aosp12/external/libxaac/decoder/armv8/ixheaacd_inv_dit_fft_8pt.s

175 lines
5.7 KiB
ArmAsm

//VOID ixheaacd_inv_dit_fft_8pt(WORD32 *y,
// WORD32 *real,
// WORD32 *imag)
.macro push_v_regs
stp q8, q9, [sp, #-32]!
stp q10, q11, [sp, #-32]!
stp q12, q13, [sp, #-32]!
stp q14, q15, [sp, #-32]!
.endm
.macro pop_v_regs
ldp q14, q15, [sp], #32
ldp q12, q13, [sp], #32
ldp q10, q11, [sp], #32
ldp q8, q9, [sp], #32
.endm
.text
.global ixheaacd_inv_dit_fft_8pt_armv8
ixheaacd_inv_dit_fft_8pt_armv8:
push_v_regs
MOV w3, #0x5A820000
DUP v0.2s, w3
MOV x5, #8
ADD x6, x0, #4
//LD2 {v1.2s,v2.2s},[x0],x5
//LD2 {v3.2s,v4.2s},[x0],x5
//LD2 {v5.2s,v6.2s},[x0],x5
//LD2 {v7.2s,v8.2s},[x0],x5
LD1 {v1.s}[0], [x0], x5
LD1 {v2.s}[0], [x6], x5
LD1 {v1.s}[1], [x0], x5
LD1 {v2.s}[1], [x6], x5
LD1 {v3.s}[0], [x0], x5
LD1 {v4.s}[0], [x6], x5
LD1 {v3.s}[1], [x0], x5
LD1 {v4.s}[1], [x6], x5
LD1 {v5.s}[0], [x0], x5
LD1 {v6.s}[0], [x6], x5
LD1 {v5.s}[1], [x0], x5
LD1 {v6.s}[1], [x6], x5
LD1 {v7.s}[0], [x0], x5
LD1 {v8.s}[0], [x6], x5
LD1 {v7.s}[1], [x0], x5
LD1 {v8.s}[1], [x6], x5
//v1 - y0_2
//v2 - y1_3
//v3 - y4_6
//v4 - y5_7
//v5 - y8_10
//v6 - y9_11
//v7 - y12_14
//v8 - y13_15
SQADD v9.2s, v1.2s, v5.2s //a00_v = vqadd_s32(y0_2,y8_10);
SQADD v10.2s, v2.2s, v6.2s //a20_v = vqadd_s32(y1_3,y9_11);
SQADD v11.2s, v3.2s, v7.2s //a10_v = vqadd_s32(y4_6,y12_14);
SQADD v12.2s, v4.2s, v8.2s //a30_v = vqadd_s32(y5_7,y13_15);
SQSUB v1.2s, v1.2s, v5.2s //a0_v = vqsub_s32(y0_2,y8_10);
SQSUB v5.2s, v2.2s, v6.2s //a3_v = vqsub_s32(y1_3,y9_11);
SQSUB v2.2s, v3.2s, v7.2s //a2_v = vqsub_s32(y4_6,y12_14);
SQSUB v6.2s, v4.2s, v8.2s //a1_v = vqsub_s32(y5_7,y13_15);
SQADD v3.2s, v9.2s, v11.2s //x0_8 = vqadd_s32(a00_v,a10_v);
SQADD v7.2s, v10.2s, v12.2s //x1_9 = vqadd_s32(a20_v,a30_v);
SQSUB v4.2s, v9.2s, v11.2s //x4_12 = vqsub_s32(a00_v,a10_v);
SQSUB v8.2s, v10.2s, v12.2s //x5_13 = vqsub_s32(a20_v,a30_v);
SQADD v9.2s, v1.2s, v6.2s //x6_14 = vqadd_s32(a0_v,a1_v);
SQADD v11.2s, v5.2s, v2.2s //x3_11 = vqadd_s32(a3_v,a2_v);
SQSUB v10.2s, v1.2s, v6.2s //x2_10 = vqsub_s32(a0_v,a1_v);
SQSUB v13.2s, v5.2s, v2.2s //x7_15 = vqsub_s32(a3_v,a2_v);
UZP1 v1.2s, v3.2s, v7.2s //x0_1 = vuzp1_s32(x0_8,x1_9);
UZP2 v5.2s, v3.2s, v7.2s //x8_9 = vuzp2_s32(x0_8,x1_9);
UZP1 v6.2s, v4.2s, v8.2s //x4_5 = vuzp1_s32(x4_12,x5_13);
UZP2 v7.2s, v4.2s, v8.2s //x12_13 = vuzp2_s32(x4_12,x5_13);
REV64 v7.2s, v7.2s //x13_12 = vrev64_s32(x12_13);
SQADD v3.2s, v1.2s, v5.2s //real_imag0 = vqadd_s32(x0_1,x8_9);
SQSUB v8.2s, v1.2s, v5.2s //a00_10_v = vqsub_s32(x0_1,x8_9);
SQADD v12.2s, v6.2s, v7.2s //real_imag4 = vqadd_s32(x4_5,x13_12);
SQSUB v14.2s, v6.2s, v7.2s //a0_1_v = vqsub_s32(x4_5,x13_12);
MOV w4, v12.s[1]
MOV v12.s[1], v14.s[1]
MOV v14.s[1], w4
UZP1 v6.2s, v10.2s, v11.2s //x2_3
SQSUB v1.2s, v10.2s, v11.2s //tempr = vqsub_s32(x2_10,x3_11)
SQADD v5.2s, v10.2s, v11.2s //tempi = vqadd_s32(x2_10,x3_11)
SMULL v7.2d, v1.2s, v0.2s
SMULL v10.2d, v5.2s, v0.2s
SSHR v7.2d, v7.2d, #32 //tempr_q
SSHR v10.2d, v10.2d, #32 //tempi_q
SHL v7.4s, v7.4s, #1
SHL v10.4s, v10.4s, #1
MOV v1.s[0], v7.s[2]
MOV v1.s[1], v10.s[2] //vr_i
SQSUB v7.2s, v6.2s, v1.2s //a2_3_v = vqsub_s32(x2_3,vr_i);
SQADD v4.2s, v6.2s, v1.2s //real_imag1 = vqadd_s32(x2_3,vr_i);
SQADD v5.2s, v14.2s, v7.2s //real_imag2 = vqadd_s32(a0_1_v,a2_3_v);
UZP1 v1.2s, v9.2s, v13.2s //x6_7
SQADD v6.2s, v9.2s, v13.2s //tempr = vqadd_s32(x6_14,x7_15);
SQSUB v14.2s, v9.2s, v13.2s //tempi = vqsub_s32(x6_14,x7_15);
SMULL v9.2d, v6.2s, v0.2s
SMULL v13.2d, v14.2s, v0.2s
SSHR v9.2d, v9.2d, #32
SSHR v13.2d, v13.2d, #32
SHL v9.4s, v9.4s, #1
SHL v13.4s, v13.4s, #1
MOV v0.s[0], v9.s[2]
MOV v0.s[1], v13.s[2]
SQSUB v9.2s, v1.2s, v0.2s // a20_30_v
SQADD v13.2s, v1.2s, v0.2s //real_imag5
MOV w4, v9.s[1]
MOV v9.s[1], v13.s[1]
MOV v13.s[1], w4
SQADD v6.2s, v9.2s, v8.2s //real_imag3
ST1 {v3.s}[0], [x1], #4
ST1 {v4.s}[0], [x1], #4
ST1 {v5.s}[0], [x1], #4
ST1 {v6.s}[0], [x1], #4
ST1 {v12.s}[0], [x1], #4
ST1 {v13.s}[0], [x1], #4
ST1 {v3.s}[1], [x2], #4
ST1 {v4.s}[1], [x2], #4
ST1 {v5.s}[1], [x2], #4
ST1 {v6.s}[1], [x2], #4
ST1 {v12.s}[1], [x2], #4
ST1 {v13.s}[1], [x2], #4
//ST4 {v3.s,v4.s,v5.s,v6.s}[0],[x1],x5
//ST4 {v3.s,v4.s,v5.s,v6.s}[1],[x2],x5
//ST2 {v12.s,v13.s}[0],[x1]
//ST2 {v12.s,v13.s}[1],[x2]
pop_v_regs
ret