summaryrefslogtreecommitdiff
path: root/sys/src/ape/lib/ap/sparc/memmove.s
diff options
context:
space:
mode:
authorTaru Karttunen <taruti@taruti.net>2011-03-30 15:46:40 +0300
committerTaru Karttunen <taruti@taruti.net>2011-03-30 15:46:40 +0300
commite5888a1ffdae813d7575f5fb02275c6bb07e5199 (patch)
treed8d51eac403f07814b9e936eed0c9a79195e2450 /sys/src/ape/lib/ap/sparc/memmove.s
Import sources from 2011-03-30 iso image
Diffstat (limited to 'sys/src/ape/lib/ap/sparc/memmove.s')
-rwxr-xr-xsys/src/ape/lib/ap/sparc/memmove.s169
1 files changed, 169 insertions, 0 deletions
diff --git a/sys/src/ape/lib/ap/sparc/memmove.s b/sys/src/ape/lib/ap/sparc/memmove.s
new file mode 100755
index 000000000..8879a74e8
--- /dev/null
+++ b/sys/src/ape/lib/ap/sparc/memmove.s
@@ -0,0 +1,169 @@
+ TEXT memmove(SB), $0
+ JMP move
+
+ TEXT memcpy(SB), $0
+move:
+
+/*
+ * performance:
+ * (tba)
+ */
+
+ MOVW R7, s1+0(FP)
+ MOVW n+8(FP), R9 /* R9 is count */
+ MOVW R7, R10 /* R10 is to-pointer */
+ SUBCC R0,R9, R0
+ BGE ok
+ MOVW 0(R0), R0
+
+ok:
+ MOVW s2+4(FP), R11 /* R11 is from-pointer */
+ ADD R9,R11, R13 /* R13 is end from-pointer */
+ ADD R9,R10, R12 /* R12 is end to-pointer */
+
+/*
+ * easiest test is copy backwards if
+ * destination string has higher mem address
+ */
+ SUBCC R11,R10, R0
+ BGU back
+
+/*
+ * if not at least 4 chars,
+ * dont even mess around.
+ * 3 chars to guarantee any
+ * rounding up to a word
+ * boundary and 4 characters
+ * to get at least maybe one
+ * full word store.
+ */
+ SUBCC $4,R9, R0
+ BL fout
+
+/*
+ * test if both pointers
+ * are similarly word aligned
+ */
+ XOR R10,R11, R7
+ ANDCC $3,R7, R0
+ BNE fout
+
+/*
+ * byte at a time to word align
+ */
+f1:
+ ANDCC $3,R10, R0
+ BE f2
+ MOVB 0(R11), R16
+ ADD $1, R11
+ MOVB R16, 0(R10)
+ ADD $1, R10
+ JMP f1
+
+/*
+ * turn R9 into to-end pointer-15
+ * copy 16 at a time while theres room.
+ * R12 is smaller than R13 --
+ * there are problems if R13 is 0.
+ */
+f2:
+ SUB $15,R12, R9
+f3:
+ SUBCC R10,R9, R0
+ BLEU f4
+ MOVW 0(R11), R16
+ MOVW 4(R11), R17
+ MOVW R16, 0(R10)
+ MOVW 8(R11), R16
+ MOVW R17, 4(R10)
+ MOVW 12(R11), R17
+ ADD $16, R11
+ MOVW R16, 8(R10)
+ MOVW R17, 12(R10)
+ ADD $16, R10
+ JMP f3
+
+/*
+ * turn R9 into to-end pointer-3
+ * copy 4 at a time while theres room
+ */
+f4:
+ SUB $3,R12, R9
+f5:
+ SUBCC R10,R9, R0
+ BLEU fout
+ MOVW 0(R11), R16
+ ADD $4, R11
+ MOVW R16, 0(R10)
+ ADD $4, R10
+ JMP f5
+
+/*
+ * last loop, copy byte at a time
+ */
+fout:
+ SUBCC R11,R13, R0
+ BLEU ret
+ MOVB 0(R11), R16
+ ADD $1, R11
+ MOVB R16, 0(R10)
+ ADD $1, R10
+ JMP fout
+
+/*
+ * whole thing repeated for backwards
+ */
+back:
+ SUBCC $4,R9, R0
+ BL bout
+
+ XOR R12,R13, R7
+ ANDCC $3,R7, R0
+ BNE bout
+b1:
+ ANDCC $3,R13, R0
+ BE b2
+ MOVB -1(R13), R16
+ SUB $1, R13
+ MOVB R16, -1(R12)
+ SUB $1, R12
+ JMP b1
+b2:
+ ADD $15,R11, R9
+b3:
+ SUBCC R9,R13, R0
+ BLEU b4
+ MOVW -4(R13), R16
+ MOVW -8(R13), R17
+ MOVW R16, -4(R12)
+ MOVW -12(R13), R16
+ MOVW R17, -8(R12)
+ MOVW -16(R13), R17
+ SUB $16, R13
+ MOVW R16, -12(R12)
+ MOVW R17, -16(R12)
+ SUB $16, R12
+ JMP b3
+b4:
+ ADD $3,R11, R9
+b5:
+ SUBCC R9,R13, R0
+ BLEU bout
+ MOVW -4(R13), R16
+ SUB $4, R13
+ MOVW R16, -4(R12)
+ SUB $4, R12
+ JMP b5
+
+bout:
+ SUBCC R11,R13, R0
+ BLEU ret
+ MOVB -1(R13), R16
+ SUB $1, R13
+ MOVB R16, -1(R12)
+ SUB $1, R12
+ JMP bout
+
+ret:
+ MOVW s1+0(FP), R7
+ RETURN