diff options
Diffstat (limited to 'deps/openssl/openssl/crypto/sha/asm/sha512-ppc.pl')
| -rw-r--r--[-rwxr-xr-x] | deps/openssl/openssl/crypto/sha/asm/sha512-ppc.pl | 114 |
1 files changed, 58 insertions, 56 deletions
diff --git a/deps/openssl/openssl/crypto/sha/asm/sha512-ppc.pl b/deps/openssl/openssl/crypto/sha/asm/sha512-ppc.pl index 6b44a68e59..768a6a6fad 100755..100644 --- a/deps/openssl/openssl/crypto/sha/asm/sha512-ppc.pl +++ b/deps/openssl/openssl/crypto/sha/asm/sha512-ppc.pl @@ -40,7 +40,6 @@ $output =shift; if ($flavour =~ /64/) { $SIZE_T=8; - $LRSAVE=2*$SIZE_T; $STU="stdu"; $UCMP="cmpld"; $SHL="sldi"; @@ -48,7 +47,6 @@ if ($flavour =~ /64/) { $PUSH="std"; } elsif ($flavour =~ /32/) { $SIZE_T=4; - $LRSAVE=$SIZE_T; $STU="stwu"; $UCMP="cmplw"; $SHL="slwi"; @@ -89,8 +87,7 @@ if ($output =~ /512/) { $SHR="srwi"; } -$FRAME=32*$SIZE_T+16*$SZ; -$LOCALS=6*$SIZE_T; +$FRAME=32*$SIZE_T; $sp ="r1"; $toc="r2"; @@ -182,12 +179,13 @@ $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) @@ -208,7 +206,6 @@ $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 @@ -220,7 +217,7 @@ $func: $LD $G,`6*$SZ`($ctx) $LD $H,`7*$SZ`($ctx) - bl LPICmeup + b LPICmeup LPICedup: andi. r0,$inp,3 bne Lunaligned @@ -229,14 +226,40 @@ Laligned: $PUSH $num,`$FRAME-$SIZE_T*24`($sp) ; end pointer $PUSH $inp,`$FRAME-$SIZE_T*23`($sp) ; inp pointer bl Lsha2_block_private - b Ldone +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 +___ -; 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... +# 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.=<<___; .align 4 Lunaligned: subfic $t1,$inp,4096 @@ -255,7 +278,7 @@ Lunaligned: Lcross_page: li $t1,`16*$SZ/4` mtctr $t1 - addi r20,$sp,$LOCALS ; aligned spot below the frame + addi r20,$sp,$FRAME ; aligned spot below the frame Lmemcpy: lbz r16,0($inp) lbz r17,1($inp) @@ -270,8 +293,8 @@ Lmemcpy: bdnz Lmemcpy $PUSH $inp,`$FRAME-$SIZE_T*26`($sp) ; save real inp - addi $t1,$sp,`$LOCALS+16*$SZ` ; fictitious end pointer - addi $inp,$sp,$LOCALS ; fictitious inp pointer + addi $t1,$sp,`$FRAME+16*$SZ` ; fictitious end pointer + addi $inp,$sp,$FRAME ; 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 @@ -280,36 +303,10 @@ Lmemcpy: $POP $num,`$FRAME-$SIZE_T*25`($sp) ; restore real num addic. $num,$num,`-16*$SZ` ; num-- bne- Lunaligned + b Ldone +___ -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,$FRAME - blr - .long 0 - .byte 0,12,4,1,0x80,18,3,0 - .long 0 - +$code.=<<___; .align 4 Lsha2_block_private: ___ @@ -375,8 +372,6 @@ $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 @@ -384,15 +379,22 @@ ___ $code.=<<___; .align 6 LPICmeup: - mflr r0 - bcl 20,31,\$+4 - mflr $Tbl ; vvvvvv "distance" between . and 1st data entry - addi $Tbl,$Tbl,`64-8` - mtlr r0 + bl LPIC + addi $Tbl,$Tbl,`64-4` ; "distance" between . and last nop + b LPICedup + nop + nop + nop + nop + nop +LPIC: mflr $Tbl blr - .long 0 - .byte 0,12,0x14,0,0,0,0,0 - .space `64-9*4` + nop + nop + nop + nop + nop + nop ___ $code.=<<___ if ($SZ==8); .long 0x428a2f98,0xd728ae22,0x71374491,0x23ef65cd |