| Message ID | 20190922035458.14879-9-richard.henderson@linaro.org |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Move rom and notdirty handling to cputlb | expand |
On 9/22/19 5:54 AM, Richard Henderson wrote: > This forced inlining can result in missing symbols, > which makes a debugging build harder to follow. > > Reviewed-by: David Hildenbrand <david@redhat.com> > Reported-by: Peter Maydell <peter.maydell@linaro.org> > Signed-off-by: Richard Henderson <richard.henderson@linaro.org> > --- > include/qemu/compiler.h | 11 +++++++++++ > accel/tcg/cputlb.c | 4 ++-- > 2 files changed, 13 insertions(+), 2 deletions(-) > > diff --git a/include/qemu/compiler.h b/include/qemu/compiler.h > index 09fc44cca4..d6d400c523 100644 > --- a/include/qemu/compiler.h > +++ b/include/qemu/compiler.h > @@ -170,6 +170,17 @@ > # define QEMU_NONSTRING > #endif > > +/* > + * Forced inlining may be desired to encourage constant propagation > + * of function parameters. However, it can also make debugging harder, > + * so disable it for a non-optimizing build. > + */ > +#if defined(__OPTIMIZE__) && __has_attribute(always_inline) > +#define QEMU_ALWAYS_INLINE __attribute__((always_inline)) > +#else > +#define QEMU_ALWAYS_INLINE > +#endif > + > /* Implement C11 _Generic via GCC builtins. Example: > * > * QEMU_GENERIC(x, (float, sinf), (long double, sinl), sin) (x) > diff --git a/accel/tcg/cputlb.c b/accel/tcg/cputlb.c > index abae79650c..2222b87764 100644 > --- a/accel/tcg/cputlb.c > +++ b/accel/tcg/cputlb.c > @@ -1281,7 +1281,7 @@ static void *atomic_mmu_lookup(CPUArchState *env, target_ulong addr, > typedef uint64_t FullLoadHelper(CPUArchState *env, target_ulong addr, > TCGMemOpIdx oi, uintptr_t retaddr); > > -static inline uint64_t __attribute__((always_inline)) > +static inline uint64_t QEMU_ALWAYS_INLINE > load_helper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi, > uintptr_t retaddr, MemOp op, bool code_read, > FullLoadHelper *full_load) > @@ -1530,7 +1530,7 @@ tcg_target_ulong helper_be_ldsl_mmu(CPUArchState *env, target_ulong addr, > * Store Helpers > */ > > -static inline void __attribute__((always_inline)) > +static inline void QEMU_ALWAYS_INLINE > store_helper(CPUArchState *env, target_ulong addr, uint64_t val, > TCGMemOpIdx oi, uintptr_t retaddr, MemOp op) > { > Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
On 22/09/19 05:54, Richard Henderson wrote: > +/* > + * Forced inlining may be desired to encourage constant propagation > + * of function parameters. However, it can also make debugging harder, > + * so disable it for a non-optimizing build. > + */ > +#if defined(__OPTIMIZE__) && __has_attribute(always_inline) > +#define QEMU_ALWAYS_INLINE __attribute__((always_inline)) GCC doesn't have __has_attribute, does it? I think you can just assume that it exists and #ifdef __OPTIMIZE__. Paolo > +#else > +#define QEMU_ALWAYS_INLINE
On 9/23/19 2:45 AM, Paolo Bonzini wrote: > On 22/09/19 05:54, Richard Henderson wrote: >> +/* >> + * Forced inlining may be desired to encourage constant propagation >> + * of function parameters. However, it can also make debugging harder, >> + * so disable it for a non-optimizing build. >> + */ >> +#if defined(__OPTIMIZE__) && __has_attribute(always_inline) >> +#define QEMU_ALWAYS_INLINE __attribute__((always_inline)) > > GCC doesn't have __has_attribute, does it? It does, since at least gcc 5. And now I realize that's only a reorganization of the support and not when it was introduced. r~
On 23/09/19 18:00, Richard Henderson wrote: > On 9/23/19 2:45 AM, Paolo Bonzini wrote: >> On 22/09/19 05:54, Richard Henderson wrote: >>> +/* >>> + * Forced inlining may be desired to encourage constant propagation >>> + * of function parameters. However, it can also make debugging harder, >>> + * so disable it for a non-optimizing build. >>> + */ >>> +#if defined(__OPTIMIZE__) && __has_attribute(always_inline) >>> +#define QEMU_ALWAYS_INLINE __attribute__((always_inline)) >> >> GCC doesn't have __has_attribute, does it? > > It does, since at least gcc 5. And now I realize that's only a reorganization > of the support and not when it was introduced. Hmm, still we support 4.8 and always_inline is much older than that. So I'm not sure it's useful to test it. Paolo
On 9/23/19 9:49 AM, Paolo Bonzini wrote: > Hmm, still we support 4.8 and always_inline is much older than that. So > I'm not sure it's useful to test it. Fair enough. r~
diff --git a/include/qemu/compiler.h b/include/qemu/compiler.h index 09fc44cca4..d6d400c523 100644 --- a/include/qemu/compiler.h +++ b/include/qemu/compiler.h @@ -170,6 +170,17 @@ # define QEMU_NONSTRING #endif +/* + * Forced inlining may be desired to encourage constant propagation + * of function parameters. However, it can also make debugging harder, + * so disable it for a non-optimizing build. + */ +#if defined(__OPTIMIZE__) && __has_attribute(always_inline) +#define QEMU_ALWAYS_INLINE __attribute__((always_inline)) +#else +#define QEMU_ALWAYS_INLINE +#endif + /* Implement C11 _Generic via GCC builtins. Example: * * QEMU_GENERIC(x, (float, sinf), (long double, sinl), sin) (x) diff --git a/accel/tcg/cputlb.c b/accel/tcg/cputlb.c index abae79650c..2222b87764 100644 --- a/accel/tcg/cputlb.c +++ b/accel/tcg/cputlb.c @@ -1281,7 +1281,7 @@ static void *atomic_mmu_lookup(CPUArchState *env, target_ulong addr, typedef uint64_t FullLoadHelper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi, uintptr_t retaddr); -static inline uint64_t __attribute__((always_inline)) +static inline uint64_t QEMU_ALWAYS_INLINE load_helper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi, uintptr_t retaddr, MemOp op, bool code_read, FullLoadHelper *full_load) @@ -1530,7 +1530,7 @@ tcg_target_ulong helper_be_ldsl_mmu(CPUArchState *env, target_ulong addr, * Store Helpers */ -static inline void __attribute__((always_inline)) +static inline void QEMU_ALWAYS_INLINE store_helper(CPUArchState *env, target_ulong addr, uint64_t val, TCGMemOpIdx oi, uintptr_t retaddr, MemOp op) {