[U-Boot] [PATCH] config.mk: use memoization in cc-option macro to speed up compilation
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
This feature can be enabled by setting CACHE_CC_OPTIONS=y in the environment.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com ---
Some MAKEALL runs for ARM and MIPS (on Core 2 Duo E6600 at 2.4GHz):
time CROSS_COMPILE=/opt/codesourcery/arm-2011.03/bin/arm-none-linux-gnueabi- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL -s omap3 -v ti
real 3m0.380s user 9m30.570s sys 1m34.550s
time CACHE_CC_OPTIONS=y CROSS_COMPILE=/opt/codesourcery/arm-2011.03/bin/arm-none-linux-gnueabi- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL -s omap3 -v ti
real 1m15.661s user 4m21.510s sys 0m33.190s
time CROSS_COMPILE=/opt/codesourcery/mips-2011.03/bin/mips-linux-gnu- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL mips4kc
real 2m49.883s user 6m25.840s sys 0m58.200s
time CACHE_CC_OPTIONS=y CROSS_COMPILE=/opt/codesourcery/mips-2011.03/bin/mips-linux-gnu- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL mips4kc
real 2m18.205s user 4m59.740s sys 0m39.530s
config.mk | 14 ++++++++++++++ 1 files changed, 14 insertions(+), 0 deletions(-)
diff --git a/config.mk b/config.mk index 11b67e5..27c366f 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,22 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # +ifeq ($(CACHE_CC_OPTIONS),y) +sinclude $(OBJTREE)/include/cc-options.mk + +cc-option-cached = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ + > /dev/null 2>&1; then \ + echo 'CC_OPTIONS += $(strip $1)' \ + >> $(OBJTREE)/include/cc-options.mk; \ + echo "$(1)"; else echo "$(2)"; fi ;) + +cc-option = $(if $(filter $1,$(CC_OPTIONS)),\ + $(filter $1,$(CC_OPTIONS)),\ + $(call cc-option-cached,$1,$2)) +else cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;) +endif
# # Include the make variables (CC, etc...)
Hi Daniel,
On Wed, Nov 2, 2011 at 11:46 AM, Daniel Schwierzeck daniel.schwierzeck@googlemail.com wrote:
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
This feature can be enabled by setting CACHE_CC_OPTIONS=y in the environment.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com
Some MAKEALL runs for ARM and MIPS (on Core 2 Duo E6600 at 2.4GHz):
time CROSS_COMPILE=/opt/codesourcery/arm-2011.03/bin/arm-none-linux-gnueabi- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL -s omap3 -v ti
real 3m0.380s user 9m30.570s sys 1m34.550s
time CACHE_CC_OPTIONS=y CROSS_COMPILE=/opt/codesourcery/arm-2011.03/bin/arm-none-linux-gnueabi- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL -s omap3 -v ti
real 1m15.661s user 4m21.510s sys 0m33.190s
time CROSS_COMPILE=/opt/codesourcery/mips-2011.03/bin/mips-linux-gnu- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL mips4kc
real 2m49.883s user 6m25.840s sys 0m58.200s
time CACHE_CC_OPTIONS=y CROSS_COMPILE=/opt/codesourcery/mips-2011.03/bin/mips-linux-gnu- MAKEALL_LOGDIR=../logs/ BUILD_DIR=../test/ ./MAKEALL mips4kc
real 2m18.205s user 4m59.740s sys 0m39.530s
config.mk | 14 ++++++++++++++ 1 files changed, 14 insertions(+), 0 deletions(-)
diff --git a/config.mk b/config.mk index 11b67e5..27c366f 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,22 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # +ifeq ($(CACHE_CC_OPTIONS),y) +sinclude $(OBJTREE)/include/cc-options.mk
It would be better to put this into include/generated - it will also make git ignore it and it will be also be removed on clobber which I think you need.
+cc-option-cached = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \
- > /dev/null 2>&1; then \
- echo 'CC_OPTIONS += $(strip $1)' \
- >> $(OBJTREE)/include/cc-options.mk; \
- echo "$(1)"; else echo "$(2)"; fi ;)
+cc-option = $(if $(filter $1,$(CC_OPTIONS)),\
- $(filter $1,$(CC_OPTIONS)),\
- $(call cc-option-cached,$1,$2))
+else cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;) +endif
Tested-by: Simon Glass sjg@chromium.org
With this patch and Wolfgang's I get 1.122s for an incremental build now. Now I just need faster SPI flash and a faster car to get to work sooner.
Is there any reason not to enable this option by default?
Regards, Simon
# # Include the make variables (CC, etc...) -- 1.7.7.1
U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
Hi Simon,
On 02.11.2011 21:39, Simon Glass wrote: ...
diff --git a/config.mk b/config.mk index 11b67e5..27c366f 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,22 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # +ifeq ($(CACHE_CC_OPTIONS),y) +sinclude $(OBJTREE)/include/cc-options.mk
It would be better to put this into include/generated - it will also make git ignore it and it will be also be removed on clobber which I think you need.
The patch is still experimental. But I will move that file to include/generated in the final version.
+cc-option-cached = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ +> /dev/null 2>&1; then \
echo 'CC_OPTIONS += $(strip $1)' \+>> $(OBJTREE)/include/cc-options.mk; \
echo "$(1)"; else echo "$(2)"; fi ;)+cc-option = $(if $(filter $1,$(CC_OPTIONS)),\
$(filter $1,$(CC_OPTIONS)),\$(call cc-option-cached,$1,$2))+else cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;) +endif
Tested-by: Simon Glasssjg@chromium.org
With this patch and Wolfgang's I get 1.122s for an incremental build now. Now I just need faster SPI flash and a faster car to get to work sooner.
Is there any reason not to enable this option by default?
To see the difference between compilation times with and without this optimization and to do not break existing code. In the final version I would remove the config switch and enable the optimization by default ;)
Best regards, Daniel
Dear Daniel Schwierzeck,
In message 4EB1B7DD.40704@googlemail.com you wrote:
The patch is still experimental. But I will move that file to include/generated in the final version.
Indeed, please do.
To see the difference between compilation times with and without this optimization and to do not break existing code. In the final version I would remove the config switch and enable the optimization by default ;)
Looking forward to seeing this final version soon :-)
Thanks!!
Best regards,
Wolfgang Denk
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com --- Changes for v2: - move cache file to $(obj)/include/generated - reworked completely - cache also non-working gcc options - remove CACHE_CC_OPTIONS config switch and enable this optimization by default
config.mk | 23 +++++++++++++++++++++-- 1 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/config.mk b/config.mk index 918cffe..0da961a 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,27 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # -cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ - > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;) +CC_OPTIONS_CACHE_FILE := $(OBJTREE)/include/generated/cc_options.mk + +$(if $(wildcard $(CC_OPTIONS_CACHE_FILE)),,\ + $(shell mkdir -p $(dir $(CC_OPTIONS_CACHE_FILE)))) + +sinclude $(CC_OPTIONS_CACHE_FILE) + +_ccopt_sys = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ + > /dev/null 2>&1; then \ + echo 'CC_OPTIONS += $(strip $1)' >> $(CC_OPTIONS_CACHE_FILE); \ + echo "$(1)"; else \ + [ "x$(strip $2)" != "x" ] && \ + echo 'CC_OPTIONS_NOP += $(strip $2)' >> $(CC_OPTIONS_CACHE_FILE); \ + echo "$(2)"; fi) + +_ccopt_cached = $(if $(filter $1,$(CC_OPTIONS)),$1,) +_ccopt_nop_cached = $(if $(filter $1,$(CC_OPTIONS_NOP)),$1,) + +cc-option = $(if $(call _ccopt_cached,$1),$1,\ + $(if $(call _ccopt_nop_cached,$2),$2,\ + $(call _ccopt_sys,$1,$2)))
# # Include the make variables (CC, etc...)
On Fri, Nov 4, 2011 at 5:53 AM, Daniel Schwierzeck daniel.schwierzeck@googlemail.com wrote:
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com
Tested-by: Simon Glass sjg@chromium.org
I see a big speed-up with this:
full build 7.05s -> 4.1s incremental 2.25s -> 1.05s
Changes for v2: - move cache file to $(obj)/include/generated - reworked completely - cache also non-working gcc options - remove CACHE_CC_OPTIONS config switch and enable this optimization by default
config.mk | 23 +++++++++++++++++++++-- 1 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/config.mk b/config.mk index 918cffe..0da961a 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,27 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # -cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \
- > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;)
+CC_OPTIONS_CACHE_FILE := $(OBJTREE)/include/generated/cc_options.mk
+$(if $(wildcard $(CC_OPTIONS_CACHE_FILE)),,\
- $(shell mkdir -p $(dir $(CC_OPTIONS_CACHE_FILE))))
+sinclude $(CC_OPTIONS_CACHE_FILE)
+_ccopt_sys = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \
- > /dev/null 2>&1; then \
- echo 'CC_OPTIONS += $(strip $1)' >> $(CC_OPTIONS_CACHE_FILE); \
- echo "$(1)"; else \
- [ "x$(strip $2)" != "x" ] && \
Do shell still need that x bit?
- echo 'CC_OPTIONS_NOP += $(strip $2)' >> $(CC_OPTIONS_CACHE_FILE); \
- echo "$(2)"; fi)
+_ccopt_cached = $(if $(filter $1,$(CC_OPTIONS)),$1,)
Do you need the $(if - doesn't filter give you what you want by itself?
+_ccopt_nop_cached = $(if $(filter $1,$(CC_OPTIONS_NOP)),$1,)
+cc-option = $(if $(call _ccopt_cached,$1),$1,\
- $(if $(call _ccopt_nop_cached,$2),$2,\
- $(call _ccopt_sys,$1,$2)))
# # Include the make variables (CC, etc...) -- 1.7.7.1
Thanks for doing this!
Regards, Simon
Hi SImon,
On Fri, Nov 4, 2011 at 5:32 PM, Simon Glass sjg@chromium.org wrote:
On Fri, Nov 4, 2011 at 5:53 AM, Daniel Schwierzeck daniel.schwierzeck@googlemail.com wrote:
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com
Tested-by: Simon Glass sjg@chromium.org
I see a big speed-up with this:
full build 7.05s -> 4.1s incremental 2.25s -> 1.05s
Changes for v2: - move cache file to $(obj)/include/generated - reworked completely - cache also non-working gcc options - remove CACHE_CC_OPTIONS config switch and enable this optimization by default
config.mk | 23 +++++++++++++++++++++-- 1 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/config.mk b/config.mk index 918cffe..0da961a 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,27 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # -cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \
- > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;)
+CC_OPTIONS_CACHE_FILE := $(OBJTREE)/include/generated/cc_options.mk
+$(if $(wildcard $(CC_OPTIONS_CACHE_FILE)),,\
- $(shell mkdir -p $(dir $(CC_OPTIONS_CACHE_FILE))))
+sinclude $(CC_OPTIONS_CACHE_FILE)
+_ccopt_sys = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \
- > /dev/null 2>&1; then \
- echo 'CC_OPTIONS += $(strip $1)' >> $(CC_OPTIONS_CACHE_FILE); \
- echo "$(1)"; else \
- [ "x$(strip $2)" != "x" ] && \
Do shell still need that x bit?
Probably not but it is safer and does not really harm ;)
- echo 'CC_OPTIONS_NOP += $(strip $2)' >> $(CC_OPTIONS_CACHE_FILE); \
- echo "$(2)"; fi)
+_ccopt_cached = $(if $(filter $1,$(CC_OPTIONS)),$1,)
Do you need the $(if - doesn't filter give you what you want by itself?
You are right, the if is redundant. I will optimize this.
+_ccopt_nop_cached = $(if $(filter $1,$(CC_OPTIONS_NOP)),$1,)
+cc-option = $(if $(call _ccopt_cached,$1),$1,\
- $(if $(call _ccopt_nop_cached,$2),$2,\
- $(call _ccopt_sys,$1,$2)))
# # Include the make variables (CC, etc...) -- 1.7.7.1
Thanks for doing this!
Regards, Simon
Thanks, Daniel
Hi Albert, Wolfgang,
following code from arch/arm/config does not work correctly with my optimization
PF_CPPFLAGS_ABI := $(call cc-option,\ -mabi=aapcs-linux -mno-thumb-interwork,\ $(call cc-option,\ -mapcs-32,\ $(call cc-option,\ -mabi=apcs-gnu,\ )\ ) $(call cc-option,-mno-thumb-interwork,)\ ) PLATFORM_CPPFLAGS += $(PF_CPPFLAGS_ARM) $(PF_CPPFLAGS_ABI)
Compiling with seabord_config and latest CodeSourcery toolchain the generated cache file has following content:
CC_OPTIONS += -marm CC_OPTIONS += -mabi=apcs-gnu CC_OPTIONS_NOP += -mabi=apcs-gnu CC_OPTIONS += -mno-thumb-interwork CC_OPTIONS += -mabi=aapcs-linux -mno-thumb-interwork CC_OPTIONS += -fno-stack-protector CC_OPTIONS += -Wno-format-nonliteral CC_OPTIONS += -Wno-format-security CC_OPTIONS += -fno-toplevel-reorder
If I rewrite that code to
PF_CPPFLAGS_ABI := $(call cc-option,-mabi=aapcs-linux,) ifeq ($(PF_CPPFLAGS_ABI),) PF_CPPFLAGS_ABI := $(call cc-option,-mapcs-32,-mabi=apcs-gnu) endif PF_CPPFLAGS_THUMB := $(call cc-option,-mno-thumb-interwork,) PLATFORM_CPPFLAGS += $(PF_CPPFLAGS_ARM) $(PF_CPPFLAGS_ABI) $(PF_CPPFLAGS_THUMB)
the cache file looks better:
CC_OPTIONS += -marm CC_OPTIONS += -mabi=aapcs-linux CC_OPTIONS += -mno-thumb-interwork CC_OPTIONS += -fno-stack-protector CC_OPTIONS += -Wno-format-nonliteral CC_OPTIONS += -Wno-format-security CC_OPTIONS += -fno-toplevel-reorder
Should we change it? is the semantic still the same?
Best regards, Daniel
Dear Daniel Schwierzeck,
In message CACUy__XY1873+dT-s=-GyPeBq5hnYsmYwWt-gxEXiwY97EcGdg@mail.gmail.com you wrote:
Should we change it? is the semantic still the same?
I'm not sure. At first reading it doesn't look really the same to me.
Best regards,
Wolfgang Denk
Hi all,
Le 04/11/2011 18:56, Wolfgang Denk a écrit :
Dear Daniel Schwierzeck,
In messageCACUy__XY1873+dT-s=-GyPeBq5hnYsmYwWt-gxEXiwY97EcGdg@mail.gmail.com you wrote:
Should we change it? is the semantic still the same?
I'm not sure. At first reading it doesn't look really the same to me.
They are not, at least for ELDK4.2.
The only difference is in -mabi options, where the change would reduce "-mabi=apcs-gnu -mabi=aapcs-linux" to "-mabi=aapcs-linux".
apcs-gnu, IIUC, is 'old ABI', while 'aapcs-linux' is 'new ABI', aka eabi. Most of the toolchains I see are eabi (ELDK and CS notably). There may be 'old ABI' toolchains out there, but I don't think they are old ABI either.
Anyway, I've just tried ./MAKEALL edminiv2 with ELD42 and a couple of CS toolchains, and nowhere in the log does -mabi=apcs-gnu show up -- the gcc invocations only have -mabi=aapcs-linux.
I've also tested making ED Mini V2 with and without the patch but without Daniel's proposed change to arch/arm/config.mk, and there is no difference in build commands (except that for some reason the patch inserts multiple spaces between some gcc invocation options.
Daniel, what do you mean with "does not work correctly"?
Best regards,
Wolfgang Denk
Amicalement,
Dear Albert ARIBAUD,
In message 4EB4FF09.5070601@aribaud.net you wrote:
apcs-gnu, IIUC, is 'old ABI', while 'aapcs-linux' is 'new ABI', aka eabi. Most of the toolchains I see are eabi (ELDK and CS notably). There may be 'old ABI' toolchains out there, but I don't think they are old ABI either.
Anyway, I've just tried ./MAKEALL edminiv2 with ELD42 and a couple of CS toolchains, and nowhere in the log does -mabi=apcs-gnu show up -- the gcc invocations only have -mabi=aapcs-linux.
Try ELDK 4.1 ...
Best regards,
Wolfgang Denk
Hi Albert,
On 05.11.2011 10:16, Albert ARIBAUD wrote:
Hi all,
Le 04/11/2011 18:56, Wolfgang Denk a écrit :
Dear Daniel Schwierzeck,
In messageCACUy__XY1873+dT-s=-GyPeBq5hnYsmYwWt-gxEXiwY97EcGdg@mail.gmail.com you wrote:
Should we change it? is the semantic still the same?
I'm not sure. At first reading it doesn't look really the same to me.
They are not, at least for ELDK4.2.
The only difference is in -mabi options, where the change would reduce "-mabi=apcs-gnu -mabi=aapcs-linux" to "-mabi=aapcs-linux".
apcs-gnu, IIUC, is 'old ABI', while 'aapcs-linux' is 'new ABI', aka eabi. Most of the toolchains I see are eabi (ELDK and CS notably). There may be 'old ABI' toolchains out there, but I don't think they are old ABI either.
Anyway, I've just tried ./MAKEALL edminiv2 with ELD42 and a couple of CS toolchains, and nowhere in the log does -mabi=apcs-gnu show up -- the gcc invocations only have -mabi=aapcs-linux.
Looks like I read it wrong. So you always want "-mabi=apcs-gnu -mabi=aapcs-linux -mno-thumb-interwork" in $(PF_CPPFLAGS_ABI) with EABI? Sorry but I am not an ARM expert ;)
I've also tested making ED Mini V2 with and without the patch but without Daniel's proposed change to arch/arm/config.mk, and there is no difference in build commands (except that for some reason the patch inserts multiple spaces between some gcc invocation options.
I guess this comes from making the macros more readable. Maybe I can optimize this.
Daniel, what do you mean with "does not work correctly"?
that the generated cache file looks not right
CC_OPTIONS += -mabi=apcs-gnu CC_OPTIONS_NOP += -mabi=apcs-gnu CC_OPTIONS += -mno-thumb-interwork CC_OPTIONS += -mabi=aapcs-linux -mno-thumb-interwork
But if you want "-mabi=apcs-gnu -mabi=aapcs-linux -mno-thumb-interwork" then it should already work correctly without my change in arch/arm/config.mk.
Best regards, Daniel
Hi Daniel,
Le 05/11/2011 14:43, Daniel Schwierzeck a écrit :
Hi Albert,
On 05.11.2011 10:16, Albert ARIBAUD wrote:
Hi all,
Le 04/11/2011 18:56, Wolfgang Denk a écrit :
Dear Daniel Schwierzeck,
In messageCACUy__XY1873+dT-s=-GyPeBq5hnYsmYwWt-gxEXiwY97EcGdg@mail.gmail.com
you wrote:
Should we change it? is the semantic still the same?
I'm not sure. At first reading it doesn't look really the same to me.
They are not, at least for ELDK4.2.
The only difference is in -mabi options, where the change would reduce "-mabi=apcs-gnu -mabi=aapcs-linux" to "-mabi=aapcs-linux".
apcs-gnu, IIUC, is 'old ABI', while 'aapcs-linux' is 'new ABI', aka eabi. Most of the toolchains I see are eabi (ELDK and CS notably). There may be 'old ABI' toolchains out there, but I don't think they are old ABI either.
Anyway, I've just tried ./MAKEALL edminiv2 with ELD42 and a couple of CS toolchains, and nowhere in the log does -mabi=apcs-gnu show up -- the gcc invocations only have -mabi=aapcs-linux.
Looks like I read it wrong. So you always want "-mabi=apcs-gnu -mabi=aapcs-linux -mno-thumb-interwork" in $(PF_CPPFLAGS_ABI) with EABI? Sorry but I am not an ARM expert ;)
No, I don't want that. :)
The problem I see is having two conflicting -mabi options, -mabi=apcs-gnu and -mabi=aapcs-linux, in the same command line. There should be only one -- and it should be the same across the whole U-Boot building process.
The duplicate -mno-thumb-interwork does not worry me fronm a functional standpoint; it's just a waste of space, that's all.
I've also tested making ED Mini V2 with and without the patch but without Daniel's proposed change to arch/arm/config.mk, and there is no difference in build commands (except that for some reason the patch inserts multiple spaces between some gcc invocation options.
I guess this comes from making the macros more readable. Maybe I can optimize this.
Daniel, what do you mean with "does not work correctly"?
that the generated cache file looks not right
CC_OPTIONS += -mabi=apcs-gnu CC_OPTIONS_NOP += -mabi=apcs-gnu CC_OPTIONS += -mno-thumb-interwork CC_OPTIONS += -mabi=aapcs-linux -mno-thumb-interwork
I'm not a makefile expert, so "looks not right" is a bit meaningless to me. If "not right" means "there are several conflicting -mabi options and there are repeated -mno-thumb-interwork options" then I agree with your "does not look right" statement.
But if you want "-mabi=apcs-gnu -mabi=aapcs-linux -mno-thumb-interwork" then it should already work correctly without my change in arch/arm/config.mk.
As I said, no, I don't want that. I want the right -mabi option only, and I'd like a single -mno-thumb-interwork option.
But what I don't understand is the discrepancy which I see between the CC_OPTIONS resulting from include/generated/cc_options.mk and the actual command line options used to generate e.g. edminiv2.
Best regards, Daniel
Amicalement,
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
The old behaviour can be restored by defining the config option CONFIG_CC_OPT_CACHE_DISABLE=y.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com --- Changes for v3: - reworked handling of non-working gcc options - add config option CONFIG_CC_OPT_CACHE_DISABLE
Changes for v2: - move cache file to $(obj)/include/generated - reworked completely - cache also non-working gcc options - remove CACHE_CC_OPTIONS config switch and enable this optimization by default
config.mk | 20 ++++++++++++++++++-- 1 files changed, 18 insertions(+), 2 deletions(-)
diff --git a/config.mk b/config.mk index 918cffe..ddaa477 100644 --- a/config.mk +++ b/config.mk @@ -107,8 +107,24 @@ HOSTCFLAGS += -pedantic # Option checker (courtesy linux kernel) to ensure # only supported compiler options are used # -cc-option = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ - > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi ;) +CC_OPTIONS_CACHE_FILE := $(OBJTREE)/include/generated/cc_options.mk + +$(if $(wildcard $(CC_OPTIONS_CACHE_FILE)),,\ + $(shell mkdir -p $(dir $(CC_OPTIONS_CACHE_FILE)))) + +-include $(CC_OPTIONS_CACHE_FILE) + +cc-option-sys = $(shell if $(CC) $(CFLAGS) $(1) -S -o /dev/null -xc /dev/null \ + > /dev/null 2>&1; then \ + echo 'CC_OPTIONS += $(strip $1)' >> $(CC_OPTIONS_CACHE_FILE); \ + echo "$(1)"; fi) + +ifeq ($(CONFIG_CC_OPT_CACHE_DISABLE),y) +cc-option = $(strip $(if $(call cc-option-sys,$1),$1,$2)) +else +cc-option = $(strip $(if $(findstring $1,$(CC_OPTIONS)),$1,\ + $(if $(call cc-option-sys,$1),$1,$2))) +endif
# # Include the make variables (CC, etc...)
Dear Daniel Schwierzeck,
In message 1320679603-24847-1-git-send-email-daniel.schwierzeck@googlemail.com you wrote:
Apply memoization to cc-option macro by caching the results of the gcc calls. This macro is called very often so using cached results leads to faster compilation times.
The old behaviour can be restored by defining the config option CONFIG_CC_OPT_CACHE_DISABLE=y.
Signed-off-by: Daniel Schwierzeck daniel.schwierzeck@googlemail.com
Changes for v3:
- reworked handling of non-working gcc options
- add config option CONFIG_CC_OPT_CACHE_DISABLE
Changes for v2:
- move cache file to $(obj)/include/generated
- reworked completely
- cache also non-working gcc options
- remove CACHE_CC_OPTIONS config switch and enable this optimization by default
config.mk | 20 ++++++++++++++++++-- 1 files changed, 18 insertions(+), 2 deletions(-)
Applied, thanks.
Best regards,
Wolfgang Denk
participants (4)
-
Albert ARIBAUD -
Daniel Schwierzeck -
Simon Glass -
Wolfgang Denk