diff options
Diffstat (limited to 'sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S')
| -rw-r--r-- | sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S | 76 |
1 files changed, 76 insertions, 0 deletions
diff --git a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S index 028c6d3d74..a3535ad500 100644 --- a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S +++ b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S @@ -25,6 +25,14 @@ # define STRCAT __strcat_sse2_unaligned # endif +# ifdef __CHKP__ +# define RETURN \ + bndcu -1(%rdi, %rax), %bnd0; \ + ret +# else +# define RETURN ret +# endif + # define USE_AS_STRCAT .text @@ -37,6 +45,10 @@ ENTRY (STRCAT) /* Inline corresponding strlen file, temporary until new strcpy implementation gets merged. */ +# ifdef __CHKP__ + bndcl (%rdi), %bnd0 + bndcu (%rdi), %bnd0 +# endif xor %rax, %rax mov %edi, %ecx and $0x3f, %ecx @@ -67,84 +79,132 @@ L(align16_start): pxor %xmm1, %xmm1 pxor %xmm2, %xmm2 pxor %xmm3, %xmm3 +# ifdef __CHKP__ + bndcu 16(%rax), %bnd0 +# endif pcmpeqb 16(%rax), %xmm0 pmovmskb %xmm0, %edx test %edx, %edx jnz L(exit16) +# ifdef __CHKP__ + bndcu 32(%rax), %bnd0 +# endif pcmpeqb 32(%rax), %xmm1 pmovmskb %xmm1, %edx test %edx, %edx jnz L(exit32) +# ifdef __CHKP__ + bndcu 48(%rax), %bnd0 +# endif pcmpeqb 48(%rax), %xmm2 pmovmskb %xmm2, %edx test %edx, %edx jnz L(exit48) +# ifdef __CHKP__ + bndcu 64(%rax), %bnd0 +# endif pcmpeqb 64(%rax), %xmm3 pmovmskb %xmm3, %edx test %edx, %edx jnz L(exit64) +# ifdef __CHKP__ + bndcu 80(%rax), %bnd0 +# endif pcmpeqb 80(%rax), %xmm0 add $64, %rax pmovmskb %xmm0, %edx test %edx, %edx jnz L(exit16) +# ifdef __CHKP__ + bndcu 32(%rax), %bnd0 +# endif pcmpeqb 32(%rax), %xmm1 pmovmskb %xmm1, %edx test %edx, %edx jnz L(exit32) +# ifdef __CHKP__ + bndcu 48(%rax), %bnd0 +# endif pcmpeqb 48(%rax), %xmm2 pmovmskb %xmm2, %edx test %edx, %edx jnz L(exit48) +# ifdef __CHKP__ + bndcu 64(%rax), %bnd0 +# endif pcmpeqb 64(%rax), %xmm3 pmovmskb %xmm3, %edx test %edx, %edx jnz L(exit64) +# ifdef __CHKP__ + bndcu 80(%rax), %bnd0 +# endif pcmpeqb 80(%rax), %xmm0 add $64, %rax pmovmskb %xmm0, %edx test %edx, %edx jnz L(exit16) +# ifdef __CHKP__ + bndcu 32(%rax), %bnd0 +# endif pcmpeqb 32(%rax), %xmm1 pmovmskb %xmm1, %edx test %edx, %edx jnz L(exit32) +# ifdef __CHKP__ + bndcu 48(%rax), %bnd0 +# endif pcmpeqb 48(%rax), %xmm2 pmovmskb %xmm2, %edx test %edx, %edx jnz L(exit48) +# ifdef __CHKP__ + bndcu 64(%rax), %bnd0 +# endif pcmpeqb 64(%rax), %xmm3 pmovmskb %xmm3, %edx test %edx, %edx jnz L(exit64) +# ifdef __CHKP__ + bndcu 80(%rax), %bnd0 +# endif pcmpeqb 80(%rax), %xmm0 add $64, %rax pmovmskb %xmm0, %edx test %edx, %edx jnz L(exit16) +# ifdef __CHKP__ + bndcu 32(%rax), %bnd0 +# endif pcmpeqb 32(%rax), %xmm1 pmovmskb %xmm1, %edx test %edx, %edx jnz L(exit32) +# ifdef __CHKP__ + bndcu 48(%rax), %bnd0 +# endif pcmpeqb 48(%rax), %xmm2 pmovmskb %xmm2, %edx test %edx, %edx jnz L(exit48) +# ifdef __CHKP__ + bndcu 64(%rax), %bnd0 +# endif pcmpeqb 64(%rax), %xmm3 pmovmskb %xmm3, %edx test %edx, %edx @@ -153,6 +213,9 @@ L(align16_start): test $0x3f, %rax jz L(align64_loop) +# ifdef __CHKP__ + bndcu 80(%rax), %bnd0 +# endif pcmpeqb 80(%rax), %xmm0 add $80, %rax pmovmskb %xmm0, %edx @@ -162,6 +225,9 @@ L(align16_start): test $0x3f, %rax jz L(align64_loop) +# ifdef __CHKP__ + bndcu 16(%rax), %bnd0 +# endif pcmpeqb 16(%rax), %xmm1 add $16, %rax pmovmskb %xmm1, %edx @@ -171,6 +237,9 @@ L(align16_start): test $0x3f, %rax jz L(align64_loop) +# ifdef __CHKP__ + bndcu 16(%rax), %bnd0 +# endif pcmpeqb 16(%rax), %xmm2 add $16, %rax pmovmskb %xmm2, %edx @@ -180,6 +249,9 @@ L(align16_start): test $0x3f, %rax jz L(align64_loop) +# ifdef __CHKP__ + bndcu 16(%rax), %bnd0 +# endif pcmpeqb 16(%rax), %xmm3 add $16, %rax pmovmskb %xmm3, %edx @@ -187,8 +259,12 @@ L(align16_start): jnz L(exit) add $16, %rax + .p2align 4 L(align64_loop): +# ifdef __CHKP__ + bndcu (%rax), %bnd0 +# endif movaps (%rax), %xmm4 pminub 16(%rax), %xmm4 movaps 32(%rax), %xmm5 |