Assembly coded android_memset16 and android_memset32
Change-Id: Ife2dd406e1dcb962e5e97788c515ac96f5c52e44
diff --git a/libcutils/arch-x86/sse2-memset32-atom.S b/libcutils/arch-x86/sse2-memset32-atom.S
new file mode 100644
index 0000000..4a52484
--- /dev/null
+++ b/libcutils/arch-x86/sse2-memset32-atom.S
@@ -0,0 +1,513 @@
+/*
+ * Copyright (C) 2010 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.
+ */
+/*
+ * Contributed by: Intel Corporation
+ */
+
+#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 cfi_rel_offset
+# define cfi_rel_offset(reg, off) .cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg) .cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off) .cfi_adjust_cfa_offset off
+#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 CFI_PUSH(REG) \
+ cfi_adjust_cfa_offset (4); \
+ cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG) \
+ cfi_adjust_cfa_offset (-4); \
+ cfi_restore (REG)
+
+#define PUSH(REG) pushl REG; CFI_PUSH (REG)
+#define POP(REG) popl REG; CFI_POP (REG)
+
+#ifdef USE_AS_BZERO32
+# define DEST PARMS
+# define LEN DEST+4
+#else
+# define DEST PARMS
+# define DWDS DEST+4
+# define LEN DWDS+4
+#endif
+
+#ifdef USE_AS_WMEMSET32
+# define SETRTNVAL movl DEST(%esp), %eax
+#else
+# define SETRTNVAL
+#endif
+
+#ifdef SHARED
+# define ENTRANCE PUSH (%ebx);
+# define RETURN_END POP (%ebx); ret
+# define RETURN RETURN_END; CFI_PUSH (%ebx)
+# define PARMS 8 /* Preserve EBX. */
+# define JMPTBL(I, B) I - B
+
+/* Load an entry in a jump table into EBX and branch to it. TABLE is a
+ jump table with relative offsets. */
+# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
+ /* We first load PC into EBX. */ \
+ call __i686.get_pc_thunk.bx; \
+ /* Get the address of the jump table. */ \
+ add $(TABLE - .), %ebx; \
+ /* Get the entry and convert the relative offset to the \
+ absolute address. */ \
+ add (%ebx,%ecx,4), %ebx; \
+ /* We loaded the jump table and adjuested EDX. Go. */ \
+ jmp *%ebx
+
+ .section .gnu.linkonce.t.__i686.get_pc_thunk.bx,"ax",@progbits
+ .globl __i686.get_pc_thunk.bx
+ .hidden __i686.get_pc_thunk.bx
+ ALIGN (4)
+ .type __i686.get_pc_thunk.bx,@function
+__i686.get_pc_thunk.bx:
+ movl (%esp), %ebx
+ ret
+#else
+# define ENTRANCE
+# define RETURN_END ret
+# define RETURN RETURN_END
+# define PARMS 4
+# define JMPTBL(I, B) I
+
+/* Branch to an entry in a jump table. TABLE is a jump table with
+ absolute offsets. */
+# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
+ jmp *TABLE(,%ecx,4)
+#endif
+
+ .section .text.sse2,"ax",@progbits
+ ALIGN (4)
+ENTRY (sse2_memset32_atom)
+ ENTRANCE
+
+ movl LEN(%esp), %ecx
+#ifdef USE_AS_ANDROID
+ shr $2, %ecx
+#endif
+#ifdef USE_AS_BZERO32
+ xor %eax, %eax
+#else
+ mov DWDS(%esp), %eax
+ mov %eax, %edx
+#endif
+ movl DEST(%esp), %edx
+ cmp $16, %ecx
+ jae L(16dbwordsormore)
+
+L(write_less16dbwords):
+ lea (%edx, %ecx, 4), %edx
+ BRANCH_TO_JMPTBL_ENTRY (L(table_less16dbwords))
+
+ .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 %eax, -60(%edx)
+L(write_14dbwords):
+ movl %eax, -56(%edx)
+L(write_13dbwords):
+ movl %eax, -52(%edx)
+L(write_12dbwords):
+ movl %eax, -48(%edx)
+L(write_11dbwords):
+ movl %eax, -44(%edx)
+L(write_10dbwords):
+ movl %eax, -40(%edx)
+L(write_9dbwords):
+ movl %eax, -36(%edx)
+L(write_8dbwords):
+ movl %eax, -32(%edx)
+L(write_7dbwords):
+ movl %eax, -28(%edx)
+L(write_6dbwords):
+ movl %eax, -24(%edx)
+L(write_5dbwords):
+ movl %eax, -20(%edx)
+L(write_4dbwords):
+ movl %eax, -16(%edx)
+L(write_3dbwords):
+ movl %eax, -12(%edx)
+L(write_2dbwords):
+ movl %eax, -8(%edx)
+L(write_1dbwords):
+ movl %eax, -4(%edx)
+L(write_0dbwords):
+ SETRTNVAL
+ RETURN
+
+ ALIGN (4)
+L(16dbwordsormore):
+ test $3, %edx
+ jz L(aligned4bytes)
+ mov %eax, (%edx)
+ mov %eax, -4(%edx, %ecx, 4)
+ sub $1, %ecx
+ rol $24, %eax
+ add $1, %edx
+ test $3, %edx
+ jz L(aligned4bytes)
+ ror $8, %eax
+ add $1, %edx
+ test $3, %edx
+ jz L(aligned4bytes)
+ ror $8, %eax
+ add $1, %edx
+L(aligned4bytes):
+ shl $2, %ecx
+
+#ifdef USE_AS_BZERO32
+ pxor %xmm0, %xmm0
+#else
+ movd %eax, %xmm0
+ pshufd $0, %xmm0, %xmm0
+#endif
+ testl $0xf, %edx
+ jz L(aligned_16)
+/* ECX > 32 and EDX is not 16 byte aligned. */
+L(not_aligned_16):
+ movdqu %xmm0, (%edx)
+ movl %edx, %eax
+ and $-16, %edx
+ add $16, %edx
+ sub %edx, %eax
+ add %eax, %ecx
+ movd %xmm0, %eax
+ ALIGN (4)
+L(aligned_16):
+ cmp $128, %ecx
+ jae L(128bytesormore)
+
+L(aligned_16_less128bytes):
+ add %ecx, %edx
+ shr $2, %ecx
+ BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
+
+ ALIGN (4)
+L(128bytesormore):
+#ifdef SHARED_CACHE_SIZE
+ PUSH (%ebx)
+ mov $SHARED_CACHE_SIZE, %ebx
+#else
+# ifdef SHARED
+ call __i686.get_pc_thunk.bx
+ add $_GLOBAL_OFFSET_TABLE_, %ebx
+ mov __x86_shared_cache_size@GOTOFF(%ebx), %ebx
+# else
+ PUSH (%ebx)
+ mov __x86_shared_cache_size, %ebx
+# endif
+#endif
+ cmp %ebx, %ecx
+ jae L(128bytesormore_nt_start)
+
+#ifdef DATA_CACHE_SIZE
+ POP (%ebx)
+# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
+ cmp $DATA_CACHE_SIZE, %ecx
+#else
+# ifdef SHARED
+# define RESTORE_EBX_STATE
+ call __i686.get_pc_thunk.bx
+ add $_GLOBAL_OFFSET_TABLE_, %ebx
+ cmp __x86_data_cache_size@GOTOFF(%ebx), %ecx
+# else
+ POP (%ebx)
+# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
+ cmp __x86_data_cache_size, %ecx
+# endif
+#endif
+
+ jae L(128bytes_L2_normal)
+ subl $128, %ecx
+L(128bytesormore_normal):
+ sub $128, %ecx
+ movdqa %xmm0, (%edx)
+ movdqa %xmm0, 0x10(%edx)
+ movdqa %xmm0, 0x20(%edx)
+ movdqa %xmm0, 0x30(%edx)
+ movdqa %xmm0, 0x40(%edx)
+ movdqa %xmm0, 0x50(%edx)
+ movdqa %xmm0, 0x60(%edx)
+ movdqa %xmm0, 0x70(%edx)
+ lea 128(%edx), %edx
+ jb L(128bytesless_normal)
+
+
+ sub $128, %ecx
+ movdqa %xmm0, (%edx)
+ movdqa %xmm0, 0x10(%edx)
+ movdqa %xmm0, 0x20(%edx)
+ movdqa %xmm0, 0x30(%edx)
+ movdqa %xmm0, 0x40(%edx)
+ movdqa %xmm0, 0x50(%edx)
+ movdqa %xmm0, 0x60(%edx)
+ movdqa %xmm0, 0x70(%edx)
+ lea 128(%edx), %edx
+ jae L(128bytesormore_normal)
+
+L(128bytesless_normal):
+ lea 128(%ecx), %ecx
+ add %ecx, %edx
+ shr $2, %ecx
+ BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
+
+ ALIGN (4)
+L(128bytes_L2_normal):
+ prefetcht0 0x380(%edx)
+ prefetcht0 0x3c0(%edx)
+ sub $128, %ecx
+ movdqa %xmm0, (%edx)
+ movaps %xmm0, 0x10(%edx)
+ movaps %xmm0, 0x20(%edx)
+ movaps %xmm0, 0x30(%edx)
+ movaps %xmm0, 0x40(%edx)
+ movaps %xmm0, 0x50(%edx)
+ movaps %xmm0, 0x60(%edx)
+ movaps %xmm0, 0x70(%edx)
+ add $128, %edx
+ cmp $128, %ecx
+ jae L(128bytes_L2_normal)
+
+L(128bytesless_L2_normal):
+ add %ecx, %edx
+ shr $2, %ecx
+ BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
+
+ RESTORE_EBX_STATE
+L(128bytesormore_nt_start):
+ sub %ebx, %ecx
+ mov %ebx, %eax
+ and $0x7f, %eax
+ add %eax, %ecx
+ movd %xmm0, %eax
+ ALIGN (4)
+L(128bytesormore_shared_cache_loop):
+ prefetcht0 0x3c0(%edx)
+ prefetcht0 0x380(%edx)
+ sub $0x80, %ebx
+ movdqa %xmm0, (%edx)
+ movdqa %xmm0, 0x10(%edx)
+ movdqa %xmm0, 0x20(%edx)
+ movdqa %xmm0, 0x30(%edx)
+ movdqa %xmm0, 0x40(%edx)
+ movdqa %xmm0, 0x50(%edx)
+ movdqa %xmm0, 0x60(%edx)
+ movdqa %xmm0, 0x70(%edx)
+ add $0x80, %edx
+ cmp $0x80, %ebx
+ jae L(128bytesormore_shared_cache_loop)
+ cmp $0x80, %ecx
+ jb L(shared_cache_loop_end)
+
+ ALIGN (4)
+L(128bytesormore_nt):
+ sub $0x80, %ecx
+ movntdq %xmm0, (%edx)
+ movntdq %xmm0, 0x10(%edx)
+ movntdq %xmm0, 0x20(%edx)
+ movntdq %xmm0, 0x30(%edx)
+ movntdq %xmm0, 0x40(%edx)
+ movntdq %xmm0, 0x50(%edx)
+ movntdq %xmm0, 0x60(%edx)
+ movntdq %xmm0, 0x70(%edx)
+ add $0x80, %edx
+ cmp $0x80, %ecx
+ jae L(128bytesormore_nt)
+ sfence
+L(shared_cache_loop_end):
+#if defined DATA_CACHE_SIZE || !defined SHARED
+ POP (%ebx)
+#endif
+ add %ecx, %edx
+ shr $2, %ecx
+ BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
+
+ .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(%edx)
+L(aligned_16_96bytes):
+ movdqa %xmm0, -96(%edx)
+L(aligned_16_80bytes):
+ movdqa %xmm0, -80(%edx)
+L(aligned_16_64bytes):
+ movdqa %xmm0, -64(%edx)
+L(aligned_16_48bytes):
+ movdqa %xmm0, -48(%edx)
+L(aligned_16_32bytes):
+ movdqa %xmm0, -32(%edx)
+L(aligned_16_16bytes):
+ movdqa %xmm0, -16(%edx)
+L(aligned_16_0bytes):
+ SETRTNVAL
+ RETURN
+
+ ALIGN (4)
+L(aligned_16_116bytes):
+ movdqa %xmm0, -116(%edx)
+L(aligned_16_100bytes):
+ movdqa %xmm0, -100(%edx)
+L(aligned_16_84bytes):
+ movdqa %xmm0, -84(%edx)
+L(aligned_16_68bytes):
+ movdqa %xmm0, -68(%edx)
+L(aligned_16_52bytes):
+ movdqa %xmm0, -52(%edx)
+L(aligned_16_36bytes):
+ movdqa %xmm0, -36(%edx)
+L(aligned_16_20bytes):
+ movdqa %xmm0, -20(%edx)
+L(aligned_16_4bytes):
+ movl %eax, -4(%edx)
+ SETRTNVAL
+ RETURN
+
+ ALIGN (4)
+L(aligned_16_120bytes):
+ movdqa %xmm0, -120(%edx)
+L(aligned_16_104bytes):
+ movdqa %xmm0, -104(%edx)
+L(aligned_16_88bytes):
+ movdqa %xmm0, -88(%edx)
+L(aligned_16_72bytes):
+ movdqa %xmm0, -72(%edx)
+L(aligned_16_56bytes):
+ movdqa %xmm0, -56(%edx)
+L(aligned_16_40bytes):
+ movdqa %xmm0, -40(%edx)
+L(aligned_16_24bytes):
+ movdqa %xmm0, -24(%edx)
+L(aligned_16_8bytes):
+ movq %xmm0, -8(%edx)
+ SETRTNVAL
+ RETURN
+
+ ALIGN (4)
+L(aligned_16_124bytes):
+ movdqa %xmm0, -124(%edx)
+L(aligned_16_108bytes):
+ movdqa %xmm0, -108(%edx)
+L(aligned_16_92bytes):
+ movdqa %xmm0, -92(%edx)
+L(aligned_16_76bytes):
+ movdqa %xmm0, -76(%edx)
+L(aligned_16_60bytes):
+ movdqa %xmm0, -60(%edx)
+L(aligned_16_44bytes):
+ movdqa %xmm0, -44(%edx)
+L(aligned_16_28bytes):
+ movdqa %xmm0, -28(%edx)
+L(aligned_16_12bytes):
+ movq %xmm0, -12(%edx)
+ movl %eax, -4(%edx)
+ SETRTNVAL
+ RETURN
+
+END (sse2_memset32_atom)