Update idct NEON optimization to not use narrowing saturating shift
Change-Id: Iae517017217dbacd638d40fcfeeb0f4bba7b8b8b
This commit is contained in:
parent
228c9940ea
commit
018a2adcb1
@ -77,10 +77,10 @@ static INLINE void IDCT4x4_1D(int16x4_t *d0s16, int16x4_t *d1s16,
|
||||
q15s32 = vmlsl_s16(q15s32, d19s16, *d0s16);
|
||||
q10s32 = vmlal_s16(q10s32, d19s16, *d2s16);
|
||||
|
||||
d26s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d27s16 = vqrshrn_n_s32(q14s32, 14);
|
||||
d29s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d28s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d26s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d27s16 = vrshrn_n_s32(q14s32, 14);
|
||||
d29s16 = vrshrn_n_s32(q15s32, 14);
|
||||
d28s16 = vrshrn_n_s32(q10s32, 14);
|
||||
|
||||
q13s16 = vcombine_s16(d26s16, d27s16);
|
||||
q14s16 = vcombine_s16(d28s16, d29s16);
|
||||
@ -125,10 +125,10 @@ static INLINE void IADST4x4_1D(int16x4_t *d3s16, int16x4_t *d4s16,
|
||||
q14s32 = vaddq_s32(q11s32, q12s32);
|
||||
q10s32 = vsubq_s32(q10s32, q12s32);
|
||||
|
||||
d16s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d17s16 = vqrshrn_n_s32(q14s32, 14);
|
||||
d18s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d19s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d16s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d17s16 = vrshrn_n_s32(q14s32, 14);
|
||||
d18s16 = vrshrn_n_s32(q15s32, 14);
|
||||
d19s16 = vrshrn_n_s32(q10s32, 14);
|
||||
|
||||
*q8s16 = vcombine_s16(d16s16, d17s16);
|
||||
*q9s16 = vcombine_s16(d18s16, d19s16);
|
||||
|
@ -76,10 +76,10 @@ static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q5s32 = vmlsl_s16(q5s32, d22s16, d3s16);
|
||||
q6s32 = vmlsl_s16(q6s32, d23s16, d3s16);
|
||||
|
||||
d8s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d9s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d8s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d9s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q6s32, 14);
|
||||
q4s16 = vcombine_s16(d8s16, d9s16);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
|
||||
@ -93,10 +93,10 @@ static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q9s32 = vmlal_s16(q9s32, d22s16, d2s16);
|
||||
q13s32 = vmlal_s16(q13s32, d23s16, d2s16);
|
||||
|
||||
d14s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d15s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d14s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d15s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q13s32, 14);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
q7s16 = vcombine_s16(d14s16, d15s16);
|
||||
|
||||
@ -115,10 +115,10 @@ static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
d0s16 = vdup_n_s16(cospi_24_64);
|
||||
d1s16 = vdup_n_s16(cospi_8_64);
|
||||
|
||||
d18s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d19s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d22s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d23s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d18s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d19s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d22s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d23s16 = vrshrn_n_s32(q15s32, 14);
|
||||
*q9s16 = vcombine_s16(d18s16, d19s16);
|
||||
*q11s16 = vcombine_s16(d22s16, d23s16);
|
||||
|
||||
@ -132,10 +132,10 @@ static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q8s32 = vmlal_s16(q8s32, d28s16, d0s16);
|
||||
q12s32 = vmlal_s16(q12s32, d29s16, d0s16);
|
||||
|
||||
d26s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d27s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d30s16 = vqrshrn_n_s32(q8s32, 14);
|
||||
d31s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d26s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d27s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d30s16 = vrshrn_n_s32(q8s32, 14);
|
||||
d31s16 = vrshrn_n_s32(q12s32, 14);
|
||||
*q13s16 = vcombine_s16(d26s16, d27s16);
|
||||
*q15s16 = vcombine_s16(d30s16, d31s16);
|
||||
|
||||
@ -165,10 +165,10 @@ static INLINE void IDCT8x8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q11s32 = vmlal_s16(q11s32, d26s16, d16s16);
|
||||
q12s32 = vmlal_s16(q12s32, d27s16, d16s16);
|
||||
|
||||
d10s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q9s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q11s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q12s32, 14);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -242,8 +242,8 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q1s32 = vsubq_s32(q1s32, q5s32);
|
||||
q2s32 = vsubq_s32(q2s32, q6s32);
|
||||
|
||||
d22s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d23s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d22s16 = vrshrn_n_s32(q11s32, 14);
|
||||
d23s16 = vrshrn_n_s32(q12s32, 14);
|
||||
*q11s16 = vcombine_s16(d22s16, d23s16);
|
||||
|
||||
q12s32 = vaddq_s32(q3s32, q7s32);
|
||||
@ -251,12 +251,12 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q3s32 = vsubq_s32(q3s32, q7s32);
|
||||
q4s32 = vsubq_s32(q4s32, q8s32);
|
||||
|
||||
d2s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d3s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d24s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d25s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d6s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d7s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d2s16 = vrshrn_n_s32(q1s32, 14);
|
||||
d3s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d24s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d25s16 = vrshrn_n_s32(q15s32, 14);
|
||||
d6s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d7s16 = vrshrn_n_s32(q4s32, 14);
|
||||
*q12s16 = vcombine_s16(d24s16, d25s16);
|
||||
|
||||
d0s16 = vdup_n_s16(cospi_10_64);
|
||||
@ -291,10 +291,10 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q2s32 = vsubq_s32(q2s32, q10s32);
|
||||
q6s32 = vsubq_s32(q6s32, q9s32);
|
||||
|
||||
d28s16 = vqrshrn_n_s32(q14s32, 14);
|
||||
d29s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d4s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d28s16 = vrshrn_n_s32(q14s32, 14);
|
||||
d29s16 = vrshrn_n_s32(q15s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q6s32, 14);
|
||||
*q14s16 = vcombine_s16(d28s16, d29s16);
|
||||
|
||||
q9s32 = vaddq_s32(q4s32, q0s32);
|
||||
@ -305,10 +305,10 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
d30s16 = vdup_n_s16(cospi_8_64);
|
||||
d31s16 = vdup_n_s16(cospi_24_64);
|
||||
|
||||
d18s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d19s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d8s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d9s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d18s16 = vrshrn_n_s32(q9s32, 14);
|
||||
d19s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d8s16 = vrshrn_n_s32(q4s32, 14);
|
||||
d9s16 = vrshrn_n_s32(q5s32, 14);
|
||||
*q9s16 = vcombine_s16(d18s16, d19s16);
|
||||
|
||||
q5s32 = vmull_s16(d2s16, d30s16);
|
||||
@ -341,10 +341,10 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q5s32 = vsubq_s32(q5s32, q1s32);
|
||||
q6s32 = vsubq_s32(q6s32, q3s32);
|
||||
|
||||
d18s16 = vqrshrn_n_s32(q14s32, 14);
|
||||
d19s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d18s16 = vrshrn_n_s32(q14s32, 14);
|
||||
d19s16 = vrshrn_n_s32(q15s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q6s32, 14);
|
||||
*q9s16 = vcombine_s16(d18s16, d19s16);
|
||||
|
||||
q1s32 = vaddq_s32(q7s32, q10s32);
|
||||
@ -352,10 +352,10 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q7s32 = vsubq_s32(q7s32, q10s32);
|
||||
q0s32 = vsubq_s32(q0s32, q2s32);
|
||||
|
||||
d28s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d29s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d14s16 = vqrshrn_n_s32(q7s32, 14);
|
||||
d15s16 = vqrshrn_n_s32(q0s32, 14);
|
||||
d28s16 = vrshrn_n_s32(q1s32, 14);
|
||||
d29s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d14s16 = vrshrn_n_s32(q7s32, 14);
|
||||
d15s16 = vrshrn_n_s32(q0s32, 14);
|
||||
*q14s16 = vcombine_s16(d28s16, d29s16);
|
||||
|
||||
d30s16 = vdup_n_s16(cospi_16_64);
|
||||
@ -374,10 +374,10 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q13s32 = vmlsl_s16(q13s32, d24s16, d30s16);
|
||||
q1s32 = vmlsl_s16(q1s32, d25s16, d30s16);
|
||||
|
||||
d4s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d24s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d25s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d24s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d25s16 = vrshrn_n_s32(q1s32, 14);
|
||||
q2s16 = vcombine_s16(d4s16, d5s16);
|
||||
*q12s16 = vcombine_s16(d24s16, d25s16);
|
||||
|
||||
@ -391,10 +391,10 @@ static INLINE void IADST8X8_1D(int16x8_t *q8s16, int16x8_t *q9s16,
|
||||
q11s32 = vmlsl_s16(q11s32, d14s16, d30s16);
|
||||
q0s32 = vmlsl_s16(q0s32, d15s16, d30s16);
|
||||
|
||||
d20s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d21s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q0s32, 14);
|
||||
d20s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d21s16 = vrshrn_n_s32(q1s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q11s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q0s32, 14);
|
||||
*q10s16 = vcombine_s16(d20s16, d21s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
|
@ -100,12 +100,12 @@
|
||||
vdup.16 d3, r12 ; duplicate cospi_20_64
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d8, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d9, q3, #14 ; >> 14
|
||||
vrshrn.s32 d8, q2, #14 ; >> 14
|
||||
vrshrn.s32 d9, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d14, q5, #14 ; >> 14
|
||||
vqrshrn.s32 d15, q6, #14 ; >> 14
|
||||
vrshrn.s32 d14, q5, #14 ; >> 14
|
||||
vrshrn.s32 d15, q6, #14 ; >> 14
|
||||
|
||||
; preloading to avoid stall
|
||||
; cospi_16_64 = 11585
|
||||
@ -131,12 +131,12 @@
|
||||
vmlal.s16 q15, d23, d2
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d10, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q3, #14 ; >> 14
|
||||
vrshrn.s32 d10, q2, #14 ; >> 14
|
||||
vrshrn.s32 d11, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q15, #14 ; >> 14
|
||||
vrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vrshrn.s32 d13, q15, #14 ; >> 14
|
||||
|
||||
; stage 4
|
||||
vdup.16 d30, r3 ; cospi_16_64
|
||||
@ -164,12 +164,12 @@
|
||||
vsub.s32 q1, q11, q1
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d16, q3, #14 ; >> 14
|
||||
vqrshrn.s32 d17, q12, #14 ; >> 14
|
||||
vrshrn.s32 d16, q3, #14 ; >> 14
|
||||
vrshrn.s32 d17, q12, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d18, q13, #14 ; >> 14
|
||||
vqrshrn.s32 d19, q1, #14 ; >> 14
|
||||
vrshrn.s32 d18, q13, #14 ; >> 14
|
||||
vrshrn.s32 d19, q1, #14 ; >> 14
|
||||
|
||||
; step1[2] * cospi_24_64 - step1[3] * cospi_8_64;
|
||||
; step1[2] * cospi_8_64
|
||||
@ -189,12 +189,12 @@
|
||||
vmlsl.s16 q13, d29, d31
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d22, q0, #14 ; >> 14
|
||||
vqrshrn.s32 d23, q1, #14 ; >> 14
|
||||
vrshrn.s32 d22, q0, #14 ; >> 14
|
||||
vrshrn.s32 d23, q1, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d20, q12, #14 ; >> 14
|
||||
vqrshrn.s32 d21, q13, #14 ; >> 14
|
||||
vrshrn.s32 d20, q12, #14 ; >> 14
|
||||
vrshrn.s32 d21, q13, #14 ; >> 14
|
||||
|
||||
vsub.s16 q13, q4, q5 ; step2[5] = step1[4] - step1[5];
|
||||
vadd.s16 q4, q4, q5 ; step2[4] = step1[4] + step1[5];
|
||||
@ -229,15 +229,15 @@
|
||||
vadd.s32 q10, q10, q12
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d10, q6, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q13, #14 ; >> 14
|
||||
vrshrn.s32 d10, q6, #14 ; >> 14
|
||||
vrshrn.s32 d11, q13, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q10, #14 ; >> 14
|
||||
vrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vrshrn.s32 d13, q10, #14 ; >> 14
|
||||
|
||||
; stage 6
|
||||
vadd.s16 q8, q0, q15 ; step2[0] = step1[0] + step1[7];
|
||||
vadd.s16 q8, q0, q15 ; step2[0] = step1[0] + step1[7];
|
||||
vadd.s16 q9, q1, q6 ; step2[1] = step1[1] + step1[6];
|
||||
vadd.s16 q10, q2, q5 ; step2[2] = step1[2] + step1[5];
|
||||
vadd.s16 q11, q3, q4 ; step2[3] = step1[3] + step1[4];
|
||||
@ -327,12 +327,12 @@
|
||||
vdup.16 d31, r12 ; duplicate cospi_18_64
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d0, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d1, q3, #14 ; >> 14
|
||||
vrshrn.s32 d0, q2, #14 ; >> 14
|
||||
vrshrn.s32 d1, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d14, q1, #14 ; >> 14
|
||||
vqrshrn.s32 d15, q4, #14 ; >> 14
|
||||
vrshrn.s32 d14, q1, #14 ; >> 14
|
||||
vrshrn.s32 d15, q4, #14 ; >> 14
|
||||
|
||||
; preloading to avoid stall
|
||||
; cospi_22_64 = 7723
|
||||
@ -361,12 +361,12 @@
|
||||
vdup.16 d31, r12 ; duplicate cospi_10_64
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d2, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d3, q3, #14 ; >> 14
|
||||
vrshrn.s32 d2, q2, #14 ; >> 14
|
||||
vrshrn.s32 d3, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d12, q4, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q5, #14 ; >> 14
|
||||
vrshrn.s32 d12, q4, #14 ; >> 14
|
||||
vrshrn.s32 d13, q5, #14 ; >> 14
|
||||
|
||||
; step1[10] * cospi_22_64
|
||||
vmull.s16 q11, d20, d30
|
||||
@ -395,12 +395,12 @@
|
||||
vdup.16 d31, r12 ; duplicate cospi_26_64
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d4, q11, #14 ; >> 14
|
||||
vqrshrn.s32 d5, q12, #14 ; >> 14
|
||||
vrshrn.s32 d4, q11, #14 ; >> 14
|
||||
vrshrn.s32 d5, q12, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d11, q5, #14 ; >> 14
|
||||
vqrshrn.s32 d10, q4, #14 ; >> 14
|
||||
vrshrn.s32 d11, q5, #14 ; >> 14
|
||||
vrshrn.s32 d10, q4, #14 ; >> 14
|
||||
|
||||
; step1[11] * cospi_6_64
|
||||
vmull.s16 q10, d28, d30
|
||||
@ -422,12 +422,12 @@
|
||||
vadd.s16 q0, q0, q1 ; step1[8]=step2[8]+step2[9]
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d6, q10, #14 ; >> 14
|
||||
vqrshrn.s32 d7, q11, #14 ; >> 14
|
||||
vrshrn.s32 d6, q10, #14 ; >> 14
|
||||
vrshrn.s32 d7, q11, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d8, q12, #14 ; >> 14
|
||||
vqrshrn.s32 d9, q13, #14 ; >> 14
|
||||
vrshrn.s32 d8, q12, #14 ; >> 14
|
||||
vrshrn.s32 d9, q13, #14 ; >> 14
|
||||
|
||||
; stage 3
|
||||
vsub.s16 q10, q3, q2 ; step1[10]=-step2[10]+step2[11]
|
||||
@ -468,12 +468,12 @@
|
||||
vdup.16 d30, r12 ; duplicate -cospi_8_64
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d12, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q3, #14 ; >> 14
|
||||
vrshrn.s32 d12, q2, #14 ; >> 14
|
||||
vrshrn.s32 d13, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d2, q4, #14 ; >> 14
|
||||
vqrshrn.s32 d3, q5, #14 ; >> 14
|
||||
vrshrn.s32 d2, q4, #14 ; >> 14
|
||||
vrshrn.s32 d3, q5, #14 ; >> 14
|
||||
|
||||
vmov.s16 q3, q11
|
||||
vmov.s16 q4, q12
|
||||
@ -495,12 +495,12 @@
|
||||
vmlal.s16 q9, d27, d31
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d4, q11, #14 ; >> 14
|
||||
vqrshrn.s32 d5, q12, #14 ; >> 14
|
||||
vrshrn.s32 d4, q11, #14 ; >> 14
|
||||
vrshrn.s32 d5, q12, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d10, q8, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q9, #14 ; >> 14
|
||||
vrshrn.s32 d10, q8, #14 ; >> 14
|
||||
vrshrn.s32 d11, q9, #14 ; >> 14
|
||||
|
||||
; stage 5
|
||||
vadd.s16 q8, q0, q3 ; step1[8] = step2[8]+step2[11];
|
||||
@ -535,12 +535,12 @@
|
||||
vadd.s32 q4, q4, q1
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d4, q5, #14 ; >> 14
|
||||
vqrshrn.s32 d5, q6, #14 ; >> 14
|
||||
vrshrn.s32 d4, q5, #14 ; >> 14
|
||||
vrshrn.s32 d5, q6, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d10, q10, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q4, #14 ; >> 14
|
||||
vrshrn.s32 d10, q10, #14 ; >> 14
|
||||
vrshrn.s32 d11, q4, #14 ; >> 14
|
||||
|
||||
; step1[11] * cospi_16_64
|
||||
vmull.s16 q0, d22, d14
|
||||
@ -559,12 +559,12 @@
|
||||
vadd.s32 q6, q6, q1
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d6, q10, #14 ; >> 14
|
||||
vqrshrn.s32 d7, q4, #14 ; >> 14
|
||||
vrshrn.s32 d6, q10, #14 ; >> 14
|
||||
vrshrn.s32 d7, q4, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d8, q13, #14 ; >> 14
|
||||
vqrshrn.s32 d9, q6, #14 ; >> 14
|
||||
vrshrn.s32 d8, q13, #14 ; >> 14
|
||||
vrshrn.s32 d9, q6, #14 ; >> 14
|
||||
|
||||
mov r4, #16 ; pass1_output stride
|
||||
ldr r3, [sp] ; load skip_adding
|
||||
@ -833,12 +833,12 @@ end_idct16x16_pass2
|
||||
vadd.s32 q10, q10, q12
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d11, q15, #14 ; >> 14
|
||||
vqrshrn.s32 d10, q6, #14 ; >> 14
|
||||
vrshrn.s32 d11, q15, #14 ; >> 14
|
||||
vrshrn.s32 d10, q6, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q10, #14 ; >> 14
|
||||
vrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vrshrn.s32 d13, q10, #14 ; >> 14
|
||||
|
||||
; stage 6
|
||||
vadd.s16 q2, q8, q7 ; step2[0] = step1[0] + step1[7];
|
||||
@ -950,12 +950,12 @@ end_idct16x16_pass2
|
||||
vdup.16 d30, r12 ; duplicate -cospi_8_64
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d2, q12, #14 ; >> 14
|
||||
vqrshrn.s32 d3, q5, #14 ; >> 14
|
||||
vrshrn.s32 d2, q12, #14 ; >> 14
|
||||
vrshrn.s32 d3, q5, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d12, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q11, #14 ; >> 14
|
||||
vrshrn.s32 d12, q2, #14 ; >> 14
|
||||
vrshrn.s32 d13, q11, #14 ; >> 14
|
||||
|
||||
; - step1[13] * cospi_8_64
|
||||
vmull.s16 q10, d8, d30
|
||||
@ -974,12 +974,12 @@ end_idct16x16_pass2
|
||||
vmlal.s16 q9, d9, d31
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d4, q10, #14 ; >> 14
|
||||
vqrshrn.s32 d5, q13, #14 ; >> 14
|
||||
vrshrn.s32 d4, q10, #14 ; >> 14
|
||||
vrshrn.s32 d5, q13, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d10, q8, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q9, #14 ; >> 14
|
||||
vrshrn.s32 d10, q8, #14 ; >> 14
|
||||
vrshrn.s32 d11, q9, #14 ; >> 14
|
||||
|
||||
; stage 5
|
||||
vadd.s16 q8, q0, q3 ; step1[8] = step2[8]+step2[11];
|
||||
@ -1014,12 +1014,12 @@ end_idct16x16_pass2
|
||||
vadd.s32 q1, q4, q1
|
||||
|
||||
; dct_const_round_shift(temp1)
|
||||
vqrshrn.s32 d4, q5, #14 ; >> 14
|
||||
vqrshrn.s32 d5, q6, #14 ; >> 14
|
||||
vrshrn.s32 d4, q5, #14 ; >> 14
|
||||
vrshrn.s32 d5, q6, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(temp2)
|
||||
vqrshrn.s32 d10, q0, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q1, #14 ; >> 14
|
||||
vrshrn.s32 d10, q0, #14 ; >> 14
|
||||
vrshrn.s32 d11, q1, #14 ; >> 14
|
||||
|
||||
; step1[11] * cospi_16_64
|
||||
vmull.s16 q0, d22, d14
|
||||
@ -1038,12 +1038,12 @@ end_idct16x16_pass2
|
||||
vadd.s32 q6, q6, q1
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d6, q10, #14 ; >> 14
|
||||
vqrshrn.s32 d7, q4, #14 ; >> 14
|
||||
vrshrn.s32 d6, q10, #14 ; >> 14
|
||||
vrshrn.s32 d7, q4, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift((step1[11] + step1[12]) * cospi_16_64);
|
||||
vqrshrn.s32 d8, q13, #14 ; >> 14
|
||||
vqrshrn.s32 d9, q6, #14 ; >> 14
|
||||
vrshrn.s32 d8, q13, #14 ; >> 14
|
||||
vrshrn.s32 d9, q6, #14 ; >> 14
|
||||
|
||||
mov r4, #16 ; pass1_output stride
|
||||
ldr r3, [sp] ; load skip_adding
|
||||
|
@ -85,10 +85,10 @@ void vpx_idct16x16_256_add_neon_pass1(const int16_t *in, int16_t *out) {
|
||||
d2s16 = vdup_n_s16((int16_t)cospi_12_64);
|
||||
d3s16 = vdup_n_s16((int16_t)cospi_20_64);
|
||||
|
||||
d8s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d9s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d14s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d15s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d8s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d9s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d14s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d15s16 = vrshrn_n_s32(q6s32, 14);
|
||||
q4s16 = vcombine_s16(d8s16, d9s16);
|
||||
q7s16 = vcombine_s16(d14s16, d15s16);
|
||||
|
||||
@ -102,10 +102,10 @@ void vpx_idct16x16_256_add_neon_pass1(const int16_t *in, int16_t *out) {
|
||||
q9s32 = vmlal_s16(q9s32, d22s16, d2s16);
|
||||
q15s32 = vmlal_s16(q15s32, d23s16, d2s16);
|
||||
|
||||
d10s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q15s32, 14);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -125,10 +125,10 @@ void vpx_idct16x16_256_add_neon_pass1(const int16_t *in, int16_t *out) {
|
||||
q13s32 = vsubq_s32(q2s32, q0s32);
|
||||
q1s32 = vsubq_s32(q11s32, q1s32);
|
||||
|
||||
d16s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d17s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d18s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d19s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d16s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d17s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d18s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d19s16 = vrshrn_n_s32(q1s32, 14);
|
||||
q8s16 = vcombine_s16(d16s16, d17s16);
|
||||
q9s16 = vcombine_s16(d18s16, d19s16);
|
||||
|
||||
@ -142,10 +142,10 @@ void vpx_idct16x16_256_add_neon_pass1(const int16_t *in, int16_t *out) {
|
||||
q12s32 = vmlsl_s16(q12s32, d28s16, d31s16);
|
||||
q13s32 = vmlsl_s16(q13s32, d29s16, d31s16);
|
||||
|
||||
d22s16 = vqrshrn_n_s32(q0s32, 14);
|
||||
d23s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d20s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d21s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d22s16 = vrshrn_n_s32(q0s32, 14);
|
||||
d23s16 = vrshrn_n_s32(q1s32, 14);
|
||||
d20s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d21s16 = vrshrn_n_s32(q13s32, 14);
|
||||
q10s16 = vcombine_s16(d20s16, d21s16);
|
||||
q11s16 = vcombine_s16(d22s16, d23s16);
|
||||
|
||||
@ -176,10 +176,10 @@ void vpx_idct16x16_256_add_neon_pass1(const int16_t *in, int16_t *out) {
|
||||
q9s32 = vaddq_s32(q9s32, q11s32);
|
||||
q10s32 = vaddq_s32(q10s32, q12s32);
|
||||
|
||||
d10s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q6s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q10s32, 14);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -289,10 +289,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q1s32 = vmlal_s16(q1s32, d30s16, d12s16);
|
||||
q4s32 = vmlal_s16(q4s32, d31s16, d12s16);
|
||||
|
||||
d0s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d1s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d14s16 = vqrshrn_n_s32(q1s32, 14);
|
||||
d15s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d0s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d1s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d14s16 = vrshrn_n_s32(q1s32, 14);
|
||||
d15s16 = vrshrn_n_s32(q4s32, 14);
|
||||
q0s16 = vcombine_s16(d0s16, d1s16);
|
||||
q7s16 = vcombine_s16(d14s16, d15s16);
|
||||
|
||||
@ -309,10 +309,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q4s32 = vmlal_s16(q4s32, d22s16, d30s16);
|
||||
q5s32 = vmlal_s16(q5s32, d23s16, d30s16);
|
||||
|
||||
d2s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d3s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d2s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d3s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q4s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q5s32, 14);
|
||||
q1s16 = vcombine_s16(d2s16, d3s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -329,10 +329,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q4s32 = vmlal_s16(q4s32, d26s16, d30s16);
|
||||
q5s32 = vmlal_s16(q5s32, d27s16, d30s16);
|
||||
|
||||
d4s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q11s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q4s32, 14);
|
||||
q2s16 = vcombine_s16(d4s16, d5s16);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
|
||||
@ -349,10 +349,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q12s32 = vmlal_s16(q12s32, d18s16, d30s16);
|
||||
q13s32 = vmlal_s16(q13s32, d19s16, d30s16);
|
||||
|
||||
d6s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d7s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d8s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d9s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d6s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d7s16 = vrshrn_n_s32(q11s32, 14);
|
||||
d8s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d9s16 = vrshrn_n_s32(q13s32, 14);
|
||||
q3s16 = vcombine_s16(d6s16, d7s16);
|
||||
q4s16 = vcombine_s16(d8s16, d9s16);
|
||||
|
||||
@ -389,10 +389,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q4s32 = vmlsl_s16(q4s32, d18s16, d30s16);
|
||||
q5s32 = vmlsl_s16(q5s32, d19s16, d30s16);
|
||||
|
||||
d12s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q3s32, 14);
|
||||
d2s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d3s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q3s32, 14);
|
||||
d2s16 = vrshrn_n_s32(q4s32, 14);
|
||||
d3s16 = vrshrn_n_s32(q5s32, 14);
|
||||
q1s16 = vcombine_s16(d2s16, d3s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -410,10 +410,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q8s32 = vmlal_s16(q8s32, d26s16, d31s16);
|
||||
q9s32 = vmlal_s16(q9s32, d27s16, d31s16);
|
||||
|
||||
d4s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q8s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q11s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q8s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q9s32, 14);
|
||||
q2s16 = vcombine_s16(d4s16, d5s16);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
|
||||
@ -449,10 +449,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q10s32 = vaddq_s32(q3s32, q0s32);
|
||||
q4s32 = vaddq_s32(q4s32, q1s32);
|
||||
|
||||
d4s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q6s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q4s32, 14);
|
||||
q2s16 = vcombine_s16(d4s16, d5s16);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
|
||||
@ -466,10 +466,10 @@ void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
|
||||
q13s32 = vaddq_s32(q13s32, q0s32);
|
||||
q6s32 = vaddq_s32(q6s32, q1s32);
|
||||
|
||||
d6s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d7s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d8s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d9s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d6s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d7s16 = vrshrn_n_s32(q4s32, 14);
|
||||
d8s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d9s16 = vrshrn_n_s32(q6s32, 14);
|
||||
q3s16 = vcombine_s16(d6s16, d7s16);
|
||||
q4s16 = vcombine_s16(d8s16, d9s16);
|
||||
|
||||
@ -823,10 +823,10 @@ void vpx_idct16x16_10_add_neon_pass1(const tran_low_t *in, int16_t *out) {
|
||||
q9s32 = vaddq_s32(q9s32, q11s32);
|
||||
q10s32 = vaddq_s32(q10s32, q12s32);
|
||||
|
||||
d11s16 = vqrshrn_n_s32(q15s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q15s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q6s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q9s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q10s32, 14);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -934,10 +934,10 @@ void vpx_idct16x16_10_add_neon_pass2(const tran_low_t *src, int16_t *out,
|
||||
q2s32 = vmlal_s16(q2s32, d14s16, d30s16);
|
||||
q11s32 = vmlal_s16(q11s32, d15s16, d30s16);
|
||||
|
||||
d2s16 = vqrshrn_n_s32(q12s32, 14);
|
||||
d3s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d12s16 = vqrshrn_n_s32(q2s32, 14);
|
||||
d13s16 = vqrshrn_n_s32(q11s32, 14);
|
||||
d2s16 = vrshrn_n_s32(q12s32, 14);
|
||||
d3s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d12s16 = vrshrn_n_s32(q2s32, 14);
|
||||
d13s16 = vrshrn_n_s32(q11s32, 14);
|
||||
q1s16 = vcombine_s16(d2s16, d3s16);
|
||||
q6s16 = vcombine_s16(d12s16, d13s16);
|
||||
|
||||
@ -952,10 +952,10 @@ void vpx_idct16x16_10_add_neon_pass2(const tran_low_t *src, int16_t *out,
|
||||
q8s32 = vmlal_s16(q8s32, d8s16, d31s16);
|
||||
q9s32 = vmlal_s16(q9s32, d9s16, d31s16);
|
||||
|
||||
d4s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q8s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q9s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q8s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q9s32, 14);
|
||||
q2s16 = vcombine_s16(d4s16, d5s16);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
|
||||
@ -990,10 +990,10 @@ void vpx_idct16x16_10_add_neon_pass2(const tran_low_t *src, int16_t *out,
|
||||
q0s32 = vaddq_s32(q3s32, q0s32);
|
||||
q4s32 = vaddq_s32(q4s32, q1s32);
|
||||
|
||||
d4s16 = vqrshrn_n_s32(q5s32, 14);
|
||||
d5s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d10s16 = vqrshrn_n_s32(q0s32, 14);
|
||||
d11s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d4s16 = vrshrn_n_s32(q5s32, 14);
|
||||
d5s16 = vrshrn_n_s32(q6s32, 14);
|
||||
d10s16 = vrshrn_n_s32(q0s32, 14);
|
||||
d11s16 = vrshrn_n_s32(q4s32, 14);
|
||||
q2s16 = vcombine_s16(d4s16, d5s16);
|
||||
q5s16 = vcombine_s16(d10s16, d11s16);
|
||||
|
||||
@ -1007,10 +1007,10 @@ void vpx_idct16x16_10_add_neon_pass2(const tran_low_t *src, int16_t *out,
|
||||
q13s32 = vaddq_s32(q13s32, q0s32);
|
||||
q6s32 = vaddq_s32(q6s32, q1s32);
|
||||
|
||||
d6s16 = vqrshrn_n_s32(q10s32, 14);
|
||||
d7s16 = vqrshrn_n_s32(q4s32, 14);
|
||||
d8s16 = vqrshrn_n_s32(q13s32, 14);
|
||||
d9s16 = vqrshrn_n_s32(q6s32, 14);
|
||||
d6s16 = vrshrn_n_s32(q10s32, 14);
|
||||
d7s16 = vrshrn_n_s32(q4s32, 14);
|
||||
d8s16 = vrshrn_n_s32(q13s32, 14);
|
||||
d9s16 = vrshrn_n_s32(q6s32, 14);
|
||||
q3s16 = vcombine_s16(d6s16, d7s16);
|
||||
q4s16 = vcombine_s16(d8s16, d9s16);
|
||||
|
||||
|
@ -146,8 +146,8 @@ static INLINE void DO_BUTTERFLY(int16x8_t q14s16, int16x8_t q13s16,
|
||||
q11s32 = vaddq_s32(q12s32, q11s32);
|
||||
q10s32 = vaddq_s32(q10s32, q15s32);
|
||||
|
||||
*qAs16 = vcombine_s16(vqrshrn_n_s32(q8s32, 14), vqrshrn_n_s32(q9s32, 14));
|
||||
*qBs16 = vcombine_s16(vqrshrn_n_s32(q11s32, 14), vqrshrn_n_s32(q10s32, 14));
|
||||
*qAs16 = vcombine_s16(vrshrn_n_s32(q8s32, 14), vrshrn_n_s32(q9s32, 14));
|
||||
*qBs16 = vcombine_s16(vrshrn_n_s32(q11s32, 14), vrshrn_n_s32(q10s32, 14));
|
||||
}
|
||||
|
||||
static INLINE void idct32_transpose_pair(const int16_t *input, int16_t *t_buf) {
|
||||
|
@ -88,10 +88,10 @@
|
||||
vmlal.s16 q1, d19, d22
|
||||
|
||||
; dct_const_round_shift
|
||||
vqrshrn.s32 d26, q13, #14
|
||||
vqrshrn.s32 d27, q14, #14
|
||||
vqrshrn.s32 d29, q15, #14
|
||||
vqrshrn.s32 d28, q1, #14
|
||||
vrshrn.s32 d26, q13, #14
|
||||
vrshrn.s32 d27, q14, #14
|
||||
vrshrn.s32 d29, q15, #14
|
||||
vrshrn.s32 d28, q1, #14
|
||||
|
||||
; stage 2
|
||||
; output[0] = step[0] + step[3];
|
||||
@ -139,10 +139,10 @@
|
||||
vmlal.s16 q1, d19, d22
|
||||
|
||||
; dct_const_round_shift
|
||||
vqrshrn.s32 d26, q13, #14
|
||||
vqrshrn.s32 d27, q14, #14
|
||||
vqrshrn.s32 d29, q15, #14
|
||||
vqrshrn.s32 d28, q1, #14
|
||||
vrshrn.s32 d26, q13, #14
|
||||
vrshrn.s32 d27, q14, #14
|
||||
vrshrn.s32 d29, q15, #14
|
||||
vrshrn.s32 d28, q1, #14
|
||||
|
||||
; stage 2
|
||||
; output[0] = step[0] + step[3];
|
||||
|
@ -47,12 +47,12 @@
|
||||
vmlsl.s16 q6, d23, d3
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d8, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d9, q3, #14 ; >> 14
|
||||
vrshrn.s32 d8, q2, #14 ; >> 14
|
||||
vrshrn.s32 d9, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d10, q5, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q6, #14 ; >> 14
|
||||
vrshrn.s32 d10, q5, #14 ; >> 14
|
||||
vrshrn.s32 d11, q6, #14 ; >> 14
|
||||
|
||||
; input[1] * cospi_4_64
|
||||
vmull.s16 q2, d18, d1
|
||||
@ -71,15 +71,15 @@
|
||||
vmlal.s16 q13, d23, d2
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d14, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d15, q3, #14 ; >> 14
|
||||
vrshrn.s32 d14, q2, #14 ; >> 14
|
||||
vrshrn.s32 d15, q3, #14 ; >> 14
|
||||
|
||||
; stage 2 & stage 3 - even half
|
||||
vdup.16 d0, r7 ; duplicate cospi_16_64
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q13, #14 ; >> 14
|
||||
vrshrn.s32 d12, q9, #14 ; >> 14
|
||||
vrshrn.s32 d13, q13, #14 ; >> 14
|
||||
|
||||
; input[0] * cospi_16_64
|
||||
vmull.s16 q2, d16, d0
|
||||
@ -101,12 +101,12 @@
|
||||
vdup.16 d1, r9 ; duplicate cospi_8_64
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d18, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d19, q3, #14 ; >> 14
|
||||
vrshrn.s32 d18, q2, #14 ; >> 14
|
||||
vrshrn.s32 d19, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d22, q13, #14 ; >> 14
|
||||
vqrshrn.s32 d23, q15, #14 ; >> 14
|
||||
vrshrn.s32 d22, q13, #14 ; >> 14
|
||||
vrshrn.s32 d23, q15, #14 ; >> 14
|
||||
|
||||
; input[1] * cospi_24_64 - input[3] * cospi_8_64
|
||||
; input[1] * cospi_24_64
|
||||
@ -126,12 +126,12 @@
|
||||
vmlal.s16 q12, d29, d0
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d26, q2, #14 ; >> 14
|
||||
vqrshrn.s32 d27, q3, #14 ; >> 14
|
||||
vrshrn.s32 d26, q2, #14 ; >> 14
|
||||
vrshrn.s32 d27, q3, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d30, q8, #14 ; >> 14
|
||||
vqrshrn.s32 d31, q12, #14 ; >> 14
|
||||
vrshrn.s32 d30, q8, #14 ; >> 14
|
||||
vrshrn.s32 d31, q12, #14 ; >> 14
|
||||
|
||||
vadd.s16 q0, q9, q15 ; output[0] = step[0] + step[3]
|
||||
vadd.s16 q1, q11, q13 ; output[1] = step[1] + step[2]
|
||||
@ -164,12 +164,12 @@
|
||||
vmlal.s16 q12, d27, d16
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d10, q9, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q10, #14 ; >> 14
|
||||
vrshrn.s32 d10, q9, #14 ; >> 14
|
||||
vrshrn.s32 d11, q10, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d12, q11, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q12, #14 ; >> 14
|
||||
vrshrn.s32 d12, q11, #14 ; >> 14
|
||||
vrshrn.s32 d13, q12, #14 ; >> 14
|
||||
|
||||
; stage 4
|
||||
vadd.s16 q8, q0, q7 ; output[0] = step1[0] + step1[7];
|
||||
@ -423,12 +423,12 @@
|
||||
vmlal.s16 q12, d27, d16
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d10, q9, #14 ; >> 14
|
||||
vqrshrn.s32 d11, q10, #14 ; >> 14
|
||||
vrshrn.s32 d10, q9, #14 ; >> 14
|
||||
vrshrn.s32 d11, q10, #14 ; >> 14
|
||||
|
||||
; dct_const_round_shift(input_dc * cospi_16_64)
|
||||
vqrshrn.s32 d12, q11, #14 ; >> 14
|
||||
vqrshrn.s32 d13, q12, #14 ; >> 14
|
||||
vrshrn.s32 d12, q11, #14 ; >> 14
|
||||
vrshrn.s32 d13, q12, #14 ; >> 14
|
||||
|
||||
; stage 4
|
||||
vadd.s16 q8, q0, q7 ; output[0] = step1[0] + step1[7];
|
||||
|
Loading…
Reference in New Issue
Block a user