diff options
Diffstat (limited to 'libcutils/arch-x86_64/android_memset32.S')
-rw-r--r-- | libcutils/arch-x86_64/android_memset32.S | 373 |
1 files changed, 373 insertions, 0 deletions
diff --git a/libcutils/arch-x86_64/android_memset32.S b/libcutils/arch-x86_64/android_memset32.S new file mode 100644 index 0000000..1514aa2 --- /dev/null +++ b/libcutils/arch-x86_64/android_memset32.S @@ -0,0 +1,373 @@ +/* + * Copyright (C) 2014 The Android Open Source Project + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "cache.h" + +#ifndef MEMSET +# define MEMSET android_memset32 +#endif + +#ifndef L +# define L(label) .L##label +#endif + +#ifndef ALIGN +# define ALIGN(n) .p2align n +#endif + +#ifndef cfi_startproc +# define cfi_startproc .cfi_startproc +#endif + +#ifndef cfi_endproc +# define cfi_endproc .cfi_endproc +#endif + +#ifndef ENTRY +# define ENTRY(name) \ + .type name, @function; \ + .globl name; \ + .p2align 4; \ +name: \ + cfi_startproc +#endif + +#ifndef END +# define END(name) \ + cfi_endproc; \ + .size name, .-name +#endif + +#define JMPTBL(I, B) I - B + +/* Branch to an entry in a jump table. TABLE is a jump table with + relative offsets. INDEX is a register contains the index into the + jump table. SCALE is the scale of INDEX. */ +#define BRANCH_TO_JMPTBL_ENTRY(TABLE, INDEX, SCALE) \ + lea TABLE(%rip), %r11; \ + movslq (%r11, INDEX, SCALE), INDEX; \ + lea (%r11, INDEX), INDEX; \ + jmp *INDEX + + .section .text.sse2,"ax",@progbits + ALIGN (4) +ENTRY (MEMSET) // Address in rdi + shr $2, %rdx // Count in rdx + movl %esi, %ecx // Pattern in ecx + + cmp $16, %rdx + jae L(16dbwordsormore) + +L(write_less16dbwords): + lea (%rdi, %rdx, 4), %rdi + BRANCH_TO_JMPTBL_ENTRY (L(table_less16dbwords), %rdx, 4) + + .pushsection .rodata.sse2,"a",@progbits + ALIGN (2) +L(table_less16dbwords): + .int JMPTBL (L(write_0dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_1dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_2dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_3dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_4dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_5dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_6dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_7dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_8dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_9dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_10dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_11dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_12dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_13dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_14dbwords), L(table_less16dbwords)) + .int JMPTBL (L(write_15dbwords), L(table_less16dbwords)) + .popsection + + ALIGN (4) +L(write_15dbwords): + movl %ecx, -60(%rdi) +L(write_14dbwords): + movl %ecx, -56(%rdi) +L(write_13dbwords): + movl %ecx, -52(%rdi) +L(write_12dbwords): + movl %ecx, -48(%rdi) +L(write_11dbwords): + movl %ecx, -44(%rdi) +L(write_10dbwords): + movl %ecx, -40(%rdi) +L(write_9dbwords): + movl %ecx, -36(%rdi) +L(write_8dbwords): + movl %ecx, -32(%rdi) +L(write_7dbwords): + movl %ecx, -28(%rdi) +L(write_6dbwords): + movl %ecx, -24(%rdi) +L(write_5dbwords): + movl %ecx, -20(%rdi) +L(write_4dbwords): + movl %ecx, -16(%rdi) +L(write_3dbwords): + movl %ecx, -12(%rdi) +L(write_2dbwords): + movl %ecx, -8(%rdi) +L(write_1dbwords): + movl %ecx, -4(%rdi) +L(write_0dbwords): + ret + + ALIGN (4) +L(16dbwordsormore): + test $3, %edi + jz L(aligned4bytes) + mov %ecx, (%rdi) + mov %ecx, -4(%rdi, %rdx, 4) + sub $1, %rdx + rol $24, %ecx + add $1, %rdi + test $3, %edi + jz L(aligned4bytes) + ror $8, %ecx + add $1, %rdi + test $3, %edi + jz L(aligned4bytes) + ror $8, %ecx + add $1, %rdi +L(aligned4bytes): + shl $2, %rdx + + /* Fill xmm0 with the pattern. */ + movd %ecx, %xmm0 + pshufd $0, %xmm0, %xmm0 + + testl $0xf, %edi + jz L(aligned_16) +/* RDX > 32 and RDI is not 16 byte aligned. */ + movdqu %xmm0, (%rdi) + mov %rdi, %rsi + and $-16, %rdi + add $16, %rdi + sub %rdi, %rsi + add %rsi, %rdx + + ALIGN (4) +L(aligned_16): + cmp $128, %rdx + jge L(128bytesormore) + +L(aligned_16_less128bytes): + add %rdx, %rdi + shr $2, %rdx + BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes), %rdx, 4) + + ALIGN (4) +L(128bytesormore): + cmp $SHARED_CACHE_SIZE, %rdx + jg L(128bytesormore_nt) + +L(128bytesormore_normal): + sub $128, %rdx + movdqa %xmm0, (%rdi) + movdqa %xmm0, 0x10(%rdi) + movdqa %xmm0, 0x20(%rdi) + movdqa %xmm0, 0x30(%rdi) + movdqa %xmm0, 0x40(%rdi) + movdqa %xmm0, 0x50(%rdi) + movdqa %xmm0, 0x60(%rdi) + movdqa %xmm0, 0x70(%rdi) + lea 128(%rdi), %rdi + cmp $128, %rdx + jl L(128bytesless_normal) + + sub $128, %rdx + movdqa %xmm0, (%rdi) + movdqa %xmm0, 0x10(%rdi) + movdqa %xmm0, 0x20(%rdi) + movdqa %xmm0, 0x30(%rdi) + movdqa %xmm0, 0x40(%rdi) + movdqa %xmm0, 0x50(%rdi) + movdqa %xmm0, 0x60(%rdi) + movdqa %xmm0, 0x70(%rdi) + lea 128(%rdi), %rdi + cmp $128, %rdx + jl L(128bytesless_normal) + + sub $128, %rdx + movdqa %xmm0, (%rdi) + movdqa %xmm0, 0x10(%rdi) + movdqa %xmm0, 0x20(%rdi) + movdqa %xmm0, 0x30(%rdi) + movdqa %xmm0, 0x40(%rdi) + movdqa %xmm0, 0x50(%rdi) + movdqa %xmm0, 0x60(%rdi) + movdqa %xmm0, 0x70(%rdi) + lea 128(%rdi), %rdi + cmp $128, %rdx + jl L(128bytesless_normal) + + sub $128, %rdx + movdqa %xmm0, (%rdi) + movdqa %xmm0, 0x10(%rdi) + movdqa %xmm0, 0x20(%rdi) + movdqa %xmm0, 0x30(%rdi) + movdqa %xmm0, 0x40(%rdi) + movdqa %xmm0, 0x50(%rdi) + movdqa %xmm0, 0x60(%rdi) + movdqa %xmm0, 0x70(%rdi) + lea 128(%rdi), %rdi + cmp $128, %rdx + jge L(128bytesormore_normal) + +L(128bytesless_normal): + add %rdx, %rdi + shr $2, %rdx + BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes), %rdx, 4) + + ALIGN (4) +L(128bytesormore_nt): + sub $128, %rdx + movntdq %xmm0, (%rdi) + movntdq %xmm0, 0x10(%rdi) + movntdq %xmm0, 0x20(%rdi) + movntdq %xmm0, 0x30(%rdi) + movntdq %xmm0, 0x40(%rdi) + movntdq %xmm0, 0x50(%rdi) + movntdq %xmm0, 0x60(%rdi) + movntdq %xmm0, 0x70(%rdi) + lea 128(%rdi), %rdi + cmp $128, %rdx + jge L(128bytesormore_nt) + + sfence + add %rdx, %rdi + shr $2, %rdx + BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes), %rdx, 4) + + .pushsection .rodata.sse2,"a",@progbits + ALIGN (2) +L(table_16_128bytes): + .int JMPTBL (L(aligned_16_0bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_4bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_8bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_12bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_16bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_20bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_24bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_28bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_32bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_36bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_40bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_44bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_48bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_52bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_56bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_60bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_64bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_68bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_72bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_76bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_80bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_84bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_88bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_92bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_96bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_100bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_104bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_108bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_112bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_116bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_120bytes), L(table_16_128bytes)) + .int JMPTBL (L(aligned_16_124bytes), L(table_16_128bytes)) + .popsection + + ALIGN (4) +L(aligned_16_112bytes): + movdqa %xmm0, -112(%rdi) +L(aligned_16_96bytes): + movdqa %xmm0, -96(%rdi) +L(aligned_16_80bytes): + movdqa %xmm0, -80(%rdi) +L(aligned_16_64bytes): + movdqa %xmm0, -64(%rdi) +L(aligned_16_48bytes): + movdqa %xmm0, -48(%rdi) +L(aligned_16_32bytes): + movdqa %xmm0, -32(%rdi) +L(aligned_16_16bytes): + movdqa %xmm0, -16(%rdi) +L(aligned_16_0bytes): + ret + + ALIGN (4) +L(aligned_16_116bytes): + movdqa %xmm0, -116(%rdi) +L(aligned_16_100bytes): + movdqa %xmm0, -100(%rdi) +L(aligned_16_84bytes): + movdqa %xmm0, -84(%rdi) +L(aligned_16_68bytes): + movdqa %xmm0, -68(%rdi) +L(aligned_16_52bytes): + movdqa %xmm0, -52(%rdi) +L(aligned_16_36bytes): + movdqa %xmm0, -36(%rdi) +L(aligned_16_20bytes): + movdqa %xmm0, -20(%rdi) +L(aligned_16_4bytes): + movl %ecx, -4(%rdi) + ret + + ALIGN (4) +L(aligned_16_120bytes): + movdqa %xmm0, -120(%rdi) +L(aligned_16_104bytes): + movdqa %xmm0, -104(%rdi) +L(aligned_16_88bytes): + movdqa %xmm0, -88(%rdi) +L(aligned_16_72bytes): + movdqa %xmm0, -72(%rdi) +L(aligned_16_56bytes): + movdqa %xmm0, -56(%rdi) +L(aligned_16_40bytes): + movdqa %xmm0, -40(%rdi) +L(aligned_16_24bytes): + movdqa %xmm0, -24(%rdi) +L(aligned_16_8bytes): + movq %xmm0, -8(%rdi) + ret + + ALIGN (4) +L(aligned_16_124bytes): + movdqa %xmm0, -124(%rdi) +L(aligned_16_108bytes): + movdqa %xmm0, -108(%rdi) +L(aligned_16_92bytes): + movdqa %xmm0, -92(%rdi) +L(aligned_16_76bytes): + movdqa %xmm0, -76(%rdi) +L(aligned_16_60bytes): + movdqa %xmm0, -60(%rdi) +L(aligned_16_44bytes): + movdqa %xmm0, -44(%rdi) +L(aligned_16_28bytes): + movdqa %xmm0, -28(%rdi) +L(aligned_16_12bytes): + movq %xmm0, -12(%rdi) + movl %ecx, -4(%rdi) + ret + +END (MEMSET) |