On Thu, Aug 10, 2006 at 12:03:09AM -0700, Nathaniel Smith wrote:
Umm... anyone want to test out their asm chops?
1.3 GHz Athlon Thunderbird:
Botan mainline:71.01 Mbytes/sec
Botan w/attached: 103.80 Mbytes/sec
Botan w/OpenSSL: 133.14 Mbytes/sec
2 GHz P4-M:
Botan mainline:49.78 Mbytes/sec
Botan w/attached: 63.98 Mbytes/sec
Botan w/OpenSSL: 180.77 Mbytes/sec
Obviously this could be scheduled much better for the P4 (and the
Athlon, for that matter), however I don't know much about Netburst
instruction scheduling. Was only testing on the Athlon until the very
end, so it's possible this is a completely wrong approach for P4
performance, dunno.
-Jack
.file sha1core.S
.text
.p2align 4,,15
.global sha160_core
.type sha160_core, @function
sha160_core:
pushl %ebp
pushl %edi
pushl %esi
pushl %ebx
movl24(%esp), %ebp # byte input[64]
movl28(%esp), %edi # u32bit W[80]
movl$0, %esi # loop counter
.p2align 4,,7
.LOAD_INPUT_LOOP:
movl0(%ebp), %eax
bswapl %eax
movl4(%ebp), %ebx
bswapl %ebx
movl %eax, 0(%edi,%esi,4)
movl8(%ebp), %ecx
bswapl %ecx
movl %ebx, 4(%edi,%esi,4)
movl 12(%ebp), %edx
bswapl %edx
movl %ecx, 8(%edi,%esi,4)
movl %edx, 12(%edi,%esi,4)
addl $4, %esi
addl $16, %ebp
cmpl $16, %esi
jne.LOAD_INPUT_LOOP
leal64(%edi), %ebp
.p2align 4,,7
.EXPANSION_LOOP:
addl $4, %esi
xorl %eax, %eax
movl -4(%ebp), %ebx
movl -8(%ebp), %ecx
movl -12(%ebp), %edx
xorl -20(%ebp), %eax
xorl -24(%ebp), %ebx
xorl -28(%ebp), %ecx
xorl -32(%ebp), %edx
xorl -44(%ebp), %eax
xorl -48(%ebp), %ebx
xorl -52(%ebp), %ecx
xorl -52(%ebp), %eax
xorl -56(%ebp), %edx
xorl -56(%ebp), %ebx
xorl -60(%ebp), %ecx
xorl -64(%ebp), %edx
roll $1, %edx
roll $1, %ecx
movl %edx, (%ebp)
roll $1, %ebx
movl %ecx, 4(%ebp)
xorl %edx, %eax
movl %ebx, 8(%ebp)
roll $1, %eax
movl %eax, 12(%ebp)
addl $16, %ebp
cmpl $80, %esi
jne .EXPANSION_LOOP
movl 20(%esp), %ebp
movl 0(%ebp), %eax
movl 4(%ebp), %ebx
movl 8(%ebp), %ecx
movl 12(%ebp), %edx
movl 16(%ebp), %esi
#define MAGIC1 0x5A827999
#define MAGIC2 0x6ED9EBA1
#define MAGIC3 0x8F1BBCDC
#define MAGIC4 0xCA62C1D6
#define F1(A, B, C, D, E, TEMP, MSG) \
addl 4*MSG(%edi), E ; \
movl C, TEMP ; \
roll $5, A ; \
xorl D, TEMP ; \
addl A, E; \
andl B, TEMP ; \
rorl $2, B ; \
xorl D, TEMP ; \
leal MAGIC1(E,TEMP,1), E ; \
rorl $5, A ;
#define F2_OR_F4(A, B, C, D, E, TEMP, MSG, MAGIC) \
addl 4*MSG(%edi), E ; \
movl B, TEMP ; \
roll $5, A ; \
xorl D, TEMP ; \
addl A, E; \
xorl C, TEMP ; \
rorl $2, B ; \
leal MAGIC(E,TEMP,1), E ; \
rorl $5, A ;
#define F3(A, B, C, D, E, TEMP, MSG) \
addl 4*MSG(%edi), E ; \
movl B, TEMP ; \
roll $5, A ; \
orl C, TEMP ; \
movl B, (%edi) ; \
andl D, TEMP ; \
andl C, (%edi) ; \
orl (%edi), TEMP; \
addl A, E; \
leal MAGIC3(E,TEMP,1), E ; \
rorl $2, B ; \
rorl $5, A ;
#define F2(A, B, C, D, E, TEMP, MSG) \
F2_OR_F4(A, B, C, D, E, TEMP, MSG, MAGIC2)
#define F4(A, B, C, D, E, TEMP, MSG) \
F2_OR_F4(A, B, C, D, E, TEMP, MSG, MAGIC4)
#define F_BLOCK(F, MSG) \
F(%eax, %ebx, %ecx, %edx, %esi, %ebp, (MSG+0)) \
F(%esi, %eax, %ebx, %ecx, %edx, %ebp, (MSG+1)) \
F(%edx, %esi, %eax, %ebx, %ecx, %ebp, (MSG+2)) \
F(%ecx, %edx, %esi, %eax, %ebx, %ebp, (MSG+3)) \
F(%ebx, %ecx, %edx, %esi, %eax, %ebp, (MSG+4))
F_BLOCK(F1, 0)
F_BLOCK(F1, 5)
F_BLOCK(F1, 10)
F_BLOCK(F1, 15)
F_BLOCK(F2, 20)
F_BLOCK(F2, 25)
F_BLOCK(F2, 30)
F_BLOCK(F2, 35)
F_BLOCK(F3, 40)
F_BLOCK(F3, 45)
F_BLOCK(F3, 50)
F_BLOCK(F3, 55)
F_BLOCK(F4, 60)
F_BLOCK(F4, 65)
F_BLOCK(F4, 70)
F_BLOCK(F4, 75)
movl 20(%esp), %ebp
addl %eax, 0(%ebp)
addl %ebx, 4(%ebp)
addl %ecx, 8(%ebp)
addl %edx, 12(%ebp)
addl %esi, 16(%ebp)
popl%ebx
popl%esi
popl%edi
popl%ebp
ret
/*
* SHA-160 Source File
