diff options
Diffstat (limited to 'sysdeps/x86_64/rawmemchr.S')
| -rw-r--r-- | sysdeps/x86_64/rawmemchr.S | 53 |
1 files changed, 46 insertions, 7 deletions
diff --git a/sysdeps/x86_64/rawmemchr.S b/sysdeps/x86_64/rawmemchr.S index f4d559155c..2f4cb25e00 100644 --- a/sysdeps/x86_64/rawmemchr.S +++ b/sysdeps/x86_64/rawmemchr.S @@ -20,11 +20,23 @@ #include <sysdep.h> +#ifdef __CHKP__ +# define RETURN \ + bndcu (%rax), %bnd0; \ + ret +#else +# define RETURN ret +#endif + .text ENTRY (rawmemchr) movd %rsi, %xmm1 mov %rdi, %rcx +#ifdef __CHKP__ + bndcl (%rdi), %bnd0 +#endif + punpcklbw %xmm1, %xmm1 punpcklbw %xmm1, %xmm1 @@ -63,7 +75,7 @@ L(crosscache): add %rdi, %rax add %rcx, %rax - ret + RETURN .p2align 4 L(unaligned_no_match): @@ -71,24 +83,36 @@ L(unaligned_no_match): .p2align 4 L(loop_prolog): +#ifdef __CHKP__ + bndcu (%rdi), %bnd0 +#endif movdqa (%rdi), %xmm0 pcmpeqb %xmm1, %xmm0 pmovmskb %xmm0, %eax test %eax, %eax jnz L(matches) +#ifdef __CHKP__ + bndcu 16(%rdi), %bnd0 +#endif movdqa 16(%rdi), %xmm2 pcmpeqb %xmm1, %xmm2 pmovmskb %xmm2, %eax test %eax, %eax jnz L(matches16) +#ifdef __CHKP__ + bndcu 32(%rdi), %bnd0 +#endif movdqa 32(%rdi), %xmm3 pcmpeqb %xmm1, %xmm3 pmovmskb %xmm3, %eax test %eax, %eax jnz L(matches32) +#ifdef __CHKP__ + bndcu 48(%rdi), %bnd0 +#endif movdqa 48(%rdi), %xmm4 pcmpeqb %xmm1, %xmm4 add $64, %rdi @@ -99,24 +123,36 @@ L(loop_prolog): test $0x3f, %rdi jz L(align64_loop) +#ifdef __CHKP__ + bndcu (%rdi), %bnd0 +#endif movdqa (%rdi), %xmm0 pcmpeqb %xmm1, %xmm0 pmovmskb %xmm0, %eax test %eax, %eax jnz L(matches) +#ifdef __CHKP__ + bndcu 16(%rdi), %bnd0 +#endif movdqa 16(%rdi), %xmm2 pcmpeqb %xmm1, %xmm2 pmovmskb %xmm2, %eax test %eax, %eax jnz L(matches16) +#ifdef __CHKP__ + bndcu 32(%rdi), %bnd0 +#endif movdqa 32(%rdi), %xmm3 pcmpeqb %xmm1, %xmm3 pmovmskb %xmm3, %eax test %eax, %eax jnz L(matches32) +#ifdef __CHKP__ + bndcu 48(%rdi), %bnd0 +#endif movdqa 48(%rdi), %xmm3 pcmpeqb %xmm1, %xmm3 pmovmskb %xmm3, %eax @@ -129,6 +165,9 @@ L(loop_prolog): .p2align 4 L(align64_loop): +#ifdef __CHKP__ + bndcu (%rdi), %bnd0 +#endif movdqa (%rdi), %xmm0 movdqa 16(%rdi), %xmm2 movdqa 32(%rdi), %xmm3 @@ -170,36 +209,36 @@ L(align64_loop): pmovmskb %xmm1, %eax bsf %eax, %eax lea 48(%rdi, %rax), %rax - ret + RETURN .p2align 4 L(matches0): bsf %eax, %eax lea -16(%rax, %rdi), %rax - ret + RETURN .p2align 4 L(matches): bsf %eax, %eax add %rdi, %rax - ret + RETURN .p2align 4 L(matches16): bsf %eax, %eax lea 16(%rax, %rdi), %rax - ret + RETURN .p2align 4 L(matches32): bsf %eax, %eax lea 32(%rax, %rdi), %rax - ret + RETURN .p2align 4 L(return_null): xor %rax, %rax - ret + RETURN END (rawmemchr) |