Message ID | 1428019673-920-1-git-send-email-cota@braap.org |
---|---|
State | New |
Headers | show |
Am 03.04.2015 um 02:07 schrieb Emilio G. Cota: > This brings down the size of the struct from 56 to 32 bytes on 64-bit, > and to 20 bytes on 32-bit. This leads to memory savings: > > Before: > $ find . -name 'tcg.o' | xargs size > text data bss dec hex filename > 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o > 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o > 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o > 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o > 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o > > After: > $ find . -name 'tcg.o' | xargs size > text data bss dec hex filename > 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o > 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o > 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o > 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o > 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o > > Note that using an entire byte for some enums that need less than > that wastes a few bits (noticeable in 32 bits, where we use > 20 bytes instead of 16) but avoids extraction code, which overall > is a win--I've tested several variations of the patch, and the appended > is the best performer for OpenSSL's bntest by a very small margin: > > Before: > $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null > [...] > Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): > > 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) > 772 context-switches # 0.073 K/sec ( +- 2.03% ) > 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) > 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) > 10.552871687 seconds time elapsed ( +- 0.39% ) > > After: > $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null > Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): > > 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) > 739 context-switches # 0.071 K/sec ( +- 1.71% ) > 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) > 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) > 10.473900411 seconds time elapsed ( +- 0.30% ) > > Suggested-by: Stefan Weil <sw@weilnetz.de> > Suggested-by: Richard Henderson <rth@twiddle.net> > Signed-off-by: Emilio G. Cota <cota@braap.org> > --- > tcg/tcg.h | 26 ++++++++++++++------------ > 1 file changed, 14 insertions(+), 12 deletions(-) > > diff --git a/tcg/tcg.h b/tcg/tcg.h > index add7f75..7f95132 100644 > --- a/tcg/tcg.h > +++ b/tcg/tcg.h > @@ -417,20 +417,19 @@ static inline TCGCond tcg_high_cond(TCGCond c) > } > } > > -#define TEMP_VAL_DEAD 0 > -#define TEMP_VAL_REG 1 > -#define TEMP_VAL_MEM 2 > -#define TEMP_VAL_CONST 3 > +typedef enum TCGTempVal { > + TEMP_VAL_DEAD, > + TEMP_VAL_REG, > + TEMP_VAL_MEM, > + TEMP_VAL_CONST, > +} TCGTempVal; > > -/* XXX: optimize memory layout */ > typedef struct TCGTemp { > - TCGType base_type; > - TCGType type; > - int val_type; > - int reg; > - tcg_target_long val; > - int mem_reg; > - intptr_t mem_offset; > + unsigned int reg:8; > + unsigned int mem_reg:8; > + TCGTempVal val_type:8; > + TCGType base_type:8; > + TCGType type:8; > unsigned int fixed_reg:1; > unsigned int mem_coherent:1; > unsigned int mem_allocated:1; > @@ -438,6 +437,9 @@ typedef struct TCGTemp { > basic blocks. Otherwise, it is not > preserved across basic blocks. */ > unsigned int temp_allocated:1; /* never used for code gen */ > + > + tcg_target_long val; > + intptr_t mem_offset; > const char *name; > } TCGTemp; Thanks for doing those tests. There are some smaller cosmetics which might be changed, too (uint8_t for unsigned int with 8 bit, bool for boolean bit values), but I think your patch is a real gain. Reviewed-by: Stefan Weil <sw@weilnetz.de>
On 04/02/2015 05:07 PM, Emilio G. Cota wrote: > After: > $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null > Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): > > 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) > 739 context-switches # 0.071 K/sec ( +- 1.71% ) > 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) > 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) > 10.473900411 seconds time elapsed ( +- 0.30% ) > > Suggested-by: Stefan Weil <sw@weilnetz.de> > Suggested-by: Richard Henderson <rth@twiddle.net> > Signed-off-by: Emilio G. Cota <cota@braap.org> > --- > tcg/tcg.h | 26 ++++++++++++++------------ > 1 file changed, 14 insertions(+), 12 deletions(-) Reviewed-by: Richard Henderson <rth@twiddle.net> I'll put this in a queue for 2.4. r~
Emilio G. Cota <cota@braap.org> writes: > This brings down the size of the struct from 56 to 32 bytes on 64-bit, > and to 20 bytes on 32-bit. This leads to memory savings: > > Before: > $ find . -name 'tcg.o' | xargs size > text data bss dec hex filename > 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o > 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o > 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o > 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o > 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o > > After: > $ find . -name 'tcg.o' | xargs size > text data bss dec hex filename > 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o > 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o > 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o > 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o > 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o > > Note that using an entire byte for some enums that need less than > that wastes a few bits (noticeable in 32 bits, where we use > 20 bytes instead of 16) but avoids extraction code, which overall > is a win--I've tested several variations of the patch, and the appended > is the best performer for OpenSSL's bntest by a very small margin: > > Before: > $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null > [...] > Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): > > 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) > 772 context-switches # 0.073 K/sec ( +- 2.03% ) > 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) > 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) > 10.552871687 seconds time elapsed ( +- 0.39% ) > > After: > $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null > Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): > > 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) > 739 context-switches # 0.071 K/sec ( +- 1.71% ) > 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) > 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) > 10.473900411 seconds time elapsed > ( +- 0.30% ) I'll take that as a win condition ;-) Reviewed-by: Alex Bennée <alex.bennee@linaro.org> > > Suggested-by: Stefan Weil <sw@weilnetz.de> > Suggested-by: Richard Henderson <rth@twiddle.net> > Signed-off-by: Emilio G. Cota <cota@braap.org> > --- > tcg/tcg.h | 26 ++++++++++++++------------ > 1 file changed, 14 insertions(+), 12 deletions(-) > > diff --git a/tcg/tcg.h b/tcg/tcg.h > index add7f75..7f95132 100644 > --- a/tcg/tcg.h > +++ b/tcg/tcg.h > @@ -417,20 +417,19 @@ static inline TCGCond tcg_high_cond(TCGCond c) > } > } > > -#define TEMP_VAL_DEAD 0 > -#define TEMP_VAL_REG 1 > -#define TEMP_VAL_MEM 2 > -#define TEMP_VAL_CONST 3 > +typedef enum TCGTempVal { > + TEMP_VAL_DEAD, > + TEMP_VAL_REG, > + TEMP_VAL_MEM, > + TEMP_VAL_CONST, > +} TCGTempVal; > > -/* XXX: optimize memory layout */ > typedef struct TCGTemp { > - TCGType base_type; > - TCGType type; > - int val_type; > - int reg; > - tcg_target_long val; > - int mem_reg; > - intptr_t mem_offset; > + unsigned int reg:8; > + unsigned int mem_reg:8; > + TCGTempVal val_type:8; > + TCGType base_type:8; > + TCGType type:8; > unsigned int fixed_reg:1; > unsigned int mem_coherent:1; > unsigned int mem_allocated:1; > @@ -438,6 +437,9 @@ typedef struct TCGTemp { > basic blocks. Otherwise, it is not > preserved across basic blocks. */ > unsigned int temp_allocated:1; /* never used for code gen */ > + > + tcg_target_long val; > + intptr_t mem_offset; > const char *name; > } TCGTemp;
diff --git a/tcg/tcg.h b/tcg/tcg.h index add7f75..7f95132 100644 --- a/tcg/tcg.h +++ b/tcg/tcg.h @@ -417,20 +417,19 @@ static inline TCGCond tcg_high_cond(TCGCond c) } } -#define TEMP_VAL_DEAD 0 -#define TEMP_VAL_REG 1 -#define TEMP_VAL_MEM 2 -#define TEMP_VAL_CONST 3 +typedef enum TCGTempVal { + TEMP_VAL_DEAD, + TEMP_VAL_REG, + TEMP_VAL_MEM, + TEMP_VAL_CONST, +} TCGTempVal; -/* XXX: optimize memory layout */ typedef struct TCGTemp { - TCGType base_type; - TCGType type; - int val_type; - int reg; - tcg_target_long val; - int mem_reg; - intptr_t mem_offset; + unsigned int reg:8; + unsigned int mem_reg:8; + TCGTempVal val_type:8; + TCGType base_type:8; + TCGType type:8; unsigned int fixed_reg:1; unsigned int mem_coherent:1; unsigned int mem_allocated:1; @@ -438,6 +437,9 @@ typedef struct TCGTemp { basic blocks. Otherwise, it is not preserved across basic blocks. */ unsigned int temp_allocated:1; /* never used for code gen */ + + tcg_target_long val; + intptr_t mem_offset; const char *name; } TCGTemp;
This brings down the size of the struct from 56 to 32 bytes on 64-bit, and to 20 bytes on 32-bit. This leads to memory savings: Before: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o After: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o Note that using an entire byte for some enums that need less than that wastes a few bits (noticeable in 32 bits, where we use 20 bytes instead of 16) but avoids extraction code, which overall is a win--I've tested several variations of the patch, and the appended is the best performer for OpenSSL's bntest by a very small margin: Before: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null [...] Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) 772 context-switches # 0.073 K/sec ( +- 2.03% ) 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) 10.552871687 seconds time elapsed ( +- 0.39% ) After: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) 739 context-switches # 0.071 K/sec ( +- 1.71% ) 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) 10.473900411 seconds time elapsed ( +- 0.30% ) Suggested-by: Stefan Weil <sw@weilnetz.de> Suggested-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> --- tcg/tcg.h | 26 ++++++++++++++------------ 1 file changed, 14 insertions(+), 12 deletions(-)