diff options
| author | Justin Ruggles <justin.ruggles@gmail.com> | 2011-01-13 15:28:06 -0500 |
|---|---|---|
| committer | Mans Rullgard <mans@mansr.com> | 2011-01-22 17:53:27 +0000 |
| commit | 6eabb0d3ad42b91c1b4c298718c29961f7c1653a (patch) | |
| tree | 0cb7ebc7b25fcb4bf3f91fe2735ff9f264dff015 /libavcodec/arm/dsputil_neon.S | |
| parent | fcb7e535dd9ad142c079af62af9c1d0f4b001057 (diff) | |
Change DSPContext.vector_fmul() from dst=dst*src to dest=src0*src1.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Diffstat (limited to 'libavcodec/arm/dsputil_neon.S')
| -rw-r--r-- | libavcodec/arm/dsputil_neon.S | 45 |
1 files changed, 22 insertions, 23 deletions
diff --git a/libavcodec/arm/dsputil_neon.S b/libavcodec/arm/dsputil_neon.S index 2bcdb397f9..42fb38de52 100644 --- a/libavcodec/arm/dsputil_neon.S +++ b/libavcodec/arm/dsputil_neon.S @@ -738,42 +738,41 @@ function ff_float_to_int16_interleave_neon, export=1 endfunc function ff_vector_fmul_neon, export=1 - mov r3, r0 - subs r2, r2, #8 - vld1.64 {d0-d3}, [r0,:128]! - vld1.64 {d4-d7}, [r1,:128]! + subs r3, r3, #8 + vld1.64 {d0-d3}, [r1,:128]! + vld1.64 {d4-d7}, [r2,:128]! vmul.f32 q8, q0, q2 vmul.f32 q9, q1, q3 beq 3f - bics ip, r2, #15 + bics ip, r3, #15 beq 2f 1: subs ip, ip, #16 - vld1.64 {d0-d1}, [r0,:128]! - vld1.64 {d4-d5}, [r1,:128]! + vld1.64 {d0-d1}, [r1,:128]! + vld1.64 {d4-d5}, [r2,:128]! vmul.f32 q10, q0, q2 - vld1.64 {d2-d3}, [r0,:128]! - vld1.64 {d6-d7}, [r1,:128]! + vld1.64 {d2-d3}, [r1,:128]! + vld1.64 {d6-d7}, [r2,:128]! vmul.f32 q11, q1, q3 - vst1.64 {d16-d19},[r3,:128]! - vld1.64 {d0-d1}, [r0,:128]! - vld1.64 {d4-d5}, [r1,:128]! + vst1.64 {d16-d19},[r0,:128]! + vld1.64 {d0-d1}, [r1,:128]! + vld1.64 {d4-d5}, [r2,:128]! vmul.f32 q8, q0, q2 - vld1.64 {d2-d3}, [r0,:128]! - vld1.64 {d6-d7}, [r1,:128]! + vld1.64 {d2-d3}, [r1,:128]! + vld1.64 {d6-d7}, [r2,:128]! vmul.f32 q9, q1, q3 - vst1.64 {d20-d23},[r3,:128]! + vst1.64 {d20-d23},[r0,:128]! bne 1b - ands r2, r2, #15 + ands r3, r3, #15 beq 3f -2: vld1.64 {d0-d1}, [r0,:128]! - vld1.64 {d4-d5}, [r1,:128]! - vst1.64 {d16-d17},[r3,:128]! +2: vld1.64 {d0-d1}, [r1,:128]! + vld1.64 {d4-d5}, [r2,:128]! + vst1.64 {d16-d17},[r0,:128]! vmul.f32 q8, q0, q2 - vld1.64 {d2-d3}, [r0,:128]! - vld1.64 {d6-d7}, [r1,:128]! - vst1.64 {d18-d19},[r3,:128]! + vld1.64 {d2-d3}, [r1,:128]! + vld1.64 {d6-d7}, [r2,:128]! + vst1.64 {d18-d19},[r0,:128]! vmul.f32 q9, q1, q3 -3: vst1.64 {d16-d19},[r3,:128]! +3: vst1.64 {d16-d19},[r0,:128]! bx lr endfunc |