about summary refs log tree commit diff
path: root/sysdeps/x86_64/multiarch/rawmemchr-sse2.S
diff options
context:
space:
mode:
Diffstat (limited to 'sysdeps/x86_64/multiarch/rawmemchr-sse2.S')
-rw-r--r--sysdeps/x86_64/multiarch/rawmemchr-sse2.S198
1 files changed, 188 insertions, 10 deletions
diff --git a/sysdeps/x86_64/multiarch/rawmemchr-sse2.S b/sysdeps/x86_64/multiarch/rawmemchr-sse2.S
index 3841c14c34..e2c2e20d85 100644
--- a/sysdeps/x86_64/multiarch/rawmemchr-sse2.S
+++ b/sysdeps/x86_64/multiarch/rawmemchr-sse2.S
@@ -16,14 +16,192 @@
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
-/* Define multiple versions only for the definition in libc. */
-#if IS_IN (libc)
-# define __rawmemchr __rawmemchr_sse2
-
-# undef weak_alias
-# define weak_alias(__rawmemchr, rawmemchr)
-# undef libc_hidden_def
-# define libc_hidden_def(__rawmemchr)
-#endif
+#include <isa-level.h>
+#include <sysdep.h>
+
+/* MINIMUM_X86_ISA_LEVEL <= 2 because there is no V2 implementation
+   so we need this to build for ISA V2 builds. */
+#if ISA_SHOULD_BUILD (2)
+
+# ifndef RAWMEMCHR
+#  define RAWMEMCHR	__rawmemchr_sse2
+# endif
+
+	.text
+ENTRY (RAWMEMCHR)
+	movd	%rsi, %xmm1
+	mov	%rdi, %rcx
+
+	punpcklbw %xmm1, %xmm1
+	punpcklbw %xmm1, %xmm1
+
+	and	$63, %rcx
+	pshufd	$0, %xmm1, %xmm1
+
+	cmp	$48, %rcx
+	ja	L(crosscache)
+
+	movdqu	(%rdi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+
+	jnz	L(matches)
+	add	$16, %rdi
+	and	$-16, %rdi
+	jmp	L(loop_prolog)
+
+	.p2align 4
+L(crosscache):
+	and	$15, %rcx
+	and	$-16, %rdi
+	movdqa	(%rdi), %xmm0
+
+	pcmpeqb	%xmm1, %xmm0
+/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+/* Remove the leading bytes.  */
+	sar	%cl, %eax
+	test	%eax, %eax
+	je	L(unaligned_no_match)
+/* Check which byte is a match.  */
+	bsf	%eax, %eax
+
+	add	%rdi, %rax
+	add	%rcx, %rax
+	ret
+
+	.p2align 4
+L(unaligned_no_match):
+	add	$16, %rdi
+
+	.p2align 4
+L(loop_prolog):
+	movdqa	(%rdi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+
+	movdqa	16(%rdi), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches16)
+
+	movdqa	32(%rdi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	movdqa	48(%rdi), %xmm4
+	pcmpeqb	%xmm1, %xmm4
+	add	$64, %rdi
+	pmovmskb %xmm4, %eax
+	test	%eax, %eax
+	jnz	L(matches0)
+
+	test	$0x3f, %rdi
+	jz	L(align64_loop)
+
+	movdqa	(%rdi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+
+	movdqa	16(%rdi), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches16)
 
-#include "../rawmemchr.S"
+	movdqa	32(%rdi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	movdqa	48(%rdi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+
+	add	$64, %rdi
+	test	%eax, %eax
+	jnz	L(matches0)
+
+	and	$-64, %rdi
+
+	.p2align 4
+L(align64_loop):
+	movdqa	(%rdi), %xmm0
+	movdqa	16(%rdi), %xmm2
+	movdqa	32(%rdi), %xmm3
+	movdqa	48(%rdi), %xmm4
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm1, %xmm2
+	pcmpeqb	%xmm1, %xmm3
+	pcmpeqb	%xmm1, %xmm4
+
+	pmaxub	%xmm0, %xmm3
+	pmaxub	%xmm2, %xmm4
+	pmaxub	%xmm3, %xmm4
+	pmovmskb %xmm4, %eax
+
+	add	$64, %rdi
+
+	test	%eax, %eax
+	jz	L(align64_loop)
+
+	sub	$64, %rdi
+
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches16)
+
+	movdqa	32(%rdi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+
+	pcmpeqb	48(%rdi), %xmm1
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	pmovmskb %xmm1, %eax
+	bsf	%eax, %eax
+	lea	48(%rdi, %rax), %rax
+	ret
+
+	.p2align 4
+L(matches0):
+	bsf	%eax, %eax
+	lea	-16(%rax, %rdi), %rax
+	ret
+
+	.p2align 4
+L(matches):
+	bsf	%eax, %eax
+	add	%rdi, %rax
+	ret
+
+	.p2align 4
+L(matches16):
+	bsf	%eax, %eax
+	lea	16(%rax, %rdi), %rax
+	ret
+
+	.p2align 4
+L(matches32):
+	bsf	%eax, %eax
+	lea	32(%rax, %rdi), %rax
+	ret
+
+END (RAWMEMCHR)
+#endif