Commit b2875087 authored by Andy Polyakov's avatar Andy Polyakov
Browse files

ghash-sparcv9.pl: fix Makefile rule and add performance data for T1.

parent d9218e11

crypto/modes/Makefile

+1 −1
Original line number Diff line number Diff line
@@ -50,7 +50,7 @@ ghash-x86.s: asm/ghash-x86.pl 
ghash-x86_64.s:	asm/ghash-x86_64.pl

	$(PERL) asm/ghash-x86_64.pl $(PERLASM_SCHEME) > $@

ghash-sparcv9.s:	asm/ghash-sparcv9.pl

	$(PERL) asm/ghash-sparcv9.pl $(CFLAGS) > $@

	$(PERL) asm/ghash-sparcv9.pl $@ $(CFLAGS)

ghash-alpha.s:	asm/ghash-alpha.pl

	$(PERL) $< | $(CC) -E - | tee $@ > /dev/null

ghash-parisc.s:	asm/ghash-parisc.pl

crypto/modes/asm/ghash-sparcv9.pl

+8 −1
Original line number Diff line number Diff line
@@ -20,6 +20,13 @@ 
# 32-bit build	81.0		48.6		11.8	(+586%/+311%)

# 64-bit build	27.5		20.3		11.8	(+133%/+72%)

#

# Here is data collected on UltraSPARC T1 system running Linux:

#

#		gcc 4.4.1			this assembler

#

# 32-bit build	566				50	(+1000%)

# 64-bit build	56				50	(+12%)

#

# I don't quite understand why difference between 32-bit and 64-bit

# compiler-generated code is so big. Compilers *were* instructed to

# generate code for UltraSPARC and should have used 64-bit registers
@@ -28,7 +35,7 @@ 
# module;-) Loops are aggressively modulo-scheduled in respect to

# references to input data and Z.hi updates to achieve 12 cycles

# timing. To anchor to something else, sha1-sparcv9.pl spends 11.6

# cycles to process one byte [on UltraSPARC pre-Tx CPU].

# cycles to process one byte on UltraSPARC pre-Tx CPU and ~24 on T1.



$bits=32;

for (@ARGV)     { $bits=64 if (/\-m64/ || /\-xarch\=v9/); }