[v5,2/2] x86: Add bounds `x86_non_temporal_threshold`
Commit Message
The lower-bound (131072) and upper-bound (SIZE_MAX / 16) are assumed
by memmove-vec-unaligned-erms.
The lower-bound is needed because memmove-vec-unaligned-erms unrolls
the loop aggressively in the L(large_memset_4x) case.
The upper-bound is needed because memmove-vec-unaligned-erms
right-shifts the value of `x86_non_temporal_threshold` by
LOG_4X_MEMCPY_THRESH (4) which without a bound may overflow.
The lack of lower-bound can be a correctness issue. The lack of
upper-bound cannot.
---
manual/tunables.texi | 2 +-
sysdeps/x86/dl-cacheinfo.h | 7 ++++++-
2 files changed, 7 insertions(+), 2 deletions(-)
Comments
On Wed, Jun 15, 2022 at 11:32 AM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> The lower-bound (131072) and upper-bound (SIZE_MAX / 16) are assumed
> by memmove-vec-unaligned-erms.
>
> The lower-bound is needed because memmove-vec-unaligned-erms unrolls
> the loop aggressively in the L(large_memset_4x) case.
>
> The upper-bound is needed because memmove-vec-unaligned-erms
> right-shifts the value of `x86_non_temporal_threshold` by
> LOG_4X_MEMCPY_THRESH (4) which without a bound may overflow.
>
> The lack of lower-bound can be a correctness issue. The lack of
> upper-bound cannot.
> ---
> manual/tunables.texi | 2 +-
> sysdeps/x86/dl-cacheinfo.h | 7 ++++++-
> 2 files changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/manual/tunables.texi b/manual/tunables.texi
> index 1482412078..a420ed6206 100644
> --- a/manual/tunables.texi
> +++ b/manual/tunables.texi
> @@ -47,7 +47,7 @@ glibc.malloc.mxfast: 0x0 (min: 0x0, max: 0xffffffffffffffff)
> glibc.elision.skip_lock_busy: 3 (min: -2147483648, max: 2147483647)
> glibc.malloc.top_pad: 0x0 (min: 0x0, max: 0xffffffffffffffff)
> glibc.cpu.x86_rep_stosb_threshold: 0x800 (min: 0x1, max: 0xffffffffffffffff)
> -glibc.cpu.x86_non_temporal_threshold: 0xc0000 (min: 0x0, max: 0xffffffffffffffff)
> +glibc.cpu.x86_non_temporal_threshold: 0xc0000 (min: 0x20000, max: 0x0fffffffffffffff)
> glibc.cpu.x86_shstk:
> glibc.cpu.hwcap_mask: 0x6 (min: 0x0, max: 0xffffffffffffffff)
> glibc.malloc.mmap_max: 0 (min: -2147483648, max: 2147483647)
> diff --git a/sysdeps/x86/dl-cacheinfo.h b/sysdeps/x86/dl-cacheinfo.h
> index cc3b840f9c..b4ff385ae1 100644
> --- a/sysdeps/x86/dl-cacheinfo.h
> +++ b/sysdeps/x86/dl-cacheinfo.h
> @@ -931,8 +931,13 @@ dl_init_cacheinfo (struct cpu_features *cpu_features)
>
> TUNABLE_SET_WITH_BOUNDS (x86_data_cache_size, data, 0, SIZE_MAX);
> TUNABLE_SET_WITH_BOUNDS (x86_shared_cache_size, shared, 0, SIZE_MAX);
> + /* SIZE_MAX >> 4 because memmove-vec-unaligned-erms right-shifts the value of
> + 'x86_non_temporal_threshold' by `LOG_4X_MEMCPY_THRESH` (4) and it is best
> + if that operation cannot overflow. 0x20000 (131072) because the
> + L(large_memset_4x) case aggressively unrolls the loop. Both values are
How is 0x20000 computed? Shouldn't it depend on vector size?
> + reflected in the manual. */
> TUNABLE_SET_WITH_BOUNDS (x86_non_temporal_threshold, non_temporal_threshold,
> - 0, SIZE_MAX);
> + 0x20000, SIZE_MAX >> 4);
> TUNABLE_SET_WITH_BOUNDS (x86_rep_movsb_threshold, rep_movsb_threshold,
> minimum_rep_movsb_threshold, SIZE_MAX);
> TUNABLE_SET_WITH_BOUNDS (x86_rep_stosb_threshold, rep_stosb_threshold, 1,
> --
> 2.34.1
>
@@ -47,7 +47,7 @@ glibc.malloc.mxfast: 0x0 (min: 0x0, max: 0xffffffffffffffff)
glibc.elision.skip_lock_busy: 3 (min: -2147483648, max: 2147483647)
glibc.malloc.top_pad: 0x0 (min: 0x0, max: 0xffffffffffffffff)
glibc.cpu.x86_rep_stosb_threshold: 0x800 (min: 0x1, max: 0xffffffffffffffff)
-glibc.cpu.x86_non_temporal_threshold: 0xc0000 (min: 0x0, max: 0xffffffffffffffff)
+glibc.cpu.x86_non_temporal_threshold: 0xc0000 (min: 0x20000, max: 0x0fffffffffffffff)
glibc.cpu.x86_shstk:
glibc.cpu.hwcap_mask: 0x6 (min: 0x0, max: 0xffffffffffffffff)
glibc.malloc.mmap_max: 0 (min: -2147483648, max: 2147483647)
@@ -931,8 +931,13 @@ dl_init_cacheinfo (struct cpu_features *cpu_features)
TUNABLE_SET_WITH_BOUNDS (x86_data_cache_size, data, 0, SIZE_MAX);
TUNABLE_SET_WITH_BOUNDS (x86_shared_cache_size, shared, 0, SIZE_MAX);
+ /* SIZE_MAX >> 4 because memmove-vec-unaligned-erms right-shifts the value of
+ 'x86_non_temporal_threshold' by `LOG_4X_MEMCPY_THRESH` (4) and it is best
+ if that operation cannot overflow. 0x20000 (131072) because the
+ L(large_memset_4x) case aggressively unrolls the loop. Both values are
+ reflected in the manual. */
TUNABLE_SET_WITH_BOUNDS (x86_non_temporal_threshold, non_temporal_threshold,
- 0, SIZE_MAX);
+ 0x20000, SIZE_MAX >> 4);
TUNABLE_SET_WITH_BOUNDS (x86_rep_movsb_threshold, rep_movsb_threshold,
minimum_rep_movsb_threshold, SIZE_MAX);
TUNABLE_SET_WITH_BOUNDS (x86_rep_stosb_threshold, rep_stosb_threshold, 1,