aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--benchtests/bench-malloc-simple.c2
-rw-r--r--elf/Makefile21
-rw-r--r--malloc/arena.c13
-rw-r--r--malloc/malloc.c2
-rw-r--r--sysdeps/aarch64/fpu/coshf_sve.c6
-rw-r--r--sysdeps/aarch64/fpu/expf_sve.c6
-rw-r--r--sysdeps/aarch64/fpu/sv_expf_inline.h30
7 files changed, 34 insertions, 46 deletions
diff --git a/benchtests/bench-malloc-simple.c b/benchtests/bench-malloc-simple.c
index f93d76e..1d7989e 100644
--- a/benchtests/bench-malloc-simple.c
+++ b/benchtests/bench-malloc-simple.c
@@ -35,7 +35,7 @@
multi-threaded using thread-arena, and main arena with SINGLE_THREAD_P
false. */
-#define NUM_ITERS 200000
+#define NUM_ITERS 5000000
#define NUM_ALLOCS 4
#define MAX_ALLOCS 1600
diff --git a/elf/Makefile b/elf/Makefile
index 2bce1ed..3d60000 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -456,6 +456,7 @@ tests += \
tst-noload \
tst-non-directory-path \
tst-null-argv \
+ tst-origin \
tst-p_align1 \
tst-p_align2 \
tst-p_align3 \
@@ -763,6 +764,7 @@ modules-names += \
libmarkermod5-3 \
libmarkermod5-4 \
libmarkermod5-5 \
+ liborigin-mod \
libtracemod1-1 \
libtracemod2-1 \
libtracemod3-1 \
@@ -1421,6 +1423,10 @@ $(objpfx)tst-_dl_addr_inside_object: $(objpfx)dl-addr-obj.os
CFLAGS-tst-_dl_addr_inside_object.c += $(PIE-ccflag)
endif
+ifeq ($(run-built-tests),yes)
+tests-special += $(objpfx)tst-origin.out
+endif
+
include ../Rules
ifeq (yes,$(build-shared))
@@ -3443,19 +3449,10 @@ $(objpfx)tst-dlopen-constructor-null: \
$(objpfx)tst-dlopen-constructor-null-mod2.so: \
$(objpfx)tst-dlopen-constructor-null-mod1.so
-ifeq ($(run-built-tests),yes)
-tests-special += $(objpfx)tst-origin.out
-endif
CFLAGS-tst-origin.c += $(no-stack-protector)
-$(objpfx)tst-origin: $(objpfx)tst-origin.o $(objpfx)liborigin-mod.so
- $(LINK.o) -o $@ -B$(csu-objpfx) $(LDFLAGS.so) $< \
- -Wl,-rpath,\$$ORIGIN \
- -L$(subst :, -L,$(rpath-link)) -Wl,--no-as-needed -lorigin-mod
-$(objpfx)liborigin-mod.so: $(objpfx)liborigin-mod.os
- $(LINK.o) -shared -o $@ -B$(csu-objpfx) $(LDFLAGS.so) \
- $(LDFLAGS-soname-fname) \
- $<
-$(objpfx)tst-origin.out: tst-origin.sh $(objpfx)tst-origin
+CFLAGS-liborigin-mod.c += $(no-stack-protector)
+LDFLAGS-tst-origin += -Wl,-rpath,\$$ORIGIN -L$(subst :, -L,$(rpath-link)) -lorigin-mod
+$(objpfx)tst-origin.out: tst-origin.sh $(objpfx)liborigin-mod.so $(objpfx)tst-origin
$(SHELL) \
$< \
'$(common-objpfx)' \
diff --git a/malloc/arena.c b/malloc/arena.c
index 353b634..405ae82 100644
--- a/malloc/arena.c
+++ b/malloc/arena.c
@@ -40,19 +40,20 @@
mmap threshold, so that requests with a size just below that
threshold can be fulfilled without creating too many heaps. */
-/* When huge pages are used to create new arenas, the maximum and minimum
- size are based on the runtime defined huge page size. */
+/* HEAP_MAX_SIZE should be larger than the huge page size, otherwise heaps will
+ use not huge pages. It is a constant so arena_for_chunk() is efficient. */
static inline size_t
heap_min_size (void)
{
- return mp_.hp_pagesize == 0 ? HEAP_MIN_SIZE : mp_.hp_pagesize;
+ return mp_.hp_pagesize == 0 || mp_.hp_pagesize > HEAP_MAX_SIZE
+ ? HEAP_MIN_SIZE : mp_.hp_pagesize;
}
static inline size_t
heap_max_size (void)
{
- return mp_.hp_pagesize == 0 ? HEAP_MAX_SIZE : mp_.hp_pagesize * 4;
+ return HEAP_MAX_SIZE;
}
/***************************************************************************/
@@ -313,7 +314,7 @@ ptmalloc_init (void)
TUNABLE_GET (mxfast, size_t, TUNABLE_CALLBACK (set_mxfast));
TUNABLE_GET (hugetlb, size_t, TUNABLE_CALLBACK (set_hugetlb));
- if (mp_.hp_pagesize > 0)
+ if (mp_.hp_pagesize > 0 && mp_.hp_pagesize <= heap_max_size ())
{
/* Force mmap for main arena instead of sbrk, so MAP_HUGETLB is always
tried. Also tune the mmap threshold, so allocation smaller than the
@@ -460,7 +461,7 @@ alloc_new_heap (size_t size, size_t top_pad, size_t pagesize,
static heap_info *
new_heap (size_t size, size_t top_pad)
{
- if (__glibc_unlikely (mp_.hp_pagesize != 0))
+ if (mp_.hp_pagesize != 0 && mp_.hp_pagesize <= heap_max_size ())
{
heap_info *h = alloc_new_heap (size, top_pad, mp_.hp_pagesize,
mp_.hp_flags);
diff --git a/malloc/malloc.c b/malloc/malloc.c
index 931ca48..55fb2ab 100644
--- a/malloc/malloc.c
+++ b/malloc/malloc.c
@@ -298,7 +298,7 @@
# define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ)
/* When "x" is from chunksize(). */
-# define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT)
+# define csize2tidx(x) (((x) - MINSIZE) / MALLOC_ALIGNMENT)
/* When "x" is a user-provided size. */
# define usize2tidx(x) csize2tidx (request2size (x))
diff --git a/sysdeps/aarch64/fpu/coshf_sve.c b/sysdeps/aarch64/fpu/coshf_sve.c
index fb8e06c..8056055 100644
--- a/sysdeps/aarch64/fpu/coshf_sve.c
+++ b/sysdeps/aarch64/fpu/coshf_sve.c
@@ -39,9 +39,9 @@ special_case (svfloat32_t x, svfloat32_t half_e, svfloat32_t half_over_e,
}
/* Single-precision vector cosh, using vector expf.
- Maximum error is 2.77 ULP:
- _ZGVsMxv_coshf(-0x1.5b38f4p+1) got 0x1.e45946p+2
- want 0x1.e4594cp+2. */
+ Maximum error is 2.56 +0.5 ULP:
+ _ZGVsMxv_coshf(-0x1.5b40f4p+1) got 0x1.e47748p+2
+ want 0x1.e4774ep+2. */
svfloat32_t SV_NAME_F1 (cosh) (svfloat32_t x, svbool_t pg)
{
const struct data *d = ptr_barrier (&data);
diff --git a/sysdeps/aarch64/fpu/expf_sve.c b/sysdeps/aarch64/fpu/expf_sve.c
index f9249db..c361997 100644
--- a/sysdeps/aarch64/fpu/expf_sve.c
+++ b/sysdeps/aarch64/fpu/expf_sve.c
@@ -40,9 +40,9 @@ special_case (svfloat32_t x, svbool_t special, const struct sv_expf_data *d)
}
/* Optimised single-precision SVE exp function.
- Worst-case error is 1.04 ulp:
- SV_NAME_F1 (exp)(0x1.a8eda4p+1) got 0x1.ba74bcp+4
- want 0x1.ba74bap+4. */
+ Worst-case error is 0.88 +0.50 ULP:
+ _ZGVsMxv_expf(-0x1.bba276p-6) got 0x1.f25288p-1
+ want 0x1.f2528ap-1. */
svfloat32_t SV_NAME_F1 (exp) (svfloat32_t x, const svbool_t pg)
{
const struct data *d = ptr_barrier (&data);
diff --git a/sysdeps/aarch64/fpu/sv_expf_inline.h b/sysdeps/aarch64/fpu/sv_expf_inline.h
index 16b81fc..e2d2e90 100644
--- a/sysdeps/aarch64/fpu/sv_expf_inline.h
+++ b/sysdeps/aarch64/fpu/sv_expf_inline.h
@@ -24,50 +24,40 @@
struct sv_expf_data
{
- float c1, c3, inv_ln2;
- float ln2_lo, c0, c2, c4;
- float ln2_hi, shift;
+ float ln2_hi, ln2_lo, c1, null;
+ float inv_ln2, shift;
};
-/* Coefficients copied from the polynomial in AdvSIMD variant, reversed for
- compatibility with polynomial helpers. Shift is 1.5*2^17 + 127. */
+/* Shift is 1.5*2^17 + 127. */
#define SV_EXPF_DATA \
{ \
- /* Coefficients copied from the polynomial in AdvSIMD variant. */ \
- .c0 = 0x1.ffffecp-1f, .c1 = 0x1.fffdb6p-2f, .c2 = 0x1.555e66p-3f, \
- .c3 = 0x1.573e2ep-5f, .c4 = 0x1.0e4020p-7f, .inv_ln2 = 0x1.715476p+0f, \
- .ln2_hi = 0x1.62e4p-1f, .ln2_lo = 0x1.7f7d1cp-20f, \
- .shift = 0x1.803f8p17f, \
+ .c1 = 0.5f, .inv_ln2 = 0x1.715476p+0f, .ln2_hi = 0x1.62e4p-1f, \
+ .ln2_lo = 0x1.7f7d1cp-20f, .shift = 0x1.803f8p17f, \
}
-#define C(i) sv_f32 (d->poly[i])
-
static inline svfloat32_t
expf_inline (svfloat32_t x, const svbool_t pg, const struct sv_expf_data *d)
{
/* exp(x) = 2^n (1 + poly(r)), with 1 + poly(r) in [1/sqrt(2),sqrt(2)]
x = ln2*n + r, with r in [-ln2/2, ln2/2]. */
- svfloat32_t lane_consts = svld1rq (svptrue_b32 (), &d->ln2_lo);
+ svfloat32_t lane_consts = svld1rq (svptrue_b32 (), &d->ln2_hi);
/* n = round(x/(ln2/N)). */
svfloat32_t z = svmad_x (pg, sv_f32 (d->inv_ln2), x, d->shift);
svfloat32_t n = svsub_x (pg, z, d->shift);
/* r = x - n*ln2/N. */
- svfloat32_t r = svmsb_x (pg, sv_f32 (d->ln2_hi), n, x);
+ svfloat32_t r = x;
r = svmls_lane (r, n, lane_consts, 0);
+ r = svmls_lane (r, n, lane_consts, 1);
/* scale = 2^(n/N). */
svfloat32_t scale = svexpa (svreinterpret_u32 (z));
- /* poly(r) = exp(r) - 1 ~= C0 r + C1 r^2 + C2 r^3 + C3 r^4 + C4 r^5. */
- svfloat32_t p12 = svmla_lane (sv_f32 (d->c1), r, lane_consts, 2);
- svfloat32_t p34 = svmla_lane (sv_f32 (d->c3), r, lane_consts, 3);
+ /* poly(r) = exp(r) - 1 ~= r + 0.5 r^2. */
svfloat32_t r2 = svmul_x (svptrue_b32 (), r, r);
- svfloat32_t p14 = svmla_x (pg, p12, p34, r2);
- svfloat32_t p0 = svmul_lane (r, lane_consts, 1);
- svfloat32_t poly = svmla_x (pg, p0, r2, p14);
+ svfloat32_t poly = svmla_lane (r, r2, lane_consts, 2);
return svmla_x (pg, scale, scale, poly);
}
generated by cgit v1.1 at 2025-03-19 04:23:04 +0000