| Message ID | 20220511150725.744021-1-kda@semihalf.com (mailing list archive) |
|---|---|
| State | Accepted |
| Delegated to: | David Marchand |
| Headers | show |
| Series | [v2,1/1] test/ring: remove excessive inlining | expand |
| Context | Check | Description |
|---|---|---|
| ci/iol-x86_64-compile-testing | success | Testing PASS |
| ci/iol-abi-testing | success | Testing PASS |
| ci/iol-x86_64-unit-testing | success | Testing PASS |
| ci/iol-aarch64-compile-testing | success | Testing PASS |
| ci/iol-aarch64-unit-testing | success | Testing PASS |
| ci/iol-intel-Functional | success | Functional Testing PASS |
| ci/iol-intel-Performance | success | Performance Testing PASS |
| ci/iol-mellanox-Performance | success | Performance Testing PASS |
| ci/github-robot: build | success | github build: passed |
| ci/intel-Testing | success | Testing PASS |
| ci/Intel-compilation | success | Compilation OK |
| ci/checkpatch | success | coding style OK |
On Wed, May 11, 2022 at 05:07:25PM +0200, Stanislaw Kardach wrote: > Forcing inlining in test_ring_enqueue and test_ring_dequeue can cause > the compiled code to grow extensively when compiled with no optimization > (-O0 or -Og). This is default in the meson's debug configuration. This > can collide with compiler bugs and cause issues during linking of unit > tests where the api_type or esize are non-const variables causing > inlining cascade. In perf tests this is not the case in perf-tests as > esize and api_type are const values. > > One such case was discovered when porting DPDK to RISC-V. GCC 11.2 (and > no fix still in 12.1) is generating a short relative jump instruction > (J <offset>) for goto and for loops. When loop body grows extensively in > ring test, the target offset goes beyond supported offfset of +/- 1MB > from PC. This is an obvious bug in the GCC as RISC-V has a > two-instruction construct to jump to any absolute address (AUIPC+JALR). > > However there is no reason to force inlining as the test code works > perfectly fine without it. > > GCC has a bug report for a similar case (with conditionals): > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93062 > > Fixes: a9fe152363 test/ring: add custom element size functional tests > > Signed-off-by: Stanislaw Kardach <kda@semihalf.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
<snip> > > Forcing inlining in test_ring_enqueue and test_ring_dequeue can cause the > compiled code to grow extensively when compiled with no optimization > (-O0 or -Og). This is default in the meson's debug configuration. This can collide > with compiler bugs and cause issues during linking of unit tests where the > api_type or esize are non-const variables causing inlining cascade. In perf tests > this is not the case in perf-tests as esize and api_type are const values. > > One such case was discovered when porting DPDK to RISC-V. GCC 11.2 (and no > fix still in 12.1) is generating a short relative jump instruction (J <offset>) for > goto and for loops. When loop body grows extensively in ring test, the target > offset goes beyond supported offfset of +/- 1MB from PC. This is an obvious > bug in the GCC as RISC-V has a two-instruction construct to jump to any > absolute address (AUIPC+JALR). > > However there is no reason to force inlining as the test code works perfectly > fine without it. > > GCC has a bug report for a similar case (with conditionals): > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93062 > > Fixes: a9fe152363 test/ring: add custom element size functional tests > > Signed-off-by: Stanislaw Kardach <kda@semihalf.com> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
11/05/2022 16:07, Stanislaw Kardach пишет: > Forcing inlining in test_ring_enqueue and test_ring_dequeue can cause > the compiled code to grow extensively when compiled with no optimization > (-O0 or -Og). This is default in the meson's debug configuration. This > can collide with compiler bugs and cause issues during linking of unit > tests where the api_type or esize are non-const variables causing > inlining cascade. In perf tests this is not the case in perf-tests as > esize and api_type are const values. > > One such case was discovered when porting DPDK to RISC-V. GCC 11.2 (and > no fix still in 12.1) is generating a short relative jump instruction > (J <offset>) for goto and for loops. When loop body grows extensively in > ring test, the target offset goes beyond supported offfset of +/- 1MB > from PC. This is an obvious bug in the GCC as RISC-V has a > two-instruction construct to jump to any absolute address (AUIPC+JALR). > > However there is no reason to force inlining as the test code works > perfectly fine without it. > > GCC has a bug report for a similar case (with conditionals): > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93062 > > Fixes: a9fe152363 test/ring: add custom element size functional tests > > Signed-off-by: Stanislaw Kardach <kda@semihalf.com> > --- > app/test/test_ring.h | 6 +++--- > 1 file changed, 3 insertions(+), 3 deletions(-) > > diff --git a/app/test/test_ring.h b/app/test/test_ring.h > index c8bfec8399..45c263f3ff 100644 > --- a/app/test/test_ring.h > +++ b/app/test/test_ring.h > @@ -97,7 +97,7 @@ test_ring_copy_from(struct rte_ring_zc_data *zcd, void *dst, int esize, > } > } > > -static __rte_always_inline unsigned int > +static inline unsigned int > test_ring_enqueue(struct rte_ring *r, void **obj, int esize, unsigned int n, > unsigned int api_type) > { > @@ -158,7 +158,7 @@ test_ring_enqueue(struct rte_ring *r, void **obj, int esize, unsigned int n, > } > } > > -static __rte_always_inline unsigned int > +static inline unsigned int > test_ring_dequeue(struct rte_ring *r, void **obj, int esize, unsigned int n, > unsigned int api_type) > { > @@ -222,7 +222,7 @@ test_ring_dequeue(struct rte_ring *r, void **obj, int esize, unsigned int n, > /* This function is placed here as it is required for both > * performance and functional tests. > */ > -static __rte_always_inline void * > +static inline void * > test_ring_calloc(unsigned int rsize, int esize) > { > unsigned int sz; Acked-by: Konstantin Ananyev <konstantin.v.ananyev@yandex.ru>
On Wed, May 11, 2022 at 5:49 PM Bruce Richardson <bruce.richardson@intel.com> wrote: > > On Wed, May 11, 2022 at 05:07:25PM +0200, Stanislaw Kardach wrote: > > Forcing inlining in test_ring_enqueue and test_ring_dequeue can cause > > the compiled code to grow extensively when compiled with no optimization > > (-O0 or -Og). This is default in the meson's debug configuration. This > > can collide with compiler bugs and cause issues during linking of unit > > tests where the api_type or esize are non-const variables causing > > inlining cascade. In perf tests this is not the case in perf-tests as > > esize and api_type are const values. > > > > One such case was discovered when porting DPDK to RISC-V. GCC 11.2 (and > > no fix still in 12.1) is generating a short relative jump instruction > > (J <offset>) for goto and for loops. When loop body grows extensively in > > ring test, the target offset goes beyond supported offfset of +/- 1MB > > from PC. This is an obvious bug in the GCC as RISC-V has a > > two-instruction construct to jump to any absolute address (AUIPC+JALR). > > > > However there is no reason to force inlining as the test code works > > perfectly fine without it. > > > > GCC has a bug report for a similar case (with conditionals): > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93062 > > > > Fixes: a9fe152363 test/ring: add custom element size functional tests Fixes: a9fe152363e2 ("test/ring: add custom element size functional tests") > > > > Signed-off-by: Stanislaw Kardach <kda@semihalf.com> > Acked-by: Bruce Richardson <bruce.richardson@intel.com> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Acked-by: Konstantin Ananyev <konstantin.v.ananyev@yandex.ru> Applied, thanks.
diff --git a/app/test/test_ring.h b/app/test/test_ring.h index c8bfec8399..45c263f3ff 100644 --- a/app/test/test_ring.h +++ b/app/test/test_ring.h @@ -97,7 +97,7 @@ test_ring_copy_from(struct rte_ring_zc_data *zcd, void *dst, int esize, } } -static __rte_always_inline unsigned int +static inline unsigned int test_ring_enqueue(struct rte_ring *r, void **obj, int esize, unsigned int n, unsigned int api_type) { @@ -158,7 +158,7 @@ test_ring_enqueue(struct rte_ring *r, void **obj, int esize, unsigned int n, } } -static __rte_always_inline unsigned int +static inline unsigned int test_ring_dequeue(struct rte_ring *r, void **obj, int esize, unsigned int n, unsigned int api_type) { @@ -222,7 +222,7 @@ test_ring_dequeue(struct rte_ring *r, void **obj, int esize, unsigned int n, /* This function is placed here as it is required for both * performance and functional tests. */ -static __rte_always_inline void * +static inline void * test_ring_calloc(unsigned int rsize, int esize) { unsigned int sz;
Forcing inlining in test_ring_enqueue and test_ring_dequeue can cause the compiled code to grow extensively when compiled with no optimization (-O0 or -Og). This is default in the meson's debug configuration. This can collide with compiler bugs and cause issues during linking of unit tests where the api_type or esize are non-const variables causing inlining cascade. In perf tests this is not the case in perf-tests as esize and api_type are const values. One such case was discovered when porting DPDK to RISC-V. GCC 11.2 (and no fix still in 12.1) is generating a short relative jump instruction (J <offset>) for goto and for loops. When loop body grows extensively in ring test, the target offset goes beyond supported offfset of +/- 1MB from PC. This is an obvious bug in the GCC as RISC-V has a two-instruction construct to jump to any absolute address (AUIPC+JALR). However there is no reason to force inlining as the test code works perfectly fine without it. GCC has a bug report for a similar case (with conditionals): https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93062 Fixes: a9fe152363 test/ring: add custom element size functional tests Signed-off-by: Stanislaw Kardach <kda@semihalf.com> --- app/test/test_ring.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)