* [PATCH v2 0/3] valgrind cleanup + fix for armv7a without vfp/neon
@ 2016-01-20 2:47 Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 1/3] valgrind: let valgrind determine its own optimisation flags Andre McCurdy
` (2 more replies)
0 siblings, 3 replies; 6+ messages in thread
From: Andre McCurdy @ 2016-01-20 2:47 UTC (permalink / raw)
To: openembedded-core
v2: fix issue with intdiv test if TUNE_CCARGS sets -march instead of -mcpu
Andre McCurdy (3):
valgrind: let valgrind determine its own optimisation flags
valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests
valgrind: avoid neon for targets which don't support it
...d-neon-for-targets-which-don-t-support-it.patch | 33 +++++++++++
.../valgrind/remove-arm-variant-specific.patch | 66 ----------------------
...opriate-march-mcpu-mfpu-for-ARM-test-apps.patch | 44 +++++++++++++++
meta/recipes-devtools/valgrind/valgrind_3.11.0.bb | 12 ++--
4 files changed, 85 insertions(+), 70 deletions(-)
create mode 100644 meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
delete mode 100644 meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
create mode 100644 meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
--
1.9.1
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v2 1/3] valgrind: let valgrind determine its own optimisation flags
2016-01-20 2:47 [PATCH v2 0/3] valgrind cleanup + fix for armv7a without vfp/neon Andre McCurdy
@ 2016-01-20 2:47 ` Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 2/3] valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 3/3] valgrind: avoid neon for targets which don't support it Andre McCurdy
2 siblings, 0 replies; 6+ messages in thread
From: Andre McCurdy @ 2016-01-20 2:47 UTC (permalink / raw)
To: openembedded-core
Valgrind likes to control its own optimisation flags. It generally
defaults to -O2 but uses -O0 for some specific test apps etc. Passing
our own flags (via CFLAGS) means we interfere with that.
Giving valgrind control of optimisation is hopefully an even better
solution than the previous one of forcing -O0 for all tests.
http://git.openembedded.org/openembedded-core/commit/?h=master-next&id=98c4a3ffb8dca10739be600e8d6df7fb6aa4958f
Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
---
meta/recipes-devtools/valgrind/valgrind_3.11.0.bb | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
index 70d84fe..8e9b72c 100644
--- a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
+++ b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
@@ -44,6 +44,11 @@ EXTRA_OECONF_append_arm = " --host=armv7${HOST_VENDOR}-${HOST_OS}"
EXTRA_OEMAKE = "-w"
+# valgrind likes to control its own optimisation flags. It generally defaults
+# to -O2 but uses -O0 for some specific test apps etc. Passing our own flags
+# (via CFLAGS) means we interfere with that.
+SELECTED_OPTIMIZATION = ""
+
CFLAGS_append_libc-uclibc = " -D__UCLIBC__ "
do_install_append () {
@@ -63,10 +68,9 @@ RDEPENDS_${PN}-ptest += " sed perl glibc-utils perl-module-file-glob"
INSANE_SKIP_${PN}-ptest += "file-rdeps"
do_compile_ptest() {
- oe_runmake check CFLAGS="${CFLAGS} -O0" CXXFLAGS="${CXXFLAGS} -O0"
+ oe_runmake check
}
-
do_install_ptest() {
chmod +x ${B}/tests/vg_regtest
@@ -107,4 +111,3 @@ do_install_ptest() {
# handle multilib
sed -i s:@libdir@:${libdir}:g ${D}${PTEST_PATH}/run-ptest
}
-
--
1.9.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v2 2/3] valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests
2016-01-20 2:47 [PATCH v2 0/3] valgrind cleanup + fix for armv7a without vfp/neon Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 1/3] valgrind: let valgrind determine its own optimisation flags Andre McCurdy
@ 2016-01-20 2:47 ` Andre McCurdy
2016-01-20 3:00 ` Rongqing Li
2016-01-20 2:47 ` [PATCH v2 3/3] valgrind: avoid neon for targets which don't support it Andre McCurdy
2 siblings, 1 reply; 6+ messages in thread
From: Andre McCurdy @ 2016-01-20 2:47 UTC (permalink / raw)
To: openembedded-core
The intdiv test has been partially fixed upstream and the
vcvt_fixed_float_VFP test can be fixed with a similar approach, ie
ensuring that it is always compiled with appropriate
-march/-mcpu/-mfpu flags to support the instructions being tested.
For tests requiring armv7ve instructions, ensure that we set both
-march=armv7ve and -mcpu=cortex-a15 (since some TUNE_CCARGS may set
-march=armv7-a and adding -mcpu=cortex-a15 alone is not enough to
over-ride that).
See similar cases in none/tests/arm/Makefile.am
Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
---
.../valgrind/remove-arm-variant-specific.patch | 66 ----------------------
...opriate-march-mcpu-mfpu-for-ARM-test-apps.patch | 44 +++++++++++++++
meta/recipes-devtools/valgrind/valgrind_3.11.0.bb | 2 +-
3 files changed, 45 insertions(+), 67 deletions(-)
delete mode 100644 meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
create mode 100644 meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
diff --git a/meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch b/meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
deleted file mode 100644
index 2319ab9..0000000
--- a/meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
+++ /dev/null
@@ -1,66 +0,0 @@
-Remove arm tests that don't compile
-
-Upstream-Status: Pending
-
-Corrects the original commit for the patch that removed ARM ptest CFLAGS
-settings. Since the flags could be set by a user, the flags should
-be kept in place during compilation. By keeping the original up-stream
-CFLAGS for the tests, then additional tests successfully compile
-for all tested ARM tunings.
-
-However, there were still two tests listed below that did not compile
-for any beaglebone tuning that is valid for valgrind. With the updated
-patch, the set of excluded ARM ptests and their respective build
-failures are:
- intdiv - fails for all beaglebone tunings with 2 errors:
- {standard input}:(40 or 41): Error: selected processor does not
- support Thumb mode `udiv r3,r9,r10'
- {standard input}:(72 or 73): Error: selected processor does not
- support Thumb mode `sdiv r3,r9,r10'
-
- vcvt_fixed_float_VFP - fails for all beaglebone tunings in one of
- two ways:
- with neon tuning (-mfpu=neon) fails with Internal Compiler Error
- without neon tuning fails with 3 errors:
- {standard input}:33: Error: selected FPU does not support
- instruction -- `vcvt.f32.s32 s15,s15,#1'
- {standard input}:58: Error: selected FPU does not support
- instruction -- `vcvt.f32.s32 s15,s15,#32'
- {standard input}:136: Error: selected FPU does not support
- instruction -- `vcvt.f32.u32 s15,s15,#1'
-
-After applying this commit, the valgrind ARM ptests compile without
-errors for tunings:
- armv7[t][hf][b][-neon] cortexa8[t][hf][-neon]
-where the tuning [option] was successfully compiled, both with
-and without the 'option', and in combination with all other options.
-
-Signed-off-by: Dave Lerner <dave.lerner@windriver.com>
-
-Index: valgrind-3.10.1/none/tests/arm/Makefile.am
-===================================================================
---- valgrind-3.10.1.orig/none/tests/arm/Makefile.am 2015-05-19 15:11:59.224842927 -0500
-+++ valgrind-3.10.1/none/tests/arm/Makefile.am 2015-05-19 15:14:20.808847028 -0500
-@@ -17,9 +17,13 @@
- vfp.stdout.exp vfp.stderr.exp vfp.vgtest \
- vfpv4_fma.stdout.exp vfpv4_fma.stderr.exp vfpv4_fma.vgtest
-
-+# Remove the following tests which cause compiler errors for all tunings
-+# available for beagle bone (see remove-arm-variant-specific.patch):
-+# intdiv
-+# vcvt_fixed_float_VFP
-+
- check_PROGRAMS = \
- allexec \
-- intdiv \
- ldrt \
- ldrt_arm \
- neon128 \
-@@ -27,7 +31,6 @@
- v6intARM \
- v6intThumb \
- v6media \
-- vcvt_fixed_float_VFP \
- vfp \
- vfpv4_fma
-
diff --git a/meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch b/meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
new file mode 100644
index 0000000..adea405
--- /dev/null
+++ b/meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
@@ -0,0 +1,44 @@
+From d134dafc2f11e0d247420a0ba360bcdef77b4093 Mon Sep 17 00:00:00 2001
+From: Andre McCurdy <armccurdy@gmail.com>
+Date: Tue, 19 Jan 2016 16:00:00 -0800
+Subject: [PATCH] use appropriate -march/-mcpu/-mfpu for ARM test apps
+
+Ensure that test apps in none/tests/arm are compiled with appropriate
+-march/-mcpu/-mfpu flags to support the instructions being tested.
+The aim is to build all tests, even ones which may not run correctly
+on all target CPUs.
+
+For tests requiring armv7ve instructions, ensure that we set both
+-march=armv7ve and -mcpu=cortex-a15 (since some TUNE_CCARGS may set
+-march=armv7-a and adding -mcpu=cortex-a15 alone is not enough to
+over-ride that).
+
+See similar cases in none/tests/arm/Makefile.am
+
+Upstream-Status: Pending
+
+Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
+---
+ none/tests/arm/Makefile.am | 6 ++++--
+ 1 file changed, 4 insertions(+), 2 deletions(-)
+
+diff --git a/none/tests/arm/Makefile.am b/none/tests/arm/Makefile.am
+index 4507a20..825290f 100644
+--- a/none/tests/arm/Makefile.am
++++ b/none/tests/arm/Makefile.am
+@@ -62,8 +62,10 @@ neon64_CFLAGS = $(AM_CFLAGS) -g -O0 -mcpu=cortex-a8 \
+ -mfpu=neon \
+ -mthumb
+
+-intdiv_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a15 -mthumb
++intdiv_CFLAGS = $(AM_CFLAGS) -g -march=armv7ve -mcpu=cortex-a15 -mthumb
+ ldrt_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a8 -mthumb
+ ldrt_arm_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a8 -marm
+
+-vfpv4_fma_CFLAGS = $(AM_CFLAGS) -g -O0 -mcpu=cortex-a15 -mfpu=vfpv4 -marm
++vcvt_fixed_float_VFP_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a8 -mfpu=vfpv3
++
++vfpv4_fma_CFLAGS = $(AM_CFLAGS) -g -O0 -march=armv7ve -mcpu=cortex-a15 -mfpu=vfpv4 -marm
+--
+1.9.1
+
diff --git a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
index 8e9b72c..4f7c39a 100644
--- a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
+++ b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
@@ -15,13 +15,13 @@ DEPENDS = "${@bb.utils.contains('DISTRO_FEATURES', 'x11', '${X11DEPENDS}', '', d
SRC_URI = "http://www.valgrind.org/downloads/valgrind-${PV}.tar.bz2 \
file://fixed-perl-path.patch \
file://Added-support-for-PPC-instructions-mfatbu-mfatbl.patch \
- file://remove-arm-variant-specific.patch \
file://run-ptest \
file://11_mips-link-tool.patch \
file://0002-remove-rpath.patch \
file://0004-Fix-out-of-tree-builds.patch \
file://0005-Modify-vg_test-wrapper-to-support-PTEST-formats.patch \
file://0001-Remove-tests-that-fail-to-build-on-some-PPC32-config.patch \
+ file://use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch \
"
SRC_URI[md5sum] = "4ea62074da73ae82e0162d6550d3f129"
--
1.9.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v2 3/3] valgrind: avoid neon for targets which don't support it
2016-01-20 2:47 [PATCH v2 0/3] valgrind cleanup + fix for armv7a without vfp/neon Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 1/3] valgrind: let valgrind determine its own optimisation flags Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 2/3] valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests Andre McCurdy
@ 2016-01-20 2:47 ` Andre McCurdy
2016-01-20 3:00 ` Rongqing Li
2 siblings, 1 reply; 6+ messages in thread
From: Andre McCurdy @ 2016-01-20 2:47 UTC (permalink / raw)
To: openembedded-core
The sh-mem-random.c test app tries to use neon loads and stores to
test 64-bit float copies when building for ARM. Allow it to do so if
possible, but fallback to C when building for ARM targets which don't
support neon.
Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
---
...d-neon-for-targets-which-don-t-support-it.patch | 33 ++++++++++++++++++++++
meta/recipes-devtools/valgrind/valgrind_3.11.0.bb | 1 +
2 files changed, 34 insertions(+)
create mode 100644 meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
diff --git a/meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch b/meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
new file mode 100644
index 0000000..5fcfec0
--- /dev/null
+++ b/meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
@@ -0,0 +1,33 @@
+From 8facc29c3c56e6cf9cfef70986cf73876044a3fb Mon Sep 17 00:00:00 2001
+From: Andre McCurdy <armccurdy@gmail.com>
+Date: Tue, 19 Jan 2016 16:42:36 -0800
+Subject: [PATCH] avoid neon for targets which don't support it
+
+The sh-mem-random.c test app tries to use neon loads and stores to
+test 64-bit float copies when building for ARM. Allow it to do so if
+possible, but fallback to C when building for ARM targets which don't
+support neon.
+
+Upstream-Status: Pending
+
+Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
+---
+ memcheck/tests/sh-mem-random.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/memcheck/tests/sh-mem-random.c b/memcheck/tests/sh-mem-random.c
+index ae82248..816e139 100644
+--- a/memcheck/tests/sh-mem-random.c
++++ b/memcheck/tests/sh-mem-random.c
+@@ -191,7 +191,7 @@ void do_test_at ( U1* arr )
+ "emms"
+ : : "r"(arr+dst), "r"(arr+src) : "memory"
+ );
+-#elif defined(__linux__) && defined(__arm__) && !defined(__aarch64__)
++#elif defined(__linux__) && defined(__arm__) && defined(__ARM_NEON__) && !defined(__aarch64__)
+ /* On arm32, many compilers generate a 64-bit float move
+ using two 32 bit integer registers, which completely
+ defeats this test. Hence force a 64-bit NEON load and
+--
+1.9.1
+
diff --git a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
index 4f7c39a..7ef1e48 100644
--- a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
+++ b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
@@ -22,6 +22,7 @@ SRC_URI = "http://www.valgrind.org/downloads/valgrind-${PV}.tar.bz2 \
file://0005-Modify-vg_test-wrapper-to-support-PTEST-formats.patch \
file://0001-Remove-tests-that-fail-to-build-on-some-PPC32-config.patch \
file://use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch \
+ file://avoid-neon-for-targets-which-don-t-support-it.patch \
"
SRC_URI[md5sum] = "4ea62074da73ae82e0162d6550d3f129"
--
1.9.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH v2 3/3] valgrind: avoid neon for targets which don't support it
2016-01-20 2:47 ` [PATCH v2 3/3] valgrind: avoid neon for targets which don't support it Andre McCurdy
@ 2016-01-20 3:00 ` Rongqing Li
0 siblings, 0 replies; 6+ messages in thread
From: Rongqing Li @ 2016-01-20 3:00 UTC (permalink / raw)
To: Andre McCurdy, openembedded-core
On 2016年01月20日 10:47, Andre McCurdy wrote:
> The sh-mem-random.c test app tries to use neon loads and stores to
> test 64-bit float copies when building for ARM. Allow it to do so if
> possible, but fallback to C when building for ARM targets which don't
> support neon.
>
> Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
it work for me, thank Andre McCurdy
Reported-and-Tested-by: Roy Li <rongqing.li@windriver.com>
-Roy
> ---
> ...d-neon-for-targets-which-don-t-support-it.patch | 33 ++++++++++++++++++++++
> meta/recipes-devtools/valgrind/valgrind_3.11.0.bb | 1 +
> 2 files changed, 34 insertions(+)
> create mode 100644 meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
>
> diff --git a/meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch b/meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
> new file mode 100644
> index 0000000..5fcfec0
> --- /dev/null
> +++ b/meta/recipes-devtools/valgrind/valgrind/avoid-neon-for-targets-which-don-t-support-it.patch
> @@ -0,0 +1,33 @@
> +From 8facc29c3c56e6cf9cfef70986cf73876044a3fb Mon Sep 17 00:00:00 2001
> +From: Andre McCurdy <armccurdy@gmail.com>
> +Date: Tue, 19 Jan 2016 16:42:36 -0800
> +Subject: [PATCH] avoid neon for targets which don't support it
> +
> +The sh-mem-random.c test app tries to use neon loads and stores to
> +test 64-bit float copies when building for ARM. Allow it to do so if
> +possible, but fallback to C when building for ARM targets which don't
> +support neon.
> +
> +Upstream-Status: Pending
> +
> +Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
> +---
> + memcheck/tests/sh-mem-random.c | 2 +-
> + 1 file changed, 1 insertion(+), 1 deletion(-)
> +
> +diff --git a/memcheck/tests/sh-mem-random.c b/memcheck/tests/sh-mem-random.c
> +index ae82248..816e139 100644
> +--- a/memcheck/tests/sh-mem-random.c
> ++++ b/memcheck/tests/sh-mem-random.c
> +@@ -191,7 +191,7 @@ void do_test_at ( U1* arr )
> + "emms"
> + : : "r"(arr+dst), "r"(arr+src) : "memory"
> + );
> +-#elif defined(__linux__) && defined(__arm__) && !defined(__aarch64__)
> ++#elif defined(__linux__) && defined(__arm__) && defined(__ARM_NEON__) && !defined(__aarch64__)
> + /* On arm32, many compilers generate a 64-bit float move
> + using two 32 bit integer registers, which completely
> + defeats this test. Hence force a 64-bit NEON load and
> +--
> +1.9.1
> +
> diff --git a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
> index 4f7c39a..7ef1e48 100644
> --- a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
> +++ b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
> @@ -22,6 +22,7 @@ SRC_URI = "http://www.valgrind.org/downloads/valgrind-${PV}.tar.bz2 \
> file://0005-Modify-vg_test-wrapper-to-support-PTEST-formats.patch \
> file://0001-Remove-tests-that-fail-to-build-on-some-PPC32-config.patch \
> file://use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch \
> + file://avoid-neon-for-targets-which-don-t-support-it.patch \
> "
>
> SRC_URI[md5sum] = "4ea62074da73ae82e0162d6550d3f129"
>
--
Best Reagrds,
Roy | RongQing Li
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH v2 2/3] valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests
2016-01-20 2:47 ` [PATCH v2 2/3] valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests Andre McCurdy
@ 2016-01-20 3:00 ` Rongqing Li
0 siblings, 0 replies; 6+ messages in thread
From: Rongqing Li @ 2016-01-20 3:00 UTC (permalink / raw)
To: Andre McCurdy, openembedded-core
On 2016年01月20日 10:47, Andre McCurdy wrote:
> The intdiv test has been partially fixed upstream and the
> vcvt_fixed_float_VFP test can be fixed with a similar approach, ie
> ensuring that it is always compiled with appropriate
> -march/-mcpu/-mfpu flags to support the instructions being tested.
>
> For tests requiring armv7ve instructions, ensure that we set both
> -march=armv7ve and -mcpu=cortex-a15 (since some TUNE_CCARGS may set
> -march=armv7-a and adding -mcpu=cortex-a15 alone is not enough to
> over-ride that).
>
> See similar cases in none/tests/arm/Makefile.am
>
> Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
Tested-by: Roy Li <rongqing.li@windriver.com>
thanks
-Roy
> ---
> .../valgrind/remove-arm-variant-specific.patch | 66 ----------------------
> ...opriate-march-mcpu-mfpu-for-ARM-test-apps.patch | 44 +++++++++++++++
> meta/recipes-devtools/valgrind/valgrind_3.11.0.bb | 2 +-
> 3 files changed, 45 insertions(+), 67 deletions(-)
> delete mode 100644 meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
> create mode 100644 meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
>
> diff --git a/meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch b/meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
> deleted file mode 100644
> index 2319ab9..0000000
> --- a/meta/recipes-devtools/valgrind/valgrind/remove-arm-variant-specific.patch
> +++ /dev/null
> @@ -1,66 +0,0 @@
> -Remove arm tests that don't compile
> -
> -Upstream-Status: Pending
> -
> -Corrects the original commit for the patch that removed ARM ptest CFLAGS
> -settings. Since the flags could be set by a user, the flags should
> -be kept in place during compilation. By keeping the original up-stream
> -CFLAGS for the tests, then additional tests successfully compile
> -for all tested ARM tunings.
> -
> -However, there were still two tests listed below that did not compile
> -for any beaglebone tuning that is valid for valgrind. With the updated
> -patch, the set of excluded ARM ptests and their respective build
> -failures are:
> - intdiv - fails for all beaglebone tunings with 2 errors:
> - {standard input}:(40 or 41): Error: selected processor does not
> - support Thumb mode `udiv r3,r9,r10'
> - {standard input}:(72 or 73): Error: selected processor does not
> - support Thumb mode `sdiv r3,r9,r10'
> -
> - vcvt_fixed_float_VFP - fails for all beaglebone tunings in one of
> - two ways:
> - with neon tuning (-mfpu=neon) fails with Internal Compiler Error
> - without neon tuning fails with 3 errors:
> - {standard input}:33: Error: selected FPU does not support
> - instruction -- `vcvt.f32.s32 s15,s15,#1'
> - {standard input}:58: Error: selected FPU does not support
> - instruction -- `vcvt.f32.s32 s15,s15,#32'
> - {standard input}:136: Error: selected FPU does not support
> - instruction -- `vcvt.f32.u32 s15,s15,#1'
> -
> -After applying this commit, the valgrind ARM ptests compile without
> -errors for tunings:
> - armv7[t][hf][b][-neon] cortexa8[t][hf][-neon]
> -where the tuning [option] was successfully compiled, both with
> -and without the 'option', and in combination with all other options.
> -
> -Signed-off-by: Dave Lerner <dave.lerner@windriver.com>
> -
> -Index: valgrind-3.10.1/none/tests/arm/Makefile.am
> -===================================================================
> ---- valgrind-3.10.1.orig/none/tests/arm/Makefile.am 2015-05-19 15:11:59.224842927 -0500
> -+++ valgrind-3.10.1/none/tests/arm/Makefile.am 2015-05-19 15:14:20.808847028 -0500
> -@@ -17,9 +17,13 @@
> - vfp.stdout.exp vfp.stderr.exp vfp.vgtest \
> - vfpv4_fma.stdout.exp vfpv4_fma.stderr.exp vfpv4_fma.vgtest
> -
> -+# Remove the following tests which cause compiler errors for all tunings
> -+# available for beagle bone (see remove-arm-variant-specific.patch):
> -+# intdiv
> -+# vcvt_fixed_float_VFP
> -+
> - check_PROGRAMS = \
> - allexec \
> -- intdiv \
> - ldrt \
> - ldrt_arm \
> - neon128 \
> -@@ -27,7 +31,6 @@
> - v6intARM \
> - v6intThumb \
> - v6media \
> -- vcvt_fixed_float_VFP \
> - vfp \
> - vfpv4_fma
> -
> diff --git a/meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch b/meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
> new file mode 100644
> index 0000000..adea405
> --- /dev/null
> +++ b/meta/recipes-devtools/valgrind/valgrind/use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch
> @@ -0,0 +1,44 @@
> +From d134dafc2f11e0d247420a0ba360bcdef77b4093 Mon Sep 17 00:00:00 2001
> +From: Andre McCurdy <armccurdy@gmail.com>
> +Date: Tue, 19 Jan 2016 16:00:00 -0800
> +Subject: [PATCH] use appropriate -march/-mcpu/-mfpu for ARM test apps
> +
> +Ensure that test apps in none/tests/arm are compiled with appropriate
> +-march/-mcpu/-mfpu flags to support the instructions being tested.
> +The aim is to build all tests, even ones which may not run correctly
> +on all target CPUs.
> +
> +For tests requiring armv7ve instructions, ensure that we set both
> +-march=armv7ve and -mcpu=cortex-a15 (since some TUNE_CCARGS may set
> +-march=armv7-a and adding -mcpu=cortex-a15 alone is not enough to
> +over-ride that).
> +
> +See similar cases in none/tests/arm/Makefile.am
> +
> +Upstream-Status: Pending
> +
> +Signed-off-by: Andre McCurdy <armccurdy@gmail.com>
> +---
> + none/tests/arm/Makefile.am | 6 ++++--
> + 1 file changed, 4 insertions(+), 2 deletions(-)
> +
> +diff --git a/none/tests/arm/Makefile.am b/none/tests/arm/Makefile.am
> +index 4507a20..825290f 100644
> +--- a/none/tests/arm/Makefile.am
> ++++ b/none/tests/arm/Makefile.am
> +@@ -62,8 +62,10 @@ neon64_CFLAGS = $(AM_CFLAGS) -g -O0 -mcpu=cortex-a8 \
> + -mfpu=neon \
> + -mthumb
> +
> +-intdiv_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a15 -mthumb
> ++intdiv_CFLAGS = $(AM_CFLAGS) -g -march=armv7ve -mcpu=cortex-a15 -mthumb
> + ldrt_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a8 -mthumb
> + ldrt_arm_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a8 -marm
> +
> +-vfpv4_fma_CFLAGS = $(AM_CFLAGS) -g -O0 -mcpu=cortex-a15 -mfpu=vfpv4 -marm
> ++vcvt_fixed_float_VFP_CFLAGS = $(AM_CFLAGS) -g -mcpu=cortex-a8 -mfpu=vfpv3
> ++
> ++vfpv4_fma_CFLAGS = $(AM_CFLAGS) -g -O0 -march=armv7ve -mcpu=cortex-a15 -mfpu=vfpv4 -marm
> +--
> +1.9.1
> +
> diff --git a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
> index 8e9b72c..4f7c39a 100644
> --- a/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
> +++ b/meta/recipes-devtools/valgrind/valgrind_3.11.0.bb
> @@ -15,13 +15,13 @@ DEPENDS = "${@bb.utils.contains('DISTRO_FEATURES', 'x11', '${X11DEPENDS}', '', d
> SRC_URI = "http://www.valgrind.org/downloads/valgrind-${PV}.tar.bz2 \
> file://fixed-perl-path.patch \
> file://Added-support-for-PPC-instructions-mfatbu-mfatbl.patch \
> - file://remove-arm-variant-specific.patch \
> file://run-ptest \
> file://11_mips-link-tool.patch \
> file://0002-remove-rpath.patch \
> file://0004-Fix-out-of-tree-builds.patch \
> file://0005-Modify-vg_test-wrapper-to-support-PTEST-formats.patch \
> file://0001-Remove-tests-that-fail-to-build-on-some-PPC32-config.patch \
> + file://use-appropriate-march-mcpu-mfpu-for-ARM-test-apps.patch \
> "
>
> SRC_URI[md5sum] = "4ea62074da73ae82e0162d6550d3f129"
>
--
Best Reagrds,
Roy | RongQing Li
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2016-01-20 3:00 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-01-20 2:47 [PATCH v2 0/3] valgrind cleanup + fix for armv7a without vfp/neon Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 1/3] valgrind: let valgrind determine its own optimisation flags Andre McCurdy
2016-01-20 2:47 ` [PATCH v2 2/3] valgrind: re-enable ARM intdiv and vcvt_fixed_float_VFP tests Andre McCurdy
2016-01-20 3:00 ` Rongqing Li
2016-01-20 2:47 ` [PATCH v2 3/3] valgrind: avoid neon for targets which don't support it Andre McCurdy
2016-01-20 3:00 ` Rongqing Li
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.