shithub: openh264

ref: 2907cb2875354bb8e36633cf61d7654fa7554f3e
dir: /codec/encoder/core/arm/svc_motion_estimation.S/

View raw version
/*!
 * \copy
 *     Copyright (c)  2013, Cisco Systems
 *     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.
 *
 *     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.
 *
 */

#ifdef  HAVE_NEON
.text
#include "arm_arch_common_macro.S"


WELS_ASM_FUNC_BEGIN SumOf8x8SingleBlock_neon
    vld1.64 {d0}, [r0], r1
    vld1.64 {d1}, [r0], r1
    vld1.64 {d2}, [r0], r1
    vld1.64 {d3}, [r0], r1
    vld1.64 {d4}, [r0], r1
    vld1.64 {d5}, [r0], r1
    vld1.64 {d6}, [r0], r1
    vld1.64 {d7}, [r0]
    vpaddl.u8 q0, q0
    vpadal.u8 q0, q1
    vpadal.u8 q0, q2
    vpadal.u8 q0, q3

    vpaddl.u16 q0, q0
    vpadd.i32 d0, d1
    vpadd.i32 d0, d0
    vmov    r0, r1, d0
WELS_ASM_FUNC_END


WELS_ASM_FUNC_BEGIN SumOf16x16SingleBlock_neon
    vld1.64 {q0}, [r0], r1
    vpaddl.u8 q0, q0
.rept 15
    vld1.64 {q1}, [r0], r1
    vpadal.u8 q0, q1
.endr
    vpaddl.u16 q0, q0
    vpadd.i32 d0, d1
    vpadd.i32 d0, d0
    vmov    r0, r1, d0
WELS_ASM_FUNC_END


WELS_ASM_FUNC_BEGIN SumOf8x8BlockOfFrame_neon
//(uint8_t* pRefPicture, const int32_t kiWidth, const int32_t kiHeight,const int32_t kiRefStride,uint16_t* pFeatureOfBlock, uint32_t pTimesOfFeatureValue[])
    stmdb sp!, {r4-r12}
    ldr	r5, [sp, #40] //pTimesOfFeatureValue
    ldr	r4, [sp, #36] //pFeatureOfBlock

    mov r8, r0
    mov r6, r1
    add r8, r6
    add r4, r4, r6, lsl #1

    mov r7, r6
_width_loop8x8_1:
    subs r0, r8, r7
    vld1.64 {d0}, [r0], r3
    vld1.64 {d1}, [r0], r3
    vld1.64 {d2}, [r0], r3
    vld1.64 {d3}, [r0], r3
    vld1.64 {d4}, [r0], r3
    vld1.64 {d5}, [r0], r3
    vld1.64 {d6}, [r0], r3
    vld1.64 {d7}, [r0]

    vpaddl.u8 q0, q0
    vpadal.u8 q0, q1
    vpadal.u8 q0, q2
    vpadal.u8 q0, q3
    vpaddl.u16 q0, q0
    vpadd.i32 d0, d1
    vpadd.i32 d0, d0

    subs r1, r4, r7, lsl #1
    vst1.16 {d0[0]}, [r1] // sum -> pFeatureOfBlock[i]
    vmov    r0, r1, d0
    add r1, r5, r0, lsl #2
    ldr r0, [r1]
    add r0, #1
    str r0, [r1]

    subs r7, #1
    bne _width_loop8x8_1

    add r8, r3
    add r4, r4, r6, lsl #1
    subs r2, #1
    beq _SumOf8x8BlockOfFrame_end


_height_loop8x8:
    mov r7, r6
_width_loop8x8_2:
    subs r0, r8, r7
    subs r1, r4, r7, lsl #1

    subs r9, r1, r6, lsl #1 // last line of pFeatureOfBlock[i]
    ldrh  r10, [r9] // sum of last line of pFeatureOfBlock[i]

    subs r11, r0, r3
    vld1.64 {d1}, [r11]
    add r0, r11, r3, lsl #3
    vld1.64 {d0}, [r0] //

    vpaddl.u8 q0, q0
    vpadd.u16 d0, d0, d1
    vpaddl.u16 d0, d0
    vmov r11, r12, d0
    subs r10, r12
    add r0, r10, r11

    strh r0, [r1] // sum -> pFeatureOfBlock[i]

    add r1, r5, r0, lsl #2
    ldr r0, [r1]
    add r0, #1
    str r0, [r1]
    subs r7, #1
    bne _width_loop8x8_2

    add r8, r3
    add r4, r4, r6, lsl #1
    subs r2, #1
    bne _height_loop8x8
_SumOf8x8BlockOfFrame_end:
    ldmia sp!, {r4-r12}
WELS_ASM_FUNC_END

WELS_ASM_FUNC_BEGIN SumOf16x16BlockOfFrame_neon
//(uint8_t* pRefPicture, const int32_t kiWidth, const int32_t kiHeight,const int32_t kiRefStride,uint16_t* pFeatureOfBlock, uint32_t pTimesOfFeatureValue[])
    stmdb sp!, {r4-r12}
    ldr	r5, [sp, #40] //pTimesOfFeatureValue
    ldr	r4, [sp, #36] //pFeatureOfBlock

    mov r8, r0
    mov r6, r1
    add r8, r6
    add r4, r4, r6, lsl #1

    mov r7, r6
_width_loop16x16_1:
    subs r0, r8, r7
    vld1.64 {q0}, [r0], r3
    vpaddl.u8 q0, q0
.rept 15
    vld1.64 {q1}, [r0], r3
    vpadal.u8 q0, q1
.endr
    vpaddl.u16 q0, q0
    vpadd.i32 d0, d1
    vpadd.i32 d0, d0

    subs r1, r4, r7, lsl #1
    vst1.16 {d0[0]}, [r1] // sum -> pFeatureOfBlock[i]
    vmov    r0, r1, d0
    add r1, r5, r0, lsl #2
    ldr r0, [r1]
    add r0, #1
    str r0, [r1]

    subs r7, #1
    bne _width_loop16x16_1
    add r8, r3
    add r4, r4, r6, lsl #1
    subs r2, #1
    beq _SumOf16x16BlockOfFrame_neon_end

_height_loop16x16:
    mov r7, r6
_width_loop16x16_2:
    subs r0, r8, r7
    subs r1, r4, r7, lsl #1
    subs r9, r1, r6, lsl #1 // last line of pFeatureOfBlock[i]
    ldrh  r10, [r9] // sum of last line of pFeatureOfBlock[i]

    subs r11, r0, r3
    vld1.64 {q1}, [r11]
    add r0, r11, r3, lsl #4
    vld1.64 {q0}, [r0] //

    vpaddl.u8 q0, q0
    vpaddl.u8 q1, q1
    vpadd.u16 d0, d0, d1
    vpadd.u16 d1, d2, d3
    vpadd.u16 d0, d0, d1
    vpaddl.u16 d0, d0

    vmov r11, r12, d0
    subs r10, r12
    add r0, r10, r11

    strh r0, [r1] // sum -> pFeatureOfBlock[i]
    add r1, r5, r0, lsl #2
    ldr r0, [r1]
    add r0, #1
    str r0, [r1]

    subs r7, #1
    bne _width_loop16x16_2

    add r8, r3
    add r4, r4, r6, lsl #1
    subs r2, #1
    bne _height_loop16x16
_SumOf16x16BlockOfFrame_neon_end:
    ldmia sp!, {r4-r12}
WELS_ASM_FUNC_END

WELS_ASM_FUNC_BEGIN InitializeHashforFeature_neon
// (uint32_t* pTimesOfFeatureValue, uint16_t* pBuf, const int32_t kiListSize, uint16_t** pLocationOfFeature, uint16_t** pFeatureValuePointerList);
    stmdb sp!, {r4-r7}
    ldr	r4, [sp, #16] //pFeatureValuePointerList
    bic r5, r2, #3
_hash_assign_loop_x4:
    vld1.64 {q0}, [r0]!
    vshl.u32 q0, q0, #2
    vceq.u32 q1, q0, #0
    vand.i32 d2, d2, d3
    vmov r6, r7, d2
    and r6, r6, r7
    cmp r6, #0xffffffff
    beq _hash_assign_with_copy_x4

    veor q1, q1
    vext.32 q2, q1, q0, #3
    vext.32 q3, q1, q0, #2
    vext.32 q4, q1, q0, #1
    vadd.u32 q0, q0, q2
    vadd.u32 q0, q0, q3
    vadd.u32 q0, q0, q4
    vext.32 q2, q1, q0, #3
    vdup.32  q3, r1
    vadd.u32 q2, q2, q3
    vst1.64 {q2}, [r3]!
    vst1.64 {q2}, [r4]!
    vmov.32 r6, d1[1]
    add r1, r1, r6
    b _assign_next

_hash_assign_with_copy_x4:
    vdup.32  q2, r1
    vst1.64 {q2}, [r3]!
    vst1.64 {q2}, [r4]!

_assign_next:
	subs r5, r5, #4
	bne _hash_assign_loop_x4

    and r5, r2, #3
    cmp r5, #0
    beq _hash_assign_end
_hash_assign_loop_x4_rem:
    str r1, [r3], #4
    str r1, [r4], #4
    ldr r7, [r0], #4
    lsl r7, r7, #2
    add r1, r1, r7
    subs r5, r5, #1
    bne _hash_assign_loop_x4_rem
_hash_assign_end:

    ldmia sp!, {r4-r7}
WELS_ASM_FUNC_END

.align 4
mv_x_inc_x4: .short 0x10, 0x10, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00
mv_y_inc_x4: .short 0x04, 0x04, 0x04, 0x04, 0x00, 0x00, 0x00, 0x00
mx_x_offset_x4: .short 0x00, 0x04, 0x08, 0x0c, 0x00, 0x00, 0x00, 0x00

WELS_ASM_FUNC_BEGIN FillQpelLocationByFeatureValue_neon
// void  (uint16_t* pFeatureOfBlock, const int32_t kiWidth, const int32_t kiHeight, uint16_t** pFeatureValuePointerList)
    stmdb sp!, {r4-r8}
    vpush		{q4-q7}
    adr r7, mv_x_inc_x4
    vld1.64 {q7}, [r7]
    adr r7, mv_y_inc_x4
    vld1.64 {q6}, [r7]
    adr r7, mx_x_offset_x4
    vld1.64 {q5}, [r7]
    veor q4, q4
    veor q3, q3
    vdup.32 q8, r3
_hash_height_loop:
    mov r7, r1
    vmov q2, q5 //mx_x_offset_x4
_hash_width_loop:
    vld1.64 {d0}, [r0]!
    vshll.u16 q0, d0, #2
    vadd.u32 q0, q8
    vmov q1, q2
    vmov q4, q3
    vzip.16 q1, q4

    vmov.32 r4, d0[0]
    ldr r5, [r4]
    vmov.32 r6, d2[0]
    str r6, [r5]
    add r5, r5, #4
    pld [r5] // cache miss?
    str r5, [r4]

    vmov.32 r4, d0[1]
    ldr r5, [r4]
    vmov.32 r6, d2[1]
    str r6, [r5]
    add r5, r5, #4
    pld [r5] // cache miss?
    str r5, [r4]

    vmov.32 r4, d1[0]
    ldr r5, [r4]
    vmov.32 r6, d3[0]
    str r6, [r5]
    add r5, r5, #4
    pld [r5] // cache miss?
    str r5, [r4]

    vmov.32 r4, d1[1]
    ldr r5, [r4]
    vmov.32 r6, d3[1]
    str r6, [r5]
    add r5, r5, #4
    pld [r5] // cache miss?
    str r5, [r4]

    vadd.u16 q2, q2, q7
    subs r7, #4
    bne _hash_width_loop

    vadd.u16 q3, q3, q6
    subs r2, #1
    bne _hash_height_loop

    vpop		{q4-q7}
    ldmia sp!, {r4-r8}
WELS_ASM_FUNC_END
#endif