x86-64: Optimize strrchr/wcsrchr with AVX2

Message ID 20170601181401.GD28627@lucon.org
State New, archived
Headers

Commit Message

Lu, Hongjiu June 1, 2017, 6:14 p.m. UTC
  Optimize strrchr/wcsrchr with AVX2 to check 32 bytes with vector
instructions.  It is as fast as SSE2 version for small data sizes
and up to 1X faster for large data sizes on Haswell.  Select AVX2
version on AVX2 machines where vzeroupper is preferred and AVX
unaligned load is fast.

Any comments?

H.J.
--
	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
	strrchr-avx2 and wcsrchr-avx2.
	* sysdeps/x86_64/multiarch/ifunc-impl-list.c
	(__libc_ifunc_impl_list): Add tests for __strrchr_avx2,
	__strrchr_sse2, __wcsrchr_avx2 and __wcsrchr_sse2.
	* sysdeps/x86_64/multiarch/strrchr-avx2.S: New file.
	* sysdeps/x86_64/multiarch/strrchr.S: Likewise.
	* sysdeps/x86_64/multiarch/wcsrchr-avx2.S: Likewise.
	* sysdeps/x86_64/multiarch/wcsrchr.S: Likewise.
---
 sysdeps/x86_64/multiarch/Makefile          |   2 +
 sysdeps/x86_64/multiarch/ifunc-impl-list.c |  14 ++
 sysdeps/x86_64/multiarch/strrchr-avx2.S    | 235 +++++++++++++++++++++++++++++
 sysdeps/x86_64/multiarch/strrchr.S         |  63 ++++++++
 sysdeps/x86_64/multiarch/wcsrchr-avx2.S    |   3 +
 sysdeps/x86_64/multiarch/wcsrchr.S         |  55 +++++++
 6 files changed, 372 insertions(+)
 create mode 100644 sysdeps/x86_64/multiarch/strrchr-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/strrchr.S
 create mode 100644 sysdeps/x86_64/multiarch/wcsrchr-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/wcsrchr.S
  

Comments

Adhemerval Zanella June 1, 2017, 6:42 p.m. UTC | #1
On 01/06/2017 15:14, H.J. Lu wrote:
> Optimize strrchr/wcsrchr with AVX2 to check 32 bytes with vector
> instructions.  It is as fast as SSE2 version for small data sizes
> and up to 1X faster for large data sizes on Haswell.  Select AVX2
> version on AVX2 machines where vzeroupper is preferred and AVX
> unaligned load is fast.
> 
> Any comments?
> 
> H.J.
> --
> 	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
> 	strrchr-avx2 and wcsrchr-avx2.
> 	* sysdeps/x86_64/multiarch/ifunc-impl-list.c
> 	(__libc_ifunc_impl_list): Add tests for __strrchr_avx2,
> 	__strrchr_sse2, __wcsrchr_avx2 and __wcsrchr_sse2.
> 	* sysdeps/x86_64/multiarch/strrchr-avx2.S: New file.
> 	* sysdeps/x86_64/multiarch/strrchr.S: Likewise.

I think this could be an opportunity to avoid adding more assembly ifunc
written in assembly and use C implementation instead.
  

Patch

diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index aee4820..5d7825a 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -15,6 +15,7 @@  sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 \
 		   memmove-ssse3-back \
 		   memmove-avx512-no-vzeroupper strcasecmp_l-ssse3 \
 		   strchr-avx2 strchrnul-avx2 \
+		   strrchr-avx2 \
 		   strlen-avx2 strnlen-avx2 \
 		   strncase_l-ssse3 strcat-ssse3 strncat-ssse3\
 		   strcpy-ssse3 strncpy-ssse3 stpcpy-ssse3 stpncpy-ssse3 \
@@ -40,5 +41,6 @@  sysdep_routines += wmemcmp-sse4 wmemcmp-ssse3 wmemcmp-c \
 		   wmemcmp-avx2 \
 		   wcscpy-ssse3 wcscpy-c \
 		   wcschr-avx2 \
+		   wcsrchr-avx2 \
 		   wcslen-avx2 wcsnlen-avx2
 endif
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index 9f1ec4b..0112621 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -250,6 +250,13 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __strchrnul_avx2)
 	      IFUNC_IMPL_ADD (array, i, strchrnul, 1, __strchrnul_sse2))
 
+  /* Support sysdeps/x86_64/multiarch/strrchr.S.  */
+  IFUNC_IMPL (i, name, strrchr,
+	      IFUNC_IMPL_ADD (array, i, strrchr,
+			      HAS_ARCH_FEATURE (AVX2_Usable),
+			      __strrchr_avx2)
+	      IFUNC_IMPL_ADD (array, i, strrchr, 1, __strrchr_sse2))
+
   /* Support sysdeps/x86_64/multiarch/strcmp.S.  */
   IFUNC_IMPL (i, name, strcmp,
 	      IFUNC_IMPL_ADD (array, i, strcmp, HAS_CPU_FEATURE (SSE4_2),
@@ -341,6 +348,13 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __wcschr_avx2)
 	      IFUNC_IMPL_ADD (array, i, wcschr, 1, __wcschr_sse2))
 
+  /* Support sysdeps/x86_64/multiarch/wcsrchr.S.  */
+  IFUNC_IMPL (i, name, wcsrchr,
+	      IFUNC_IMPL_ADD (array, i, wcsrchr,
+			      HAS_ARCH_FEATURE (AVX2_Usable),
+			      __wcsrchr_avx2)
+	      IFUNC_IMPL_ADD (array, i, wcsrchr, 1, __wcsrchr_sse2))
+
   /* Support sysdeps/x86_64/multiarch/wcscpy.S.  */
   IFUNC_IMPL (i, name, wcscpy,
 	      IFUNC_IMPL_ADD (array, i, wcscpy, HAS_CPU_FEATURE (SSSE3),
diff --git a/sysdeps/x86_64/multiarch/strrchr-avx2.S b/sysdeps/x86_64/multiarch/strrchr-avx2.S
new file mode 100644
index 0000000..36ef660
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strrchr-avx2.S
@@ -0,0 +1,235 @@ 
+/* strrchr/wcsrchr optimized with AVX2.
+   Copyright (C) 2017 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#if IS_IN (libc)
+
+# include <sysdep.h>
+
+# ifndef STRRCHR
+#  define STRRCHR	__strrchr_avx2
+# endif
+
+# ifdef USE_AS_WCSRCHR
+#  define VPBROADCAST	vpbroadcastd
+#  define VPCMPEQ	vpcmpeqd
+# else
+#  define VPBROADCAST	vpbroadcastb
+#  define VPCMPEQ	vpcmpeqb
+# endif
+
+# ifndef VZEROUPPER
+#  define VZEROUPPER	vzeroupper
+# endif
+
+# define VEC_SIZE	32
+
+	.section .text.avx,"ax",@progbits
+ENTRY (STRRCHR)
+	movd	%esi, %xmm4
+	movl	%edi, %ecx
+	/* Broadcast CHAR to YMM4.  */
+	VPBROADCAST %xmm4, %ymm4
+	vpxor	%ymm0, %ymm0, %ymm0
+
+	/* Check if we may cross page boundary with one vector load.  */
+	andl	$(2 * VEC_SIZE - 1), %ecx
+	cmpl	$VEC_SIZE, %ecx
+	ja	L(cros_page_boundary)
+
+	vmovdqu	(%rdi), %ymm1
+	VPCMPEQ	%ymm1, %ymm0, %ymm2
+	VPCMPEQ	%ymm1, %ymm4, %ymm3
+	vpmovmskb %ymm2, %ecx
+	vpmovmskb %ymm3, %eax
+	addq	$VEC_SIZE, %rdi
+
+	testl	%eax, %eax
+	jnz	L(first_vec)
+
+	testl	%ecx, %ecx
+	jnz	L(return_null)
+
+	andq	$-VEC_SIZE, %rdi
+	xorl	%edx, %edx
+	jmp	L(aligned_loop)
+
+	.p2align 4
+L(first_vec):
+	/* Check if there is a nul CHAR.  */
+	testl	%ecx, %ecx
+	jnz	L(char_and_nul_in_first_vec)
+
+	/* Remember the match and keep searching.  */
+	movl	%eax, %edx
+	movq	%rdi, %rsi
+	andq	$-VEC_SIZE, %rdi
+	jmp	L(aligned_loop)
+
+	.p2align 4
+L(cros_page_boundary):
+	andl	$(VEC_SIZE - 1), %ecx
+	andq	$-VEC_SIZE, %rdi
+	vmovdqa	(%rdi), %ymm1
+	VPCMPEQ	%ymm1, %ymm0, %ymm2
+	VPCMPEQ	%ymm1, %ymm4, %ymm3
+	vpmovmskb %ymm2, %edx
+	vpmovmskb %ymm3, %eax
+	shrl	%cl, %edx
+	shrl	%cl, %eax
+	addq	$VEC_SIZE, %rdi
+
+	/* Check if there is a CHAR.  */
+	testl	%eax, %eax
+	jnz	L(found_char)
+
+	testl	%edx, %edx
+	jnz	L(return_null)
+
+	jmp	L(aligned_loop)
+
+	.p2align 4
+L(found_char):
+	testl	%edx, %edx
+	jnz	L(char_and_nul)
+
+	/* Remember the match and keep searching.  */
+	movl	%eax, %edx
+	leaq	(%rdi, %rcx), %rsi
+
+	.p2align 4
+L(aligned_loop):
+	vmovdqa	(%rdi), %ymm1
+	VPCMPEQ	%ymm1, %ymm0, %ymm2
+	addq	$VEC_SIZE, %rdi
+	VPCMPEQ	%ymm1, %ymm4, %ymm3
+	vpmovmskb %ymm2, %ecx
+	vpmovmskb %ymm3, %eax
+	orl	%eax, %ecx
+	jnz	L(char_nor_null)
+
+	vmovdqa	(%rdi), %ymm1
+	VPCMPEQ	%ymm1, %ymm0, %ymm2
+	add	$VEC_SIZE, %rdi
+	VPCMPEQ	%ymm1, %ymm4, %ymm3
+	vpmovmskb %ymm2, %ecx
+	vpmovmskb %ymm3, %eax
+	orl	%eax, %ecx
+	jnz	L(char_nor_null)
+
+	vmovdqa	(%rdi), %ymm1
+	VPCMPEQ	%ymm1, %ymm0, %ymm2
+	addq	$VEC_SIZE, %rdi
+	VPCMPEQ	%ymm1, %ymm4, %ymm3
+	vpmovmskb %ymm2, %ecx
+	vpmovmskb %ymm3, %eax
+	orl	%eax, %ecx
+	jnz	L(char_nor_null)
+
+	vmovdqa	(%rdi), %ymm1
+	VPCMPEQ	%ymm1, %ymm0, %ymm2
+	addq	$VEC_SIZE, %rdi
+	VPCMPEQ	%ymm1, %ymm4, %ymm3
+	vpmovmskb %ymm2, %ecx
+	vpmovmskb %ymm3, %eax
+	orl	%eax, %ecx
+	jz	L(aligned_loop)
+
+	.p2align 4
+L(char_nor_null):
+	/* Find a CHAR or a nul CHAR in a loop.  */
+	testl	%eax, %eax
+	jnz	L(match)
+L(return_value):
+	testl	%edx, %edx
+	jz	L(return_null)
+	movl	%edx, %eax
+	movq	%rsi, %rdi
+
+# ifdef USE_AS_WCSRCHR
+	/* Keep the first bit for each matching CHAR for bsr.  */
+	andl	$0x11111111, %eax
+# endif
+	bsrl	%eax, %eax
+	leaq	-VEC_SIZE(%rdi, %rax), %rax
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(match):
+	/* Find a CHAR.  Check if there is a nul CHAR.  */
+	vpmovmskb %ymm2, %ecx
+	testl	%ecx, %ecx
+	jnz	L(find_nul)
+
+	/* Remember the match and keep searching.  */
+	movl	%eax, %edx
+	movq	%rdi, %rsi
+	jmp	L(aligned_loop)
+
+	.p2align 4
+L(find_nul):
+# ifdef USE_AS_WCSRCHR
+	/* Keep the first bit for each matching CHAR for bsr.  */
+	andl	$0x11111111, %ecx
+	andl	$0x11111111, %eax
+# endif
+	/* Mask out any matching bits after the nul CHAR.  */
+	movl	%ecx, %r8d
+	subl	$1, %r8d
+	xorl	%ecx, %r8d
+	andl	%r8d, %eax
+	testl	%eax, %eax
+	/* If there is no CHAR here, return the remembered one.  */
+	jz	L(return_value)
+	bsrl	%eax, %eax
+	leaq	-VEC_SIZE(%rdi, %rax), %rax
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(char_and_nul):
+	/* Find both a CHAR and a nul CHAR.  */
+	addq	%rcx, %rdi
+	movl	%edx, %ecx
+L(char_and_nul_in_first_vec):
+# ifdef USE_AS_WCSRCHR
+	/* Keep the first bit for each matching CHAR for bsr.  */
+	andl	$0x11111111, %ecx
+	andl	$0x11111111, %eax
+# endif
+	/* Mask out any matching bits after the nul CHAR.  */
+	movl	%ecx, %r8d
+	subl	$1, %r8d
+	xorl	%ecx, %r8d
+	andl	%r8d, %eax
+	testl	%eax, %eax
+	/* Return null pointer if the nul CHAR comes first.  */
+	jz	L(return_null)
+	bsrl	%eax, %eax
+	leaq	-VEC_SIZE(%rdi, %rax), %rax
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(return_null):
+	xorl	%eax, %eax
+	VZEROUPPER
+	ret
+
+END (STRRCHR)
+#endif
diff --git a/sysdeps/x86_64/multiarch/strrchr.S b/sysdeps/x86_64/multiarch/strrchr.S
new file mode 100644
index 0000000..e5b49a9
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strrchr.S
@@ -0,0 +1,63 @@ 
+/* Multiple versions of strrchr
+   Copyright (C) 2009-2017 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+
+/* Define multiple versions only for the definition in libc.  */
+#if IS_IN (libc)
+	.text
+ENTRY(strrchr)
+	.type	strrchr, @gnu_indirect_function
+	LOAD_RTLD_GLOBAL_RO_RDX
+	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+	jnz	1f
+	HAS_ARCH_FEATURE (AVX2_Usable)
+	jz	1f
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	1f
+	leaq	__strrchr_avx2(%rip), %rax
+	ret
+1:
+	leaq	__strrchr_sse2(%rip), %rax
+	ret
+END(strrchr)
+
+
+
+# undef ENTRY
+# define ENTRY(name) \
+	.type __strrchr_sse2, @function; \
+	.align 16; \
+	.globl __strrchr_sse2; \
+	.hidden __strrchr_sse2; \
+	__strrchr_sse2: cfi_startproc; \
+	CALL_MCOUNT
+# undef END
+# define END(name) \
+	cfi_endproc; .size __strrchr_sse2, .-__strrchr_sse2
+# undef libc_hidden_builtin_def
+/* It doesn't make sense to send libc-internal strrchr calls through a PLT.
+   The speedup we get from using SSE4.2 instruction is likely eaten away
+   by the indirect call in the PLT.  */
+# define libc_hidden_builtin_def(name) \
+	.globl __GI_strrchr; __GI_strrchr = __strrchr_sse2
+#endif
+
+#include "../strrchr.S"
diff --git a/sysdeps/x86_64/multiarch/wcsrchr-avx2.S b/sysdeps/x86_64/multiarch/wcsrchr-avx2.S
new file mode 100644
index 0000000..cf8a239
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsrchr-avx2.S
@@ -0,0 +1,3 @@ 
+#define STRRCHR __wcsrchr_avx2
+#define USE_AS_WCSRCHR 1
+#include "strrchr-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsrchr.S b/sysdeps/x86_64/multiarch/wcsrchr.S
new file mode 100644
index 0000000..59276cb
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsrchr.S
@@ -0,0 +1,55 @@ 
+/* Multiple versions of wcsrchr
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2017 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+	.text
+ENTRY(wcsrchr)
+	.type	wcsrchr, @gnu_indirect_function
+	LOAD_RTLD_GLOBAL_RO_RDX
+	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+	jnz	1f
+	HAS_ARCH_FEATURE (AVX2_Usable)
+	jz	1f
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	1f
+	leaq	__wcsrchr_avx2(%rip), %rax
+	ret
+
+1:	leaq	__wcsrchr_sse2(%rip), %rax
+	ret
+END(wcsrchr)
+
+# undef ENTRY
+# define ENTRY(name) \
+	.type __wcsrchr_sse2, @function; \
+	.p2align 4; \
+	.globl __wcsrchr_sse2; \
+	.hidden __wcsrchr_sse2; \
+	__wcsrchr_sse2: cfi_startproc; \
+	CALL_MCOUNT
+# undef END
+# define END(name) \
+	cfi_endproc; .size __wcsrchr_sse2, .-__wcsrchr_sse2
+#endif
+
+#include "../wcsrchr.S"