// uint32_t quant_c(const int16_t* coef, const int32_t quantScale, int32_t* deltaU, int16_t* qCoef, int qBits, int add, int numCoeff)
function x265_quant_neon
mov w9, #1 //x9的低32位 = 1
lsl w9, w9, w4 //w9 = 1 << qBits;
dup v0.2s, w9 //16bits为单位 = w9 2s ?
neg w9, w4 //w4 = 0 - w9
dup v1.4s, w9 //v1, 4字节signed = w9
add w9, w9, #8 //w8 += 8
dup v2.4s, w9 //v2 4字节signed = w9
dup v3.4s, w5 //v3 4字节signed = w5 参数add
mov w11, w1 //w11 = w1 //w11 = quantScale
lsr w6, w6, #2 //w6 = w6 >> 2 numCoeff 参数， 一组4个系数
eor v4.16b, v4.16b, v4.16b //v4 = 0
eor w10, w10, w10 //w10 = 0;
eor v17.16b, v17.16b, v17.16b //v17 = 0
.loop_quant:
ld1 {v18.4h}, [x0], #8 //coef 加载 4个系数，到v18的低64位
//ld1 {v7.4s}, [x1], #16
dup v7.4s, w11 //v7 4字节为单位的signed = quantScale
sxtl v6.4s, v18.4h //v18 本来是16bits extern 到 32bits 存入v6
cmlt v5.4s, v6.4s, #0 //v5的值小于寄存器v6的值，判断v6 系数是否 > 0
abs v6.4s, v6.4s //求绝对值
mul v6.4s, v6.4s, v7.4s //系数 = 系数 * qscale int tmplevel = abs(level) * quantScale;
add v7.4s, v6.4s, v3.4s //得到的结果加上偏移， tmplevel += add
sshl v7.4s, v7.4s, v1.4s //逻辑左移 qBits 位
mls v6.4s, v7.4s, v0.s[0] //level << qBits level <<= qBits 存入v6
sshl v16.4s, v6.4s, v2.4s //level << qBits >> qBits8 左移 -qBits8 >> 右移qBits8
st1 {v16.4s}, [x2], #16 //deltaU[blockpos] = ((tmplevel - (level << qBits)) >> qBits8); 数据存入deltaU
// numsig
cmeq v16.4s, v7.4s, v17.4s //v7和v17做比较，结果存储到v16, v17 = 0在这里
add v4.4s, v4.4s, v16.4s //系数个数累加
add w10, w10, #4 //处理了四个系数
// level *= sign
eor v16.16b, v7.16b, v5.16b //符号位通过异或 这里是相当于取反
sub v16.4s, v16.4s, v5.4s // 然后加1
sqxtn v5.4h, v16.4s //v16 4字节宽，缩小到2字节， 存储到v5
st1 {v5.4h}, [x3], #8 //存入量化后的系数，数组
subs w6, w6, #1 //w6 = w6 - 1 处理完一组4个系数
b.ne .loop_quant
addv s4, v4.4s
mov w9, v4.s[0]
add w0, w10, w9
ret
endfunc