diff options
| author | Tero Rintaluoma <teror@google.com> | 2011-03-30 13:45:59 +0300 |
|---|---|---|
| committer | Tero Rintaluoma <teror@google.com> | 2011-04-01 10:06:44 +0300 |
| commit | cec76a36d687ceb62b3cb27095ae3a3e0c98ac42 (patch) | |
| tree | 12f185cf5df90598ab70720f024544e35ce7432d /vp8/encoder/arm/neon | |
| parent | 4be062bbc3466cfd542a7485f5a2a8eadac4f24e (diff) | |
| download | libvpx-cec76a36d687ceb62b3cb27095ae3a3e0c98ac42.tar libvpx-cec76a36d687ceb62b3cb27095ae3a3e0c98ac42.tar.gz libvpx-cec76a36d687ceb62b3cb27095ae3a3e0c98ac42.tar.bz2 libvpx-cec76a36d687ceb62b3cb27095ae3a3e0c98ac42.zip | |
Wrapper function removed from vp8_subtract_b_neon function call
Address calculations moved from encodemb_arm.c file to neon
optimized assembly function to save cycles in function calls.
- vp8_subtract_b_neon_func replaced with vp8_subtract_b_neon
that contains all needed address calculations
- unnecessary file encodemb_arm.c removed
- consistent with ARMv6 optimized version
Change-Id: I6cbc1a2670b56c2077f59995fcf8f70786b4990b
Diffstat (limited to 'vp8/encoder/arm/neon')
| -rw-r--r-- | vp8/encoder/arm/neon/subtract_neon.asm | 53 |
1 files changed, 33 insertions, 20 deletions
diff --git a/vp8/encoder/arm/neon/subtract_neon.asm b/vp8/encoder/arm/neon/subtract_neon.asm index 3ea00f8b9..68c295062 100644 --- a/vp8/encoder/arm/neon/subtract_neon.asm +++ b/vp8/encoder/arm/neon/subtract_neon.asm @@ -8,45 +8,58 @@ ; be found in the AUTHORS file in the root of the source tree. ; - - EXPORT |vp8_subtract_b_neon_func| + EXPORT |vp8_subtract_b_neon| EXPORT |vp8_subtract_mby_neon| EXPORT |vp8_subtract_mbuv_neon| + INCLUDE asm_enc_offsets.asm + ARM REQUIRE8 PRESERVE8 AREA ||.text||, CODE, READONLY, ALIGN=2 -;========================================= -;void vp8_subtract_b_neon_func(short *diff, unsigned char *src, unsigned char *pred, int stride, int pitch); -|vp8_subtract_b_neon_func| PROC - ldr r12, [sp] ;load pitch - - vld1.8 {d0}, [r1], r3 ;load src - vld1.8 {d1}, [r2], r12 ;load pred - vld1.8 {d2}, [r1], r3 - vld1.8 {d3}, [r2], r12 - vld1.8 {d4}, [r1], r3 - vld1.8 {d5}, [r2], r12 - vld1.8 {d6}, [r1], r3 - vld1.8 {d7}, [r2], r12 + +;void vp8_subtract_b_neon(BLOCK *be, BLOCKD *bd, int pitch) +|vp8_subtract_b_neon| PROC + + stmfd sp!, {r4-r7} + + ldr r3, [r0, #vp8_block_base_src] + ldr r4, [r0, #vp8_block_src] + ldr r5, [r0, #vp8_block_src_diff] + ldr r3, [r3] + ldr r6, [r0, #vp8_block_src_stride] + add r3, r3, r4 ; src = *base_src + src + ldr r7, [r1, #vp8_blockd_predictor] + + vld1.8 {d0}, [r3], r6 ;load src + vld1.8 {d1}, [r7], r2 ;load pred + vld1.8 {d2}, [r3], r6 + vld1.8 {d3}, [r7], r2 + vld1.8 {d4}, [r3], r6 + vld1.8 {d5}, [r7], r2 + vld1.8 {d6}, [r3], r6 + vld1.8 {d7}, [r7], r2 vsubl.u8 q10, d0, d1 vsubl.u8 q11, d2, d3 vsubl.u8 q12, d4, d5 vsubl.u8 q13, d6, d7 - mov r12, r12, lsl #1 + mov r2, r2, lsl #1 - vst1.16 {d20}, [r0], r12 ;store diff - vst1.16 {d22}, [r0], r12 - vst1.16 {d24}, [r0], r12 - vst1.16 {d26}, [r0], r12 + vst1.16 {d20}, [r5], r2 ;store diff + vst1.16 {d22}, [r5], r2 + vst1.16 {d24}, [r5], r2 + vst1.16 {d26}, [r5], r2 + ldmfd sp!, {r4-r7} bx lr + ENDP + ;========================================== ;void vp8_subtract_mby_neon(short *diff, unsigned char *src, unsigned char *pred, int stride) |vp8_subtract_mby_neon| PROC |