rpms/ruby
7
0
Fork 0
Code Issues Pull Requests Releases Activity
17f011b0ca
ruby/ruby-3.0.0-preview1-Enable-arm64-optimizations-that-exist-for-power-x86.patch

179 lines
6.2 KiB
Diff
Raw Normal View History

RHEL 9.0.0 Alpha bootstrap The content of this branch was automatically imported from Fedora ELN with the following as its source: https://src.fedoraproject.org/rpms/ruby#be9961f6d007f12e7ac37b59800127a9d735c257
2020-10-15 04:55:41 +00:00
From c5806d668f84a86e9e6a522f84b8aa6cb4cdaae9 Mon Sep 17 00:00:00 2001
From: Ali Saidi <alisaidi@amazon.com>
Date: Wed, 5 Aug 2020 20:46:28 -0500
Subject: [PATCH 1/3] Enable unaligned accesses on arm64
64-bit Arm platforms support unaligned accesses.
Running the string benchmarks this change improves performance
by an average of 1.04x, min .96x, max 1.21x, median 1.01x
---
include/ruby/defines.h | 2 +-
regint.h | 2 +-
siphash.c | 2 +-
st.c | 2 +-
4 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/include/ruby/defines.h b/include/ruby/defines.h
index 49f673ef936a..0193275e8b78 100644
--- a/include/ruby/defines.h
+++ b/include/ruby/defines.h
@@ -485,7 +485,7 @@
#ifndef UNALIGNED_WORD_ACCESS
# if defined(__i386) || defined(__i386__) || defined(_M_IX86) || \
defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || \
- defined(__powerpc64__) || \
+ defined(__powerpc64__) || defined(__aarch64__) || \
defined(__mc68020__)
# define UNALIGNED_WORD_ACCESS 1
# else
diff --git a/regint.h b/regint.h
index a2f5bbba1d1f..0740429688bc 100644
--- a/regint.h
+++ b/regint.h
@@ -52,7 +52,7 @@
#ifndef UNALIGNED_WORD_ACCESS
# if defined(__i386) || defined(__i386__) || defined(_M_IX86) || \
defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || \
- defined(__powerpc64__) || \
+ defined(__powerpc64__) || defined(__aarch64__) || \
defined(__mc68020__)
# define UNALIGNED_WORD_ACCESS 1
# else
diff --git a/siphash.c b/siphash.c
index 153d2c690ab9..ddf8ee245d81 100644
--- a/siphash.c
+++ b/siphash.c
@@ -30,7 +30,7 @@
#ifndef UNALIGNED_WORD_ACCESS
# if defined(__i386) || defined(__i386__) || defined(_M_IX86) || \
defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || \
- defined(__powerpc64__) || \
+ defined(__powerpc64__) || defined(__aarch64__) || \
defined(__mc68020__)
# define UNALIGNED_WORD_ACCESS 1
# endif
diff --git a/st.c b/st.c
index c11535ef9779..8be466bf733f 100644
--- a/st.c
+++ b/st.c
@@ -1815,7 +1815,7 @@ st_values_check(st_table *tab, st_data_t *values, st_index_t size,
#ifndef UNALIGNED_WORD_ACCESS
# if defined(__i386) || defined(__i386__) || defined(_M_IX86) || \
defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || \
- defined(__powerpc64__) || \
+ defined(__powerpc64__) || defined(__aarch64__) || \
defined(__mc68020__)
# define UNALIGNED_WORD_ACCESS 1
# endif
From 79b7b9143fda0f33fc9375980cecc61eb42c6f66 Mon Sep 17 00:00:00 2001
From: Ali Saidi <alisaidi@amazon.com>
Date: Wed, 5 Aug 2020 21:04:37 -0500
Subject: [PATCH 2/3] arm64 enable gc optimizations
Similar to x86 and powerpc optimizations.
| |compare-ruby|built-ruby|
|:------|-----------:|---------:|
|hash1 | 0.225| 0.237|
| | -| 1.05x|
|hash2 | 0.110| 0.110|
| | 1.00x| -|
---
gc.c | 13 +++++++++++++
gc.h | 2 ++
2 files changed, 15 insertions(+)
diff --git a/gc.c b/gc.c
index 22972dfc806c..788f06f1586e 100644
--- a/gc.c
+++ b/gc.c
@@ -1153,6 +1153,19 @@ tick(void)
return val;
}
+#elif defined(__aarch64__) && defined(__GNUC__)
+typedef unsigned long tick_t;
+#define PRItick "lu"
+
+static __inline__ tick_t
+tick(void)
+{
+ unsigned long val;
+ __asm__ __volatile__ ("mrs %0, cntvct_el0", : "=r" (val));
+ return val;
+}
+
+
#elif defined(_WIN32) && defined(_MSC_VER)
#include <intrin.h>
typedef unsigned __int64 tick_t;
diff --git a/gc.h b/gc.h
index 6568079c54e5..47a4ca19a0c5 100644
--- a/gc.h
+++ b/gc.h
@@ -8,6 +8,8 @@
#define SET_MACHINE_STACK_END(p) __asm__ __volatile__ ("movl\t%%esp, %0" : "=r" (*(p)))
#elif defined(__powerpc64__) && defined(__GNUC__)
#define SET_MACHINE_STACK_END(p) __asm__ __volatile__ ("mr\t%0, %%r1" : "=r" (*(p)))
+#elif defined(__aarch64__) && defined(__GNUC__)
+#define SET_MACHINE_STACK_END(p) __asm__ __volatile__ ("mov\t%0, sp" : "=r" (*(p)))
#else
NOINLINE(void rb_gc_set_stack_end(VALUE **stack_end_p));
#define SET_MACHINE_STACK_END(p) rb_gc_set_stack_end(p)
From c985b8c6868a380e44e285368af4a4f414ce3309 Mon Sep 17 00:00:00 2001
From: Ali Saidi <alisaidi@amazon.com>
Date: Wed, 5 Aug 2020 21:15:55 -0500
Subject: [PATCH 3/3] vm_exec.c: improve performance for arm64
| |compare-ruby|built-ruby|
|:------------------------------|-----------:|---------:|
|vm_array | 26.501M| 27.959M|
| | -| 1.06x|
|vm_attr_ivar | 21.606M| 31.429M|
| | -| 1.45x|
|vm_attr_ivar_set | 21.178M| 26.113M|
| | -| 1.23x|
|vm_backtrace | 6.621| 6.668|
| | -| 1.01x|
|vm_bigarray | 26.205M| 29.958M|
| | -| 1.14x|
|vm_bighash | 504.155k| 479.306k|
| | 1.05x| -|
|vm_block | 16.692M| 21.315M|
| | -| 1.28x|
|block_handler_type_iseq | 5.083| 7.004|
| | -| 1.38x|
---
vm_exec.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/vm_exec.c b/vm_exec.c
index ce2e053ee745..7aa56f6ad620 100644
--- a/vm_exec.c
+++ b/vm_exec.c
@@ -27,6 +27,9 @@ static void vm_insns_counter_count_insn(int insn) {}
#elif defined(__GNUC__) && defined(__powerpc64__)
#define DECL_SC_REG(type, r, reg) register type reg_##r __asm__("r" reg)
+#elif defined(__GNUC__) && defined(__aarch64__)
+#define DECL_SC_REG(type, r, reg) register type reg_##r __asm__("x" reg)
+
#else
#define DECL_SC_REG(type, r, reg) register type reg_##r
#endif
@@ -74,6 +77,11 @@ vm_exec_core(rb_execution_context_t *ec, VALUE initial)
DECL_SC_REG(rb_control_frame_t *, cfp, "15");
#define USE_MACHINE_REGS 1
+#elif defined(__GNUC__) && defined(__aarch64__)
+ DECL_SC_REG(const VALUE *, pc, "19");
+ DECL_SC_REG(rb_control_frame_t *, cfp, "20");
+#define USE_MACHINE_REGS 1
+
#else
register rb_control_frame_t *reg_cfp;
const VALUE *reg_pc;
Reference in New Issue Copy Permalink