blob: 937d14bfe73fd93a77557a8909085d566b6bac50 [file] [log] [blame]
/***************************************************************************
Copyright (c) 2009-2012 Code Aurora Forum. All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of Code Aurora nor the names of its contributors may
be used to endorse or promote products derived from this software
without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
POSSIBILITY OF SUCH DAMAGE.
***************************************************************************/
/***************************************************************************
* Neon memmove: Attempts to do a memmove with Neon registers if possible,
* Inputs:
* dest: The destination buffer
* src: The source buffer
* n: The size of the buffer to transfer
* Outputs:
*
***************************************************************************/
#include <machine/cpu-features.h>
#if defined(KRAIT_NEON_OPTIMIZATION) || defined(SPARROW_NEON_OPTIMIZATION)
/*
* These can be overridden in:
* device/<vendor>/<board>/BoardConfig.mk
* by setting the following:
* TARGET_USE_KRAIT_BIONIC_OPTIMIZATION := true
* TARGET_USE_KRAIT_PLD_SET := true
* TARGET_KRAIT_BIONIC_PLDOFFS := <pldoffset>
* TARGET_KRAIT_BIONIC_PLDSIZE := <pldsize>
* TARGET_KRAIT_BIONIC_PLDTHRESH := <pldthreshold>
*/
#ifndef PLDOFFS
#define PLDOFFS (10)
#endif
#ifndef PLDTHRESH
#define PLDTHRESH (PLDOFFS)
#endif
#if (PLDOFFS < 5)
#error Routine does not support offsets less than 5
#endif
#if (PLDTHRESH < PLDOFFS)
#error PLD threshold must be greater than or equal to the PLD offset
#endif
#ifndef PLDSIZE
#define PLDSIZE (64)
#endif
#define NOP_OPCODE (0xe320f000)
.code 32
.align 5
.global memmove
.type memmove, %function
.global _memmove_words
.type _memmove_words, %function
.global bcopy
.type bcopy, %function
bcopy:
mov r12, r0
mov r0, r1
mov r1, r12
.balignl 64, NOP_OPCODE, 4*2
memmove:
_memmove_words:
.Lneon_memmove_cmf:
subs r12, r0, r1
bxeq lr
cmphi r2, r12
bls memcpy /* Use memcpy for non-overlapping areas */
push {r0}
.Lneon_back_to_front_copy:
add r0, r0, r2
add r1, r1, r2
cmp r2, #4
bgt .Lneon_b2f_gt4
cmp r2, #0
.Lneon_b2f_smallcopy_loop:
beq .Lneon_memmove_done
ldrb r12, [r1, #-1]!
subs r2, r2, #1
strb r12, [r0, #-1]!
b .Lneon_b2f_smallcopy_loop
.Lneon_b2f_gt4:
sub r3, r0, r1
cmp r2, r3
movle r12, r2
movgt r12, r3
cmp r12, #64
bge .Lneon_b2f_copy_64
cmp r12, #32
bge .Lneon_b2f_copy_32
cmp r12, #8
bge .Lneon_b2f_copy_8
cmp r12, #4
bge .Lneon_b2f_copy_4
b .Lneon_b2f_copy_1
.Lneon_b2f_copy_64:
sub r1, r1, #64 /* Predecrement */
sub r0, r0, #64
movs r12, r2, lsr #6
cmp r12, #PLDTHRESH
ble .Lneon_b2f_copy_64_loop_nopld
sub r12, #PLDOFFS
pld [r1, #-(PLDOFFS-5)*PLDSIZE]
pld [r1, #-(PLDOFFS-4)*PLDSIZE]
pld [r1, #-(PLDOFFS-3)*PLDSIZE]
pld [r1, #-(PLDOFFS-2)*PLDSIZE]
pld [r1, #-(PLDOFFS-1)*PLDSIZE]
.balignl 64, NOP_OPCODE, 4*2
.Lneon_b2f_copy_64_loop_outer:
pld [r1, #-(PLDOFFS)*PLDSIZE]
vld1.32 {q0, q1}, [r1]!
vld1.32 {q2, q3}, [r1]
subs r12, r12, #1
vst1.32 {q0, q1}, [r0]!
sub r1, r1, #96 /* Post-fixup and predecrement */
vst1.32 {q2, q3}, [r0]
sub r0, r0, #96
bne .Lneon_b2f_copy_64_loop_outer
mov r12, #PLDOFFS
.balignl 64, NOP_OPCODE, 4*2
.Lneon_b2f_copy_64_loop_nopld:
vld1.32 {q8, q9}, [r1]!
vld1.32 {q10, q11}, [r1]
subs r12, r12, #1
vst1.32 {q8, q9}, [r0]!
sub r1, r1, #96 /* Post-fixup and predecrement */
vst1.32 {q10, q11}, [r0]
sub r0, r0, #96
bne .Lneon_b2f_copy_64_loop_nopld
ands r2, r2, #0x3f
beq .Lneon_memmove_done
add r1, r1, #64 /* Post-fixup */
add r0, r0, #64
cmp r2, #32
blt .Lneon_b2f_copy_finish
.Lneon_b2f_copy_32:
mov r12, r2, lsr #5
.Lneon_b2f_copy_32_loop:
sub r1, r1, #32 /* Predecrement */
sub r0, r0, #32
vld1.32 {q0,q1}, [r1]
subs r12, r12, #1
vst1.32 {q0,q1}, [r0]
bne .Lneon_b2f_copy_32_loop
ands r2, r2, #0x1f
beq .Lneon_memmove_done
.Lneon_b2f_copy_finish:
.Lneon_b2f_copy_8:
movs r12, r2, lsr #0x3
beq .Lneon_b2f_copy_4
.balignl 64, NOP_OPCODE, 4*2
.Lneon_b2f_copy_8_loop:
sub r1, r1, #8 /* Predecrement */
sub r0, r0, #8
vld1.32 {d0}, [r1]
subs r12, r12, #1
vst1.32 {d0}, [r0]
bne .Lneon_b2f_copy_8_loop
ands r2, r2, #0x7
beq .Lneon_memmove_done
.Lneon_b2f_copy_4:
movs r12, r2, lsr #0x2
beq .Lneon_b2f_copy_1
.Lneon_b2f_copy_4_loop:
ldr r3, [r1, #-4]!
subs r12, r12, #1
str r3, [r0, #-4]!
bne .Lneon_b2f_copy_4_loop
ands r2, r2, #0x3
.Lneon_b2f_copy_1:
cmp r2, #0
beq .Lneon_memmove_done
.balignl 64, NOP_OPCODE, 4*2
.Lneon_b2f_copy_1_loop:
ldrb r12, [r1, #-1]!
subs r2, r2, #1
strb r12, [r0, #-1]!
bne .Lneon_b2f_copy_1_loop
.Lneon_memmove_done:
pop {r0}
bx lr
.end
#elif defined(SCORPION_NEON_OPTIMIZATION)
/*
* These can be overridden in:
* device/<vendor>/<board>/BoardConfig.mk
* by setting the following:
* TARGET_USE_SCORPION_BIONIC_OPTIMIZATION := true
* TARGET_USE_SCORPION_PLD_SET := true
* TARGET_SCORPION_BIONIC_PLDOFFS := <pldoffset>
* TARGET_SCORPION_BIONIC_PLDSIZE := <pldsize>
*/
#ifndef PLDOFFS
#define PLDOFFS (6)
#endif
#ifndef PLDSIZE
#define PLDSIZE (128) /* L2 cache line size */
#endif
.code 32
.align 5
.global memmove
.type memmove, %function
.global bcopy
.type bcopy, %function
bcopy:
mov r12, r0
mov r0, r1
mov r1, r12
memmove:
push {r0}
/*
* The requirements for memmove state that the function should
* operate as if data were being copied from the source to a
* buffer, then to the destination. This is to allow a user
* to copy data from a source and target that overlap.
*
* We can't just do byte copies front-to-back automatically, since
* there's a good chance we may have an overlap (why else would someone
* intentionally use memmove then?).
*
* We'll break this into two parts. Front-to-back, or back-to-front
* copies.
*/
.Lneon_memmove_cmf:
cmp r0, r1
blt .Lneon_front_to_back_copy
bgt .Lneon_back_to_front_copy
b .Lneon_memmove_done
/* #############################################################
* Front to Back copy
*/
.Lneon_front_to_back_copy:
/*
* For small copies, just do a quick memcpy. We can do this for
* front-to-back copies, aligned or unaligned, since we're only
* doing 1 byte at a time...
*/
cmp r2, #4
bgt .Lneon_f2b_gt4
cmp r2, #0
.Lneon_f2b_smallcopy_loop:
beq .Lneon_memmove_done
ldrb r12, [r1], #1
subs r2, r2, #1
strb r12, [r0], #1
b .Lneon_f2b_smallcopy_loop
.Lneon_f2b_gt4:
/* The window size is in r3. */
sub r3, r1, r0
/* #############################################################
* Front to Back copy
*/
/*
* Note that we can't just route based on the size in r2. If that's
* larger than the overlap window in r3, we could potentially
* (and likely!) destroy data we're copying.
*/
cmp r2, r3
movle r12, r2
movgt r12, r3
cmp r12, #256
bge .Lneon_f2b_copy_128
cmp r12, #64
bge .Lneon_f2b_copy_32
cmp r12, #16
bge .Lneon_f2b_copy_16
cmp r12, #8
bge .Lneon_f2b_copy_8
cmp r12, #4
bge .Lneon_f2b_copy_4
b .Lneon_f2b_copy_1
nop
.Lneon_f2b_copy_128:
mov r12, r2, lsr #7
cmp r12, #PLDOFFS
ble .Lneon_f2b_copy_128_loop_nopld
sub r12, #PLDOFFS
pld [r1, #(PLDOFFS-1)*PLDSIZE]
.Lneon_f2b_copy_128_loop_outer:
pld [r1, #(PLDOFFS*PLDSIZE)]
vld1.32 {q0,q1}, [r1]!
vld1.32 {q2,q3}, [r1]!
vld1.32 {q8,q9}, [r1]!
vld1.32 {q10,q11}, [r1]!
subs r12, r12, #1
vst1.32 {q0,q1}, [r0]!
vst1.32 {q2,q3}, [r0]!
vst1.32 {q8,q9}, [r0]!
vst1.32 {q10,q11}, [r0]!
bne .Lneon_f2b_copy_128_loop_outer
mov r12, #PLDOFFS
.Lneon_f2b_copy_128_loop_nopld:
vld1.32 {q0,q1}, [r1]!
vld1.32 {q2,q3}, [r1]!
vld1.32 {q8,q9}, [r1]!
vld1.32 {q10,q11}, [r1]!
subs r12, r12, #1
vst1.32 {q0,q1}, [r0]!
vst1.32 {q2,q3}, [r0]!
vst1.32 {q8,q9}, [r0]!
vst1.32 {q10,q11}, [r0]!
bne .Lneon_f2b_copy_128_loop_nopld
ands r2, r2, #0x7f
beq .Lneon_memmove_done
cmp r2, #32
bge .Lneon_f2b_copy_32
b .Lneon_f2b_copy_finish
.Lneon_f2b_copy_32:
mov r12, r2, lsr #5
.Lneon_f2b_copy_32_loop:
vld1.32 {q0,q1}, [r1]!
subs r12, r12, #1
vst1.32 {q0,q1}, [r0]!
bne .Lneon_f2b_copy_32_loop
ands r2, r2, #0x1f
beq .Lneon_memmove_done
.Lneon_f2b_copy_finish:
.Lneon_f2b_copy_16:
movs r12, r2, lsr #4
beq .Lneon_f2b_copy_8
.Lneon_f2b_copy_16_loop:
vld1.32 {q0}, [r1]!
subs r12, r12, #1
vst1.32 {q0}, [r0]!
bne .Lneon_f2b_copy_16_loop
ands r2, r2, #0xf
beq .Lneon_memmove_done
.Lneon_f2b_copy_8:
movs r12, r2, lsr #3
beq .Lneon_f2b_copy_4
.Lneon_f2b_copy_8_loop:
vld1.32 {d0}, [r1]!
subs r12, r12, #1
vst1.32 {d0}, [r0]!
bne .Lneon_f2b_copy_8_loop
ands r2, r2, #0x7
beq .Lneon_memmove_done
.Lneon_f2b_copy_4:
movs r12, r2, lsr #2
beq .Lneon_f2b_copy_1
.Lneon_f2b_copy_4_loop:
ldr r3, [r1], #4
subs r12, r12, #1
str r3, [r0], #4
bne .Lneon_f2b_copy_4_loop
ands r2, r2, #0x3
nop
.Lneon_f2b_copy_1:
cmp r2, #0
beq .Lneon_memmove_done
.Lneon_f2b_copy_1_loop:
ldrb r12, [r1], #1
subs r2, r2, #1
strb r12, [r0], #1
bne .Lneon_f2b_copy_1_loop
.Lneon_f2b_finish:
b .Lneon_memmove_done
/* #############################################################
* Back to Front copy
*/
.Lneon_back_to_front_copy:
/*
* Here, we'll want to shift to the end of the buffers. This
* actually points us one past where we need to go, but since
* we'll pre-decrement throughout, this will be fine.
*/
add r0, r0, r2
add r1, r1, r2
cmp r2, #4
bgt .Lneon_b2f_gt4
cmp r2, #0
.Lneon_b2f_smallcopy_loop:
beq .Lneon_memmove_done
ldrb r12, [r1, #-1]!
subs r2, r2, #1
strb r12, [r0, #-1]!
b .Lneon_b2f_smallcopy_loop
.Lneon_b2f_gt4:
/*
* The minimum of the overlap window size and the copy size
* is in r3.
*/
sub r3, r0, r1
/*
* #############################################################
* Back to Front copy -
*/
cmp r2, r3
movle r12, r2
movgt r12, r3
cmp r12, #256
bge .Lneon_b2f_copy_128
cmp r12, #64
bge .Lneon_b2f_copy_32
cmp r12, #8
bge .Lneon_b2f_copy_8
cmp r12, #4
bge .Lneon_b2f_copy_4
b .Lneon_b2f_copy_1
nop
.Lneon_b2f_copy_128:
movs r12, r2, lsr #7
cmp r12, #PLDOFFS
ble .Lneon_b2f_copy_128_loop_nopld
sub r12, #PLDOFFS
pld [r1, #-(PLDOFFS-1)*PLDSIZE]
.Lneon_b2f_copy_128_loop_outer:
pld [r1, #-(PLDOFFS*PLDSIZE)]
sub r1, r1, #128
sub r0, r0, #128
vld1.32 {q0, q1}, [r1]!
vld1.32 {q2, q3}, [r1]!
vld1.32 {q8, q9}, [r1]!
vld1.32 {q10, q11}, [r1]!
subs r12, r12, #1
vst1.32 {q0, q1}, [r0]!
vst1.32 {q2, q3}, [r0]!
vst1.32 {q8, q9}, [r0]!
vst1.32 {q10, q11}, [r0]!
sub r1, r1, #128
sub r0, r0, #128
bne .Lneon_b2f_copy_128_loop_outer
mov r12, #PLDOFFS
.Lneon_b2f_copy_128_loop_nopld:
sub r1, r1, #128
sub r0, r0, #128
vld1.32 {q0, q1}, [r1]!
vld1.32 {q2, q3}, [r1]!
vld1.32 {q8, q9}, [r1]!
vld1.32 {q10, q11}, [r1]!
subs r12, r12, #1
vst1.32 {q0, q1}, [r0]!
vst1.32 {q2, q3}, [r0]!
vst1.32 {q8, q9}, [r0]!
vst1.32 {q10, q11}, [r0]!
sub r1, r1, #128
sub r0, r0, #128
bne .Lneon_b2f_copy_128_loop_nopld
ands r2, r2, #0x7f
beq .Lneon_memmove_done
cmp r2, #32
bge .Lneon_b2f_copy_32
b .Lneon_b2f_copy_finish
.Lneon_b2f_copy_32:
mov r12, r2, lsr #5
.Lneon_b2f_copy_32_loop:
sub r1, r1, #32
sub r0, r0, #32
vld1.32 {q0,q1}, [r1]
subs r12, r12, #1
vst1.32 {q0,q1}, [r0]
bne .Lneon_b2f_copy_32_loop
ands r2, r2, #0x1f
beq .Lneon_memmove_done
.Lneon_b2f_copy_finish:
.Lneon_b2f_copy_8:
movs r12, r2, lsr #0x3
beq .Lneon_b2f_copy_4
.Lneon_b2f_copy_8_loop:
sub r1, r1, #8
sub r0, r0, #8
vld1.32 {d0}, [r1]
subs r12, r12, #1
vst1.32 {d0}, [r0]
bne .Lneon_b2f_copy_8_loop
ands r2, r2, #0x7
beq .Lneon_memmove_done
.Lneon_b2f_copy_4:
movs r12, r2, lsr #0x2
beq .Lneon_b2f_copy_1
.Lneon_b2f_copy_4_loop:
ldr r3, [r1, #-4]!
subs r12, r12, #1
str r3, [r0, #-4]!
bne .Lneon_b2f_copy_4_loop
ands r2, r2, #0x3
nop
.Lneon_b2f_copy_1:
cmp r2, #0
beq .Lneon_memmove_done
.Lneon_b2f_copy_1_loop:
ldrb r12, [r1, #-1]!
subs r2, r2, #1
strb r12, [r0, #-1]!
bne .Lneon_b2f_copy_1_loop
.Lneon_memmove_done:
pop {r0}
bx lr
.end
#endif /* SCORPION_NEON_OPTIMIZATION */