 	$LD	$n0,0($n0)	; pull n0[0] value
 	addi	$num,$num,-2	; adjust $num for counter register
 
 	$LD	$m0,0($bp)	; m0=bp[0]
 	$LD	$aj,0($ap)	; ap[0]
-	addi	$tp,$sp,$FRAME
+	addi	$tp,$sp,$LOCALS
 	$UMULL	$lo0,$aj,$m0	; ap[0]*bp[0]
 	$UMULH	$hi0,$aj,$m0
 
@@ -210,8 +212,8 @@ L1st:
 Louter:
 	$LDX	$m0,$bp,$i	; m0=bp[i]
 	$LD	$aj,0($ap)	; ap[0]
-	addi	$tp,$sp,$FRAME
-	$LD	$tj,$FRAME($sp)	; tp[0]
+	addi	$tp,$sp,$LOCALS
+	$LD	$tj,$LOCALS($sp); tp[0]
 	$UMULL	$lo0,$aj,$m0	; ap[0]*bp[i]
 	$UMULH	$hi0,$aj,$m0
 	$LD	$aj,$BNSZ($ap)	; ap[1]
@@ -278,7 +280,7 @@ Linner:
 
 	addi	$num,$num,2	; restore $num
 	subfc	$j,$j,$j	; j=0 and "clear" XER[CA]
-	addi	$tp,$sp,$FRAME
+	addi	$tp,$sp,$LOCALS
 	mtctr	$num
 
 .align	4
@@ -304,23 +306,27 @@ Lcopy:				; copy or in-place refresh
 	addi	$j,$j,$BNSZ
 	bdnz-	Lcopy
 
-	$POP	r14,`4*$SIZE_T`($sp)
-	$POP	r15,`5*$SIZE_T`($sp)
-	$POP	r16,`6*$SIZE_T`($sp)
-	$POP	r17,`7*$SIZE_T`($sp)
-	$POP	r18,`8*$SIZE_T`($sp)
-	$POP	r19,`9*$SIZE_T`($sp)
-	$POP	r20,`10*$SIZE_T`($sp)
-	$POP	r21,`11*$SIZE_T`($sp)
-	$POP	r22,`12*$SIZE_T`($sp)
-	$POP	r23,`13*$SIZE_T`($sp)
-	$POP	r24,`14*$SIZE_T`($sp)
-	$POP	r25,`15*$SIZE_T`($sp)
-	$POP	$sp,0($sp)
+	$POP	$tj,0($sp)
 	li	r3,1
+	$POP	r20,`-12*$SIZE_T`($tj)
+	$POP	r21,`-11*$SIZE_T`($tj)
+	$POP	r22,`-10*$SIZE_T`($tj)
+	$POP	r23,`-9*$SIZE_T`($tj)
+	$POP	r24,`-8*$SIZE_T`($tj)
+	$POP	r25,`-7*$SIZE_T`($tj)
+	$POP	r26,`-6*$SIZE_T`($tj)
+	$POP	r27,`-5*$SIZE_T`($tj)
+	$POP	r28,`-4*$SIZE_T`($tj)
+	$POP	r29,`-3*$SIZE_T`($tj)
+	$POP	r30,`-2*$SIZE_T`($tj)
+	$POP	r31,`-1*$SIZE_T`($tj)
+	mr	$sp,$tj
 	blr
 	.long	0
-.asciz  "Montgomery Multiplication for PPC, CRYPTOGAMS by <appro\@fy.chalmers.se>"
+	.byte	0,12,4,0,0x80,12,6,0
+	.long	0
+
+.asciz  "Montgomery Multiplication for PPC, CRYPTOGAMS by <appro\@openssl.org>"
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;

crypto/bn/asm/ppc.pl

@@ -389,7 +389,9 @@ $data=<<EOF;
 	$ST		r9,`6*$BNSZ`(r3)	#r[6]=c1
 	$ST		r10,`7*$BNSZ`(r3)	#r[7]=c2
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -814,8 +816,9 @@ $data=<<EOF;
 
 
 	blr
-
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -966,7 +969,9 @@ $data=<<EOF;
 	$ST	r10,`6*$BNSZ`(r3)	#r[6]=c1
 	$ST	r11,`7*$BNSZ`(r3)	#r[7]=c2
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1502,7 +1507,9 @@ $data=<<EOF;
 	$ST	r12,`14*$BNSZ`(r3)	#r[14]=c3;
 	$ST	r10,`15*$BNSZ`(r3)	#r[15]=c1;
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1550,8 +1557,9 @@ Lppcasm_sub_adios:
 	subfze	r3,r0		# if carry bit is set then r3 = 0 else -1
 	andi.	r3,r3,1         # keep only last bit.
 	blr
-	.long	0x00000000
-
+	.long	0
+	.byte	0,12,0x14,0,0,0,4,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1594,7 +1602,9 @@ Lppcasm_add_mainloop:
 Lppcasm_add_adios:	
 	addze	r3,r0			#return carry bit.
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,4,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1707,7 +1717,9 @@ Lppcasm_div8:
 Lppcasm_div9:
 	or	r3,r8,r0
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1746,8 +1758,9 @@ Lppcasm_sqr_mainloop:
 	bdnz-	Lppcasm_sqr_mainloop
 Lppcasm_sqr_adios:	
 	blr
-	.long	0x00000000
-
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1850,7 +1863,9 @@ Lppcasm_mw_REM:
 Lppcasm_mw_OVER:	
 	addi	r3,r12,0
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,4,0
+	.long	0
 
 #
 #	NOTE:	The following label name should be changed to
@@ -1973,7 +1988,9 @@ Lppcasm_maw_leftover:
 Lppcasm_maw_adios:	
 	addi	r3,r12,0
 	blr
-	.long	0x00000000
+	.long	0
+	.byte	0,12,0x14,0,0,0,4,0
+	.long	0
 	.align	4
 EOF
 $data =~ s/\`([^\`]*)\`/eval $1/gem;

crypto/bn/asm/ppc64-mont.pl

@@ -70,7 +70,6 @@ $flavour = shift;
 if ($flavour =~ /32/) {
 	$SIZE_T=4;
 	$RZONE=	224;
-	$FRAME=	$SIZE_T*12+8*12;
 	$fname=	"bn_mul_mont_fpu64";
 
 	$STUX=	"stwux";	# store indexed and update
@@ -79,7 +78,6 @@ if ($flavour =~ /32/) {
 } elsif ($flavour =~ /64/) {
 	$SIZE_T=8;
 	$RZONE=	288;
-	$FRAME=	$SIZE_T*12+8*12;
 	$fname=	"bn_mul_mont_fpu64";
 
 	# same as above, but 64-bit mnemonics...
@@ -95,7 +93,7 @@ die "can't locate ppc-xlate.pl";
 
 open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
 
-$FRAME=($FRAME+63)&~63;
+$FRAME=64;	# padded frame header
 $TRANSFER=16*8;
 
 $carry="r0";
@@ -112,16 +110,16 @@ $tp="r10";
 $j="r11";
 $i="r12";
 # non-volatile registers
-$nap_d="r14";	# interleaved ap and np in double format
-$a0="r15";	# ap[0]
-$t0="r16";	# temporary registers
-$t1="r17";
-$t2="r18";
-$t3="r19";
-$t4="r20";
-$t5="r21";
-$t6="r22";
-$t7="r23";
+$nap_d="r22";	# interleaved ap and np in double format
+$a0="r23";	# ap[0]
+$t0="r24";	# temporary registers
+$t1="r25";
+$t2="r26";
+$t3="r27";
+$t4="r28";
+$t5="r29";
+$t6="r30";
+$t7="r31";
 
 # PPC offers enough register bank capacity to unroll inner loops twice
 #
@@ -151,28 +149,17 @@ $ba="f0";	$bb="f1";	$bc="f2";	$bd="f3";
 $na="f4";	$nb="f5";	$nc="f6";	$nd="f7";
 $dota="f8";	$dotb="f9";
 $A0="f10";	$A1="f11";	$A2="f12";	$A3="f13";
-$N0="f14";	$N1="f15";	$N2="f16";	$N3="f17";
-$T0a="f18";	$T0b="f19";
-$T1a="f20";	$T1b="f21";
-$T2a="f22";	$T2b="f23";
-$T3a="f24";	$T3b="f25";
+$N0="f20";	$N1="f21";	$N2="f22";	$N3="f23";
+$T0a="f24";	$T0b="f25";
+$T1a="f26";	$T1b="f27";
+$T2a="f28";	$T2b="f29";
+$T3a="f30";	$T3b="f31";
 
 # sp----------->+-------------------------------+
 #		| saved sp			|
 #		+-------------------------------+
-#		|				|
-#		+-------------------------------+
-#		| 10 saved gpr, r14-r23		|
-#		.				.
-#		.				.
-#   +12*size_t	+-------------------------------+
-#		| 12 saved fpr, f14-f25		|
-#		.				.
 #		.				.
-#   +12*8	+-------------------------------+
-#		| padding to 64 byte boundary	|
-#		.				.
-#   +X		+-------------------------------+
+#   +64		+-------------------------------+
 #		| 16 gpr<->fpr transfer zone	|
 #		.				.
 #		.				.
@@ -192,6 +179,16 @@ $T3a="f24";	$T3b="f25";
 #		.				.
 #		.				.
 #		+-------------------------------+
+#		.				.
+#   -12*size_t	+-------------------------------+
+#		| 10 saved gpr, r22-r31		|
+#		.				.
+#		.				.
+#   -12*8	+-------------------------------+
+#		| 12 saved fpr, f20-f31		|
+#		.				.
+#		.				.
+#		+-------------------------------+
 
 $code=<<___;
 .machine "any"
@@ -215,30 +212,31 @@ $code=<<___;
 	subf	$tp,$tp,$sp	; $sp-$tp
 	and	$tp,$tp,$i	; minimize TLB usage
 	subf	$tp,$sp,$tp	; $tp-$sp
+	mr	$i,$sp
 	$STUX	$sp,$sp,$tp	; alloca
 
-	$PUSH	r14,`2*$SIZE_T`($sp)
-	$PUSH	r15,`3*$SIZE_T`($sp)
-	$PUSH	r16,`4*$SIZE_T`($sp)
-	$PUSH	r17,`5*$SIZE_T`($sp)
-	$PUSH	r18,`6*$SIZE_T`($sp)
-	$PUSH	r19,`7*$SIZE_T`($sp)
-	$PUSH	r20,`8*$SIZE_T`($sp)
-	$PUSH	r21,`9*$SIZE_T`($sp)
-	$PUSH	r22,`10*$SIZE_T`($sp)
-	$PUSH	r23,`11*$SIZE_T`($sp)
-	stfd	f14,`12*$SIZE_T+0`($sp)
-	stfd	f15,`12*$SIZE_T+8`($sp)
-	stfd	f16,`12*$SIZE_T+16`($sp)
-	stfd	f17,`12*$SIZE_T+24`($sp)
-	stfd	f18,`12*$SIZE_T+32`($sp)
-	stfd	f19,`12*$SIZE_T+40`($sp)
-	stfd	f20,`12*$SIZE_T+48`($sp)
-	stfd	f21,`12*$SIZE_T+56`($sp)
-	stfd	f22,`12*$SIZE_T+64`($sp)
-	stfd	f23,`12*$SIZE_T+72`($sp)
-	stfd	f24,`12*$SIZE_T+80`($sp)
-	stfd	f25,`12*$SIZE_T+88`($sp)
+	$PUSH	r22,`-12*8-10*$SIZE_T`($i)
+	$PUSH	r23,`-12*8-9*$SIZE_T`($i)
+	$PUSH	r24,`-12*8-8*$SIZE_T`($i)
+	$PUSH	r25,`-12*8-7*$SIZE_T`($i)
+	$PUSH	r26,`-12*8-6*$SIZE_T`($i)
+	$PUSH	r27,`-12*8-5*$SIZE_T`($i)
+	$PUSH	r28,`-12*8-4*$SIZE_T`($i)
+	$PUSH	r29,`-12*8-3*$SIZE_T`($i)
+	$PUSH	r30,`-12*8-2*$SIZE_T`($i)
+	$PUSH	r31,`-12*8-1*$SIZE_T`($i)
+	stfd	f20,`-12*8`($i)
+	stfd	f21,`-11*8`($i)
+	stfd	f22,`-10*8`($i)
+	stfd	f23,`-9*8`($i)
+	stfd	f24,`-8*8`($i)
+	stfd	f25,`-7*8`($i)
+	stfd	f26,`-6*8`($i)
+	stfd	f27,`-5*8`($i)
+	stfd	f28,`-4*8`($i)
+	stfd	f29,`-3*8`($i)
+	stfd	f30,`-2*8`($i)
+	stfd	f31,`-1*8`($i)
 ___
 $code.=<<___ if ($SIZE_T==8);
 	ld	$a0,0($ap)	; pull ap[0] value
@@ -1052,33 +1050,37 @@ Lcopy:				; copy or in-place refresh
 ___
 
 $code.=<<___;
-	$POP	r14,`2*$SIZE_T`($sp)
-	$POP	r15,`3*$SIZE_T`($sp)
-	$POP	r16,`4*$SIZE_T`($sp)
-	$POP	r17,`5*$SIZE_T`($sp)
-	$POP	r18,`6*$SIZE_T`($sp)
-	$POP	r19,`7*$SIZE_T`($sp)
-	$POP	r20,`8*$SIZE_T`($sp)
-	$POP	r21,`9*$SIZE_T`($sp)
-	$POP	r22,`10*$SIZE_T`($sp)
-	$POP	r23,`11*$SIZE_T`($sp)
-	lfd	f14,`12*$SIZE_T+0`($sp)
-	lfd	f15,`12*$SIZE_T+8`($sp)
-	lfd	f16,`12*$SIZE_T+16`($sp)
-	lfd	f17,`12*$SIZE_T+24`($sp)
-	lfd	f18,`12*$SIZE_T+32`($sp)
-	lfd	f19,`12*$SIZE_T+40`($sp)
-	lfd	f20,`12*$SIZE_T+48`($sp)
-	lfd	f21,`12*$SIZE_T+56`($sp)
-	lfd	f22,`12*$SIZE_T+64`($sp)
-	lfd	f23,`12*$SIZE_T+72`($sp)
-	lfd	f24,`12*$SIZE_T+80`($sp)
-	lfd	f25,`12*$SIZE_T+88`($sp)
-	$POP	$sp,0($sp)
+	$POP	$i,0($sp)
 	li	r3,1	; signal "handled"
+	$POP	r22,`-12*8-10*$SIZE_T`($i)
+	$POP	r23,`-12*8-9*$SIZE_T`($i)
+	$POP	r24,`-12*8-8*$SIZE_T`($i)
+	$POP	r25,`-12*8-7*$SIZE_T`($i)
+	$POP	r26,`-12*8-6*$SIZE_T`($i)
+	$POP	r27,`-12*8-5*$SIZE_T`($i)
+	$POP	r28,`-12*8-4*$SIZE_T`($i)
+	$POP	r29,`-12*8-3*$SIZE_T`($i)
+	$POP	r30,`-12*8-2*$SIZE_T`($i)
+	$POP	r31,`-12*8-1*$SIZE_T`($i)
+	lfd	f20,`-12*8`($i)
+	lfd	f21,`-11*8`($i)
+	lfd	f22,`-10*8`($i)
+	lfd	f23,`-9*8`($i)
+	lfd	f24,`-8*8`($i)
+	lfd	f25,`-7*8`($i)
+	lfd	f26,`-6*8`($i)
+	lfd	f27,`-5*8`($i)
+	lfd	f28,`-4*8`($i)
+	lfd	f29,`-3*8`($i)
+	lfd	f30,`-2*8`($i)
+	lfd	f31,`-1*8`($i)
+	mr	$sp,$i
 	blr
 	.long	0
-.asciz  "Montgomery Multiplication for PPC64, CRYPTOGAMS by <appro\@fy.chalmers.se>"
+	.byte	0,12,4,0,0x8c,10,6,0
+	.long	0
+
+.asciz  "Montgomery Multiplication for PPC64, CRYPTOGAMS by <appro\@openssl.org>"
 ___
 
 $code =~ s/\`([^\`]*)\`/eval $1/gem;

crypto/ppccpuid.pl

@@ -29,12 +29,16 @@ $code=<<___;
 	fcfid	f1,f1
 	extrdi	r0,r0,32,0
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .globl	.OPENSSL_altivec_probe
 .align	4
 .OPENSSL_altivec_probe:
 	.long	0x10000484	# vor	v0,v0,v0
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .globl	.OPENSSL_wipe_cpu
 .align	4
@@ -65,6 +69,8 @@ $code=<<___;
 	fmr	f12,f31
 	fmr	f13,f31
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .globl	.OPENSSL_atomic_add
 .align	4
@@ -75,6 +81,9 @@ Ladd:	lwarx	r5,0,r3
 	bne-	Ladd
 	$SIGNX	r3,r0
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+	.long	0
 
 .globl	.OPENSSL_rdtsc
 .align	4
@@ -82,6 +91,8 @@ Ladd:	lwarx	r5,0,r3
 	mftb	r3
 	mftbu	r4
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 
 .globl	.OPENSSL_cleanse
 .align	4
@@ -111,6 +122,9 @@ Laligned:
 	andi.	r4,r4,3
 	bne	Little
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+	.long	0
 ___
 {
 my ($out,$cnt,$max)=("r3","r4","r5");
@@ -145,6 +159,9 @@ Loop:	mftb	$tick
 
 	mr	r3,$cnt
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+	.long	0
 
 .globl	.OPENSSL_instrument_bus2
 .align	4
@@ -193,6 +210,9 @@ Ldone2:
 	srwi	$cnt,$cnt,2
 	sub	r3,r0,$cnt
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+	.long	0
 ___
 }
 

crypto/sha/asm/sha1-ppc.pl

@@ -24,12 +24,14 @@ $flavour = shift;
 
 if ($flavour =~ /64/) {
 	$SIZE_T	=8;
+	$LRSAVE	=2*$SIZE_T;
 	$UCMP	="cmpld";
 	$STU	="stdu";
 	$POP	="ld";
 	$PUSH	="std";
 } elsif ($flavour =~ /32/) {
 	$SIZE_T	=4;
+	$LRSAVE	=$SIZE_T;
 	$UCMP	="cmplw";
 	$STU	="stwu";
 	$POP	="lwz";
@@ -43,7 +45,8 @@ die "can't locate ppc-xlate.pl";
 
 open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
 
-$FRAME=24*$SIZE_T;
+$FRAME=24*$SIZE_T+64;
+$LOCALS=6*$SIZE_T;
 
 $K  ="r0";
 $sp ="r1";
@@ -162,9 +165,8 @@ $code=<<___;
 .globl	.sha1_block_data_order
 .align	4
 .sha1_block_data_order:
+	$STU	$sp,-$FRAME($sp)
 	mflr	r0
-	$STU	$sp,`-($FRAME+64)`($sp)
-	$PUSH	r0,`$FRAME-$SIZE_T*18`($sp)
 	$PUSH	r15,`$FRAME-$SIZE_T*17`($sp)
 	$PUSH	r16,`$FRAME-$SIZE_T*16`($sp)
 	$PUSH	r17,`$FRAME-$SIZE_T*15`($sp)
@@ -182,6 +184,7 @@ $code=<<___;
 	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
 	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
 	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
 	lwz	$A,0($ctx)
 	lwz	$B,4($ctx)
 	lwz	$C,8($ctx)
@@ -192,37 +195,14 @@ $code=<<___;
 Laligned:
 	mtctr	$num
 	bl	Lsha1_block_private
-Ldone:
-	$POP	r0,`$FRAME-$SIZE_T*18`($sp)
-	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
-	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
-	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
-	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
-	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
-	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
-	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
-	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
-	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
-	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
-	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
-	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
-	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
-	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
-	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
-	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
-	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
-	mtlr	r0
-	addi	$sp,$sp,`$FRAME+64`
-	blr
-___
+	b	Ldone
 
-# PowerPC specification allows an implementation to be ill-behaved
-# upon unaligned access which crosses page boundary. "Better safe
-# than sorry" principle makes me treat it specially. But I don't
-# look for particular offending word, but rather for 64-byte input
-# block which crosses the boundary. Once found that block is aligned
-# and hashed separately...
-$code.=<<___;
+; PowerPC specification allows an implementation to be ill-behaved
+; upon unaligned access which crosses page boundary. "Better safe
+; than sorry" principle makes me treat it specially. But I don't
+; look for particular offending word, but rather for 64-byte input
+; block which crosses the boundary. Once found that block is aligned
+; and hashed separately...
 .align	4
 Lunaligned:
 	subfic	$t1,$inp,4096
@@ -237,7 +217,7 @@ Lunaligned:
 Lcross_page:
 	li	$t1,16
 	mtctr	$t1
-	addi	r20,$sp,$FRAME	; spot below the frame
+	addi	r20,$sp,$LOCALS	; spot within the frame
 Lmemcpy:
 	lbz	r16,0($inp)
 	lbz	r17,1($inp)
@@ -251,15 +231,40 @@ Lmemcpy:
 	addi	r20,r20,4
 	bdnz	Lmemcpy
 
-	$PUSH	$inp,`$FRAME-$SIZE_T*19`($sp)
+	$PUSH	$inp,`$FRAME-$SIZE_T*18`($sp)
 	li	$t1,1
-	addi	$inp,$sp,$FRAME
+	addi	$inp,$sp,$LOCALS
 	mtctr	$t1
 	bl	Lsha1_block_private
-	$POP	$inp,`$FRAME-$SIZE_T*19`($sp)
+	$POP	$inp,`$FRAME-$SIZE_T*18`($sp)
 	addic.	$num,$num,-1
 	bne-	Lunaligned
-	b	Ldone
+
+Ldone:
+	$POP	r0,`$FRAME+$LRSAVE`($sp)
+	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
+	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
+	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
+	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
+	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
+	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
+	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
+	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
+	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
+	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
+	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
+	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
+	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
+	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
+	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
+	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
+	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
+	mtlr	r0
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,18,3,0
+	.long	0
 ___
 
 # This is private block function, which uses tailored calling
@@ -309,6 +314,8 @@ $code.=<<___;
 	addi	$inp,$inp,`16*4`
 	bdnz-	Lsha1_block_private
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 ___
 $code.=<<___;
 .asciz	"SHA1 block transform for PPC, CRYPTOGAMS by <appro\@fy.chalmers.se>"

crypto/sha/asm/sha512-ppc.pl

@@ -40,6 +40,7 @@ $output =shift;
 
 if ($flavour =~ /64/) {
 	$SIZE_T=8;
+	$LRSAVE=2*$SIZE_T;
 	$STU="stdu";
 	$UCMP="cmpld";
 	$SHL="sldi";
@@ -47,6 +48,7 @@ if ($flavour =~ /64/) {
 	$PUSH="std";
 } elsif ($flavour =~ /32/) {
 	$SIZE_T=4;
+	$LRSAVE=$SIZE_T;
 	$STU="stwu";
 	$UCMP="cmplw";
 	$SHL="slwi";
@@ -87,7 +89,8 @@ if ($output =~ /512/) {
 	$SHR="srwi";
 }
 
-$FRAME=32*$SIZE_T;
+$FRAME=32*$SIZE_T+16*$SZ;
+$LOCALS=6*$SIZE_T;
 
 $sp ="r1";
 $toc="r2";
@@ -179,13 +182,12 @@ $code=<<___;
 .globl	$func
 .align	6
 $func:
+	$STU	$sp,-$FRAME($sp)
 	mflr	r0
-	$STU	$sp,`-($FRAME+16*$SZ)`($sp)
 	$SHL	$num,$num,`log(16*$SZ)/log(2)`
 
 	$PUSH	$ctx,`$FRAME-$SIZE_T*22`($sp)
 
-	$PUSH	r0,`$FRAME-$SIZE_T*21`($sp)
 	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
 	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
 	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
@@ -206,6 +208,7 @@ $func:
 	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
 	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
 	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
 
 	$LD	$A,`0*$SZ`($ctx)
 	mr	$inp,r4				; incarnate $inp
@@ -217,7 +220,7 @@ $func:
 	$LD	$G,`6*$SZ`($ctx)
 	$LD	$H,`7*$SZ`($ctx)
 
-	b	LPICmeup
+	bl	LPICmeup
 LPICedup:
 	andi.	r0,$inp,3
 	bne	Lunaligned
@@ -226,40 +229,14 @@ Laligned:
 	$PUSH	$num,`$FRAME-$SIZE_T*24`($sp)	; end pointer
 	$PUSH	$inp,`$FRAME-$SIZE_T*23`($sp)	; inp pointer
 	bl	Lsha2_block_private
-Ldone:
-	$POP	r0,`$FRAME-$SIZE_T*21`($sp)
-	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
-	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
-	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
-	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
-	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
-	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
-	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
-	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
-	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
-	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
-	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
-	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
-	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
-	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
-	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
-	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
-	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
-	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
-	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
-	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
-	mtlr	r0
-	addi	$sp,$sp,`$FRAME+16*$SZ`
-	blr
-___
+	b	Ldone
 
-# PowerPC specification allows an implementation to be ill-behaved
-# upon unaligned access which crosses page boundary. "Better safe
-# than sorry" principle makes me treat it specially. But I don't
-# look for particular offending word, but rather for the input
-# block which crosses the boundary. Once found that block is aligned
-# and hashed separately...
-$code.=<<___;
+; PowerPC specification allows an implementation to be ill-behaved
+; upon unaligned access which crosses page boundary. "Better safe
+; than sorry" principle makes me treat it specially. But I don't
+; look for particular offending word, but rather for the input
+; block which crosses the boundary. Once found that block is aligned
+; and hashed separately...
 .align	4
 Lunaligned:
 	subfic	$t1,$inp,4096
@@ -278,7 +255,7 @@ Lunaligned:
 Lcross_page:
 	li	$t1,`16*$SZ/4`
 	mtctr	$t1
-	addi	r20,$sp,$FRAME			; aligned spot below the frame
+	addi	r20,$sp,$LOCALS			; aligned spot below the frame
 Lmemcpy:
 	lbz	r16,0($inp)
 	lbz	r17,1($inp)
@@ -293,8 +270,8 @@ Lmemcpy:
 	bdnz	Lmemcpy
 
 	$PUSH	$inp,`$FRAME-$SIZE_T*26`($sp)	; save real inp
-	addi	$t1,$sp,`$FRAME+16*$SZ`		; fictitious end pointer
-	addi	$inp,$sp,$FRAME			; fictitious inp pointer
+	addi	$t1,$sp,`$LOCALS+16*$SZ`	; fictitious end pointer
+	addi	$inp,$sp,$LOCALS		; fictitious inp pointer
 	$PUSH	$num,`$FRAME-$SIZE_T*25`($sp)	; save real num
 	$PUSH	$t1,`$FRAME-$SIZE_T*24`($sp)	; end pointer
 	$PUSH	$inp,`$FRAME-$SIZE_T*23`($sp)	; inp pointer
@@ -303,10 +280,36 @@ Lmemcpy:
 	$POP	$num,`$FRAME-$SIZE_T*25`($sp)	; restore real num
 	addic.	$num,$num,`-16*$SZ`		; num--
 	bne-	Lunaligned
-	b	Ldone
-___
 
-$code.=<<___;
+Ldone:
+	$POP	r0,`$FRAME+$LRSAVE`($sp)
+	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
+	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
+	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
+	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
+	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
+	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
+	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
+	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
+	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
+	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
+	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
+	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
+	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
+	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
+	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
+	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
+	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
+	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
+	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
+	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
+	mtlr	r0
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,18,3,0
+	.long	0
+
 .align	4
 Lsha2_block_private:
 ___
@@ -372,6 +375,8 @@ $code.=<<___;
 	$ST	$H,`7*$SZ`($ctx)
 	bne	Lsha2_block_private
 	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
 ___
 
 # Ugly hack here, because PPC assembler syntax seem to vary too
@@ -379,22 +384,15 @@ ___
 $code.=<<___;
 .align	6
 LPICmeup:
-	bl	LPIC
-	addi	$Tbl,$Tbl,`64-4`	; "distance" between . and last nop
-	b	LPICedup
-	nop
-	nop
-	nop
-	nop
-	nop
-LPIC:	mflr	$Tbl
+	mflr	r0
+	bcl	20,31,\$+4
+	mflr	$Tbl	; vvvvvv "distance" between . and 1st data entry
+	addi	$Tbl,$Tbl,`64-8`
+	mtlr	r0
 	blr
-	nop
-	nop
-	nop
-	nop
-	nop
-	nop
+	.long	0
+	.byte	0,12,0x14,0,0,0,0,0
+	.space	`64-9*4`
 ___
 $code.=<<___ if ($SZ==8);
 	.long	0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd