Mercurial > libavcodec.hg
diff arm/dsputil_neon_s.S @ 8698:24a7b5d0eb27 libavcodec
ARM: NEON optimised vector_fmul_window
| author | mru |
|---|---|
| date | Fri, 30 Jan 2009 23:13:19 +0000 |
| parents | 307b176f91e7 |
| children | 9ea1ea6db616 |
line wrap: on
line diff
--- a/arm/dsputil_neon_s.S Fri Jan 30 23:13:15 2009 +0000 +++ b/arm/dsputil_neon_s.S Fri Jan 30 23:13:19 2009 +0000 @@ -649,3 +649,53 @@ 3: vst1.64 {d16-d19},[r3,:128]! bx lr .endfunc + +function ff_vector_fmul_window_neon, export=1 + vld1.32 {d16[],d17[]}, [sp,:32] + push {r4,r5,lr} + ldr lr, [sp, #16] + sub r2, r2, #8 + sub r5, lr, #2 + add r2, r2, r5, lsl #2 + add r4, r3, r5, lsl #3 + add ip, r0, r5, lsl #3 + mov r5, #-16 + vld1.64 {d0,d1}, [r1,:128]! + vld1.64 {d2,d3}, [r2,:128], r5 + vld1.64 {d4,d5}, [r3,:128]! + vld1.64 {d6,d7}, [r4,:128], r5 +1: subs lr, lr, #4 + vmov q11, q8 + vmla.f32 d22, d0, d4 + vmov q10, q8 + vmla.f32 d23, d1, d5 + vrev64.32 q3, q3 + vmla.f32 d20, d0, d7 + vrev64.32 q1, q1 + vmla.f32 d21, d1, d6 + beq 2f + vmla.f32 d22, d3, d7 + vld1.64 {d0,d1}, [r1,:128]! + vmla.f32 d23, d2, d6 + vld1.64 {d18,d19},[r2,:128], r5 + vmls.f32 d20, d3, d4 + vld1.64 {d24,d25},[r3,:128]! + vmls.f32 d21, d2, d5 + vld1.64 {d6,d7}, [r4,:128], r5 + vmov q1, q9 + vrev64.32 q11, q11 + vmov q2, q12 + vswp d22, d23 + vst1.64 {d20,d21},[r0,:128]! + vst1.64 {d22,d23},[ip,:128], r5 + b 1b +2: vmla.f32 d22, d3, d7 + vmla.f32 d23, d2, d6 + vmls.f32 d20, d3, d4 + vmls.f32 d21, d2, d5 + vrev64.32 q11, q11 + vswp d22, d23 + vst1.64 {d20,d21},[r0,:128]! + vst1.64 {d22,d23},[ip,:128], r5 + pop {r4,r5,pc} + .endfunc