summaryrefslogtreecommitdiff
path: root/arch/sparc/lib/memcpy.S
diff options
context:
space:
mode:
Diffstat (limited to 'arch/sparc/lib/memcpy.S')
-rw-r--r--arch/sparc/lib/memcpy.S541
1 files changed, 541 insertions, 0 deletions
diff --git a/arch/sparc/lib/memcpy.S b/arch/sparc/lib/memcpy.S
new file mode 100644
index 000000000..4d8c49751
--- /dev/null
+++ b/arch/sparc/lib/memcpy.S
@@ -0,0 +1,541 @@
+/* memcpy.S: Sparc optimized memcpy and memmove code
+ * Hand optimized from GNU libc's memcpy and memmove
+ * Copyright (C) 1991,1996 Free Software Foundation
+ * Copyright (C) 1995 Linus Torvalds (Linus.Torvalds@helsinki.fi)
+ * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
+ * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be)
+ * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
+ */
+
+#define FUNC(x) \
+ .globl x; \
+ .type x,@function; \
+ .align 4; \
+x:
+
+/* Both these macros have to start with exactly the same insn */
+#define MOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \
+ ldd [%src + (offset) + 0x00], %t0; \
+ ldd [%src + (offset) + 0x08], %t2; \
+ ldd [%src + (offset) + 0x10], %t4; \
+ ldd [%src + (offset) + 0x18], %t6; \
+ st %t0, [%dst + (offset) + 0x00]; \
+ st %t1, [%dst + (offset) + 0x04]; \
+ st %t2, [%dst + (offset) + 0x08]; \
+ st %t3, [%dst + (offset) + 0x0c]; \
+ st %t4, [%dst + (offset) + 0x10]; \
+ st %t5, [%dst + (offset) + 0x14]; \
+ st %t6, [%dst + (offset) + 0x18]; \
+ st %t7, [%dst + (offset) + 0x1c];
+
+#define MOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \
+ ldd [%src + (offset) + 0x00], %t0; \
+ ldd [%src + (offset) + 0x08], %t2; \
+ ldd [%src + (offset) + 0x10], %t4; \
+ ldd [%src + (offset) + 0x18], %t6; \
+ std %t0, [%dst + (offset) + 0x00]; \
+ std %t2, [%dst + (offset) + 0x08]; \
+ std %t4, [%dst + (offset) + 0x10]; \
+ std %t6, [%dst + (offset) + 0x18];
+
+#define MOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \
+ ldd [%src - (offset) - 0x10], %t0; \
+ ldd [%src - (offset) - 0x08], %t2; \
+ st %t0, [%dst - (offset) - 0x10]; \
+ st %t1, [%dst - (offset) - 0x0c]; \
+ st %t2, [%dst - (offset) - 0x08]; \
+ st %t3, [%dst - (offset) - 0x04];
+
+#define MOVE_LASTALIGNCHUNK(src, dst, offset, t0, t1, t2, t3) \
+ ldd [%src - (offset) - 0x10], %t0; \
+ ldd [%src - (offset) - 0x08], %t2; \
+ std %t0, [%dst - (offset) - 0x10]; \
+ std %t2, [%dst - (offset) - 0x08];
+
+#define MOVE_SHORTCHUNK(src, dst, offset, t0, t1) \
+ ldub [%src - (offset) - 0x02], %t0; \
+ ldub [%src - (offset) - 0x01], %t1; \
+ stb %t0, [%dst - (offset) - 0x02]; \
+ stb %t1, [%dst - (offset) - 0x01];
+
+/* Both these macros have to start with exactly the same insn */
+#define RMOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \
+ ldd [%src - (offset) - 0x20], %t0; \
+ ldd [%src - (offset) - 0x18], %t2; \
+ ldd [%src - (offset) - 0x10], %t4; \
+ ldd [%src - (offset) - 0x08], %t6; \
+ st %t0, [%dst - (offset) - 0x20]; \
+ st %t1, [%dst - (offset) - 0x1c]; \
+ st %t2, [%dst - (offset) - 0x18]; \
+ st %t3, [%dst - (offset) - 0x14]; \
+ st %t4, [%dst - (offset) - 0x10]; \
+ st %t5, [%dst - (offset) - 0x0c]; \
+ st %t6, [%dst - (offset) - 0x08]; \
+ st %t7, [%dst - (offset) - 0x04];
+
+#define RMOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \
+ ldd [%src - (offset) - 0x20], %t0; \
+ ldd [%src - (offset) - 0x18], %t2; \
+ ldd [%src - (offset) - 0x10], %t4; \
+ ldd [%src - (offset) - 0x08], %t6; \
+ std %t0, [%dst - (offset) - 0x20]; \
+ std %t2, [%dst - (offset) - 0x18]; \
+ std %t4, [%dst - (offset) - 0x10]; \
+ std %t6, [%dst - (offset) - 0x08];
+
+#define RMOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \
+ ldd [%src + (offset) + 0x00], %t0; \
+ ldd [%src + (offset) + 0x08], %t2; \
+ st %t0, [%dst + (offset) + 0x00]; \
+ st %t1, [%dst + (offset) + 0x04]; \
+ st %t2, [%dst + (offset) + 0x08]; \
+ st %t3, [%dst + (offset) + 0x0c];
+
+#define RMOVE_SHORTCHUNK(src, dst, offset, t0, t1) \
+ ldub [%src + (offset) + 0x00], %t0; \
+ ldub [%src + (offset) + 0x01], %t1; \
+ stb %t0, [%dst + (offset) + 0x00]; \
+ stb %t1, [%dst + (offset) + 0x01];
+
+#define SMOVE_CHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, prev, shil, shir, offset2) \
+ ldd [%src + (offset) + 0x00], %t0; \
+ ldd [%src + (offset) + 0x08], %t2; \
+ srl %t0, shir, %t5; \
+ srl %t1, shir, %t6; \
+ sll %t0, shil, %t0; \
+ or %t5, %prev, %t5; \
+ sll %t1, shil, %prev; \
+ or %t6, %t0, %t0; \
+ srl %t2, shir, %t1; \
+ srl %t3, shir, %t6; \
+ sll %t2, shil, %t2; \
+ or %t1, %prev, %t1; \
+ std %t4, [%dst + (offset) + (offset2) - 0x04]; \
+ std %t0, [%dst + (offset) + (offset2) + 0x04]; \
+ sll %t3, shil, %prev; \
+ or %t6, %t2, %t4;
+
+#define SMOVE_ALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, prev, shil, shir, offset2) \
+ ldd [%src + (offset) + 0x00], %t0; \
+ ldd [%src + (offset) + 0x08], %t2; \
+ srl %t0, shir, %t4; \
+ srl %t1, shir, %t5; \
+ sll %t0, shil, %t6; \
+ or %t4, %prev, %t0; \
+ sll %t1, shil, %prev; \
+ or %t5, %t6, %t1; \
+ srl %t2, shir, %t4; \
+ srl %t3, shir, %t5; \
+ sll %t2, shil, %t6; \
+ or %t4, %prev, %t2; \
+ sll %t3, shil, %prev; \
+ or %t5, %t6, %t3; \
+ std %t0, [%dst + (offset) + (offset2) + 0x00]; \
+ std %t2, [%dst + (offset) + (offset2) + 0x08];
+
+ .text
+ .align 4
+
+0:
+ retl
+ nop ! Only bcopy returns here and it retuns void...
+
+#ifdef __KERNEL__
+FUNC(amemmove)
+FUNC(__memmove)
+#endif
+FUNC(memmove)
+ cmp %o0, %o1
+ mov %o0, %g7
+ bleu 9f
+ sub %o0, %o1, %o4
+
+ add %o1, %o2, %o3
+ cmp %o3, %o0
+ bleu 0f
+ andcc %o4, 3, %o5
+
+ add %o1, %o2, %o1
+ add %o0, %o2, %o0
+ sub %o1, 1, %o1
+ sub %o0, 1, %o0
+
+1: /* reverse_bytes */
+
+ ldub [%o1], %o4
+ subcc %o2, 1, %o2
+ stb %o4, [%o0]
+ sub %o1, 1, %o1
+ bne 1b
+ sub %o0, 1, %o0
+
+ retl
+ mov %g7, %o0
+
+/* NOTE: This code is executed just for the cases,
+ where %src (=%o1) & 3 is != 0.
+ We need to align it to 4. So, for (%src & 3)
+ 1 we need to do ldub,lduh
+ 2 lduh
+ 3 just ldub
+ so even if it looks weird, the branches
+ are correct here. -jj
+ */
+78: /* dword_align */
+
+ andcc %o1, 1, %g0
+ be 4f
+ andcc %o1, 2, %g0
+
+ ldub [%o1], %g2
+ add %o1, 1, %o1
+ stb %g2, [%o0]
+ sub %o2, 1, %o2
+ bne 3f
+ add %o0, 1, %o0
+4:
+ lduh [%o1], %g2
+ add %o1, 2, %o1
+ sth %g2, [%o0]
+ sub %o2, 2, %o2
+ b 3f
+ add %o0, 2, %o0
+
+FUNC(memcpy) /* %o0=dst %o1=src %o2=len */
+
+ sub %o0, %o1, %o4
+ mov %o0, %g7
+9:
+ andcc %o4, 3, %o5
+0:
+ bne 86f
+ cmp %o2, 15
+
+ bleu 90f
+ andcc %o1, 3, %g0
+
+ bne 78b
+3:
+ andcc %o1, 4, %g0
+
+ be 2f
+ mov %o2, %g1
+
+ ld [%o1], %o4
+ sub %g1, 4, %g1
+ st %o4, [%o0]
+ add %o1, 4, %o1
+ add %o0, 4, %o0
+2:
+ andcc %g1, 0xffffff80, %g0
+ be 3f
+ andcc %o0, 4, %g0
+
+ be 82f + 4
+5:
+ MOVE_BIGCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5)
+ MOVE_BIGCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5)
+ MOVE_BIGCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5)
+ MOVE_BIGCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5)
+ sub %g1, 128, %g1
+ add %o1, 128, %o1
+ cmp %g1, 128
+ bge 5b
+ add %o0, 128, %o0
+3:
+ andcc %g1, 0x70, %g4
+ be 80f
+ andcc %g1, 8, %g0
+
+ sethi %hi(80f), %o5
+ srl %g4, 1, %o4
+ add %g4, %o4, %o4
+ add %o1, %g4, %o1
+ sub %o5, %o4, %o5
+ jmpl %o5 + %lo(80f), %g0
+ add %o0, %g4, %o0
+
+79: /* memcpy_table */
+
+ MOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g4, g5)
+ MOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g4, g5)
+ MOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g4, g5)
+ MOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g4, g5)
+ MOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g4, g5)
+ MOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g4, g5)
+ MOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g4, g5)
+
+80: /* memcpy_table_end */
+ be 81f
+ andcc %g1, 4, %g0
+
+ ldd [%o1], %g2
+ add %o0, 8, %o0
+ st %g2, [%o0 - 0x08]
+ add %o1, 8, %o1
+ st %g3, [%o0 - 0x04]
+
+81: /* memcpy_last7 */
+
+ be 1f
+ andcc %g1, 2, %g0
+
+ ld [%o1], %g2
+ add %o1, 4, %o1
+ st %g2, [%o0]
+ add %o0, 4, %o0
+1:
+ be 1f
+ andcc %g1, 1, %g0
+
+ lduh [%o1], %g2
+ add %o1, 2, %o1
+ sth %g2, [%o0]
+ add %o0, 2, %o0
+1:
+ be 1f
+ nop
+
+ ldub [%o1], %g2
+ stb %g2, [%o0]
+1:
+ retl
+ mov %g7, %o0
+
+82: /* ldd_std */
+ MOVE_BIGALIGNCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5)
+ MOVE_BIGALIGNCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5)
+ MOVE_BIGALIGNCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5)
+ MOVE_BIGALIGNCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5)
+ subcc %g1, 128, %g1
+ add %o1, 128, %o1
+ cmp %g1, 128
+ bge 82b
+ add %o0, 128, %o0
+
+ andcc %g1, 0x70, %g4
+ be 84f
+ andcc %g1, 8, %g0
+
+ sethi %hi(84f), %o5
+ add %o1, %g4, %o1
+ sub %o5, %g4, %o5
+ jmpl %o5 + %lo(84f), %g0
+ add %o0, %g4, %o0
+
+83: /* amemcpy_table */
+
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x60, g2, g3, g4, g5)
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x50, g2, g3, g4, g5)
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x40, g2, g3, g4, g5)
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x30, g2, g3, g4, g5)
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x20, g2, g3, g4, g5)
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x10, g2, g3, g4, g5)
+ MOVE_LASTALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5)
+
+84: /* amemcpy_table_end */
+ be 85f
+ andcc %g1, 4, %g0
+
+ ldd [%o1], %g2
+ add %o0, 8, %o0
+ std %g2, [%o0 - 0x08]
+ add %o1, 8, %o1
+85: /* amemcpy_last7 */
+ be 1f
+ andcc %g1, 2, %g0
+
+ ld [%o1], %g2
+ add %o1, 4, %o1
+ st %g2, [%o0]
+ add %o0, 4, %o0
+1:
+ be 1f
+ andcc %g1, 1, %g0
+
+ lduh [%o1], %g2
+ add %o1, 2, %o1
+ sth %g2, [%o0]
+ add %o0, 2, %o0
+1:
+ be 1f
+ nop
+
+ ldub [%o1], %g2
+ stb %g2, [%o0]
+1:
+ retl
+ mov %g7, %o0
+
+86: /* non_aligned */
+ cmp %o2, 6
+ bleu 88f
+ nop
+
+ save %sp, -96, %sp
+ andcc %i0, 3, %g0
+ be 61f
+ andcc %i0, 1, %g0
+ be 60f
+ andcc %i0, 2, %g0
+
+ ldub [%i1], %g5
+ add %i1, 1, %i1
+ stb %g5, [%i0]
+ sub %i2, 1, %i2
+ bne 61f
+ add %i0, 1, %i0
+60:
+ ldub [%i1], %g3
+ add %i1, 2, %i1
+ stb %g3, [%i0]
+ sub %i2, 2, %i2
+ ldub [%i1 - 1], %g3
+ add %i0, 2, %i0
+ stb %g3, [%i0 - 1]
+61:
+ and %i1, 3, %g2
+ and %i2, 0xc, %g3
+ and %i1, -4, %i1
+ cmp %g3, 4
+ sll %g2, 3, %g4
+ mov 32, %g2
+ be 4f
+ sub %g2, %g4, %l0
+
+ blu 3f
+ cmp %g3, 0x8
+
+ be 2f
+ srl %i2, 2, %g3
+
+ ld [%i1], %i3
+ add %i0, -8, %i0
+ ld [%i1 + 4], %i4
+ b 8f
+ add %g3, 1, %g3
+2:
+ ld [%i1], %i4
+ add %i0, -12, %i0
+ ld [%i1 + 4], %i5
+ add %g3, 2, %g3
+ b 9f
+ add %i1, -4, %i1
+3:
+ ld [%i1], %g1
+ add %i0, -4, %i0
+ ld [%i1 + 4], %i3
+ srl %i2, 2, %g3
+ b 7f
+ add %i1, 4, %i1
+4:
+ ld [%i1], %i5
+ cmp %i2, 7
+ ld [%i1 + 4], %g1
+ srl %i2, 2, %g3
+ bleu 10f
+ add %i1, 8, %i1
+
+ ld [%i1], %i3
+ add %g3, -1, %g3
+5:
+ sll %i5, %g4, %g2
+ srl %g1, %l0, %g5
+ or %g2, %g5, %g2
+ st %g2, [%i0]
+7:
+ ld [%i1 + 4], %i4
+ sll %g1, %g4, %g2
+ srl %i3, %l0, %g5
+ or %g2, %g5, %g2
+ st %g2, [%i0 + 4]
+8:
+ ld [%i1 + 8], %i5
+ sll %i3, %g4, %g2
+ srl %i4, %l0, %g5
+ or %g2, %g5, %g2
+ st %g2, [%i0 + 8]
+9:
+ ld [%i1 + 12], %g1
+ sll %i4, %g4, %g2
+ srl %i5, %l0, %g5
+ addcc %g3, -4, %g3
+ or %g2, %g5, %g2
+ add %i1, 16, %i1
+ st %g2, [%i0 + 12]
+ add %i0, 16, %i0
+ bne,a 5b
+ ld [%i1], %i3
+10:
+ sll %i5, %g4, %g2
+ srl %g1, %l0, %g5
+ srl %l0, 3, %g3
+ or %g2, %g5, %g2
+ sub %i1, %g3, %i1
+ andcc %i2, 2, %g0
+ st %g2, [%i0]
+ be 1f
+ andcc %i2, 1, %g0
+
+ ldub [%i1], %g2
+ add %i1, 2, %i1
+ stb %g2, [%i0 + 4]
+ add %i0, 2, %i0
+ ldub [%i1 - 1], %g2
+ stb %g2, [%i0 + 3]
+1:
+ be 1f
+ nop
+ ldub [%i1], %g2
+ stb %g2, [%i0 + 4]
+1:
+ ret
+ restore %g7, %g0, %o0
+
+88: /* short_end */
+
+ and %o2, 0xe, %o3
+20:
+ sethi %hi(89f), %o5
+ sll %o3, 3, %o4
+ add %o0, %o3, %o0
+ sub %o5, %o4, %o5
+ add %o1, %o3, %o1
+ jmpl %o5 + %lo(89f), %g0
+ andcc %o2, 1, %g0
+
+ MOVE_SHORTCHUNK(o1, o0, 0x0c, g2, g3)
+ MOVE_SHORTCHUNK(o1, o0, 0x0a, g2, g3)
+ MOVE_SHORTCHUNK(o1, o0, 0x08, g2, g3)
+ MOVE_SHORTCHUNK(o1, o0, 0x06, g2, g3)
+ MOVE_SHORTCHUNK(o1, o0, 0x04, g2, g3)
+ MOVE_SHORTCHUNK(o1, o0, 0x02, g2, g3)
+ MOVE_SHORTCHUNK(o1, o0, 0x00, g2, g3)
+
+89: /* short_table_end */
+
+ be 1f
+ nop
+
+ ldub [%o1], %g2
+ stb %g2, [%o0]
+1:
+ retl
+ mov %g7, %o0
+
+90: /* short_aligned_end */
+ bne 88b
+ andcc %o2, 8, %g0
+
+ be 1f
+ andcc %o2, 4, %g0
+
+ ld [%o1 + 0x00], %g2
+ ld [%o1 + 0x04], %g3
+ add %o1, 8, %o1
+ st %g2, [%o0 + 0x00]
+ st %g3, [%o0 + 0x04]
+ add %o0, 8, %o0
+1:
+ b 81b
+ mov %o2, %g1