generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

idct16x16,NEON,cosmetics: normalize fn signatures

Browse Source 
+ remove unused parameters from vpx_idct16x16_10_add_neon_pass2

Change-Id: Ie5912a4abdd308fab589380bca054a2e7234a2c4
This commit is contained in:
James Zern 2016-11-23 17:21:26 -08:00
parent 12566c3d0f
commit c6641782c3
3 changed files with 86 additions and 97 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
vpx_dsp/arm/idct16x16_add_neon.asm
View File

@ -36,12 +36,12 @@
MEND
AREA Block, CODE, READONLY ; name this block of code
;void |vpx_idct16x16_256_add_neon_pass1|(int16_t *input,
;void |vpx_idct16x16_256_add_neon_pass1|(const int16_t *input,
; int16_t *output, int output_stride)
;
; r0 int16_t input
; r0 const int16_t *input
; r1 int16_t *output
; r2 int output_stride)
; r2 int output_stride
; idct16 stage1 - stage6 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
@ -267,19 +267,19 @@
bx lr
ENDP ; |vpx_idct16x16_256_add_neon_pass1|
;void vpx_idct16x16_256_add_neon_pass2(int16_t *src,
;void vpx_idct16x16_256_add_neon_pass2(const int16_t *src,
; int16_t *output,
; int16_t *pass1Output,
; int16_t *pass1_output,
; int16_t skip_adding,
; uint8_t *dest,
; int dest_stride)
;
; r0 int16_t *src
; r1 int16_t *output,
; r2 int16_t *pass1Output,
; r3 int16_t skip_adding,
; r4 uint8_t *dest,
; r5 int dest_stride)
; r0 const int16_t *src
; r1 int16_t *output
; r2 int16_t *pass1_output
; r3 int16_t skip_adding
; r4 uint8_t *dest
; r5 int dest_stride
; idct16 stage1 - stage7 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
@ -578,7 +578,7 @@
vqrshrn.s32 d8, q13, #14 ; >> 14
vqrshrn.s32 d9, q6, #14 ; >> 14
mov r4, #16 ; pass1Output stride
mov r4, #16 ; pass1_output stride
ldr r3, [sp] ; load skip_adding
cmp r3, #0 ; check if need adding dest data
beq skip_adding_dest
@ -767,12 +767,12 @@ end_idct16x16_pass2
bx lr
ENDP ; |vpx_idct16x16_256_add_neon_pass2|
;void |vpx_idct16x16_10_add_neon_pass1|(int16_t *input,
;void |vpx_idct16x16_10_add_neon_pass1|(const int16_t *input,
; int16_t *output, int output_stride)
;
; r0 int16_t input
; r0 const int16_t *input
; r1 int16_t *output
; r2 int output_stride)
; r2 int output_stride
; idct16 stage1 - stage6 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
@ -884,19 +884,12 @@ end_idct16x16_pass2
bx lr
ENDP ; |vpx_idct16x16_10_add_neon_pass1|
;void vpx_idct16x16_10_add_neon_pass2(int16_t *src,
; int16_t *output,
; int16_t *pass1Output,
; int16_t skip_adding,
; uint8_t *dest,
; int dest_stride)
;void vpx_idct16x16_10_add_neon_pass2(const int16_t *src, int16_t *output,
; int16_t *pass1_output)
;
; r0 int16_t *src
; r1 int16_t *output,
; r2 int16_t *pass1Output,
; r3 int16_t skip_adding,
; r4 uint8_t *dest,
; r5 int dest_stride)
; r0 const int16_t *src
; r1 int16_t *output
; r2 int16_t *pass1_output
; idct16 stage1 - stage7 on all the elements loaded in q8-q15. The output
; will be stored back into q8-q15 registers. This function will touch q0-q7
@ -1076,7 +1069,7 @@ end_idct16x16_pass2
vqrshrn.s32 d8, q13, #14 ; >> 14
vqrshrn.s32 d9, q6, #14 ; >> 14
mov r4, #16 ; pass1Output stride
mov r4, #16 ; pass1_output stride
ldr r3, [sp] ; load skip_adding
; stage 7

111
vpx_dsp/arm/idct16x16_add_neon.c
View File

@ -14,7 +14,7 @@
#include "vpx_dsp/arm/transpose_neon.h"
#include "vpx_dsp/txfm_common.h"
void vpx_idct16x16_256_add_neon_pass1(int16_t *in, int16_t *out,
void vpx_idct16x16_256_add_neon_pass1(const int16_t *in, int16_t *out,
int output_stride) {
int16x4_t d0s16, d1s16, d2s16, d3s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
@ -249,9 +249,10 @@ void vpx_idct16x16_256_add_neon_pass1(int16_t *in, int16_t *out,
vst1_u64((uint64_t *)out, d31u64);
}
void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
int16_t *pass1Output, int16_t skip_adding,
uint8_t *dest, int dest_stride) {
void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *out,
int16_t *pass1_output,
int16_t skip_adding, uint8_t *dest,
int dest_stride) {
uint8_t *d;
uint8x8_t d12u8, d13u8;
int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
@ -514,10 +515,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
if (skip_adding != 0) {
d = dest;
// load the data in pass1
q0s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q1s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q0s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q1s16 = vld1q_s16(pass1_output);
pass1_output += 8;
d12s64 = vld1_s64((int64_t *)dest);
dest += dest_stride;
d13s64 = vld1_s64((int64_t *)dest);
@ -540,10 +541,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q14s16 = vsubq_s16(q1s16, q14s16);
q15s16 = vsubq_s16(q0s16, q15s16);
q10s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q11s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q10s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q11s16 = vld1q_s16(pass1_output);
pass1_output += 8;
d12s64 = vld1_s64((int64_t *)dest);
dest += dest_stride;
d13s64 = vld1_s64((int64_t *)dest);
@ -565,10 +566,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q4s16 = vsubq_s16(q11s16, q4s16);
q5s16 = vsubq_s16(q10s16, q5s16);
q0s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q1s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q0s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q1s16 = vld1q_s16(pass1_output);
pass1_output += 8;
d12s64 = vld1_s64((int64_t *)dest);
dest += dest_stride;
d13s64 = vld1_s64((int64_t *)dest);
@ -590,9 +591,9 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q2s16 = vsubq_s16(q1s16, q2s16);
q3s16 = vsubq_s16(q0s16, q3s16);
q10s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q11s16 = vld1q_s16(pass1Output);
q10s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q11s16 = vld1q_s16(pass1_output);
d12s64 = vld1_s64((int64_t *)dest);
dest += dest_stride;
d13s64 = vld1_s64((int64_t *)dest);
@ -679,10 +680,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
d12u8 = vqmovun_s16(vreinterpretq_s16_u16(q15u16));
vst1_u64((uint64_t *)d, vreinterpret_u64_u8(d12u8));
} else { // skip_adding_dest
q0s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q1s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q0s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q1s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q0s16, q15s16);
q13s16 = vaddq_s16(q1s16, q14s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -700,10 +701,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q14s16 = vsubq_s16(q1s16, q14s16);
q15s16 = vsubq_s16(q0s16, q15s16);
q10s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q11s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q10s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q11s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q10s16, q5s16);
q13s16 = vaddq_s16(q11s16, q4s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -721,10 +722,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q4s16 = vsubq_s16(q11s16, q4s16);
q5s16 = vsubq_s16(q10s16, q5s16);
q0s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q1s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q0s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q1s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q0s16, q3s16);
q13s16 = vaddq_s16(q1s16, q2s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -742,10 +743,10 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
q2s16 = vsubq_s16(q1s16, q2s16);
q3s16 = vsubq_s16(q0s16, q3s16);
q10s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q11s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q10s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q11s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q10s16, q9s16);
q13s16 = vaddq_s16(q11s16, q8s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -797,7 +798,7 @@ void vpx_idct16x16_256_add_neon_pass2(int16_t *src, int16_t *out,
}
}
void vpx_idct16x16_10_add_neon_pass1(int16_t *in, int16_t *out,
void vpx_idct16x16_10_add_neon_pass1(const int16_t *in, int16_t *out,
int output_stride) {
int16x4_t d4s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
@ -932,9 +933,8 @@ void vpx_idct16x16_10_add_neon_pass1(int16_t *in, int16_t *out,
vst1_u64((uint64_t *)out, d31u64);
}
void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
int16_t *pass1Output, int16_t skip_adding,
uint8_t *dest, int dest_stride) {
void vpx_idct16x16_10_add_neon_pass2(const int16_t *src, int16_t *out,
int16_t *pass1_output) {
int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
int16x4_t d8s16, d9s16, d10s16, d11s16, d12s16, d13s16, d14s16, d15s16;
int16x4_t d20s16, d21s16, d22s16, d23s16;
@ -947,9 +947,6 @@ void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
int32x4_t q0s32, q1s32, q2s32, q3s32, q4s32, q5s32, q6s32, q8s32, q9s32;
int32x4_t q10s32, q11s32, q12s32, q13s32;
int16x8x2_t q0x2s16;
(void)skip_adding;
(void)dest;
(void)dest_stride;
q0x2s16 = vld2q_s16(src);
q8s16 = q0x2s16.val[0];
@ -1093,10 +1090,10 @@ void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
q4s16 = vcombine_s16(d8s16, d9s16);
// stage 7
q0s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q1s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q0s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q1s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q0s16, q15s16);
q13s16 = vaddq_s16(q1s16, q14s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -1114,10 +1111,10 @@ void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
q14s16 = vsubq_s16(q1s16, q14s16);
q15s16 = vsubq_s16(q0s16, q15s16);
q10s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q11s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q10s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q11s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q10s16, q5s16);
q13s16 = vaddq_s16(q11s16, q4s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -1135,10 +1132,10 @@ void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
q4s16 = vsubq_s16(q11s16, q4s16);
q5s16 = vsubq_s16(q10s16, q5s16);
q0s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q1s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q0s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q1s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q12s16 = vaddq_s16(q0s16, q3s16);
q13s16 = vaddq_s16(q1s16, q2s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));
@ -1156,9 +1153,9 @@ void vpx_idct16x16_10_add_neon_pass2(int16_t *src, int16_t *out,
q2s16 = vsubq_s16(q1s16, q2s16);
q3s16 = vsubq_s16(q0s16, q3s16);
q10s16 = vld1q_s16(pass1Output);
pass1Output += 8;
q11s16 = vld1q_s16(pass1Output);
q10s16 = vld1q_s16(pass1_output);
pass1_output += 8;
q11s16 = vld1q_s16(pass1_output);
q12s16 = vaddq_s16(q10s16, q9s16);
q13s16 = vaddq_s16(q11s16, q8s16);
d24u64 = vreinterpret_u64_s16(vget_low_s16(q12s16));

11
vpx_dsp/arm/idct16x16_neon.c
View File

@ -14,13 +14,13 @@
void vpx_idct16x16_256_add_neon_pass1(const int16_t *input, int16_t *output,
int output_stride);
void vpx_idct16x16_256_add_neon_pass2(const int16_t *src, int16_t *output,
int16_t *pass1Output, int16_t skip_adding,
uint8_t *dest, int dest_stride);
int16_t *pass1_output,
int16_t skip_adding, uint8_t *dest,
int dest_stride);
void vpx_idct16x16_10_add_neon_pass1(const int16_t *input, int16_t *output,
int output_stride);
void vpx_idct16x16_10_add_neon_pass2(const int16_t *src, int16_t *output,
int16_t *pass1Output, int16_t skip_adding,
uint8_t *dest, int dest_stride);
int16_t *pass1_output);
#if HAVE_NEON_ASM
/* For ARM NEON, d8-d15 are callee-saved registers, and need to be saved. */
@ -113,8 +113,7 @@ void vpx_idct16x16_10_add_neon(const int16_t *input, uint8_t *dest,
// Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
// with result in pass1(pass1_output) to calculate final result in stage 7
// which will be saved into row_idct_output.
vpx_idct16x16_10_add_neon_pass2(input + 1, row_idct_output, pass1_output, 0,
dest, dest_stride);
vpx_idct16x16_10_add_neon_pass2(input + 1, row_idct_output, pass1_output);
/* Skip Parallel idct on the lower 8 rows as they are all 0s */