diff options
Diffstat (limited to 'sysdeps/i386/i686/multiarch/memcpy-ssse3-rep.S')
-rw-r--r-- | sysdeps/i386/i686/multiarch/memcpy-ssse3-rep.S | 178 |
1 files changed, 106 insertions, 72 deletions
diff --git a/sysdeps/i386/i686/multiarch/memcpy-ssse3-rep.S b/sysdeps/i386/i686/multiarch/memcpy-ssse3-rep.S index 0547b56d7c..48a109ccd6 100644 --- a/sysdeps/i386/i686/multiarch/memcpy-ssse3-rep.S +++ b/sysdeps/i386/i686/multiarch/memcpy-ssse3-rep.S @@ -127,10 +127,8 @@ ENTRY (MEMCPY) cmp %eax, %edx jb L(copy_forward) je L(fwd_write_0bytes) - cmp $32, %ecx - jae L(memmove_bwd) - jmp L(bk_write_less32bytes_2) -L(memmove_bwd): + cmp $48, %ecx + jb L(bk_write_less48bytes) add %ecx, %eax cmp %eax, %edx movl SRC(%esp), %eax @@ -162,6 +160,7 @@ L(48bytesormore): movl %edx, %edi and $-16, %edx PUSH (%esi) + cfi_remember_state add $16, %edx movl %edi, %esi sub %edx, %edi @@ -234,6 +233,8 @@ L(shl_0_end): POP (%edi) BRANCH_TO_JMPTBL_ENTRY (L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state L(shl_0_gobble): #ifdef DATA_CACHE_SIZE_HALF @@ -252,7 +253,7 @@ L(shl_0_gobble): sub %esi, %edi cmp %edi, %ecx jae L(shl_0_gobble_mem_start) - lea -128(%ecx), %ecx + sub $128, %ecx ALIGN (4) L(shl_0_gobble_cache_loop): movdqa (%eax), %xmm0 @@ -276,9 +277,9 @@ L(shl_0_gobble_cache_loop): lea 0x80(%edx), %edx jae L(shl_0_gobble_cache_loop) - cmp $-0x40, %ecx - lea 0x80(%ecx), %ecx - jl L(shl_0_cache_less_64bytes) + add $0x80, %ecx + cmp $0x40, %ecx + jb L(shl_0_cache_less_64bytes) movdqa (%eax), %xmm0 sub $0x40, %ecx @@ -319,12 +320,13 @@ L(shl_0_cache_less_16bytes): POP (%edi) BRANCH_TO_JMPTBL_ENTRY (L(table_48bytes_fwd), %ecx, 4) - + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_0_gobble_mem_start): cmp %al, %dl je L(copy_page_by_rep) - lea -128(%ecx), %ecx + sub $128, %ecx L(shl_0_gobble_mem_loop): prefetchnta 0x1c0(%eax) prefetchnta 0x280(%eax) @@ -352,9 +354,9 @@ L(shl_0_gobble_mem_loop): lea 0x80(%edx), %edx jae L(shl_0_gobble_mem_loop) - cmp $-0x40, %ecx - lea 0x80(%ecx), %ecx - jl L(shl_0_mem_less_64bytes) + add $0x80, %ecx + cmp $0x40, %ecx + jb L(shl_0_mem_less_64bytes) movdqa (%eax), %xmm0 sub $0x40, %ecx @@ -395,14 +397,15 @@ L(shl_0_mem_less_16bytes): POP (%edi) BRANCH_TO_JMPTBL_ENTRY (L(table_48bytes_fwd), %ecx, 4) - + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_1): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -1(%eax), %eax + sub $1, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_1_loop): @@ -432,20 +435,22 @@ L(shl_1_loop): jae L(shl_1_loop) L(shl_1_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 1(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_2): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -2(%eax), %eax + sub $2, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_2_loop): @@ -475,20 +480,22 @@ L(shl_2_loop): jae L(shl_2_loop) L(shl_2_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 2(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_3): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -3(%eax), %eax + sub $3, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_3_loop): @@ -518,20 +525,22 @@ L(shl_3_loop): jae L(shl_3_loop) L(shl_3_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 3(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_4): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -4(%eax), %eax + sub $4, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_4_loop): @@ -561,20 +570,22 @@ L(shl_4_loop): jae L(shl_4_loop) L(shl_4_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 4(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_5): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -5(%eax), %eax + sub $5, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_5_loop): @@ -604,21 +615,22 @@ L(shl_5_loop): jae L(shl_5_loop) L(shl_5_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 5(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) - + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_6): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -6(%eax), %eax + sub $6, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_6_loop): @@ -648,20 +660,22 @@ L(shl_6_loop): jae L(shl_6_loop) L(shl_6_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 6(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_7): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -7(%eax), %eax + sub $7, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_7_loop): @@ -691,20 +705,22 @@ L(shl_7_loop): jae L(shl_7_loop) L(shl_7_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 7(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_8): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -8(%eax), %eax + sub $8, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_8_loop): @@ -734,20 +750,22 @@ L(shl_8_loop): jae L(shl_8_loop) L(shl_8_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 8(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_9): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -9(%eax), %eax + sub $9, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_9_loop): @@ -777,20 +795,22 @@ L(shl_9_loop): jae L(shl_9_loop) L(shl_9_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 9(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_10): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -10(%eax), %eax + sub $10, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_10_loop): @@ -820,20 +840,22 @@ L(shl_10_loop): jae L(shl_10_loop) L(shl_10_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 10(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_11): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -11(%eax), %eax + sub $11, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_11_loop): @@ -863,20 +885,22 @@ L(shl_11_loop): jae L(shl_11_loop) L(shl_11_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 11(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_12): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -12(%eax), %eax + sub $12, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_12_loop): @@ -906,20 +930,22 @@ L(shl_12_loop): jae L(shl_12_loop) L(shl_12_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 12(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_13): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -13(%eax), %eax + sub $13, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_13_loop): @@ -949,20 +975,22 @@ L(shl_13_loop): jae L(shl_13_loop) L(shl_13_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 13(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_14): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -14(%eax), %eax + sub $14, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_14_loop): @@ -992,21 +1020,22 @@ L(shl_14_loop): jae L(shl_14_loop) L(shl_14_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 14(%edi, %eax), %eax POP (%edi) BRANCH_TO_JMPTBL_ENTRY_TAIL(L(table_48bytes_fwd), %ecx, 4) - + cfi_restore_state + cfi_remember_state ALIGN (4) L(shl_15): BRANCH_TO_JMPTBL_ENTRY_VALUE(L(table_48bytes_fwd)) - lea -15(%eax), %eax + sub $15, %eax movaps (%eax), %xmm1 xor %edi, %edi - lea -32(%ecx), %ecx + sub $32, %ecx movdqu %xmm0, (%esi) POP (%esi) L(shl_15_loop): @@ -1036,7 +1065,7 @@ L(shl_15_loop): jae L(shl_15_loop) L(shl_15_end): - lea 32(%ecx), %ecx + add $32, %ecx add %ecx, %edi add %edi, %edx lea 15(%edi, %eax), %eax @@ -1240,21 +1269,23 @@ L(fwd_write_3bytes): movl DEST(%esp), %eax # endif #endif - RETURN + RETURN_END + cfi_restore_state + cfi_remember_state ALIGN (4) L(large_page): movdqu (%eax), %xmm1 - lea 16(%eax), %eax movdqu %xmm0, (%esi) movntdq %xmm1, (%edx) - lea 16(%edx), %edx - lea -16(%ecx), %ecx + add $0x10, %eax + add $0x10, %edx + sub $0x10, %ecx cmp %al, %dl je L(copy_page_by_rep) L(large_page_loop_init): POP (%esi) - lea -0x80(%ecx), %ecx + sub $0x80, %ecx POP (%edi) L(large_page_loop): prefetchnta 0x1c0(%eax) @@ -1280,9 +1311,9 @@ L(large_page_loop): movntdq %xmm7, 0x70(%edx) lea 0x80(%edx), %edx jae L(large_page_loop) - cmp $-0x40, %ecx - lea 0x80(%ecx), %ecx - jl L(large_page_less_64bytes) + add $0x80, %ecx + cmp $0x40, %ecx + jb L(large_page_less_64bytes) movdqu (%eax), %xmm0 movdqu 0x10(%eax), %xmm1 @@ -1312,6 +1343,8 @@ L(large_page_less_32bytes): sfence BRANCH_TO_JMPTBL_ENTRY (L(table_48bytes_fwd), %ecx, 4) + cfi_restore_state + cfi_remember_state ALIGN (4) L(copy_page_by_rep): mov %eax, %esi @@ -1658,8 +1691,8 @@ L(table_48_bytes_bwd): L(copy_backward): PUSH (%esi) movl %eax, %esi - lea (%ecx,%edx,1),%edx - lea (%ecx,%esi,1),%esi + add %ecx, %edx + add %ecx, %esi testl $0x3, %edx jnz L(bk_align) @@ -1698,9 +1731,10 @@ L(bk_write_less32bytes): sub %ecx, %edx sub %ecx, %eax POP (%esi) -L(bk_write_less32bytes_2): +L(bk_write_less48bytes): BRANCH_TO_JMPTBL_ENTRY (L(table_48_bytes_bwd), %ecx, 4) + CFI_PUSH (%esi) ALIGN (4) L(bk_align): cmp $8, %ecx |