SWHI t0, 0(a0)
addu a0, t1
-L(chk8w): andi t0, a2, 0x1f # 32 or more bytes left?
+L(chk8w):
+ andi t0, a2, 0x1f # 32 or more bytes left?
beq t0, a2, L(chk1w)
subu a3, a2, t0 # Yes
addu a3, a1 # a3 = end address of loop
move a2, t0 # a2 = what will be left after loop
-L(lop8w): lw t0, 0(a1) # Loop taking 8 words at a time
+L(lop8w):
+ lw t0, 0(a1) # Loop taking 8 words at a time
lw t1, 4(a1)
lw t2, 8(a1)
lw t3, 12(a1)
bne a1, a3, L(lop8w)
sw t7, -4(a0)
-L(chk1w): andi t0, a2, 0x3 # 4 or more bytes left?
+L(chk1w):
+ andi t0, a2, 0x3 # 4 or more bytes left?
beq t0, a2, L(last8)
subu a3, a2, t0 # Yes, handle them one word at a time
addu a3, a1 # a3 again end address
move a2, t0
-L(lop1w): lw t0, 0(a1)
+L(lop1w):
+ lw t0, 0(a1)
addiu a0, 4
addiu a1, 4
bne a1, a3, L(lop1w)
sw t0, -4(a0)
-L(last8): blez a2, L(lst8e) # Handle last 8 bytes, one at a time
+L(last8):
+ blez a2, L(lst8e) # Handle last 8 bytes, one at a time
addu a3, a2, a1
-L(lst8l): lb t0, 0(a1)
+L(lst8l):
+ lb t0, 0(a1)
addiu a0, 1
addiu a1, 1
bne a1, a3, L(lst8l)
sb t0, -1(a0)
-L(lst8e): jr ra # Bye, bye
+L(lst8e):
+ jr ra # Bye, bye
nop
-L(shift): subu a3, zero, a0 # Src and Dest unaligned
+L(shift):
+ subu a3, zero, a0 # Src and Dest unaligned
andi a3, 0x3 # (unoptimized case...)
beq a3, zero, L(shft1)
subu a2, a3 # a2 = bytes left
addu a1, a3
SWHI t0, 0(a0)
addu a0, a3
-L(shft1): andi t0, a2, 0x3
+L(shft1):
+ andi t0, a2, 0x3
subu a3, a2, t0
addu a3, a1
-L(shfth): LWHI t1, 0(a1) # Limp through, word by word
+L(shfth):
+ LWHI t1, 0(a1) # Limp through, word by word
LWLO t1, 3(a1)
addiu a0, 4
addiu a1, 4
bne a1, a3, L(shfth)
sw t1, -4(a0)
- b L(last8) # Handle anything which may be left
+ b L(last8) # Handle anything which may be left
move a2, t0
.set reorder
sll t0, a1, 16
or a1, t0 # a1 is now pattern in full word
-L(ueven): subu t0, zero, a0 # Unaligned address?
+L(ueven):
+ subu t0, zero, a0 # Unaligned address?
andi t0, 0x3
beq t0, zero, L(chkw)
subu a2, t0
SWHI a1, 0(a0) # Yes, handle first unaligned part
addu a0, t0 # Now both a0 and a2 are updated
-L(chkw): andi t0, a2, 0x7 # Enough left for one loop iteration?
+L(chkw):
+ andi t0, a2, 0x7 # Enough left for one loop iteration?
beq t0, a2, L(chkl)
subu a3, a2, t0
addu a3, a0 # a3 is last loop address +1
move a2, t0 # a2 is now # of bytes left after loop
-L(loopw): addiu a0, 8 # Handle 2 words pr. iteration
+L(loopw):
+ addiu a0, 8 # Handle 2 words pr. iteration
sw a1, -8(a0)
bne a0, a3, L(loopw)
sw a1, -4(a0)
-L(chkl): andi t0, a2, 0x4 # Check if there is at least a full
+L(chkl):
+ andi t0, a2, 0x4 # Check if there is at least a full
beq t0, zero, L(last8) # word remaining after the loop
subu a2, t0
sw a1, 0(a0) # Yes...
addiu a0, 4
-L(last8): blez a2, L(exit) # Handle last 8 bytes (if cnt>0)
+L(last8):
+ blez a2, L(exit) # Handle last 8 bytes (if cnt>0)
addu a3, a2, a0 # a3 is last address +1
-L(lst8l): addiu a0, 1
+L(lst8l):
+ addiu a0, 1
bne a0, a3, L(lst8l)
sb a1, -1(a0)
-L(exit): j ra # Bye, bye
+L(exit):
+ j ra # Bye, bye
nop
.set reorder
ENTRY (memcpy)
.set noreorder
- slti a4, a2, 8 # Less than 8?
- bne a4, zero, L(last8)
+ slti a4, a2, 16 # Less than 16?
+ bne a4, zero, L(last16)
move v0, a0 # Setup exit value before too late
xor a4, a1, a0 # Find a0/a1 displacement
sd t7, -8(a0)
L(chk1w):
- andi a4, a2, 0x7 # 4 or more bytes left?
- beq a4, a2, L(last8)
- PTR_SUBU a3, a2, a4 # Yes, handle them one word at a time
+ andi a4, a2, 0x7 # 8 or more bytes left?
+ beq a4, a2, L(last16)
+ PTR_SUBU a3, a2, a4 # Yes, handle them one dword at a time
PTR_ADDU a3, a1 # a3 again end address
move a2, a4
-L(lop1w):
+L(lop1w):
ld a4, 0(a1)
PTR_ADDIU a0, 8
PTR_ADDIU a1, 8
bne a1, a3, L(lop1w)
sd a4, -8(a0)
-L(last8):
- blez a2, L(lst8e) # Handle last 8 bytes, one at a time
+L(last16):
+ blez a2, L(lst16e) # Handle last 16 bytes, one at a time
PTR_ADDU a3, a2, a1
-L(lst8l):
+L(lst16l):
lb a4, 0(a1)
PTR_ADDIU a0, 1
PTR_ADDIU a1, 1
- bne a1, a3, L(lst8l)
+ bne a1, a3, L(lst16l)
sb a4, -1(a0)
-L(lst8e):
+L(lst16e):
jr ra # Bye, bye
nop
L(shift):
PTR_SUBU a3, zero, a0 # Src and Dest unaligned
andi a3, 0x7 # (unoptimized case...)
- beq a3, zero, L(shfa5)
+ beq a3, zero, L(shft1)
PTR_SUBU a2, a3 # a2 = bytes left
LDHI a4, 0(a1) # Take care of first odd part
LDLO a4, 7(a1)
PTR_ADDU a1, a3
SDHI a4, 0(a0)
PTR_ADDU a0, a3
-L(shfa5):
+L(shft1):
andi a4, a2, 0x7
PTR_SUBU a3, a2, a4
PTR_ADDU a3, a1
L(shfth):
- LDHI a5, 0(a1) # Limp through, word by word
+ LDHI a5, 0(a1) # Limp through, dword by dword
LDLO a5, 7(a1)
PTR_ADDIU a0, 8
PTR_ADDIU a1, 8
bne a1, a3, L(shfth)
sd a5, -8(a0)
- b L(last8) # Handle anything which may be left
+ b L(last16) # Handle anything which may be left
move a2, a4
.set reorder
ENTRY (memset)
.set noreorder
- slti t5, a2, 8 # Less than 8?
- bne t5, zero, L(last8)
+ slti t5, a2, 16 # Less than 16?
+ bne t5, zero, L(last16)
move v0, a0 # Setup exit value before too late
beq a1, zero, L(ueven) # If zero pattern, no need to extend
PTR_ADDU a3, a0 # a3 is last loop address +1
move a2, t4 # a2 is now # of bytes left after loop
L(loopw):
- PTR_ADDIU a0, 16 # Handle 2 words pr. iteration
+ PTR_ADDIU a0, 16 # Handle 2 dwords pr. iteration
sd a1, -16(a0)
bne a0, a3, L(loopw)
sd a1, -8(a0)
L(chkl):
andi t4, a2, 0x8 # Check if there is at least a double
- beq t4, zero, L(last8) # word remaining after the loop
+ beq t4, zero, L(last16) # word remaining after the loop
PTR_SUBU a2, t4
sd a1, 0(a0) # Yes...
PTR_ADDIU a0, 8
-L(last8):
- blez a2, L(exit) # Handle last 8 bytes (if cnt>0)
+L(last16):
+ blez a2, L(exit) # Handle last 16 bytes (if cnt>0)
PTR_ADDU a3, a2, a0 # a3 is last address +1
-L(lst8l):
+L(lst16l):
PTR_ADDIU a0, 1
- bne a0, a3, L(lst8l)
+ bne a0, a3, L(lst16l)
sb a1, -1(a0)
L(exit):
j ra # Bye, bye